JP2003032253A - In-advance countermeasure on-line diagnosis in manageable network - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a beforehand countermeasures on-line diagnosis, in a manageable network. SOLUTION: The diagnosis method for a system, made up of a plurality of mutually linked modules, includes alarm receive from the system for indicating an obstacle in one module. In response to the alarm, a causal network is constituted, and malfunctions and disorders of one or more modules with possibility of disorder are made to relate to each other, and further, each probability for the malfunction is made to relate to the conditional probability of the disorder. On the basis of the alarm and the causal network, at least one of probability of malfunction is renewed. The diagnosis of the alarm is presented, in response to the renewed probability.

Description

Detailed Description of the Invention

CROSS REFERENCE TO RELATED APPLICATIONS This application claims the benefit of US Provisional Application No. 60/214971, which is hereby incorporated by reference.

[0002]

FIELD OF THE INVENTION The present invention relates generally to methods and systems for fault diagnosis in communication networks,
In particular, it relates to a method of identifying defective components within such networks while normal communication activity is in progress.

[Prior art]

The complexity of computer networks is
As they continue to grow, so do the reliability, availability, and services required of these networks.
These factors add to the burden on diagnostic systems used in computer networks to identify and isolate network failures. network·
To prevent failures that can seriously interfere with activity, it is important to detect the intermittent and sporadic problems that are the precursors to the initial failure and pinpoint the device causing the problem. Is. In order to maintain high availability of the network, these issues must be identified while the network is online and running in normal activity mode. The service personnel can then be instructed to replace the defective element before it has been completely destroyed.

Modern networks typically provide large amounts of diagnostic information such as topology files, system-wide error logs, and component-specific trace files. Analyzing this information to identify network failures is beyond the capabilities of any person other than the highest skilled network administrator. The most automated method for network diagnostics is if-t
We try to overcome this problem by framing expert knowledge in the form of hen rules and applying this rule automatically to diagnostic information. Rules are usually heuristics and must be specifically tailored to the system to which they apply. As a result, the rules themselves
It is difficult to devise and is not generally applicable to all possible error conditions. Such rules are not universally applicable and generally must be updated when the system configuration changes.

Model-based diagnostic techniques begin with a functional model of the system in question and analyze it to identify defective components in the event of malfunction. Functional model (forward model or causal (causa
l), also referred to as a model, is often readily available as part of a system specification or reliability analysis model. The development of such models usually involves
It is a simple part of the process of system design or system analysis. Therefore, the model creation does not require the designer to be an expert in system fault diagnosis. Instead, an automated algorithm is applied to the functional model to arrive at a diagnostic conclusion. As long as the system model is updated to reflect the configuration changes, these algorithms will automatically adapt the diagnostic to the changes made.

In switched computing networks and switched communication networks, such as System Area Networks (SAN), for diagnostic applications,
Specific challenges are presented regarding its complexity and inherent uncertainty. The complexity has to deal with the large number of components used, the existence of multiple dynamic paths between devices in a network, and the large amount of information that the network carries. Uncertainty arises, inter alia, from the fact that alarm messages are carried over the network in packet form. As a result, there may be unknown delays in alarm transmission, alarms may arrive out of sequence, and some alarm packets may be lost.

One of the known paradigms in the art of model-based diagnostics in the presence of uncertainty is the Bayesian Network. Cowe
ll) et al., “Probabilistic Networks and Expert Syst
ems "(Springer-Verlag, New York, USA, 199)
9) gives a general explanation of Bayesian network theory. The same document is incorporated herein by reference. The Bayesian network is a directed acyclic graph having nodes corresponding to the region variables and a conditional probability table added to each node. Bayesian networks are also called causal networks when the orientation of the edges of the graph corresponds to the causal relationships between the nodes. The lack of edges between pairs of nodes represents the assumption that those nodes are conditionally independent. The product of the probability tables gives the joint probability distribution of the variables. Probabilities are updated as new evidence is gathered for co-occurrence of failures and malfunctions within the system under test. When a diagnostic system receives a new alarm or set of alarms, it uses the Bayesian network to automatically determine the most probable malfunction behind the alarm.

US Pat. No. 6,076,083, the disclosure of which is incorporated herein by reference, describes an exemplary application of Bayesian networks to the diagnosis of communication networks. Communication networks are represented as Bayes networks, and devices and communication links within the communication networks are represented as nodes of the Bayes networks. Faults in the communication network are identified and recorded in the form of trouble tickets, and one or more possible causes of the faults are given based on Bayesian network calculations. When the fault is corrected, the Bayesian network is updated with the knowledge acquired in correcting the fault. The updated trouble ticket information is used to automatically update the appropriate Bayesian network probability matrix. The Bayesian network of U.S. Pat. No. 6,076,083 is static and provides no provision for changes in the configuration of communication networks. Moreover, this Bayesian network models the entire communication network, and is easily out of hand when dealing with large and complex switched networks.

Another method of applying the Bayesian network to fault diagnosis in a computer system is as follows.
Pizza (Pizza, incorporated herein by reference)
a) Others, “Optimal Discrimination between Transien
t and Permanent Faults, Proceedings of the Third
IEEE High Assurance System Engineering Symposiu
m, 1998. The author proposes to apply the principles of reliability theory to distinguish between transient and permanent failures of computer system components. Reliability theory predicts the probability of failure of a given device in terms of failure rate or failure distribution over time (such as mean time between failures (MTBF)). Standard reliability theory techniques are based on sampling the device behavior under known conditions. On the other hand, in the scheme proposed by Pizza et al., The probabilities of permanent and transient failures of system components are estimated and updated by inference using Bayesian networks. But,
This scheme has only limited practical applicability.
This is because each module in the computer system is examined separately, without error propagation from one module to another, in order to reach an accurate and optimal decision regarding the probability of failure. This is not an assumption that can reasonably be made in a real world switching network.

[0010]

[Problems to be Solved by the Invention]

In a preferred embodiment of the present invention, a diagnostic method that combines Bayesian networks and reliability theory to handle large and complex switched networks in a realistic and efficient manner. And a diagnostic system. The diagnostic system maintains a local failure model for the devices in the network, along with up-to-date topological information about the network as a whole. The local fault model includes the estimated malfunction rate of the modules in the network expressed in terms of reliability theory. When an alarm (or series of alarms) is received from the network, the diagnostic system can
The estimated malfunction rate and topology information are used to build a Bayesian network that represents possible causes of alarms and their probabilities. The malfunction rate estimate is then updated based on the observed alarm and its arrival time. When the estimated malfunction rate of a given module exceeds a certain threshold, the diagnostic system declares the module suspected of failure and issues a test or replacement recommendation for the suspect module to the user of the system.

Therefore, unlike the model-based diagnostic methods known in the art, in the preferred embodiment of the present invention, a dynamic base network model is used, which model specifically receives alarms or Created in response to each of the groups of alarms. As a result, this model fully and accurately reflects the actual latest network conditions, without incurring the extremely high computational costs and memory requirements of maintaining a complete model of the entire network. In the given model produced by this diagnostic system, interaction and error propagation between connected modules are considered, unlike the above-mentioned Pizza et al. Approach, where the device model is only considered in isolation. In embodiments of the present invention, a regular pattern of network topologies, such as cascaded switches, has been identified and utilized to determine which Bayesian model must be used to correctly model error propagation between modules. It is preferred that the size of the network be limited.

In some preferred embodiments of the invention, the diagnostic system comprises two of the modules in the network.
Assess the next failure probability, ie, consider both the estimated mean failure rate and the product moment (standard deviation) of the probability distribution. The mean and product moments of the probability distributions for a given module are updated each time the Bayesian network is constructed and evaluated for the module. The use of second-order probabilities is a property of Bayesian reliability theory (separate from Bayesian networks). Bayesian reliability theory treats failure rate assessment as a process of initial assessment and correction, unlike the simpler first-order sampling-based methods used in diagnostic systems known in the art. The secondary approach is more suitable for fault diagnostic modeling.

Although a preferred embodiment is described herein with respect to fault diagnosis in switched computer networks, those skilled in the art will appreciate that the principles of the present invention apply to other types of electrical networks as well as other types of communication networks. It will be appreciated that it is equally applicable to locating faults in other systems, including systems and mechanical systems as well as medical and financial systems.

Therefore, according to a preferred embodiment of the present invention, a method for the diagnosis of a system composed of a plurality of modules linked together, wherein the system provides an alarm indicating a failure of one of the modules. And, in response to the alarm, associating the failure with a malfunction in one or more of the modules that may have led to the failure and assigning a conditional probability of the failure to a probability of each of the malfunctions. Relate to,
Configuring a causal network, updating at least one of the probabilities of the malfunction based on the alarm and the causal network, and proposing a diagnosis of the alarm in response to the updated probability. A method is provided that includes and.

Receiving the alarm comprises collecting an event report from the plurality of modules in the system and extracting the alarm from the event report, the step of collecting the event report comprising: Preferably, receiving a report of a change in the configuration of the system, and configuring the causal network preferably comprises configuring the causal network based on the changed configuration. Configuring the causal network based on the modified configuration, maintaining a database in which the configuration is recorded, and reporting the modification of the configuration for use in configuring the causal network. And responsively updating the database. .

Alternatively or additionally, the step of extracting the alarm comprises extracting a sequence of alarms occurring at times close to each other, including the alarm indicating the fault in the one of the modules. Including updating the at least one of the probabilities comprises processing the sequence of the alarms to update the probabilities. Defining the life of each of the alarms in response to an expected delay in receipt of the alarm from the system, and responsive to the respective lifespan of extracting the sequence of the alarms. And selecting the alarm to extract from the sequence. The step of selecting the alarms to extract identifies the alarms that have occurred within each lifetime of the time of occurrence of the alarms indicating the failure in the one of the modules that the causal network has been configured to respond to. Most preferably it includes a step of selecting.

Additionally or alternatively, configuring the causal network includes defining an expected alarm caused by one of the malfunctions in the one or more modules, wherein the alarm Processing the sequence of steps of updating the probability in response to the occurrence of the expected alarm in the extracted sequence of alarms.

In a preferred embodiment, the interconnected modules include a given instance of one of the modules interconnected in a regular pattern, the step of configuring the causal network comprising: Defining a template containing a group of nodes in the network corresponding to the given one of the modules; instantiating the template for the one or more modules in response to the alarm. including. The step of defining the template is the step of defining the given one of the modules.
Identifying an expected alarm caused by the one of the malfunctions in one of the two instances of instantiating the template to the network in response to the occurrence of the expected alarm. Preferably, the step of adding an instance of said template is included.

Configuring the causal network identifying a local failure condition at the one of the failed modules; and in response to the local failure condition, within the causal network, Linking the fault to one of the malfunctions occurring in the one of the modules.
Additionally or alternatively, the step of configuring the causal network comprises determining a first failure condition that occurs in the first one of the modules due to the connection with the second one of the modules in the system. Identifying, and responsive to the first failure condition, linking the failure to a second failure condition occurring in the second one of the modules within the causal network. Linking the faults, the possible cause of the second fault condition is due to another connection between the second one of the modules and a third one of the modules in the system. Determining whether or not, and responsive to said another connection, linking said fault to a third fault condition occurring in said third one of said modules within said causal network. It is preferable.

In a preferred embodiment, the step of configuring the causal network adds to the causal network one or more occurrences of the malfunction in response to the respective probabilities of the malfunction. Linking the fault to the plurality of occurrences. Linking the failure to the plurality of occurrences includes determining one or more failure conditions caused by each of the occurrences, and linking at least a portion of the failure conditions to the failure. It is preferable.

In another preferred embodiment, updating the at least one of the probabilities of the malfunction comprises assessing an average time between failures of the one or more of the modules.

The probability of the malfunction is defined with respect to a probability distribution having a mean and a product moment, and updating the at least one of the probabilities reassessing the mean and the product moment of the probability distribution. It is preferable to include. Most preferably, the probability distribution comprises a failure rate distribution and the step of reassessing the mean and the product moments comprises updating the failure rate distribution using a Bayesian reliability theory model.

Proposing the diagnosis comprises comparing one or more of the updated probabilities with a predetermined threshold and activating a diagnostic action when the one of the probabilities exceeds the threshold. It is preferable to include. Typically, invoking the diagnostic action comprises notifying a user of the system about the diagnosis, and notifying the user comprises providing a description of the diagnosis based on the causal network. Additionally or alternatively, activating the diagnostic action includes performing a diagnostic test to verify the malfunction, the diagnostic test responsive to the one of the probabilities to exceed the threshold. To be selected. The causal network is preferably modified in response to the results of the diagnostic test.

According to a preferred embodiment of the present invention, a method for the diagnosis of a system composed of a plurality of modules linked together, wherein a fault in one of the modules can lead to the fault. A causal network associating with a malfunction in two or more of the modules that have a positive effect and relating the conditional probability of the failure to a respective probability distribution of the malfunction, and an alarm from the system indicating the failure. In response, a method is also provided that includes updating the probability distribution of the malfunction and proposing a diagnosis of the alarm in response to the updated probability distribution.

According to a preferred embodiment of the invention, a device for the diagnosis of a system composed of a plurality of modules linked together, said device comprising a diagnostic processor, said diagnostic processor comprising: One or more of the modules that are coupled to receive an alarm from the system indicating a failure of one of the modules, the diagnostic processor being responsive to the alarm and leading the failure to the failure. And constructing a causal network, which relates the conditional probability of the failure to each probability of the malfunction,
Based on the alarm and the causal network,
There is additionally provided a device arranged to update at least one of the probabilities of the malfunction and propose a diagnosis of the alarm in response to the updated probabilities.

The apparatus includes a memory containing a database in which the configuration is recorded, the diagnostic processor responsive to the report of the change in the configuration for use in configuring the causal network. Are preferably combined to update

It is further preferred that the device includes a user interface and that the diagnostic processor is coupled to notify the user of the system of the diagnosis via the user interface.

Further in accordance with a preferred embodiment of the present invention an apparatus for the diagnosis of a system consisting of a plurality of modules linked together, said apparatus comprising a diagnostic processor, said diagnostic processor comprising: , Correlating a failure in one of the modules with a malfunction in two or more of the modules that may have led to the failure,
Configuring a causal network relating the conditional probabilities of the failures to respective probability distributions of the malfunctions, updating the probability distributions of the malfunctions in response to an alarm from the system indicating the failure, and updating the An apparatus is provided that is arranged to propose a diagnosis of said alarm in response to a determined probability distribution.

Further in accordance with a preferred embodiment of the present invention a computer software product for the diagnosis of a system composed of a plurality of modules linked together, said computer software product comprising:
A computer readable medium having program instructions stored thereon, the program instructions receiving, upon reading by a computer, an alarm from the system indicating a failure of one of the modules; and responsive to the alarm. A causal network that associates the fault with a malfunction in one or more of the modules that may have led to the fault and correlates the conditional probability of the fault with each probability of the malfunction. And updating at least one of the probabilities of the malfunction based on the alarm and the causal network and proposing a diagnosis of the alarm in response to the updated probability. Software products are provided.

Furthermore, according to a preferred embodiment of the present invention, a product for diagnosing a system composed of a plurality of modules linked to each other, said product being a computer-readable medium in which program instructions are stored. And, when the program instructions are read by a computer, associating with the computer a failure in one of the modules with a malfunction in two or more of the modules that may have led to the failure. Configuring a causal network that relates the conditional probability of failure to each probability distribution of the malfunction, and updating the probability distribution of the malfunction in response to an alarm from the system indicating the failure, A product is provided that allows for proposing a diagnosis of the alarm in response to an updated probability distribution. That.

[0031]

1 is a block diagram schematically illustrating a network 22, which is a manageable communication network, and a diagnostic unit 20 used to monitor the network, according to a preferred embodiment of the present invention. is there. Network 22 typically includes a system / storage area network (SAN), as is known in the art. In such networks, node 24 may include a server or other computer processor, input / output device,
Storage devices, or gateways, may be included, which are interconnected by switch 28. An example of such a network is the RS / 6000 SP system manufactured by IBM Corporation of Armonk, NY. The network 22 is said to be "manageable" in the sense that it provides the following two key features used by the diagnostic unit 20: First, the network is monitored for errors and failures such as packet corruption or device unresponsiveness and statistics that may reflect unusual functionality. Second, the network is configurable, especially with respect to the ability of the system operator or automated controller to set device parameters such as error thresholds used to determine when to generate an alarm.

The management functions of the network 22 are preferably coordinated via the node of the nodes selected to act as the primary node 26. Node 24 includes an event collector 30, which
The collector 30 is preferably implemented as a software agent running as part of the network management software running on all nodes. These agents include alarms and configuration changes,
Collect system events that occur on each node. The event collector 30 collects these events
It is sent in the form of management packets to the primary event collector 32 running on the primary node 26. Primary event collector 32 passes a stream of events to diagnostic unit 20 for processing, as described below.

For conceptual clarity, the diagnostic unit 2
Although 0 is shown as a functional block separate from the primary node 26, in the preferred embodiment of the present invention, the diagnostic unit 20 is a software program running on the primary node.
Implemented as a component. Alternatively, the diagnostic unit software can run on a separate processor that is physically separate from the primary node, and can run as a distributed application on a group of nodes or on all nodes. . This software can be downloaded electronically to the primary node or other processor, for example, via network 22 and, alternatively, can be provided on a tangible medium such as a CD-ROM.

FIG. 2 is a schematic block diagram showing details of the diagnostic unit 20 according to a preferred embodiment of the present invention. Assuming that the diagnostic unit 20 is implemented in software as noted above, the blocks shown in FIG. 2 would typically not be separate hardware elements,
Represents a functional element or process within a diagnostic software package. The stream of events collected by the primary event collector 32 is received by the event formatter and merger 40 within the diagnostic unit 20. This block is a sequence of events,
Preferably, they are arranged in chronological order based on a time stamp applied by the event collector 30 to indicate the time of occurrence of the event. Alternatively, the schedule may be based on the time of receipt of the event at the primary node 26. The event formatter and merger 40 optionally reformats the event message information received from the event collector 30 in a unified format that can be efficiently processed by subsequent blocks within the diagnostic unit 20. The event formatter and merger 40 separates the events into configuration change events and alarms (ie error reports) and provides two merged streams for processing.

The configuration tracker 42 receives the configuration change events and processes them to update the configuration database 46 based on the system model 44. The configuration database 46 stores the currently available modules, their status,
And initialized with the complete configuration of the network 22 at network startup, including topology. This database is then automatically updated in real time to reflect any changes that have occurred, eg the addition or removal of nodes 24, the enabling or disabling of ports on the switch 28. To be done. system·
In model 44, the modules used within network 22 are described, including their interconnections and hierarchy. The term "module" is generally used herein to refer to a field replaceable unit (FRU) or portion of a FRU that can be associated with a particular error report. The differentiation between the modules in system model 44 determines the granularity with which diagnostic unit 20 diagnoses the error report and localizes its source. The hierarchical system model is preferably provided to the diagnostic unit 20 by an operator of the network 22 in Extensible Markup Language (XML) format, as is known in the art.

The diagnostic engine 48 receives the alarm stream from the event formatter and merger 40 and uses this information to determine and update the reliability assessment of the module associated with each alarm. The reliability assessment is updated by configuring the Bayesian network for each alarm on the fly and using Bayesian reliability theory to assess the malfunction rate of each different malfunction of the module. The method used by the diagnostic engine is described in detail below. In configuring the Bayesian network, the diagnostic engine uses the system model 44 and the configuration database 46 as described above.
Use the information provided by. The diagnostic engine is
It also relies on a fault model 50 that describes possible faults in the network 22. A fault in this context is an abnormal condition or behavior that can occur within a given module due to local problems or unexpected inputs.

The fault model 50 is preferably supplied by the network operator, most preferably in XML format. Disability model sample DTD
(Document Type Definition) is attached to this specification as Appendix A. It typically contains global fault information along with individual fault models for all of the base modules in the system model. These basic modules are the lowest level modules in the module hierarchy.

The global fault information of fault model 50 describes all types of malfunctions possible in network 22 and their expected rates. The term "malfunction" in this context refers to the root cause of a failure within a module.
When a fault is detected on a module, the fault may be due to a malfunction occurring in the module itself or a malfunction of another module propagated in the communication traffic through the network to the module where the fault is detected. You may. The malfunction probability of failure model 50 is typically expressed in terms of failure rate, such as estimated mean time between failures (MTBF). The estimated rate is preferably accompanied by a measure of the reliability of the estimation expressed in terms of the standard deviation (or first product moment) of the probability distribution. Malfunction rate assessment can be described by a normal distribution on a logarithmic time scale. Therefore, for example, the malfunction rate assessment (10, 1) in seconds is based on the average time between malfunction occurrences of 1
0 ¹⁰ seconds and the actual time between occurrences is the interval [10 ⁸ ,
It is shown that the probability of being 10 ¹² ] seconds is 0.95. The diagnostic engine 48 infers both the mean and standard deviation as it processes alarms it receives from the network 22.

The individual fault models of each basic module include the following information: • For each malfunction that can occur in that module, whether it is detected by the module itself and leads to the generation of an alarm by that module, and whether that malfunction causes a fault condition at the output of that module. . A fault condition is the result of a fault, ie the occurrence of a malfunction that leads to the appearance of an abnormal state or behavior of a module as noted above.
A fault condition within a module that causes a fault in the module itself is referred to herein as a "local fault condition." A fault condition at the module output that causes an abnormal input state of another module is called a "connection fault condition". • For each fault condition that may appear at the module's input, whether the condition is propagated, detected, or both by the module. • Which alarms the module reports for each detected fault condition.

The recommendation and explanation generator 52 receives the malfunction assessments calculated by the diagnostic engine 48 and assesses the different modules in the network 22.
Compare with expected baseline value held in the fault model 50. When the failure rate assessment for a given module is significantly higher than its baseline value, the advisory and explanation generator 52 typically advises the user to take further diagnostic action or replace the FRU containing that module. To do. The criteria for making such a recommendation are discussed further below. The recommendations are presented via the user interface 54. Using this user interface, the user is preferably able to enter a query to the recommendation and explanation generator and in response receive a comprehensive explanation of the root cause of the recommendation. The description is preferably generated using methods known in the art for generating a description based on a Bayesian network constructed by the diagnostic engine 48. An exemplary method for this is Druzdel.
Written by Qualitative Verbal Explanations in Bayesian
Belief Networks, Artificial Intelligence and Sim
ulation of Behavior Quarterly 94 (1996), 4
Pages 3-54 and by Madigan et al.,
"Explanation in Belief Networks", Journal of Com
putational and Graphical Statistics 6, pages 160-181 (1997). Both of these publications are incorporated herein by reference.

FIG. 3 is a flow chart that schematically illustrates a method of handling alarms and generating recommendations in the diagnostic unit 20, according to a preferred embodiment of the present invention. The method is preferably activated each time the diagnostic engine 48 receives an alarm at step 60 of receiving an alarm. Alternatively, the method can be invoked in response to some type or group of alarms. In step 62 of the sequence combination, the relevant alarms occurring at short time intervals are preferably combined for processing. The methods and considerations applicable to combining alarms in a sequence for aggregation processing are described in detail below with respect to FIG.

The diagnostic engine 48 builds a Bayesian network (or causal network) applicable to the particular alarm or alarm sequence at network building step 64. A typical Bayesian network configured in response to a single alarm is shown in Figure 4,
The method used to construct this network is described in detail below with respect to FIGS. Bayesian networks are directed acyclic graphs whose nodes are
Respond to variables that include possible module malfunctions, fault conditions, and faults that lead to problem alarms. The malfunction node has a specified probability distribution based on the expected malfunction rate or the assessed malfunction rate. The probabilities of the remaining variables are described for the parents in the graph by probability tables that represent the conditional probabilities of the corresponding variables.

After constructing the graph, the diagnostic engine 48
Updates the node's probability table based on the alarms in the sequence at update step 66. By correlating different alarms that occur within a specified time frame, the diagnostic engine can adjust the conditional probabilities of the nodes and then graph them to update the malfunction rate assessments of malfunction nodes. You can In other words, for every given observed alarm A, set its probability P (A = true) to 1. The expected alarm probability is determined according to its lifetime distribution. The Bayesian network's probability table of nodes is then recomputed to be consistent with these results. In the field of Bayesian networks, this procedure is
It is called "evidence propagation".

The updated malfunction assessment serves as the basis for the advisory and explanation generator 52 to provide the advisor to the user at advisory step 68. The user can define two threshold levels applied to each module: a low threshold at which the module is flagged as "suspected" and a high threshold at which the suspicious module is reclassified as non-suspicious. preferable. These thresholds relate to the difference between the assessed malfunction rate of each module and the expected failure rate of that module based on system specifications. The user also defines the confidence level for these two thresholds. This confidence level is checked against the standard deviation value associated with the malfunction rate assessment of the module. Thus, for example, the user may
The fact that (the reciprocal of the malfunction rate) fell below 10 ⁸ was 10
It is possible to specify that a given module should be flagged as suspected of having a failure at a confidence level of%. Step 66 following an alarm sequence
MTBF assessed for the module after
Is (9,2) using the logarithmic notation described above. In such a case, the probability that the actual MTBF has dropped below the threshold of 10 ⁸ is greater than 10% and the module is flagged accordingly. Users typically set thresholds and confidence levels, weighing the severity of the consequences of a module failure during network operation, depending on the cost of replacing or otherwise servicing the FRU in question. To do.

When a given module is flagged as suspected of failing, the advisory and explanation generator 52 can apply it to that module to verify the status of that module. Inspect to determine if there is a test procedure. If so, the generator preferably invokes the procedure automatically or, instead, prompts the user to perform the procedure. The results of this procedure are preferably fed back to the diagnostic engine 48, which incorporates the results into the applicable Bayesian network and updates its malfunction rate assessment accordingly. Following this procedure, the advisory and explanation generator 52 can determine if the FRU should be replaced. Instead,
If the MTBF assessment for all possible malfunctions associated with the module in question drops below the high threshold (perhaps after receipt and processing of additional alarms from network 22), reset the suspected failure flag for that module. To do.

FIG. 4 is a graph that schematically illustrates network 70, which is an exemplary Bayesian network produced by diagnostic engine 48, in accordance with a preferred embodiment of the present invention. In this example, the diagnostic engine 48 is shown in FIG.
Observed USD received in step 60 of the method
(Unsolicited data) The network 70 is configured in response to the alarm 71. This alarm is a USD failure 7
2 occurs, which causes one receiver port of the switch 28 to be transmitted before the data,
Means that data was received that was not preceded by the beginning (BOP) character of the correct packet. There are two scenarios described in the fault model 50 that can cause this fault: Corrupted BOP-occurs in any module between the receiver portion of the preceding switch in the network 22 that sent data to the switch that reported this error and the actual receiver port in which the error was detected. there is a possibility. Local design defect-A local issue in the reporting switch other than memory corruption.

To build the network 70, the diagnostic engine 48 starts with the node corresponding to the observed USD alarm 71 and the node corresponding to the USD fault 72 that caused the alarm. Based on the failure model 50, add a node to the network that corresponds to the failure condition 74 that may have caused the USD failure 72 at the reporting switch. As noted above, these fault conditions include both bits that were destroyed on the link or in the switch itself, and local design defects that might have caused the destruction. The fault model is then used to recursively add more fault states 76 to the network 70. The added fault condition is the fault condition on the reporting switch, or one connected to it and propagated to the reporting switch, and thus one of the fault conditions 74.
All of the fault conditions on the preceding switch that may have caused one must be included. This process
Finally stop. This is because the data flow is non-circular and the network 22 is finite. Even so, the propagation of fault conditions through the entire communication network 22 should result in an unmanageably large Bayesian network 70. In the current example, the switch 28 does not retransmit the corrupted data, so propagation stops, and thus the BOP disruption may have originated in the network 22 from a location further than the receiver port of the preceding switch. There is no. In FIG. 9 below, another technique for limiting the size of a Bayesian network created in accordance with the preferred embodiment of the present invention is shown.

For each of the fault conditions 74 and 76, the diagnostic engine 48 will identify the node corresponding to the malfunction 80 that may have caused that condition to the network 7.
Add to 0. A malfunction node has a failure rate distribution associated with it, which indicates the continuous probability of a particular malfunction. To complete the network 70,
The malfunction 80 is discretized with respect to the Boolean occurrence 78. In other words, a given malfunction 80 is represented by an interval variable with a discretized failure rate distribution. For each section,
The value of the continuous failure rate distribution function is calculated (usually at the midpoint of the section) and the value of the discretized failure rate distribution of the section is given. Appearance variables are used to calculate the probability that a corresponding malfunction will occur. In other words, the occurrence variable is that item is P (malfunction occurred at time t | a <failure rate <b), where t is the time of observation of the observed alarm for which network 70 was configured. Is a Boolean variable with a conditional probability table given by The probability table is preferably determined by the estimated rate of corresponding malfunctions according to a suitable model such as Poisson arrival statistics. To reduce the complexity of the network 70, one Boolean occurrence node 7 per malfunction 80
Preferably there are only eight. Fault conditions 74 and 76 caused by a malfunction are connected to the appearance variables associated with that malfunction.

The previous switch in the network 22 and U
BO on the link to the switch that reported the SD alarm
If a P-bit break occurs (including the break on the cable and in the auxiliary component that connects the cable to the device), the broken bit is the error detection code (ED
C) Disability 82 must also be causing. This situation causes the "USD on link" failure state node to EDC
Reflected by the edges added to network 70 that connect to the failed node. EDC disorders were observed US
In addition to D alarm 71, EDC alarm 84 on the switch
Should be issued. This EDC alarm 84
Is added to the network 70 as an “expected alarm”. The arrival or non-arrival of an EDC alarm at diagnostic unit 20 is an important factor in determining the likely cause of a USD alarm, and thus in adjusting the conditional probability of a node in network 70. It is a factor.

FIG. 5 is a timing diagram that schematically illustrates the processing of a sequence of alarms 90 that the diagnostic engine 48 receives. These alarms are used to build a Bayesian network for the current alarm and to use them to evaluate the probabilities of the nodes in the network, step 6.
2 (FIG. 3). Collecting a sequence of alarms within a suitable time window, for example,
The expected alarm 84 is used to determine if it arrived with the observed alarm 71. The selection of the time window is important to correctly handle the uncertainty of alarm arrival time and the order of alarm arrival at the diagnostic unit 20.

A normal distribution over time is associated with each type of alarm to determine which alarms in a given sequence are combined for processing. This distribution, referred to as the "lifetime distribution" of alarms, represents the probability over time of arrival at the diagnostic unit 20 of an alarm associated with an event that occurred in the network 22 at time T = 0. In other words, referring to FIG. 5, due to the lifetime distribution of alarm A ", when alarm A is received at time T ₀ , alarm A" generated by the same fault condition as A is received at time T _1. Estimated probabilities are given. Normally, the life of each alarm type is specified by the user of the diagnostic unit, but instead the life can be calculated by the diagnostic unit based on the actual performance of the network 22.

In some cases, a module of network 22 accumulates a certain number of occurrences rather than raising an alarm for every occurrence of a failure, and then batch
Issue an alarm. In this case, it is necessary to multiply the single alarm life by a threshold coefficient so that the life distribution of the alarm becomes wider. Therefore, in FIG. 5, the narrow distribution 92 of the first alarm type without the threshold factor is shown.
And an intermediate distribution 94 of the second alarm type with a low threshold coefficient and a broad distribution 96 of the third alarm type with a high threshold coefficient.

Prior to processing the Bayesian network to update the probability table and malfunction rate assessment (step 66 of the method of FIG. 3), the diagnostic engine 48 causes the diagnostic engine 48 to observe the relevant observed alarms and expected values in the sequence. It is preferable to wait until you receive all of the alarms.
The length of time to wait is determined by the alarm life.
As shown in FIG. 5, the diagnostic engine 48 preferably waits until time T _END when the probability of arrival of all expected alarms is below a predetermined threshold. In that case, the alarms A ₀ , ..., A ′, A ″ are considered to be related to alarm A, but the alarm A _N arrives after T _END.
Is not considered so.

FIG. 6 illustrates a step 64 of network construction (of the method of FIG. 3) in accordance with a preferred embodiment of the present invention.
2 is a flowchart schematically showing the details of FIG. This is a recursive method and is preferably used in the construction of Bayesian networks such as the network 70 shown in FIG.
The method begins with an observed alarm A (such as observed USD alarm 71) being received at module M at time T at initialization step 100. In step 102 of network creation, the diagnostic engine 48
Creates a new Bayesian network BN and adds the node corresponding to alarm A to BN. In the find fault step 104, the engine searches the fault model 50 for an alarm to find the fault F corresponding to A.
Add the node corresponding to F to BN along with the edge (F, A).

In step 106 of finding fault conditions, diagnostic engine 48 then searches fault model 50 for fault F to find fault condition C that may have caused F. As can be seen from the example of FIG. 4, for any given fault, there will normally be multiple such fault conditions.
For each such fault condition C, the diagnostic engine 48 performs the add fault condition step 108, which may have added the node corresponding to state C on module M to BN and connected to C. Are searched for additional fault conditions. Step 108 includes a recursive routine, which is described in detail below with respect to FIG. In this step, nodes and edges corresponding to malfunctions and occurrences of malfunctions leading to each failure state are also added. In step 110 of adding edges, for each fault condition C that may have caused F, add the corresponding edge (C, F) to BN. After processing in this form all possible fault conditions C that may have caused F, the Bayesian network is complete.

FIG. 7 is a flow diagram that schematically illustrates details of the routine performed at step 108 of adding a fault condition, according to a preferred embodiment of the present invention. The routine begins by adding the node corresponding to state C of module M to BN at step 120 of adding nodes. At step 122 of the locality check, the diagnostic engine 48
Examines the fault model 50 to determine if state C is a local fault condition or a connection fault condition. In the case of a local fault condition, only the malfunction N of the module M that caused the condition C need be checked. In step 124 of malfunction detection, the diagnosis engine 48 determines that the failure model 50
Search for possible malfunctions of. For each malfunction N, at malfunction check step 126, the engine checks whether the node corresponding to N already exists in the BN. Otherwise, add node step 128.
Then, the node N is added to BN. Then, in step 129 of adding an edge, the edge (N, C) is added to BN. When all possible malfunctions have been added to the BN, step 108
Is completed.

At step 122, the handling becomes more complex when a connection failure condition is identified. In this case, in the module discovery step 130, the diagnostic engine 48 searches the system model 44 and the configuration database 46 for the module M, and the fault condition C on the connection M ′ to M.
Find one or more modules M ′ connected to M in a manner that might have caused In the add fault condition step 132, for each such module M ′, the diagnostic engine 48 causes the condition C to
Find the fault condition on M ′ and on the connection leading to M ′ that may have caused This step includes the routine described in detail below with respect to FIG. The routine of step 132 is step 1
It forms part of the recursion of the 08 routine. This routine may have led to the appearance of fault condition C, M '.
Continue until nodes and edges corresponding to all of the fault conditions above and on that connection (including other modules connected to M ', etc.) are added to the BN.

After exploring all possible connection fault conditions leading to fault condition C, diagnostic engine 48 queries fault model 50 in step 134 of Expected Fault Locator to find these fault conditions. It is determined whether there is a possibility of being connected to another failure F ′ other than the failure F found in step 104. EDC failure 82 (FIG. 4) is an example of such a failure. Step 1 for adding a failed node
At 36, add each node of such expected failure F'to the BN. In addition, the node corresponding to the expected alarm A'generated by F'is added to BN along with the edges (C, F ') and (F', A '). Additional edges may be added to the network that correspond to local failure conditions on other modules and connection failure conditions associated with modules that may have led to failure F '. The occurrence or non-occurrence of alarm A'expected within a time relative to the first alarm A (given by the specified lifetime) is written in the state probability table for the Bayesian network at step 66 (Fig. 3). Used for inclusion.

FIG. 8 is a flow chart that schematically illustrates details of the routine performed in the add fault condition step 132, in accordance with a preferred embodiment of the present invention. As noted above, this routine is executed for each of the modules M'connected to M in a way that a connection failure condition C may appear on the connection between M'and M. This routine may also be executed recursively for the module M ″ connected to M ′. At step 140 of the local fault check, the diagnostic engine 48 first causes the fault model 5
Check 0 to see if there is a local fault condition C'on M'which may have caused state C on the output of M'connected to M. If there is such a state C ′, then according to the routine of step 108, with the necessary modifications, the diagnostic engine 48 adds the node corresponding to C ′ on module M ′ to the Bayesian network BN.
This routine also leads to the addition of appropriate edges to the BN, as well as nodes corresponding to local malfunctions that may have caused C '. In step 142 of adding an edge, the edge (C ′, C) is also added to BN.

Whether a local fault condition C'is found or not, it may have caused C'caused C'between M'and another module M "connected to M '. There may also be a connection failure condition C ". This situation is
Equivalent to M'propagating to C. In propagation step 144, diagnostic engine 48 determines whether M ′ propagates to C by referencing fault model 50. If M ′ propagates to C, then at input checking step 146, the diagnostic engine 48 queries the fault model to determine the input of M ′ at which the connection fault condition C ″ may have appeared. For each such input,
Add the connection failure condition C "on M'to the BN. Again, make the necessary changes and follow the routine of step 108. In step 148 of adding edges, for each failure condition C", the edge (C Add ", C) to BN.
At this point, step 132 is complete and the Bayesian network configuration is step 134 until all recursion is complete.
Will continue in.

Since the communication network 22 must be finite, the Bayesian network illustrated by FIGS.
The method of configuring the network will eventually stop. However, the failure propagation can make the Bayesian network very large, up to the point of representing the entire communications network. Such a situation is completely out of hand and must be avoided.

Therefore, the preferred embodiment of the present invention limits the growth of the Bayesian network in step 64 by taking advantage of the regularity inherent in switching networks such as SAN. Such networks generally have a small number of different module types, which are usually arranged in regular arrangements. These structures are preferably represented by templates in the Bayesian network. All instances of a given template give the same expected alarm under a given fault condition. There may be many instances of a given template physically present in the structure of the communication network, but only when a particular instance of the template actually sees one of its expected alarms. And is instantiated, ie added to the Bayesian network.

FIG. 9 is a graph showing the regular structure of the communication network 168 and the corresponding Bayesian network 175 constructed by the diagnostic engine 48 in accordance with the preferred embodiment of the present invention. The communication network 168 in this example includes switches 1 connected in cascade.
70, 172, and 174 are included and switch 170
Is on the first layer of the cascade and switch 172 is on the second
Layer, switch 174, is on the third layer. The Bayesian network 175 configuration begins with the node corresponding to the alarm 176 observed on one of the ports of the switch 170. 6-8, the node corresponding to the fault 178 responsible for causing the alarm 176 and the node in fault 180 at the receiver port of the switch 170 that caused the fault are added to the Bayesian network 175. It This condition may have been caused by a fault condition 182 in the central queue of the switch. These are alarms 176 on the switch 170.
Is a local failure that may have caused

Alarm 176 may also be caused by a failure propagating from one of switches 172 to switch 170. Such fault propagation is a switch 1
Fault condition 184 at the transmitter port 72, fault condition 186 at the cable connecting the switches, fault condition 188 at the receiver port of switch 172, or switch 17
It can be caused by one of a series of fault conditions, including the fault condition 190 at two central queues. As with switch 170, a fault condition 188 or 190 at switch 172 causes a fault 192 at the receiver port of switch 172 leading to an expected alarm 194.

Fault conditions 184, 186, 188, and 190 indicate fault 192 and expected alarm 194.
Together, it configures a Bayesian network template corresponding to one of the switches in the communication network 168 (nodes corresponding to malfunctions and malfunctions that may lead to fault conditions are omitted here for simplicity. ). One of the switches 172 has an alarm 176
If the expected alarm 194 is issued within the appropriate time limit of the alarm, alarm 176 and expected alarm 1
There is a convincing basis to assume that 94 are related to each other.
In this case, the template corresponding to the alarm issuing switch is instantiated, i.e. it is added to the Bayesian network 175. If the expected alarm does not occur, the corresponding switch does not affect the updated malfunction assessment calculation (step 66) and omits the template from the Bayesian network without compromising the calculation. You can In this way, the Bayesian network constructed in response to a given alarm remains computationally small and manageable. When instantiating one template of the switch 172, the diagnostic engine 48 causes the third layer switch 17
It is preferable to consider the expected alarms corresponding to the layer 3 switch 174 to determine if 4 should be included in the Bayesian network 175.
However, in practice you generally only need to instantiate a few templates.

Although a preferred embodiment has been described with respect to fault diagnosis of a network 22 using a diagnostic unit 20 (using an example taken from the inventor's experience with an RS / 6000 SP system), those skilled in the art will understand that the It will be appreciated that the principles are equally applicable to locating faults in other networks and systems. Most modern communication networks, especially packet data networks, are easy to handle with fault reporting and configuration capabilities that can be used by diagnostic systems such as diagnostic unit 20. As long as all the elements of the network or system are modeled and the data flow between these elements is acyclic, a Bayesian network and Bayesian reliability theory based diagnostic model is applied in accordance with the principles of the present invention. can do. This principle is applicable not only to communication networks and computer networks (and subsystems of such networks), but also to other types of electrical and mechanical systems as well as medical and financial systems.

[0067] Appendix A-Disability Model DTD <? xml encoding = "UTF-8"?> <! ELEMENT Fault_Model_Information (g_files-list,                                    g_malfunctions-list,                                    g_alarms-list,                                  g_fault-conditions-list,                             g_fault-condition-groupings)> <! ELEMENT g_files-list (g_file) +> <! ELEMENT g_file EMPTY> <! ATTLIST g_file    name CDATA #REQUIRED    desc CDATA #IMPLIED> <! ELEMENT g_malfunctions-list (g_malfunction *)> <! ELEMENT g_malfunction (g_fault-condition-caused *)> <! ATTLIST g_malfunction    name CDATA #REQUIRED    mean CDATA #REQUIRED    deviation CDATA #REQUIRED> <! ELEMENT g_fault-condition-caused EMPTY> <! ATTLIST g_fault-condition-caused    name CDATA #REQUIRED> <! ELEMENT g_alarms-list (g_alarm *)> <! ELEMENT g_alarm EMPTY> <! ATTLIST g_alarm    name CDATA #REQUIRED    number CDATA #REQUIRED    related-fault CDATA #REQUIRED    desc CDATA #IMPLIED> <! ELEMENT g_fault-conditions-list (g_fault-condition *)> <! ELEMENT g_fault-condition (g_fault-caused *)> <! ATTLIST g_fault-condition    name CDATA #REQUIRED    connection-type (Service-Traffic | Data-Traffic |    Service-And-Data-Traffic | Sync-Traffic) #REQUIRED> <! ELEMENT g_fault-caused EMPTY> <! ATTLIST g_fault-caused    name CDATA #REQUIRED    on-input (true | false) #REQUIRED    locally (true | false) #REQUIRED> <! ELEMENT g_fault-condition-groupings    (g_fault-condition-group *)> <! ELEMENT g_fault-condition-group    (g_member-fault-condition *, g_fault-condition-group *)> <! ATTLIST g_fault-condition-group    name CDATA #REQUIRED> <! ELEMENT g_member-fault-condition EMPTY> <! ATTLIST g_member-fault-condition    name CDATA #REQUIRED> <! ELEMENT module (malfunctions, reports, triggers,    (barrier-to | propagates), converts?)> <! ATTLIST module    name CDATA #REQUIRED> <! ELEMENT malfunctions (malfunction *)> <! ELEMENT malfunction EMPTY> <! ATTLIST malfunction    name CDATA #REQUIRED> <! ELEMENT reports (data-inputs, sync-inputs,    service-inputs, local)> <! ELEMENT data-inputs (fault *)> <! ELEMENT sync-inputs (fault *)> <! ELEMENT service-inputs (fault *)> <! ELEMENT local (fault *)> <! ELEMENT fault EMPTY> <! ATTLIST fault    name CDATA #REQUIRED> <! ELEMENT converts (fault-condition-pair *)> <! ELEMENT fault-condition-pair EMPTY> <! ATTLIST fault-condition-pair    fault-cond-input CDATA #REQUIRED    fault-cond-output CDATA #REQUIRED> <! ELEMENT triggers (fault-condition *)> <! ELEMENT fault-condition EMPTY> <! ATTLIST fault-condition    name CDATA #REQUIRED> <! ELEMENT barrier-to (fault-condition *)> <! ELEMENT propagates (fault-condition *)>

In summary, the following matters will be disclosed regarding the configuration of the present invention.

(1) Multiple modules linked to each other
A method for diagnosing a system consisting of:
From the system, indicates a failure of one of the modules
Receiving an alarm and responding to the alarm
And one that may have led to the disorder
Or associated with a malfunction in multiple of the above modules,
The conditional probability of failure to each probability of the malfunction
Correlating, configuring a causal network,
Based on the alarm and the causal network,
A step of updating at least one of the probabilities of the malfunction.
Of the alarm in response to the updated probability.
Proposing a diagnosis. (2) The step of receiving the alarm includes the system
Event reports from the multiple modules in
From the event report,
Lame extraction step, as described in (1) above.
the method of. (3) Before the step of collecting the event report
To receive a report of system configuration changes
And configuring the causal network,
The causal network based on the modified configuration
The method according to (2) above, including the step of configuring. (4) The causal network based on the changed configuration
The step of configuring the
Maintaining the database and the causal network
The configuration of the modification of the configuration for use in the configuration of
Updating the database in response to a port
The method according to (3) above, which comprises: (5) The step of extracting the alarm is performed by the module.
Including said alarm indicating said failure in said one of the
The sequence of alarms that occur at times close to each other
Extracting at least one of the probabilities
To update the probability of updating one more
Processing the sequence of the alarms
The method according to (2) above. (6) Step for extracting the sequence of the alarm
Is responsible for receiving the alarm from the system.
In response to the awaited delay, the life of each of the alarms
The steps of defining life and responding to each of the aforementioned lifetimes
Select the alarm to extract from the sequence
The method according to (5) above, including the steps of: (7) The step of selecting the alarm to be extracted is
The above-mentioned mode in which the causal network was configured in response
Of the alarm indicating the failure at the one of the modules
The alert that occurred within each lifetime of the time of occurrence
The method according to (6) above, including the step of selecting the system.
Law. (8) The step of configuring the causal network is
One of the malfunctions in one or more of the modules
Defines the expected alarm triggered by
And processing the sequence of alarms.
The step of managing is the extracted sequence of alarms.
In response to the occurrence of the expected alarm in
One according to (5) above, including the step of updating the rate
Law. (9) The step of configuring the causal network is
The category of the module in the system and the module
Due to one of the malfunctions in the module in the category
Corresponding to the expected alarm caused by
Template containing a group of nodes in the network
Defining step and said extracted sequence of alarms
In response to the expected occurrence of the alarm in the
Instance the template in the causal network
The method according to (5) above, including the step of: (10) The plurality of mutually linked modules are
Where the modules are linked together in a regular pattern
Said causal net including one or more instances of
The step of structuring the work is performed at the location of the module.
A group of nodes in the network corresponding to one of the
Group containing a template,
In response to the one or more modules
And instantiating the template
The method according to (1) above, which comprises: (11) The step of defining the template is
One of the given instance of the given one of the modules
Expected to be caused by one of the above malfunctions in
Identifying the alarms that are
Instantiating the
To the network in response to the occurrence of an alarm
Above, including the step of adding a rate instance
The method according to (10). (12) The step of configuring the causal network includes
The row in the one of the modules that has failed
Cull failure condition identifying the local failure
In response to a condition, within the causal network, the module
One of the malfunctions that occurs in the one of the
Linking harm, as described in (1) above.
Method. (13) The step of configuring the causal network includes
Connection to a second one of the modules in the system
The first occurring in the first one of the modules due to
Identifying a fault condition and responding to the first fault condition.
In response, within the causal network,
Reset the fault to the second fault condition that occurs in the second one.
The method according to (1) above. (14) The step of linking the obstacle is the second obstacle.
A possible cause of the harmful condition is the second one of the modules.
One and a third one of the modules in the system
A step to determine if it is due to another connection between
And the causal network in response to the other connection.
Occurs in the third one of the modules in the network
Linking the fault to a third fault condition
The method according to (13) above, which comprises: (15) The step of configuring the causal network includes
In response to the respective probabilities of the malfunction, the malfunction
Add one or more occurrences of a work to the causal network
And the plurality of occurrences within the causal network.
Linking the disorder to life,
The method described in (1). (16) A step that links the failure to the plurality of occurrences.
One that is triggered by each of the above occurrences
Or determining a plurality of fault conditions,
Linking at least some of the conditions to the fault
The method according to (15) above, which comprises: (17) The at least one of the probabilities of the malfunction is
Updating comprises the step of updating the one or more of the modules.
Including the step of assessing the average time between disability failures,
The method according to (1) above. (18) The probability of the malfunction has an average and a product moment.
Defined with respect to the probability distribution
Updating one of the
And the step of reassessing the product moment
The method described in (1). (19) The probability distribution includes a failure rate distribution, and the average
And the step of reassessing the product moment is Bayesian reliability
A step for updating the failure rate distribution using a theoretical model.
The method according to (18) above, which comprises: (20) The step of proposing the diagnosis is updated as described above.
The step of comparing one or more of the
And a diagnosis when the one of the probabilities exceeds the threshold.
(1) above, including the step of activating an action
The method described. (21) The step of activating the diagnostic action is
Steps to notify the user of the system about the diagnosis.
The method according to (20) above, which comprises: (22) The step of notifying the user is the causal link.
Providing a description of the diagnosis based on network
The method according to (21) above, which comprises: (23) The step of activating the diagnostic action is
Steps to run diagnostic tests to verify malfunctions.
Check that the diagnostic test exceeds the threshold.
Selected in response to said one of the rates, noted in (20) above.
How to list. (24) In response to the result of the diagnostic test, the causal network
As described in (23) above, including the step of changing the network.
How to list. (25) Consists of multiple modules linked together
A method for diagnosing a
May have led to a disability in one of the
Associated with malfunctions in two or more of the modules
The conditional probability of failure is divided by the probability of each malfunction.
Steps of constructing a causal network associated with cloth
And respond to an alarm from the system indicating the fault
And updating the probability distribution of the malfunction
Of the alarm in response to the updated probability distribution.
Proposing a diagnosis. (26) The step of updating the probability distribution is the two
Assess the average time between failures of the above modules
The method according to (25) above, including the step. (27) The probability distribution has a mean and a product moment, and
Updating the probability distribution comprises:
And the step of reassessing the product moment
The method according to (25). (28) The probability distribution includes a failure rate distribution, and the average
And the step of reassessing the product moment is Bayesian reliability
A step for updating the failure rate distribution using a theoretical model.
The method according to (27) above, which comprises: (29) If the two or more modules are
Including the one of the generated modules,
The step of configuring the network before the module.
The step of identifying a local fault condition in one
In response to a local fault condition, said causal network
In which the malfunction that occurs in the one of the modules
Linking the fault to one of the
The method according to (25). (30) The two or more modules are the first module.
And a second module, the causal network
And configuring the second module in the system.
Occurs in the first module due to the connection with the module
Identifying a first fault condition to perform, said first fault
In response to a condition, in the causal network, the second
Link the fault to the second fault condition that occurs in the module
The method according to (25) above. (31) The two or more modules are a third module.
Linking the faults,
2 possible causes of the fault condition are the second module and the
Another connection in the system to the third module
The step of determining whether it is due to
In response to one connection, within the causal network,
The above-mentioned fault is added to the third fault condition that occurs in the third module.
The method according to (30) above, including the step of linking.
Law. (32) Consists of multiple modules linked together
For diagnosing a stored system, said apparatus
Includes a diagnostic processor, the diagnostic processor
The system indicates that one of the modules has failed.
And a diagnostic processor coupled to receive a ramp.
Responds to the alarm by switching the fault to the fault.
One or more of the modules that may have been traced
The conditional probability of the failure is
A causal network that is associated with each probability of malfunction
Configuring the alarm, the alarm and the causal network
Based on at least one of the probabilities of the malfunction
Updating the alarm in response to the updated probability
A device arranged to suggest a diagnosis of. (33) The diagnostic processor is the
Receive event reports from multiple modules,
Extract the alarm from the event report
The device according to (32) above, which is linked as follows. (34) The event report is the composition of the system.
Including a report of changes in the
The causal network is constructed based on the changed configuration.
The apparatus according to (33) above, wherein the apparatus is arranged to: (35) A memo including a database in which the configuration is recorded
For use in configuring the causal network, including
In the diagnostic processor, the change of the configuration
Update the database in response to reports
The device according to (34) above, which is coupled. (36) The diagnostic processor is the module of the module.
Close to each other, including the alarm indicating one of the failures
Extract the sequence of alarms that occur at the time of contact,
The sequence of the alarms to update the probability
The process according to (33) above, which is coupled to process
apparatus. (37) The life of each of the
In response to the expected delay in receiving the
Defined in terms of
Extract from the sequence in response to each lifetime
Arranged to select the alarm, above (3
The device according to 6). (38) The diagnostic processor is the causal network.
The one of the modules configured in response to
At the time of occurrence of the alarm indicating the fault in
To select the alarm that occurred within its life
The device according to (37) above, which is arranged. (39) When configuring the causal network, the diagnosis
A processor, the one or more of the modules
Expected to be caused by one of the above malfunctions in
Are arranged to define alarms that
And further the extracted sequence of alarms.
The probability in response to the occurrence of the expected alarm in
The device according to (36) above, which is arranged to update
Place (40) The category of the module in the system
And the malfunction of the module in the category
Responds to expected alarms triggered by one
Including a group of nodes in the network to
The plate is defined and the diagnostic processor activates the alarm
Of the expected aller in the extracted sequence of
In the causal network in response to the occurrence of
Arranged to instantiate the plate, above
The device according to (36). (41) The plurality of mutually linked modules are
Where the modules are linked together in a regular pattern
A module including one or more instances of
No in the network corresponding to the given one of
A template containing a group of
A processor that responds to the alarm with one or more
Install the template for the module of
The device according to (32) above, which is arranged so that
Place (42) The template is the part of the module
One of the malfunctions in one of the one instance of a given
Including the expected alarm triggered by
The diagnostic processor generates the expected alarm
In response to importing the template into the network.
Install the template by adding a stance.
(41) above, which is arranged so that it becomes an instance.
On-board equipment. (43) Before the fault occurs in the diagnostic processor
Identify a local fault condition on the one of the modules
And in response to the local fault condition, the causal net
Within the workpiece, the one generated by the one of the modules
Placed to link the fault to one of the malfunctions
The device according to (32) above. (44) The diagnostic processor is the
Due to the connection with the second one of the modules said module
The first failure condition that occurs in the first one of the
In the causal network in response to the first fault condition
And a second failure that occurs in the second one of the modules
Arranged to link the obstacle to a harmed state, above
The device according to (32). (45) The diagnostic processor may be in the second failure state.
Effective cause is that the second one of the modules and the system
Another between the third one of said modules in the stem
Determine whether it is due to one connection, and
In response to a connection, within the causal network, the module
To a third fault condition that occurs in the third one of the
Placed to link obstacles, as described in (44) above.
On-board equipment. (46) The diagnostic processor is configured to detect the malfunction.
One or more occurrences of said malfunction in response to each probability
To the causal network and add the causal network
Within the network to link the failures to the multiple occurrences.
The device according to (32) above, which is placed. (47) The diagnostic processor is provided for each of the occurrences.
Determine one or more fault conditions caused by
Link at least part of the fault condition to the fault
The device according to (46) above, which is arranged to: (48) At least one of the probabilities of the malfunction
During a failure of the one or more of the modules
The device according to (32) above, expressed as an average time. (49) The probability of the malfunction has an average and a product moment.
The diagnostic processor defined with respect to a probability distribution
Updates the mean and the product moment of the probability distribution
The device according to (32) above, which is arranged as follows. (50) The probability distribution includes a failure rate distribution, and the diagnosis
The processor uses the Bayesian reliability theory model
Arranged to update the failure rate distribution, above (49)
The device according to. (51) The diagnostic processor determines whether the updated probability
Comparing one or more with a predetermined threshold,
Trigger a diagnostic action when one exceeds the threshold
The device according to (32) above, which is arranged as follows. (52) A diagnostic interface including a user interface.
The processor can be forwarded via the user interface
Notify users of the system about diagnostics
The device according to (51) above, which is coupled. (53) The diagnostic processor uses the user interface.
Based on the causal network through the face
Arranged to provide a diagnostic description, above (52).
The device according to. (54) The diagnostic action verifies the malfunction.
Including a diagnostic test performed for
Is selected in response to the one of the probabilities above the threshold.
The device according to (51) above, which is selected. (55) The diagnostic processor determines the result of the diagnostic test.
Arranged to change the causal network in response to
The device according to (54) above. (56) Consists of multiple modules linked together
For diagnosing a stored system, said apparatus
Includes a diagnostic processor, the diagnostic processor
A failure in one of the modules may have led to the failure.
Associated with malfunctions in two or more of the modules that have the potential
And the conditional probability of the failure for each of the malfunctions.
Construct a causal network related to the probability distribution of
In response to an alarm from the system indicating the failure,
The probability distribution of the malfunction is updated, and the updated
To suggest a diagnosis of said alarm in response to a probability distribution
Is located at the device. (57) The probability distribution is the two or more modules
(56) above, showing the mean time between disability
apparatus. (58) The probability distribution has a mean and a product moment, and
The diagnostic processor responds to the alarm with the probability
Arranged to reassess the mean and the product moment of the distribution
The device according to (56) above. (59) The probability distribution includes a failure rate distribution, and the diagnosis
The processor uses the Bayesian reliability theory model
Arranged to update the failure rate distribution, above (58)
The device according to. (60) The two or more modules are
Including the one of the generated modules, the diagnostic
The processor has a local fault in the one of the modules.
Identifies the harm condition and responds to the local fault condition by
In the causal network, the one of the modules
To link the fault to one of the malfunctions that occurs
The device according to (56) above, which is disposed at. (61) The two or more modules are the first module.
And a second module, the diagnostic processor
Causes the connection to the second module in the system.
Therefore, the first failure state occurring in the first module is identified.
Separately, in response to the first fault condition, the causal network is
Second fault condition occurring in the second module in the network
Is arranged to link the obstacle to
The device according to 6). (62) The two or more modules are combined into a third module.
The diagnostic processor includes the second fault condition.
Possible causes are the second module and the third module.
Due to another connection in the system with the
And then responding to the other connection,
Occurring in the third module within the causal network
Arranged to link the fault to a third fault condition
The device according to (61) above. (63) Consists of multiple modules linked together
Computer software for diagnosing distributed systems
A product, the computer software product
Is a computer-readable medium on which the program instructions are stored.
Including the program instructions by a computer
The system, when read, on the computer
Received an alarm indicating a failure of one of the modules
And taking the fault in response to the alarm.
One or more of the models that may have led to a disability.
Associated with the malfunction of the module
A causal link that relates the rate to each probability of the malfunction.
Network, the alarm and the
Based on a causal network, the probability of the malfunction
Update at least one and respond to said updated probability
And suggest a diagnosis of the alarm.
Computer software product. (64) The program command causes the computer to
Events from the modules in the system
Receiving a report and the event report
The above-mentioned (6)
The computer software product described in 3). (65) The event report indicates the structure of the system.
Including a report of changes in the
Based on the changed configuration, the computer
The causal network is configured, as described in (6) above.
The computer software product described in 4). (66) The program command causes the computer to
In response to the report of the change in the configuration, the cause
The above configuration is recorded for use in configuring the result network.
The above, which causes the database to be recorded to be updated
The computer software product according to (65). (67) The program command causes the computer to
The alert indicating the failure in the one of the modules
Alarms that occur at times close to each other, including
To extract the sequence and to update the probability
Processing said sequence of said alarms
The computer software according to (64) above.
Products. (68) The life of each of the
In response to the expected delay in receiving the
Defined in terms of
The computer in response to each of the aforementioned lifetimes.
Select the alarm to extract from the
The computer software according to (67) above.
Products. (69) The program command causes the computer to
Said causal network configured in response to said
The alarm indicating the fault in the one of the modules
The ara occurred within each life of the occurrence time of
A method according to (68) above, wherein selection of
Computer software product. (70) The program command causes the computer to
When configuring the causal network, one of the
Triggered by one of the malfunctions in multiple modules
Defining the expected alarms that will be triggered
The expected in the extracted sequence of Rams
Updating the probability in response to the occurrence of an alarm.
The computer software according to (67), which is executed.
Clothing products. (71) The category of the module in the system
And the malfunction of the module in the category
Responds to expected alarms triggered by one
Including a group of nodes in the network to
A plate is defined and the program instructions are
Computer, within the extracted sequence of alarms
In response to the expected occurrence of the alarm, the causal network
Instantiating the template within the network
And the computer software according to (67) above.
Software products. (72) The plurality of interconnected modules are
Where the modules are linked together in a regular pattern
A module including one or more instances of
No in the network corresponding to the given one of
A template containing a group of
A gram command causes the computer to respond to the alarm.
Answering one or more of the modules
Template to instantiate the template, above
The computer software product according to (63). (73) The template is the location of the module
One of the malfunctions in one of the one instance of a given
Including the expected alarm triggered by
The program instructions cause the computer to
To the network in response to the occurrence of an alarm
By adding an instance of the template
Let's do instantiating the template
The computer software according to (72) above.
Product. (74) The program command causes the computer to
The row in the one of the modules that has failed
Cull fault condition and said local fault condition
In response to the
One of the malfunctions that occurs in the one of the
The link described in (63) above that causes linking.
Computer software products. (75) The program command causes the computer to
Connection to a second one of the modules in the system
The first occurring in the first one of the modules due to
Identifying a fault condition and responding to the first fault condition.
And within the causal network,
Link the fault to a second fault condition that occurs in the second one
The computer according to (63) above, which causes
Data software products. (76) The program command causes the computer to
A possible cause of the second fault condition is that before the module.
The second one and the third of the modules in the system
Determine whether it is due to another connection between one of
And in response to the other connection,
In the result network, the third one of the modules
Linking the fault to a third fault condition that occurs in
The computer software according to (75) above.
Software products. (77) The program command causes the computer to
In response to the respective probabilities of the malfunction, the malfunction
Add one or more occurrences of a work to the causal network
And the multiple occurrences within the causal network.
Linking the obstacles, (63) above
The computer software product described in. (78) The program command causes the computer to
One or each caused by each of the above occurrences
Determining multiple fault conditions and
Linking at least part of the disorder
Made by the computer software described in (77) above.
Goods. (79) The at least one of the probabilities of the malfunction.
During a failure of the one or more of the modules
The computer according to (63) above, expressed as an average time.
Computer software products. (80) The probability of the malfunction has an average and a product moment.
The program instructions defined with respect to a probability distribution
Of the probability distribution to the computer.
And the above-mentioned product ratio are updated.
The listed computer software product. (81) The probability distribution includes a failure rate distribution,
Gram command tells the computer that Bayesian reliability theory
Done updating the failure rate distribution using a model
The computer software according to (80) above.
A product. (82) The program command causes the computer to
Ratio one or more of the updated probabilities to a predetermined threshold
And the one of the probabilities exceeds the threshold
Sometimes triggering diagnostic actions and doing the above
The computer software product according to (63). (83) The program command causes the computer to
Notifying users of the system about the diagnosis
The computer software according to (82) above.
Software products. (84) The program command causes the computer to
User explaining the diagnosis based on the causal network
The computer according to (83) above, which is configured to
Computer software products. (85) The diagnostic action verifies the malfunction.
Including a diagnostic test performed for
Is selected in response to the one of the probabilities above the threshold.
Computer software according to (82) above, which is selected
Clothing products. (86) The program command causes the computer to
The causal network in response to the results of the diagnostic test
The computer according to (85) above, wherein
Computer software products. (87) Consists of multiple modules linked together
For diagnosing a defective system, said product
Is a computer-readable medium on which the program instructions are stored.
Including the program instructions by a computer
When read, the computer
May have led to a disability in one of the
Associated with a malfunction in two or more of the modules,
The conditional probability of failure is defined as the probability distribution of each of the malfunctions.
Configuring a causal network relating to
In response to an alarm from the system indicating a failure,
Update the probability distribution of malfunctions to
Proposing a diagnosis of said alarm in response to a rate distribution
Let the product do the work. (88) The probability distribution is the two or more modules.
(87) above, showing the mean time between disorders
Product. (89) The probability distribution has a mean and a product moment, and
Program instructions cause the computer to generate the alarm
In response to the mean of the probability distribution and the product moment
The product according to (87) above, which allows reassessment to be performed.
Goods. (90) The probability distribution includes a failure rate distribution,
Gram command tells the computer that Bayesian reliability theory
Done updating the failure rate distribution using a model
The product according to (89) above. (91) The two or more modules are
Including the one of the generated modules,
RAM instructions to the computer before the module
Identifying a local fault condition in one of the
Within the causal network in response to
Of the malfunction that occurs in the one of the modules
Linking the faults to one, said
The product according to (87). (92) The two or more modules are combined into a first module.
And a second module, the program instructions
To the second computer in the system.
Occurs in the first module due to the connection with the module
Identifying a first fault condition that
In response to the second causal network in the causal network.
Link the fault to a second fault condition that occurs in the tool
The product according to (87), which is used to do the above. (93) The two or more modules are combined into a third module.
And the program instructions include the computer
In addition, the possible causes of the second fault condition are
In the system between the module and the third module.
Determining whether it is due to one connection
In response to another connection, within the causal network
Then, in the third failure state that occurs in the third module,
As described in (87) above, which causes linking of obstacles.
Listed products.

[Brief description of drawings]

FIG. 1 is a block diagram that schematically illustrates a manageable computer network having a model-based diagnostic unit according to a preferred embodiment of the present invention.

2 is a block diagram schematically illustrating details of the diagnostic unit of FIG. 1 according to a preferred embodiment of the present invention.

FIG. 3 is a schematic flow chart of a method for network diagnosis according to a preferred embodiment of the present invention.

FIG. 4 is a graph showing an exemplary Bayesian network configured in response to an alarm in a communication network in accordance with a preferred embodiment of the present invention.

FIG. 5 is a timing diagram that schematically illustrates a method of processing a sequence of alarms according to a preferred embodiment of the present invention.

FIG. 6 is a flow diagram that schematically illustrates a method of configuring a Bayesian network in response to an alarm, according to a preferred embodiment of the present invention.

7 is a flow diagram that schematically illustrates a method of adding a fault condition to a Bayesian network configured according to the method of FIG.

8 is a flow diagram that schematically illustrates a method of adding a fault condition to a Bayesian network configured according to the method of FIG.

FIG. 9 is a graph showing a method of constructing a Bayesian network that utilizes the regularity of a modeled communication network according to a preferred embodiment of the present invention.

[Explanation of symbols]

20 diagnostic unit 40 event formatter and merger 42 configuration tracker 44 system model 46 configuration database 48 diagnostic engine 50 fault model 52 recommendation and explanation generator 54 user interface 60 receiving an alarm step 62 combining with other alarms in sequence 64 Building Bayesian Networks for Alarm Sequences Using Existing Malfunction Rate Assessments Step 66 Updating Malfunction Rate Assessments in the Network Step 68 Making Recommendations Based on Malfunction Rates

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁷ Identification code FI theme code (reference) H04L 29/14 H04L 13/00 313 (72) Inventor Igor Silashya Israel Haifa Simkin Streat 34 A ( 72) Inventor Lee Shalev Israel Zyklon Yakov Derek Sarah Street 819/5 (72) Inventor Kirill Shoiket Israel Haifa Azraj Yakov Street 11 F Term (reference) 5B048 AA18 CC15 DD11 FF02 5B089 GA21 GB02 HA10 JB17 KA12 5K030 GA12 JA10 MA01 MC06 MC08 5K035 BB04 DD01 FF01 HH07 JJ01

Claims (93)

[Claims] 1. A method for diagnosing a system composed of a plurality of interconnected modules, the method comprising: receiving from the system an alarm indicating a failure of one of the modules; and responsive to the alarm. Configuring a causal network that associates the failure with a malfunction in one or more of the modules that may have led to the failure and associates a conditional probability of the failure with a respective probability of the malfunction. And based on the alarm and the causal network,
Updating at least one of said probabilities of said malfunction, and proposing a diagnosis of said alarm in response to said updated probabilities. 2. The method of claim 1, wherein receiving the alarm comprises collecting an event report from the plurality of modules in the system and extracting the alarm from the event report. Method. 3. The step of collecting the event report includes the step of receiving a report of a configuration change of the system, and the step of configuring the causal network configures the causal network based on the changed configuration. The method of claim 2 including the step of: 4. The step of configuring the causal network based on the modified configuration maintains a database in which the configuration is recorded and the configuration of the causal network for use in configuring the causal network. Updating the database in response to the report of changes. 5. The step of extracting said alarms comprises the step of extracting a sequence of alarms occurring at times close to each other, including said alarms indicating said failure in said one of said modules, said probability The method of claim 2, wherein updating at least one of the steps comprises processing the sequence of alarms to update the probability. 6. Extracting the sequence of the alarms, responsive to an expected delay in receipt of the alarms from the system, defining the respective lifetimes of the alarms; Selecting the alarm to extract from the sequence in response to the lifetime of the alarm. 7. The step of selecting the alarm to extract comprises within each lifetime of the time of occurrence of the alarm indicating the failure in the one of the modules to which the causal network is configured in response. 7. The method of claim 6 including the step of selecting the alarm that has occurred. 8. The step of configuring the causal network comprises defining an expected alarm caused by one of the malfunctions in the one or more modules, processing the sequence of alarms. The method of claim 5, wherein the step of updating comprises updating the probability in response to the occurrence of the expected alarm in the extracted sequence of alarms. 9. The step of configuring the causal network in the network corresponding to a category of the modules in the system and an expected alarm caused by one of the malfunctions in the modules in the category. Defining a template containing a group of nodes; instantiating the template in the causal network in response to the occurrence of the expected alarm in the extracted sequence of alarms. The method according to 5. 10. The interconnected modules include multiple instances of a given one of the modules interconnected in a regular pattern, the step of configuring the causal network comprising: Defining a template containing a group of nodes in the network corresponding to the given one;
Instantiating the template for the one or more modules in response to the alarm. 11. Defining the template comprises identifying an expected alarm caused by one of the malfunctions in one of the given ones of the instances of the module, the template 11. The method of claim 10, wherein instantiating a template includes adding an instance of the template to the network in response to occurrence of the expected alarm. 12. The step of configuring the causal network comprises identifying a local fault condition at the one of the failed modules, and responsive to the local fault condition within the causal network. Linking the fault to one of the malfunctions occurring in the one of the modules. 13. The step of configuring the causal network identifies a first fault condition that occurs in a first one of the modules due to a connection with a second one of the modules in the system. 2. The method of claim 1, comprising: responsive to the first failure condition, linking the failure to a second failure condition occurring in the second one of the modules within the causal network. The method described. 14. The step of linking the faults wherein the possible causes of the second fault condition are the second one of the modules and the third one of the modules in the system.
Determining if it is due to another connection to and from the third connection, and in response to the other connection, a third failure occurring in the third one of the modules in the causal network. Linking the fault to a condition. 15. The step of configuring the causal network adds, in response to the respective probabilities of the malfunction, one or more occurrences of the malfunction to the causal network, and within the causal network. Linking the fault to a plurality of occurrences. 16. Linking the fault to the plurality of occurrences, determining one or more fault conditions caused by each of the occurrences, and linking at least some of the fault conditions to the failure. 16. The method of claim 15 including the step of: 17. The method of claim 1, wherein updating the at least one of the probabilities of the malfunction comprises assessing an average time between failures of the one or more modules. .. 18. The probability of the malfunction is defined with respect to a probability distribution having a mean and a product moment, and the step of updating the at least one of the probabilities reassess the mean and the product moment of the probability distribution. Including the steps to
The method of claim 1. 19. The probability distribution comprises a failure rate distribution, and the step of reassessing the mean and the product moments comprises updating the failure rate distribution using a Bayesian reliability theory model. Item 18. The method according to Item 18. 20. The step of proposing the diagnosis comprises comparing one or more of the updated probabilities with a predetermined threshold, and activating a diagnostic action when the one of the probabilities exceeds the threshold. The method of claim 1, comprising: 21. The method of claim 20, wherein invoking the diagnostic action comprises notifying a user of the system about the diagnostic. 22. The method of claim 21, wherein informing the user comprises providing a description of the diagnosis based on the causal network. 23. Activating the diagnostic action includes performing a diagnostic test to verify the malfunction, the diagnostic test selected in response to the one of the probabilities to exceed the threshold. 21.
The method described in. 24. Modifying the causal network in response to the results of the diagnostic test.
The method described in. 25. A method for diagnosing a system composed of a plurality of interconnected modules, wherein the failure of one of the modules is more than one that may have led to the failure. Configuring a causal network that associates a conditional probability of the fault with a respective probability distribution of the fault in association with the fault in the module; and in response to an alarm from the system indicating the fault, Updating the probability distribution, and proposing a diagnosis of the alarm in response to the updated probability distribution. 26. The method of claim 25, wherein updating the probability distribution comprises assessing an average time between failures of the two or more modules. 27. The method of claim 25, wherein the probability distribution has a mean and a product moment, and updating the probability distribution comprises reassessing the mean and the product moment of the probability distribution. the method of. 28. The probability distribution comprises a failure rate distribution, and the step of reassessing the mean and the product moments comprises updating the failure rate distribution using a Bayesian reliability theory model. Item 27. The method according to Item 27. 29. The two or more of the modules include the one of the failed modules, and the step of configuring the causal network identifies a local failure condition in the one of the modules. 26. The method of claim 25, further comprising: responsive to the local failure condition, linking the failure to one of the malfunctions occurring in the one of the modules in the causal network. Method. 30. The two or more of the modules are first
Configuring a causal network, including a module and a second module, identifying a first failure condition occurring in the first module due to a connection with the second module in the system; 26. Responsive to a first failure condition, linking the failure to a second failure condition occurring in the second module within the causal network. 31. The two or more of the modules are a third
Linking the faults, including a module,
A possible cause of the second fault condition is another cause in the system between the second module and the third module.
Determining if it is due to one connection, and linking the failure to a third failure condition occurring in the third module in the causal network in response to the another connection. 31. The method of claim 30. 32. An apparatus for diagnosing a system composed of a plurality of interconnected modules, the apparatus including a diagnostic processor, the diagnostic processor from the system to one of the modules. Coupled to receive an alarm indicative of a failure, the diagnostic processor is responsive to the alarm to associate the failure with a malfunction in one or more of the modules that may have led to the failure, the failure Constructing a causal network relating the conditional probabilities of the respective to the respective probabilities of the malfunction, updating at least one of the probabilities of the malfunction based on the alarm and the causal network, and the updated probability. An apparatus arranged to propose a diagnosis of said alarm in response to. 33. The apparatus of claim 32, wherein the diagnostic processor is linked to receive event reports from the plurality of modules in the system and extract the alarms from the event reports. 34. The event report includes a report of a configuration change of the system and the diagnostic processor is arranged to configure the causal network based on the changed configuration. The device according to. 35. A memory including a database in which the configuration is recorded, the diagnostic processor updating the database in response to the report of the change in the configuration for use in configuring the causal network. 35. The device of claim 34, wherein the device is coupled to: 36. The diagnostic processor extracts a sequence of alarms occurring at times in close proximity to each other, including the alarm indicating the fault in the one of the modules, and updating the probabilities. 34. The apparatus of claim 33, coupled to process the sequence of alarms. 37. Each lifetime is responsive to an expected delay in receipt of the alarm from the system,
37. The apparatus of claim 36, wherein the apparatus is defined for the alarms and the diagnostic processor is arranged to select the alarms to extract from the sequence in response to the respective lifetimes. 38. The diagnostic processor detects an alarm generated within a respective lifetime of the time of occurrence of the alarm indicating the fault in the one of the modules to which the causal network is configured in response. 38. The device of claim 37, wherein the device is arranged to select. 39. When configuring the causal network,
The diagnostic processor is arranged to define an expected alarm caused by the one of the malfunctions in the one or more of the modules, the diagnostic processor further being within the extracted sequence of alarms. 37. The apparatus of claim 36, arranged to update the probability in response to the occurrence of the expected alarm of. 40. A template is defined that includes a category of the modules in the system and a group of nodes in the network corresponding to an expected alarm caused by one of the malfunctions in the modules in the category. 37. The apparatus of claim 36, wherein the diagnostic processor is arranged to instantiate the template in the causal network in response to the occurrence of the expected alarm in the extracted sequence of alarms. . 41. The interlinked modules include instances of a given one of the modules interconnected in a regular pattern, corresponding to the given one of the modules. 33. A template is defined that includes a group of nodes in the network, and the diagnostic processor is arranged to instantiate the template for one or more of the modules in response to the alarm. Equipment. 42. The template includes an expected alarm caused by one of the malfunctions in one of the instances of the given one of the modules, and the diagnostic processor includes the expected alarm. 42. Arranged to instantiate the template by adding an instance of the template to the network in response to the occurrence of
The device according to. 43. The diagnostic processor identifies a local fault condition in the one of the failed modules and responds to the local fault condition in the causal network with the one of the modules. 33. The apparatus of claim 32, arranged to link the fault to one of the malfunctions that occurs in one. 44. The diagnostic processor identifies a first fault condition occurring in a first one of the modules due to a connection with a second one of the modules in the system; Arranged in response to a failure condition to link the failure to a second failure condition occurring in the second one of the modules within the causal network,
The device according to claim 32. 45. The diagnostic processor causes the possible cause of the second fault condition to be another connection between the second one of the modules and a third one of the modules in the system. Arranged to link the fault to a third fault condition occurring in the third one of the modules in the causal network in response to the another connection. Claim 44
The device according to. 46. The diagnostic processor is responsive to the respective probabilities of the malfunction to add one or more occurrences of the malfunction to the causal network and to impair the plurality of occurrences within the causal network. 33. The device of claim 32, arranged to link the. 47. The diagnostic processor is arranged to determine one or more fault conditions caused by each of the occurrences and link at least a portion of the fault conditions to the fault. The device according to. 48. The apparatus of claim 32, wherein the at least one of the probabilities of the malfunction is expressed as an average time between failures of the one or more modules. 49. The probability of the malfunction is defined with respect to a probability distribution having a mean and a product moment, and the diagnostic processor is arranged to update the mean and the product moment of the probability distribution. 32. The device according to 32. 50. The apparatus of claim 49, wherein the probability distribution comprises a failure rate distribution and the diagnostic processor is arranged to update the failure rate distribution using a Bayesian reliability theory model. . 51. The diagnostic processor is arranged to compare one or more of the updated probabilities with a predetermined threshold and trigger a diagnostic action when the one of the probabilities exceeds the threshold. The device according to claim 32. 52. The apparatus of claim 51, including a user interface, wherein the diagnostic processor is coupled to notify a user of the system about the diagnostic via the user interface. 53. The apparatus of claim 52, wherein the diagnostic processor is arranged to provide, via the user interface, an explanation of the diagnostics based on the causal network. 54. The diagnostic action comprises a diagnostic test performed to verify the malfunction, the diagnostic test being selected in response to the one of the probabilities to exceed the threshold. 51. The apparatus according to 51. 55. The apparatus of claim 54, wherein the diagnostic processor is arranged to modify the causal network in response to a result of the diagnostic test. 56. A device for diagnosing a system composed of a plurality of modules linked together, said device including a diagnostic processor, wherein the diagnostic processor detects a fault in one of the modules. A system for indicating a fault by configuring a causal network that associates a malfunction in two or more of the modules that may have led to the fault and associates a conditional probability of the fault with a respective probability distribution of the malfunction. An apparatus arranged to update the probability distribution of the malfunction in response to an alarm from and to propose a diagnosis of the alarm in response to the updated probability distribution. 57. The apparatus of claim 56, wherein the probability distribution indicates a mean time between failures of the two or more modules. 58. The probability distribution has a mean and a product moment, and the diagnostic processor is responsive to the alarm.
57. The apparatus of claim 56, arranged to reassess the mean and the product moment of the probability distribution. 59. The apparatus of claim 58, wherein the probability distribution comprises a failure rate distribution and the diagnostic processor is arranged to update the failure rate distribution using a Bayesian reliability theory model. . 60. The two or more modules include the one of the failed modules, the diagnostic processor identifying a local failure condition in the one of the modules, 57. The apparatus of claim 56, arranged to link the fault to one of the malfunctions occurring in the one of the modules in the causal network in response to a fault condition. 61. The two or more of the modules are first
A module and a second module, the diagnostic processor identifying a first fault condition occurring in the first module due to a connection with the second module in the system and responding to the first fault condition. 57. The apparatus of claim 56, wherein the apparatus is arranged to link the fault to a second fault condition occurring in the second module within the causal network. 62. The two or more of the modules comprise a third
A module, wherein the diagnostic processor is configured such that the possible causes of the second fault condition are the second module and the third
Determining whether it is due to another connection in the system with the module, and in response to the other connection, in the causal network to a third fault condition occurring in the third module. 62. The device of claim 61, arranged to link the obstacles. 63. A computer software product for diagnosing a system composed of a plurality of interconnected modules, the computer software product comprising a computer readable medium having program instructions stored thereon. When the program instructions are read by a computer, the computer may receive an alarm from the system indicating a fault in one of the modules and, in response to the alarm, may have linked the fault to the fault. Associating with a malfunction in one or more of the modules and relating the conditional probability of the failure to a respective probability of the malfunction,
Configuring a causal network, updating at least one of the probabilities of the malfunction based on the alarm and the causal network, and proposing a diagnosis of the alarm in response to the updated probability. Computer software products that let you do 64. The program instructions cause the computer to receive an event report from the plurality of modules in the system and extract the alarm from the event report. The computer software product described in. 65. The event report includes a report of a configuration change of the system, the program instructions causing the computer to configure the causal network based on the changed configuration. The computer software product of claim 64. 66. The program instructions cause the computer to respond to the report of the change in the configuration,
66. The computer software product of claim 65, which causes updating of a database in which the configuration is recorded for use in configuring the causal network. 67. The program instructions for causing the computer to extract a sequence of alarms occurring at close times of each other, including the alarm indicating the fault in the one of the modules, and the probability. 65. The computer software product of claim 64, wherein processing the sequence of alarms to update the. 68. Each lifetime is responsive to an expected delay in receipt of the alarm from the system,
68. The computer software product of claim 67, defined for the alarm, the program instructions causing the computer to select the alarm to extract from the sequence in response to the respective lifetime. 69. The program instructions occur to the computer within a respective lifetime of the occurrence of the alarm indicating the fault in the one of the modules to which the causal network is configured. 69. The computer software product of claim 68 causing selection of said alarm that has occurred. 70. The program instructions define to the computer an expected alarm caused by one of the malfunctions in the one or more modules in configuring the causal network. 68. The computer software product of claim 67, further comprising: updating the probability in response to the occurrence of the expected alarm in the extracted sequence of alarms. 71. A template is defined that includes a category of the modules in the system and a group of nodes in the network that corresponds to an expected alarm caused by one of the malfunctions in the modules in the category. 70. The method of claim 67, wherein the program instructions cause the computer to instantiate the template in the causal network in response to the occurrence of the expected alarm in the extracted sequence of alarms. The listed computer software product. 72. The interlinked modules include multiple instances of a given one of the modules interconnected in a regular pattern, corresponding to the given one of the modules. A template is defined that includes a group of nodes in the network, the program instructions causing the computer to instantiate the template for one or more of the modules in response to the alarm. 63. A computer software product according to 63. 73. The template includes an expected alarm caused by one of the malfunctions in one of the instances of the given one of the modules, the program instructions causing the computer to: 73. The computer software product of claim 72, which causes instantiation of the template by adding an instance of the template to the network in response to an expected alarm occurrence. 74. In the causal network, the program instructions identify to the computer a local fault condition on the one of the failed modules, and in response to the local fault condition. 64. The computer software product of claim 63, causing the fault to be linked to one of the malfunctions occurring in the one of the modules. 75. The program instructions identify to the computer a first fault condition occurring in a first one of the modules due to a connection with a second one of the modules in the system. 64. and, in response to the first failure condition, linking the failure to a second failure condition occurring in the second one of the modules within the causal network. The computer software product described in. 76. The program instructions provide the computer with another possible cause of the second fault condition between the second one of the modules and a third one of the modules in the system. Determining whether it is due to one connection, and in response to said another connection,
Within the causal network, the third of the modules
76. The computer of claim 75, further comprising: linking the fault to a third fault condition that occurs in one of
Software product. 77. The program instructions instructing the computer to add one or more occurrences of the malfunction to the causal network in response to the respective probabilities of the malfunction; and within the causal network. 64. The computer software product of claim 63 causing linking of the fault to multiple occurrences. 78. The program instructions cause the computer to determine one or more fault conditions caused by each of the occurrences and link at least a portion of the fault conditions to the fault. 78. The computer software product of claim 77, which causes it to run. 79. The computer software product of claim 63, wherein the at least one of the probabilities of the malfunction is expressed as an average time between failures of the one or more modules. 80. The probability of the malfunction is defined with respect to a probability distribution having a mean and a product moment, and the program instructions cause the computer to update the mean and the product moment of the probability distribution. Claim 63
The computer software product described in. 81. The method of claim 80, wherein the probability distribution comprises a failure rate distribution and the program instructions cause the computer to update the failure rate distribution using a Bayesian reliability theory model. The listed computer software product. 82. The program instructions instructing the computer to compare one or more of the updated probabilities to a predetermined threshold and to initiate a diagnostic action when the one of the probabilities exceeds the threshold. 64. The computer software product of claim 63, which causes: 83. The computer of claim 82, wherein the program instructions cause the computer to notify a user of the system about the diagnosis.
Software product. 84. The computer software product of claim 83, wherein the program instructions cause the computer to provide a user with a description of the diagnostics based on the causal network. 85. The diagnostic action comprises a diagnostic test performed to verify the malfunction, the diagnostic test being selected in response to the one of the probabilities to exceed the threshold. A computer software product according to item 82. 86. The computer software product of claim 85, wherein the program instructions cause the computer to change the causal network in response to a result of the diagnostic test. 87. A product for diagnosing a system comprising a plurality of interconnected modules, the product comprising a computer readable medium having program instructions stored thereon, the program instructions being implemented by a computer. When read, associates with the computer a failure in one of the modules with a malfunction in two or more of the modules that may have led to the failure;
Configuring a causal network that relates the conditional probability of failure to each probability distribution of the malfunction,
Updating the probability distribution of the malfunction in response to an alarm from the system indicating the fault and proposing a diagnosis of the alarm in response to the updated probability distribution. 88. The article of claim 87, wherein the probability distribution indicates a mean time between failures of the two or more modules. 89. The probability distribution has a mean and a product moment, and the program instructions cause the computer to reassess the mean and the product moment of the probability distribution in response to the alarm. 88. The product of claim 87 that is performed. 90. The method of claim 89, wherein the probability distribution comprises a failure rate distribution and the program instructions cause the computer to update the failure rate distribution using a Bayesian reliability theory model. Product listed. 91. The two or more modules include the one of the failed modules, the program instructions causing the computer to identify a local failure condition in the one of the modules. What to do
89. The article of claim 87, wherein in response to the local failure condition, linking the failure to one of the malfunctions occurring in the one of the modules within the causal network. 92. The two or more of the modules include a first
A module and a second module, wherein the program instructions identify to the computer a first fault condition that occurs in the first module due to a connection with the second module in the system; 89. The product of claim 87, wherein in response to a first failure condition, linking the failure to a second failure condition occurring in the second module within the causal network. 93. The two or more of the modules comprise a third
A module, the program instructions causing the computer to cause the second failure condition to occur in the second cause.
Determining whether it is due to another connection in the system between the module and the third module, and in response to the other connection, in the causal network, in the third module. 89. Linking the fault to a third fault condition that occurs.
Product described in.