JP5718457B2 - Service life prediction system and service life prediction method - Google Patents

Description

  The present invention relates to a maintenance part life prediction system and a maintenance part life prediction method for predicting the life of a part (hereinafter, referred to as a maintenance part) that is included in an electromechanical device and wears or deteriorates with the operating time.

  For example, an electromechanical apparatus such as an X-ray CT (Computed Tomography) apparatus, which is one of medical diagnostic apparatuses, has components such as a rotating anode X-ray tube and a capacitor that are consumed or deteriorated with operation time. . The lifetime of these parts is much shorter than that expected for the electromechanical device itself. Therefore, these parts are periodically replaced as maintenance parts according to the number of days in history or the cumulative operating time. In that case, the criteria of the replacement cycle are failure rate function λ (t), failure probability density function f (t), cumulative failure probability distribution function F (t), cumulative hazard function obtained from life test and failure field data. It is determined based on the expected lifetime using H (t) or the like.

  In Patent Document 1, a sensor that measures the usage and usage environment of a plant apparatus is provided, and for consumable parts included in the apparatus, the lifetime of the apparatus is measured based on the usage and usage environment of the apparatus measured by the sensor. An example of a plant maintenance support system that creates a prediction model and presents the replacement time of the consumable parts according to the life predicted by the prediction model is disclosed.

  Patent Document 2 discloses an example of an X-ray CT apparatus, but there is no particular description about predicting the life of consumable parts. However, even in an electromechanical apparatus such as an X-ray CT apparatus, by installing a sensor for monitoring the usage and usage environment of the apparatus, the usage and usage environment unique to the apparatus is measured, and the actual usage and It is possible to perform a life prediction suitable for the use environment, and in this case, the accuracy of the replacement cycle of the maintenance parts is improved.

Further, as a technique related to the present invention, Non-Patent Document 1 discloses a method for determining and optimizing model parameters using a Markov Chain Monte Carlo method (hereinafter abbreviated as MCMC method). Is described.
Non-Patent Document 2 describes a method of creating a mixed Weibull distribution model by a nonlinear least square method.

JP 2007-25878 A U.S. Pat. No. 7,636,416

Christophe Andrieu and 3 others, “An Introduction to MCMC for Machine Learning”, Kluwer Academic Publishers, Machine Learning, 50, p.5-p.43, 2003 Dan Ling and two others, “A method for parameter estimation of Mixed Weibull distribution”, IEEE Reliability and Maintainability Symposium 2009, p.129-p.133, Jan. 2009

  As described in Patent Document 1, if a model for predicting the life of consumable parts (maintenance parts) included in an electromechanical device is created according to the usage and usage environment of the maintenance parts, it is certain. In addition, it is possible to predict the service life of the maintenance part with high accuracy in accordance with the use method and the use environment of the maintenance part.

  However, if a service life prediction model is created for each usage and usage environment, the service life field of the maintenance parts is large enough to enable significant statistical processing for each usage and usage environment. Data or life test data is required.

  However, in general, it takes time and cost to acquire a sufficient amount of field data or life test data that allows significant statistical processing for each usage (hereinafter referred to as usage) and usage environment. Therefore, particularly for maintenance parts of electromechanical devices as new products, a sufficient amount of field data of life performance data and life test data is often not collected. Therefore, it can be said that it is actually difficult to create a prediction model of such service parts for each usage method and environment.

  SUMMARY OF THE INVENTION Accordingly, the present invention provides a maintenance part life prediction system and a maintenance part life prediction method capable of accurately predicting a maintenance part life in consideration of a usage method and a use environment with a small amount of performance data.

  The present invention is a maintenance part life prediction system that predicts the life of a maintenance part of an electromechanical device whose life varies depending on the method of use and use environment, and the elapsed time or operation time from the start of use of the maintenance part to failure. A reliability data storage unit in which reliability data configured by associating first data representing the first data and second data representing the usage method and usage environment of the maintenance component is stored, and the maintenance component The value of the meta parameter included in the parameter prediction model for predicting the value of the parameter included in the life prediction model for predicting the life of the material from the second data, the first data, the second data, and the A model creation simulator unit determined by simulation using a random number according to a probability distribution set in advance for the meta parameter, and the model A life prediction model for determining a parameter value included in the life prediction model from a parameter prediction model having a meta parameter value determined by the creation simulator unit and a value of the second data for the maintenance component to be life predicted A life prediction calculation trial unit that predicts a predicted life of the maintenance target of the life prediction target based on a life prediction model having a parameter value determined by the life prediction model creation unit. Features.

  A maintenance part life prediction system according to the present invention uses a parameter prediction model that predicts parameters included in a maintenance part life prediction model based on data (second data) representing a usage method and a use environment of the maintenance part. Predict. Then, the meta parameters included in the parameter prediction model represent the data stored in the reliability data storage unit (the first data, that is, the field data of the lifetime), the usage method and the usage environment of the maintenance parts. It is obtained by random number simulation so as to match the data (the second data).

  That is, the service life prediction model of the present invention incorporates not only the first data (life field data) but also the second data (use method and use environment data) as variables. It has become. Therefore, in the present invention, there is no need to stratify the field data stored in the reliability data storage unit according to usage method and usage environment data, and the parameters of the life prediction model are stored in the reliability data storage. It is obtained by random number simulation that matches the field data stored in the section. Therefore, in the present invention, even when there is no large amount of field data, the service life of the maintenance component can be accurately predicted.

  According to the present invention, it is possible to accurately predict the service life of a service component in consideration of the usage method and the usage environment with a small amount of actual data.

The figure which showed the example of the functional block structure of the maintenance component lifetime prediction system which concerns on embodiment of this invention. The figure which showed the example of the structure of the computer system which implement | achieves a maintenance component lifetime prediction system. The figure which showed the example of schematic structure of the medical X-ray CT apparatus used as a representative example of an electromechanical device in embodiment of this invention. The figure which showed the example of the record structure of the reliability data with respect to the medical X-ray CT apparatus accumulate | stored in a reliability data storage part. The figure which showed the example of the failure rate chart of maintenance parts, such as an X-ray tube. The figure which showed the example of the processing flow of the model creation simulation process in a model creation simulator part. FIG. 5A is a diagram showing an example of how a parameter sample sequence converges by the MCMC method. FIG. 5A shows how a parameter sample sequence converges in a section from the first time until convergence and a predetermined number of convergence sections after convergence. In the example shown, (b) shows an example of a converged parameter sample sequence in a predetermined number of convergence intervals after convergence. The figure which showed the example of the processing flow of the lifetime calculation process in the lifetime calculation part of a maintenance component lifetime prediction system. The figure which showed the example of the maintenance component lifetime prediction system screen displayed on the display apparatus of the model management computer. The figure which showed the example of the maintenance management screen displayed on the display apparatus of a maintenance management apparatus (operation computer).

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In all the drawings for explaining the embodiments, the same constituent elements are given the same reference numerals, and the repeated explanation thereof is omitted.

(1. Configuration of maintenance parts life prediction system)
FIG. 1 is a diagram showing an example of a functional block configuration of a maintenance part life prediction system according to an embodiment of the present invention. As shown in FIG. 1, the maintenance component life prediction system 100 includes blocks such as a model creation simulator unit 110, a life calculation unit 130, and a reliability data storage unit 140.

  Here, the model creation simulator unit 110 includes a model creation trial unit 111, a probabilistic parameter sample generation unit 115, a model accuracy convergence determination unit 116, a parameter sample value storage unit 121, a life prediction model storage unit 122, and the like. The model creation trial unit 111 is configured to include a life prediction model creation trial unit 112, a life prediction calculation trial unit 113, and the like. The life calculation unit 130 includes blocks such as a life prediction model creation unit 132, a life prediction calculation unit 133, and a maintenance part replacement cycle notification unit 134 therein.

  The maintenance part life prediction system 100 uses an electromechanical device 200 (for example, a medical X-ray CT device) from its use method (how to use), use environment, maintenance history, failure information (failure record data, life data). Are obtained, and based on the obtained reliability data, the service life of the maintenance parts included in the electromechanical device 200 is predicted, and the replacement time and replacement based on the predicted service life of the maintenance parts are also calculated. It has a function of outputting the cycle to the maintenance management device 135 or the like. Note that details of the functions of each block will be sequentially described in the following description of the embodiments.

  FIG. 2 is a diagram showing an example of the configuration of a computer system that implements the maintenance component life prediction system 100. As shown in FIG. 2, a model management computer 31 provided in an operation center 300 that operates and manages a maintenance part life prediction model according to an embodiment of the present invention includes a plurality of remote locations via a communication network 500. Are connected to the operation computers 41 provided in the respective sites 400. Here, the operation computer 41 is a computer that operates and controls the electromechanical device 200 such as an X-ray CT apparatus, for example, and the site 400 is installed in the electromechanical device 200 such as a hospital. It means campus.

  The model management computer 31 includes an arithmetic processing device 32 and a storage device 33. The storage device 33 includes a parameter sample value storage unit 121, a life prediction model storage unit 122, and a reliability data storage unit 140. Etc. are provided. The operation computer 41 includes an arithmetic processing device 42 and a storage device 43. In the present embodiment, the function of the model creation simulator unit 110 and the function of the life calculation unit 130 shown in FIG. 1 are realized by the model management computer 31, and the function of the maintenance management device 135 is the operation computer. 41 is realized. However, as another embodiment, the function of the model creation simulator unit 110 is realized by the model management computer 31, and the function of the life calculation unit 130 and the function of the maintenance management device 135 are realized by the operation computer 41. Also good.

(2. Configuration of X-ray CT system)
FIG. 3 is a diagram showing an example of a schematic configuration of an X-ray CT apparatus used as a representative example of the electromechanical apparatus 200 in the embodiment of the present invention. As shown in FIG. 3, the X-ray CT apparatus 200a includes a gantry 10, an X-ray tube 11, a high voltage generator 12, a collimator 13, a detector 15, a data collection unit 16, an image reconstruction calculation device 17, An operation computer 41 is included.

  In the X-ray CT apparatus 200 a, the X-ray bundle 14 generated in the X-ray tube 11 is trimmed to a required beam width by the collimator 13 and exposed to a subject (not shown) arranged at the center of the gantry 10. Be shot. The detector 15 detects the X-ray bundle 14 after passing through the subject, and the data collection unit 16 collects the detection values of the detector 15 and transfers the collected detection values to the image reconstruction calculation device 17. The image reconstruction calculation device 17 calculates the degree of X-ray absorption by the subject based on the detection value of the detector 15, and obtains an exposure image from one direction of the subject. At this time, the X-ray tube 11, the collimator 13, and the detector 15 are fixed to the rotating base 10 and rotate around the subject together with the base 10. Therefore, the image reconstruction calculation device 17 can obtain an exposure image from a plurality of directions of the subject, and reconstructs a cross-sectional image of the subject based on the exposure images from the plurality of directions.

  Since the X-ray tube 11 has a structure that generates X-rays by colliding accelerated thermoelectrons with a metal target such as tungsten, the metal target is consumed. Therefore, it can be said that the X-ray tube 11 is a typical maintenance part in the X-ray CT apparatus 200a. Also, the capacitor parts inside the high voltage generator 12 and the electronic circuit board inside the image reconstruction arithmetic unit 17 can be said to be maintenance parts.

  Therefore, the maintenance part life prediction system 100 predicts the life of a maintenance part that wears and deteriorates, such as the X-ray tube 11. In that case, the lifetime depends not only on the usage time and cumulative number of images of the X-ray CT apparatus 200a, but also on the usage method (for example, operation interval, warm-up operation time) and usage environment (ambient temperature, humidity, etc.). To do. Therefore, the maintenance part life prediction system 100 uses, for example, data such as the use time, the total number of images to be taken, the use method, and the use environment together with the failure data of the X-ray tube 11 as reliability data for clues to the life prediction. Obtained from the line CT apparatus 200 a and stored in the reliability data storage unit 140.

  FIG. 4 is a diagram illustrating an example of a record configuration of reliability data for the X-ray CT apparatus 200a accumulated in the reliability data storage unit 140. As shown in FIG. 4, the reliability data record for the X-ray CT apparatus 200a includes the apparatus number, operation start date / time, X-ray tube failure date / time, X-ray tube failure mode, X-ray tube replacement date / time, cumulative number of images to be captured, warm It includes data such as machine operation time, operation interval, installation environment temperature. Here, in order to simplify the description, only the X-ray tube 11 is a maintenance part. The reliability data record shown in FIG. 4 shows an example of a record when the X-ray tube fails. However, when preventive replacement is performed before the X-ray tube fails, a failure event monitoring abort record Thus, the X-ray tube failure date and time and the X-ray tube failure mode columns are blank. Note that abnormal conditions such as the occurrence of abnormal noise before the breakdown stops are included in the breakdown.

  Accordingly, the reliability data of one record accumulated in the reliability data storage unit 140 is the actual life data in the field of the X-ray tube 11 (that is, from the start of operation of the X-ray CT apparatus 200a or the X-ray tube 11). The time from when the X-ray tube 11 is replaced until the X-ray tube 11 breaks down or until the preventive replacement is determined as to how to use the X-ray CT apparatus 200a (total number of images, warm-up operation time, operation interval, etc. ) And data indicating the usage environment (installation environment temperature, etc.).

  Here, the life of the X-ray tube 11 is determined by the elapsed time from when the X-ray tube 11 starts to be used until it breaks down, but even if it is determined by the actually used operating time. Good. Hereinafter, the lifetime of a maintenance part may be an elapsed time or an operation time as long as it is determined in accordance with the maintenance part.

(3. General method for predicting life)
First, here, when the maintenance part life prediction system 100 manages maintenance of the X-ray CT apparatus 200a in accordance with the conventional general life prediction concept, the life of the maintenance part is predicted and a replacement cycle is set. A method will be described. That is, according to the conventional general idea, the service life prediction of the maintenance parts and the setting of the replacement cycle are performed, for example, according to the following procedure.

<Procedure 1> Based on the reliability data stored in the reliability data storage unit 140, the failure cumulative distribution F (t) according to the Weibull distribution is estimated.
For example, the maintenance part life prediction system 100 obtains the time t until failure by subtracting the operation start date and time from the X-ray tube failure date and time for the reliability data of the number N of records, and the order when the time t is arranged in ascending order. Find i. Next, according to the number N of records, the failure cumulative distribution F (t) is calculated from the experience distribution defined by the equation (1), the median rank defined by the equation (2), or the average rank defined by the equation (3). Get an estimate.

<Procedure 2> A set of data (time t until failure, estimated value of failure cumulative distribution F (t)) is plotted on Weibull probability paper, and a straight line is fitted.
<Procedure 3> The shape parameter m, the scale parameter η, and the position parameter γ of the failure cumulative distribution F (t) according to the Weibull distribution represented by Expression (4) are read from the Weibull probability paper.

<Procedure 4> Next, using the shape parameter m, the scale parameter η, and the position parameter γ obtained in Procedure 3, 100 s (0 <s <1) percent failure life (T hat) represented by Expression (5) Ask for.

  <Procedure 5> Next, based on the reliability required for the maintenance part (X-ray tube 11), a threshold value s is set as appropriate, and the life T is calculated using Equation (5). The replacement cycle is determined so that the replacement of the maintenance part is performed before the time runs out.

  In the above description, <Procedure 2> and <Procedure 3> are procedures realized mainly by human actions for convenience of explanation. In recent years, these procedures are applied to a computer such as a personal computer. Since the program to be executed has been developed, <Procedure 2> and <Procedure 3> can be performed by causing the model management computer 31 and the operation computer 41 to execute the program.

(4. Problems in general prediction methods)
However, the above-described method for predicting the service life of a maintenance part and setting a replacement cycle has a serious problem in practice. In other words, in order to predict the service life of maintenance parts, it is necessary to plot a set of data (time t until failure, estimated value of failure cumulative distribution F (t)) on Weibull probability paper and fit a straight line. However, in the actual electromechanical device 200, the usage method and usage environment are variously different, and therefore, the actual electromechanical device 200 is not necessarily a straight line on the Weibull probability paper. That is, the service life of maintenance parts often varies greatly depending on the method of use and environment of use.

  For example, the X-ray CT apparatus 200a is energized throughout the year in large general hospitals and used many times a day, but is used only once a day at most in small hospitals. Only energized when. If this is the case, it cannot be generally said which of the two methods of use is severe, but it is not surprising that there is a difference in the average life of the X-ray tube 11 included in the X-ray CT apparatus 200a.

  FIG. 5 is a diagram showing an example of a failure rate chart of maintenance parts such as the X-ray tube 11. In the failure rate chart 50 shown in FIG. 5, the horizontal axis represents the operation time t of the maintenance part, and the vertical axis represents the failure rate λ (t) of the maintenance part. The solid line, the broken line, and the alternate long and short dash line drawn therein are examples of failure rate curves when the usage method and usage environment are different for a certain maintenance component (for example, the X-ray tube 11). It represents.

Here, the failure rate λ (t) is an occurrence probability of a failure that occurs per unit time, and is expressed by equations (6-1) to (6-3) between the failure accumulation distribution F (t). Have a relationship.

  In FIG. 5, an average failure rate curve 50h (a dashed line curve) is an example of a failure rate curve obtained from an averaged life without considering the usage method and usage environment of the maintenance part (for example, the X-ray tube 11). It is. In the conventional concept of average life or failure rate statistics, only this average failure rate curve 50h is usually obtained.

  Note that the failure rate curve shown in FIG. 5 corresponds to the failure rate curve in the wear failure period after almost the accidental failure period in the so-called bathtub curve failure model. In general, the failure rate λ (t) in the wear failure period increases as the operating time increases. For the average failure rate curve 50h, a threshold value 50s of the failure rate λ (t) is preset, and the operating time t when the average failure rate curve 50h exceeds the threshold value 50s is The maintenance part replacement time 50e is determined.

  However, as described above, the service life of maintenance parts such as the X-ray tube 11 varies greatly depending on the method of use. That is, if the maintenance part is used under severe use conditions, it becomes a curve that rises earlier than the average failure rate curve 50h, as shown in the failure rate curve 50a of the method of use A in FIG. If it is used, it becomes a curve that rises later than the average failure rate curve 50h, like the failure rate curve 50c of the method of use C in FIG.

  Therefore, for example, when the maintenance part used in the usage method C (failure rate curve 50c) is replaced when the operation time t reaches the replacement time 50e, the maintenance part is still usable. Since the failure rate λ (t) is sufficiently smaller than the threshold value 50s, the maintenance cost increases as compared with the case where the failure rate λ (t) is used until the failure rate λ (t) actually reaches the threshold value 50s.

  On the other hand, when the maintenance part used in the usage method A (failure rate curve 50a) is to be replaced when the operation time t reaches the replacement time 50e, many maintenance parts are replaced. It will break down before. In this case, extra time is required for post-processing for handling a failure, resulting in a decrease in operating rate and an increase in operation cost, resulting in an increase in maintenance cost.

  Even if the same maintenance parts are used under the same use conditions, their failure rate curves are not necessarily the same if their use environments (ambient temperature, humidity, etc.) are different. For example, as shown in FIG. 5, even if the usage method A is the same, the failure rate curves 50a1 and 50a2 are often different between the usage environment A1 and the usage environment A2.

  After all, the average failure rate curve 50h is effective only in the case of the usage method B (50b) used in the most average usage conditions and usage environments. Therefore, in order to predict the life of maintenance parts according to the usage method and usage environment, and to set the replacement time optimally, it is necessary to individually model the failure rate curve according to the usage method and usage environment. is there.

  In that case, if there is a sufficient number of reliability data for each method of use and environment, the reliability data is stratified by method of use and environment, and the failure rate is determined using Weibull probability paper for each method of use and environment. What is necessary is just to obtain a curve. However, this is a big problem of a general method conventionally performed.

  That is, of the reliability data acquired for the maintenance parts, case data (field data) in which the life of the maintenance parts can be obtained due to the occurrence of a wear failure is not data of a nature that can be obtained in large numbers. Further, when the case data is stratified, the number of data after stratification is further reduced depending on the stratification. Therefore, it is difficult to draw an approximate line on the Weibull probability paper with such data. Therefore, it becomes impossible to obtain the predicted value of the lifetime. Alternatively, if an attempt is made to fit a straight line, an erroneous life or a life with a large error is required.

  As described above, in the prior art, when obtaining the life of a highly accurate maintenance part stratified by usage method and environment, it is difficult to obtain a sufficient number of reliability data sufficient to obtain it. For this reason, the problem is that it is impossible to determine the service life of maintenance parts with high accuracy.

(5. Solution in the embodiment of the present invention)
Therefore, in the present embodiment, the reliability data of the maintenance parts such as the X-ray tube 11 obtained in the field by effectively using the data itself related to the method of use and the environment of use of the electromechanical device 200, the method of use, We present a method to predict the service life of maintenance parts without stratification in the usage environment.

(5.1 Creation of integrated life prediction model)
In the present embodiment, the model creation simulator unit 110 creates an integrated life prediction model based on Expression (7) using reliability data including all failure information.

  Equation (7) is an equation representing the failure probability density distribution f (t) obtained by differentiating the failure cumulative distribution F (t) according to the Weibull distribution (see Equation (6-2)). The parameters m, η, and γ in (7) are a shape parameter, a scale parameter, and a position parameter in the Weibull distribution, respectively.

Next, with respect to these shape parameter m, scale parameter η, and position parameter γ, variables z _i representing the usage method (for example, operation interval, warm-up operation time, etc.) and the usage environment (for example, ambient temperature, humidity, etc.) The relationship of Formula (8) is set between

That is, the shape parameter m, the scale parameter η, and the position parameter γ are the common shape parameter m ₀ , the common scale parameter η ₀ , and the common position parameter γ ₀ , which are values common to the variable z _i representing the usage method and environment. And a value that is individually proportional to the variable z _i representing the method of use or environment (ie, m _i · z _i , η _i · z _i , γ _i · z _i ) And

Then, by setting a threshold value of the percentage point s (0 <s <1) for the failure probability cumulative distribution obtained by integrating the failure probability density distribution f (t) of Equation (7), Equation (9) The predicted value of the life shown in FIG. 1, that is, the predicted life T ^life hat can be obtained.

In the failure probability density distribution model shown above, regarding the effects of the usage method and the usage environment, the common effects are the parameters m ₀ , η ₀ , γ ₀ , and the individual effects are the proportional coefficients m _i , η _i , It is incorporated in an integrated form as γ _i . Therefore, even when the number of failure cases in the field is small, the common effect parameters m ₀ , η ₀ , γ ₀ can be obtained from all the failure cases, and the proportional coefficients m _i , η _i , Since γ _i can be obtained as a difference from the common effect, the lifetime can be accurately predicted.

(5.2 Issues existing in the integrated life prediction model)
In the integrated life prediction model based on Expression (9), three parameter prediction models based on Expression (8) are prepared in order to predict the three parameters of the shape parameter m, the scale parameter η, and the position parameter γ. Therefore, in order to use the parameter prediction model, m ₀ , η ₀ , γ ₀ that are parameters representing the respective intercepts in the three parameter prediction expressions represented by the equation (8), and a parameter that represents the proportional coefficient It is necessary to determine in advance m _i , η _i , and γ _i in advance (hereinafter, parameters indicating the intercept and the proportional coefficient are referred to as meta parameters).

That is, in order to use the parameter prediction model of equation (8), a total of 6 or more meta parameters (the minimum is 6 when the number of variables z _i representing the usage method or usage environment is 1) is determined in advance. It is necessary to keep it.

These meta-parameters are so-called regression model parameters that use the variable z _i representing the method of use and the environment of use as explanatory variables, so in general, statistically using the maximum likelihood method, the least square method, etc. Determined. For this purpose, reliability data including a large number of failure cases is required. However, when the amount of reliability data is small, the accuracy of meta parameters determined by the maximum likelihood method, the least square method, or the like is lowered. In addition, when categorizing and determining meta parameters according to usage methods and usage environments, the reliability data is further divided into a smaller amount of data, so that the accuracy of the determined meta parameters is further reduced.

  On the other hand, the user of the electromechanical device 200 tries to use the integrated life prediction model based on the formula (9) from the initial stage where the amount of reliability data regarding the maintenance parts is small. Otherwise, the prediction model cannot claim superiority to conventional life prediction methods using Weibull probability paper. That is, the problem to be solved in the integrated life prediction model based on the equation (9) is how to accurately determine the metaparameter of the equation (8) when only a small amount of reliability data is accumulated. That is to say.

(6. Processing in the model creation simulator)
Therefore, in order to solve this problem, in this embodiment, the method of determining the metaparameter of the parameter prediction model from the reliability data including a small amount of failure case data is not used, and the probability distribution of the metaparameter of the parameter prediction model is not adopted. Is evaluated by simulation, and the meta parameter value is determined from the average value or mode value.

  In order to realize such a method for determining a meta-parameter of a parameter prediction model, the maintenance part life prediction system 100 (see FIG. 1) in the present embodiment is provided with a model creation simulator unit 110, and further, Inside the model creation simulator unit 110, a model creation trial unit 111, a stochastic parameter sample generation unit 115, and a model accuracy convergence determination unit 116 are provided.

Then, in the model creation simulator unit 110, the probabilistic parameter sample generation unit 115 obtains probability values of six metaparameter candidate values (when the number of variables z _i is one) in the three parameter prediction models in Equation (8). And the generated meta parameter candidate value is sent to the model creation trial unit 111.

  Further, the life prediction model creation trial unit 112 of the model creation trial unit 111 performs simulation evaluation on the predicted value of the parameter and the life prediction model based on the parameter based on the meta parameter candidate value of the parameter prediction model.

  Subsequently, the life prediction calculation trial unit 113 obtains a predicted value of the failure rate λ and further obtains its prediction accuracy. At this time, the actual measurement value to be compared with the predicted value is, for example, the failure probability density distribution f () based on the section value of the estimated value of the failure cumulative distribution F (t) obtained from the equation (1), (2) or (3). An estimated value of t) is obtained.

  The model accuracy convergence determination unit 116 performs statistical convergence determination of the prediction accuracy obtained by the simulation evaluation. The processes in the stochastic parameter sample generation unit 115, the model creation trial unit 111, and the model accuracy convergence determination unit 116 are repeated until the convergence is determined in the convergence determination.

  The model creation simulator unit 110 calculates an average value or mode value from the simulation data of the convergence interval of each metaparameter candidate of the parameter prediction model, determines the calculated value as a metaparameter value, and transmits it to the life calculation unit 130 To do.

(7. Application of MCMC method)
By the way, when the metaparameter candidate values of the parameter prediction model in the stochastic parameter sample generation unit 115 are randomly generated by a simple Monte Carlo simulation method, it is long for the prediction accuracy and the probability parameter distribution of the metaparameters to converge. It takes time or does not converge. Therefore, in the present embodiment, the model creation simulator unit 110 performs a series of processes for generating metaparameter candidate values of the parameter prediction model (that is, sampling), updating the life prediction model candidates, and evaluating the convergence of the probability sample distribution. The MCMC method for the Bayes model is applied.

  If the MCMC method for the base model is applied to a series of processes in the model creation simulator unit 110, the parameter posterior probability distribution is obtained from the product of the likelihood derived from the actual value of the reliability data and the prior probability distribution of the parameter to be obtained. Furthermore, since the meta parameter candidates of the parameter prediction model can be generated (by sample sampling) based on the posterior probability distribution, the effect of fast convergence of the model accuracy can be obtained.

  FIG. 6 is a diagram illustrating an example of a processing flow of model creation simulation processing in the model creation simulator unit 110. As shown in FIG. 6, the model creation simulation process includes a parameter sample initial value setting process (step S10), a model creation trial process (step S11), and a prediction error variance calculation of a life prediction model (failure probability density model) using a Markov chain. And posterior distribution update processing (step S12), Markov chain convergence determination processing (steps S13 and 13a), convergence interval sample sequence extraction processing from Markov chain (step S14), life prediction model (failure probability) based on the sample sequence of the convergence interval Density model) parameter determination process (step S15), life prediction model (failure probability density model) parameter update process (step S16) for the life calculation unit, life prediction model parameter sample generation update process (step S17) by Markov chain, etc. Consists of.

  The model creation trial process (step S11) includes a data section initialization process (step S11a), a data record reading process (step S11b), and a life prediction model (failure probability density model) creation trial process (step S11a). Step S11c), prediction trial processing using a life prediction model (failure probability density model) (Step S11d), life prediction error calculation processing (Step S11e), data section end determination processing (Step S11f), and the like. .

  Hereinafter, with reference to FIG. 6, the details of the MCMC method and the model creation simulation process to which the MCMC method is applied will be described. First, for example, the failure probability density distribution f (t (n)) from the section value of the estimated value of the failure cumulative distribution F (t (n)) obtained from the expressions (1), (2), (3), etc. Obtain an estimate of.

Furthermore, six metaparameters of the three parameter prediction models of the equation (8) (hereinafter, the number of variables z _i is assumed to be one in order to avoid troublesome description) as a random variable θ, A base model relating to the life data T (n) shown in Expression (10) is set.

  As shown in the equation (10), the function π (left side) which is the posterior probability density function of the random variable θ based on the actual value of the lifetime is the lifetime data T (n) when the random variable θ is given. Is expressed by a product of a function L that is a likelihood function of and a function π (right side) that is a prior probability density function of a random variable θ.

  Here, assuming that the life data T (n) follows the Weibull distribution (see Expression (7)), the life data T (n) is expressed as the following Expression (11), and the likelihood function L Is expressed as in equation (12).

Further, the parameter prediction model for predicting the three parameters m, η, and γ is a regression model having the variable z ₁ (n) representing the usage method and the usage environment as explanatory variables, as in Expression (8). It shall be represented by the following formula (13). Furthermore, the prior probability density function of the six metaparameters m ₀ , η ₀ , γ ₀ , m ₁ , η ₁ , γ ₁ included in the prediction model is assumed to be represented by the distribution of Equation (14). However, it is not necessarily limited to these distributions.

Here, when there is prior knowledge regarding distribution due to failure experience in other devices or other parts, the distribution parameter values of the exponential distribution Exp and the normal distribution N in Expression (14) are set based on the prior knowledge. be able to. In this embodiment, since it is not assumed that there is prior knowledge, values that increase the width of the distribution such as 10 ⁻³ and 10 ³ are set as the distribution parameter values of the exponential distribution Exp and the normal distribution N. However, in the absence of prior knowledge, the sensitivity of these distribution parameters to the results is small, and it is not necessary to set their statistical values exactly to these values.

Next, the life prediction error E (n) is calculated by using the empirical density distribution f (n) of the life data T (n) and the probability density distribution obtained from the Weibull distribution of Expression (11). Set as follows.

  Next, the processing flow of the model creation simulation process by the MCMC method will be specifically described with reference to FIG. 6 (see FIG. 1 as appropriate).

  First, the model creation simulator unit 110 sets initial values of six meta parameters of the three parameter prediction models included in the random variable θ (life prediction model parameter sample initial value setting process: step S10). Since the sensitivity to the results of these initial values is not high, it may be set by sampling appropriately from the distributions set in advance in Equation (11) and Equation (12).

  Next, the model creation simulator unit 110 repeatedly executes the processing of step S11a to step S11f that sequentially performs sample value calculation according to the Markov chain and according to the posterior distribution of Expression (10) (model prediction trial processing: step S11). .

  In the model prediction trial process (step S11), the life prediction model creation trial unit 112 first receives and receives the current value of the random variable θ (six meta parameters) transmitted from the stochastic parameter sample generation unit 115. A parameter prediction model of Expression (13) is created using the six metaparameters, and a data section for model creation trial is initialized (data section initialization process: step S11a).

  Next, the life prediction model creation trial unit 112 uses the data on the failure occurrence time (that is, life) T (n) and the failure probability density estimated value f (n) in the set reliability data section, Variable data z (n) representing the use environment is sequentially read from the reliability data storage unit 140 (data record reading process: step S11b). Then, an attempt is made to create a life prediction model according to the three model parameters m, η, and γ derived from the read data, and the obtained life prediction model is transmitted to the life prediction calculation trial unit 113 (life prediction). Model creation trial process: Step S11c).

  Next, the lifetime prediction calculation trial unit 113 sequentially acquires the reliability data in the reliability data section set in step S11a from the reliability data storage unit 140, and substitutes it into the lifetime prediction model received earlier. The life (corresponding to the failure probability density estimated value f (n)) is predicted (prediction trial process: step S11d). Then, the lifetime prediction calculation trial unit 113 acquires the lifetime (corresponding to the failure probability density estimated value f (n)) data of the previously set section from the reliability data storage unit 140, and compares it with the predicted value for prediction. An error is calculated (life prediction error calculation process: step S11e).

  Next, the life prediction calculation trial unit 113 creates a life prediction model for all the data sections set in step S11a, and determines whether or not the life prediction trial process has ended (step S11f). As a result of the determination, if the life prediction process has not been completed for all the data sections (No in step S11f), the process returns to the previous and the processes after the data record reading process (step S11b) are repeatedly executed. .

  On the other hand, when the life prediction process has been completed for all the data sections (Yes in step S11f), the life prediction calculation trial unit 113 calculates the variance of the prediction error series obtained by the processing so far, The prediction error variance and the posterior distribution of the random variable θ are updated (prediction error variance calculation and posterior distribution update processing: step S12).

  Next, the model accuracy convergence determination unit 116 converges a chain set of sample values to a predetermined distribution from the posterior distribution of the metaparameter variance included in the sample value of the random variable θ, and after the determination of convergence, It is determined whether or not the number of Markov chain samples has been calculated (Markov chain convergence condition determination processing: step S13).

  If the convergence condition is not satisfied (No in step S13a), the probabilistic parameter sample generation unit 115 thereafter performs the process from among the metaparameters included in the random variable θ each time this step is passed. One is selected in order according to the Markov chain, and the sample value is newly generated from the posterior distribution of the formula (10) in which the parameter was updated last time, and the updated parameter value and the remaining previous parameter Is transmitted to the model creation trial unit 111 (life prediction model parameter sample generation update processing: step S17).

  Subsequent to the process of step S17, the process shifts from the stochastic parameter sample generation unit 115 to the process of the model creation trial unit 111, and the processes after step S11a are repeatedly executed.

  On the other hand, if the convergence condition for the Markov chain is satisfied in step S13 (Yes in step S13a), the model accuracy convergence determination unit 116 ends the repeated calculation processing so far, and determines the convergence interval. The sample sequence of the random variable θ is cut out (convergence interval sample sequence cut-out process: step S14).

  FIG. 7 is a diagram showing an example of how the parameter sample series converges by the MCMC method (repetitive processing up to step S13 in FIG. 6). FIG. 7A shows the interval from the first time until convergence and the convergence. An example showing how a parameter sample sequence converges in a later predetermined number of convergence intervals, and (b) is an example showing a converged parameter sample sequence in a predetermined number of convergence intervals after convergence.

  In FIG. 7A, the left chart 55a shows how the prediction error sample series converges, and the right chart 55b shows how one metaparameter sample series converges. Here, the convergence section 56 refers to a section in which convergence is determined to be confirmed, for example, a section from 2,000 times to 5,000 times.

  In FIG. 7B, the left chart 57a is a parameter sample series obtained by cutting out the convergence section 56 from the left chart 55a in FIG. 7A, and the right chart 57b represents the frequency distribution. It is. From the parameter sample series of the convergence section 56, the convergence section average value 58 is calculated.

  Referring to the processing flow of FIG. 6 again, the model creation simulator unit 110 calculates the average value or the median value based on the sample series of the convergence interval 56 of each metaparameter included in the random variable θ, and the value of the metaparameter Is determined (life prediction model parameter determination process: step S15). Thereafter, the model creation simulator unit 110 transmits the convergence interval average value 58 (see FIG. 7B) for each of the six meta parameters of the three parameter prediction models to the life prediction model creation unit 132 of the life calculation unit 130. Then, the prediction model in the life calculation unit 130 is updated (life prediction model parameter update processing: step S16).

  The content of the model creation simulation process based on the MCMC method executed by the model creation simulator unit 110 has been described above with reference to FIG. 6. Details of the actual calculation are, for example, the Metropolis Hastings algorithm, Gibbs Sampler algorithm, Slice Sampler It is executed based on an algorithm or the like (see Non-Patent Document 1, etc.).

  The model creation simulation process described above is stored in the storage device 33 as a program of the model management computer 31 installed in the operation center 300 shown in FIG. Executed. Further, when the program is executed, data in the parameter sample value storage unit 121, the life prediction model storage unit 122, and the reliability data storage unit 140 are appropriately read and written.

(8. Processing in the life calculation unit)
FIG. 8 is a diagram illustrating an example of a processing flow of a life calculation process in the life calculation unit 130 of the maintenance component life prediction system 100. Hereinafter, with reference to FIG. 8 (refer to FIG. 1 as appropriate), the processing flow of the life calculation processing in the life calculation unit 130 will be described.

  The life prediction model creation unit 132 of the life calculation unit 130 first receives the six metaparameter values of the parameter prediction model transmitted from the model creation simulator unit 110, and uses the received metaparameter values as a parameter prediction model (formula ( 13)) is updated (step S20).

Next, the life prediction model creation unit 132 obtains data (value of the variable z ₁ ) representing the usage method and usage environment regarding the maintenance part to be predicted or the electromechanical device 200 having the maintenance part from the reliability data storage unit 140. Obtain (step S21).

  Next, the life prediction model creation unit 132 substitutes data representing the use method and use environment into the parameter prediction model (formula (13)), thereby obtaining three parameters (a shape parameter m, a scale parameter η, and a position parameter). The value of γ) is obtained, and the values of the obtained three parameters are substituted into the equation (7) to determine the failure probability density distribution f (t) (step S22).

Next, the life prediction calculation unit 133 calculates the predicted life of the maintenance part based on the determined maximum value of the failure probability density distribution f (t) or the threshold value s of the percent point (step S23). . That is, when m> 1, since the failure probability density distribution f (t) has a maximum value, the operation time t at which the differential value of f (t) becomes zero is assumed to be the predicted life (predicted life value). Can be defined as a T ^Life hat. In that case, the predicted life T ^Life hat for the data z ₁ representing the usage method / use environment is given by the following equation (16).

Further, a threshold value s is set in advance for the percentage point obtained by integrating the failure probability density distribution f (t), and the operation time t when the percentage point exceeds the threshold value s is calculated as the predicted life ( (Predicted value of life) T ^Life hat. In that case, the predicted life T ^Life hat for the data z ₁ representing the usage method or usage environment is given by the following equation (17).

In this case, the life prediction calculation unit 133 acquires the threshold value s of the percent point preset for each maintenance part from, for example, the reliability data storage unit 140, and the acquired threshold value s By substituting for (17), the predicted life T ^Life hat of the maintenance part can be obtained.

Next, the maintenance part replacement cycle notifying unit 134 sets or updates the maintenance part replacement cycle based on the predicted life calculated in step S23 (step S24), and performs maintenance management of the set or updated maintenance part replacement cycle. The device 135 is notified. The maintenance part replacement cycle is usually set to a time slightly shorter than the predicted life T ^Life hat. In addition, the setting or updating of the maintenance part replacement cycle may be performed periodically at a preset period, or may be performed in synchronization with a failure occurrence event.

  The maintenance management device 135 monitors the operation time of the corresponding maintenance component of each electric machine apparatus 200 having jurisdiction in accordance with the notified maintenance component replacement cycle. When the operation time approaches the maintenance component replacement cycle, the maintenance management device 135 It is displayed on the display device of the operation computer 41 (see FIG. 2) that the part replacement date is approaching, or the maintenance worker is notified by e-mail or the like.

  In addition, the maintenance management device 135 displays a list of maintenance parts that will exceed the predicted life one after another unless they are replaced next time so that a plurality of maintenance parts can be collectively replaced by regular maintenance work as necessary. You may make it display on a display apparatus. Furthermore, the maintenance management device 135 may automatically order maintenance parts that will be required in the near future, taking into account separately the procurement lead times of registered maintenance parts. With such a function of the operation computer 41, the maintenance worker can replace the maintenance parts of the electromechanical device 200 in the maintenance part replacement cycle based on the latest predicted life of the maintenance parts without leakage. Become.

  FIG. 9 is a diagram showing an example of a maintenance component life prediction system screen displayed on the display device of the model management computer 31. As shown in FIG. 9, the maintenance component life prediction system screen 310 includes a model creation simulator subscreen 310a and a life calculation subscreen 310b.

  The model creation simulator subscreen 310a is provided with an execution button 313a for model creation simulation, as well as a data section designation subscreen 314 and a life probability distribution parameter calculation subscreen 315.

  The execution button 313a is a button representing an instruction to execute a model creation simulation. That is, when the operator clicks the execution button 313a, the model management computer 31 executes a model creation simulation process (see FIG. 6) by the model creation simulator unit 110. Note that the model management computer 31 is not limited to the time when the operator clicks the execution button 313a, but when the execution instruction is received from a computer of a host system (not shown), or a timing such as a preset date and time. Thus, this model creation simulation process may be executed.

  On the data section designation sub-screen 314, boxes 314a and 314b for displaying the start calendar date time and end calendar date time of the model creation data section are displayed, and the start calendar date time of the model creation data section is further displayed. A radio button 314c for selecting whether to set the end calendar date time automatically or manually is displayed.

  When automatic setting is selected with the radio button 314c, the model management computer 31 starts in accordance with a predetermined rule (for example, from January 1, five years ago to December 31, last year). The calendar date time and end calendar date time are automatically set, and the set start calendar date time and end calendar date time are displayed in boxes 314a and 314b. When the manual setting is selected with the radio button 314c, the model management computer 31 displays a lower subscreen (not shown) for setting the start calendar date time and the end calendar date time, The start calendar date time and end calendar date time set on the sub-screen are displayed in boxes 314a and 314b.

  In addition, the life probability distribution parameter calculation subscreen 315 includes a radio button 315a for selecting a parameter sample series convergence determination method (automatic setting or manual setting) obtained by MCMC simulation, a parameter sample series chart 315g, and a convergence section. Display boxes 315i, 315j, 315k, 315p, 315q, 315r, and the like that display the values of the probability distribution parameters (see equation (14)) determined from the parameter sample series are displayed.

  When the manual setting is selected by the radio button 315a, the model management computer 31 displays the convergence start cursor 315h on the parameter sample series chart 315g and also displays the left / right movement buttons 315c and 315d displayed at that time. A click operation is accepted, and the convergence start cursor 315h is moved right and left as appropriate in accordance with the input of the click operation. Here, the display box 315e displays the number of simulation iterations representing the position of the convergence start cursor 315h.

  Next, upon receiving an operation input for clicking the recalculation button 315f, the model management computer 31 determines the position of the convergence start cursor 315h, determines the convergence start point, and sets the convergence section. Further, the model management computer 31 calculates the parameter value of each probability distribution for which the prior distribution is defined by the equation (14) based on the average value or the mode value of the parameter sample series in the convergence section of the posterior distribution after the MCMC simulation. The determined parameter values are displayed in the display boxes 315i, 315j, 315k, 315p, 315q, and 315r, respectively.

  Further, as shown in FIG. 9, the life calculation subscreen 310b is provided with an execution button 313b for life calculation, and an object designation subscreen 316 and a calculation method selection & calculation result subscreen 317. .

  The execution button 313b is a button that represents a life calculation execution instruction. That is, when the operator clicks the execution button 313b, the model management computer 31 executes a life calculation process (see FIG. 8) by the life calculation unit 130. Note that the model management computer 31 is not limited to the time when the operator clicks the execution button 313b, but when the execution instruction is received from a computer of a host system (not shown), or a timing such as a preset date and time. Thus, this life calculation process may be executed.

  The target designation sub-screen 316 displays dialog (dialogue) boxes 316a and 316b for designating the installation location and maintenance parts of the electromechanical device 200 subject to lifetime calculation, and the usage method and usage environment of the electromechanical device 200. A box 316c to be displayed. When the operator designates the installation location and the maintenance parts via the dialog boxes 316a and 316b, the model management computer 31 uses the usage data corresponding to the maintenance parts of the electromechanical device 200 from the reliability data storage unit 140 to the installation location. The usage environment is searched, and the searched usage method and usage environment are displayed in a box 316c.

In FIG. 9, the symbol z ₁ means a variable representing the usage method / use environment in Equation (13). Then, change the value of the operator is displayed in a box 316c, when the user clicks the change button 316d, the value of the variable z ₁ shall be changed. In addition, when a list of data such as installation location, maintenance parts, usage method / use environment is created in advance, the model management computer 31 does not depend on an operator's instruction using the target designation sub-screen 316. Based on the data of the list, the life calculation is executed.

  In the calculation method selection & calculation result sub-screen 317, a radio button for selecting whether to use the maximum value of the life probability distribution of Expression (16) or the percentage point of Expression (17) as the life calculation method 317a is displayed, life calculation is performed according to the selected radio button 317a, and a predicted life as a result of the life calculation is displayed in the display box 317c. When the percentage point is selected with the radio button 317a, a box 317b for setting the threshold value s is displayed.

  The calculation method selection & calculation result sub-screen 317 further displays a transmission button 317d. When the transmission button 317d is clicked by the operator, the predicted life that is the result of the life calculation is the maintenance management device 123 ( To the operation computer 41). Note that the result of the life calculation may be a maintenance part replacement cycle (see FIG. 8, step S24) obtained based on the predicted life instead of the predicted life itself.

  FIG. 10 is a diagram showing an example of a maintenance management screen displayed on the display device of the maintenance management device 123 (operation computer 41). As shown in FIG. 10, the maintenance management screen 410 includes a maintenance component designation subscreen 411 and a maintenance plan subscreen 412.

  On the maintenance parts designation subscreen 411, a box 411a for displaying the installation location of the electromechanical device 200 that is the target of the maintenance plan, a dialog box 411b for designating maintenance parts, and the like are displayed, and further, designated by the dialog box 411b. The predicted service life of the maintenance part is displayed in the display box 411c. The predicted service life of the maintenance part displayed at that time is previously transmitted from the service life prediction system 100 (model management computer 31). The display box 411c may display a maintenance part replacement cycle instead of the predicted life of the maintenance part.

  Further, an inquiry button 411d is displayed on the maintenance part designation sub-screen 411, and when the operator clicks the inquiry button 411d, the maintenance management device 123 makes the latest life calculation to the maintenance part life prediction system 100. Queries whether there is a result. If it is found from the inquiry that there is an unacquired latest lifetime calculation result, the maintenance management device 123 acquires the latest lifetime calculation result from the maintenance component lifetime prediction system 100 and displays the display box 411c. Update the predicted life displayed in.

  The maintenance plan sub-screen 412 is provided with a display box 412a for displaying the service life arrival date of the maintenance part designated in the dialog box 411b, a dialog box 412b for selecting a scheduled maintenance plan allocation date, and a registration button 412c. The maintenance management device 123 obtains the life reached date by adding the predicted life to the previous replacement date of the maintenance part (or the first operation start date of the electromechanical device 200), and displays the obtained life reached date. This is displayed in the box 412a. The operator can set the scheduled maintenance plan allocation date by using the dialog box 412b and setting the scheduled maintenance plan allocation date by clicking the registration button 412c based on the life arrival date displayed in the display box 412a. it can.

  Here, the maintenance management device 123 creates a maintenance plan for maintenance parts designated by the operator via the maintenance management screen 410 (that is, sets a regular maintenance plan allocation date). Regardless of this, based on a list of maintenance parts created in advance, a maintenance plan for a plurality of maintenance parts is created when an instruction from a host computer (not shown) is received or at a preset date and time. It may be.

(9. Effects of the present embodiment)
In the present embodiment, the service life of the maintenance part of the electric machine device 200 (for example, the X-ray tube 11 of the X-ray CT apparatus 200a) can be predicted in consideration of the usage method and use environment of the electric machine device 200 or the maintenance part. I did it. According to the life prediction model, for example, the life prediction model is predicted by three parameters m, η, γ and elapsed time or operation time of the Weibull model, and the parameters m, η, γ are determined by the electromechanical device. It is predicted by variables representing the usage method and usage environment of 200 or maintenance parts. At this time, the meta parameters for predicting the parameters m, η, and γ are determined so as to match the field data of the service life of the maintenance part by random number simulation using the MCMC method or the like.

  Therefore, in this embodiment, since it is not necessary to stratify the field data of the lifetime accumulated in the reliability data storage unit 140, even when the field data is small, the predicted lifetime with a high accuracy of the maintenance part is obtained. be able to. In other words, even when the electromechanical device 200 and maintenance parts are newly developed products and there is no field data with sufficient lifetime, the electromechanical device 200 is used in various methods and environments. However, if any field data is collected, the service life of the maintenance part can be predicted earlier.

  As a result, it is possible to appropriately set the replacement cycle of the worn out and deteriorated part using the predicted life. Therefore, it is possible to prevent an increase in maintenance cost due to an early replacement, and it is possible to prevent a failure from occurring due to a setting of an exchange cycle that is too late and an increase in apparatus downtime.

  As described above, in the present embodiment, the electromechanical apparatus 200 is described as the X-ray CT apparatus 200a. However, the electromechanical apparatus 200 is not limited to the X-ray CT apparatus 200a, and includes components that deteriorate and wear. Any device can be used.

  For example, the electromechanical device 200 may be a machining device for polishing, grinding, and cutting, and is an automobile, a construction machine, a train, an aircraft equipped with wheels, brakes, tires, driving devices (engines, motors), and the like. It may be a power plant equipped with a turbine or the like, an analyzer having a fluorescent light source or the like, an inspection device such as an SEM equipped with an electron gun, or a hard disk device. In addition, the information processing apparatus may be an integrated circuit manufacturing apparatus using plasma or laser.

10 Gantry
11 X-ray tube (maintenance parts)
DESCRIPTION OF SYMBOLS 12 High voltage generator 13 Collimator 14 X-ray bundle 15 Detector 16 Data collection part 17 Image reconstruction calculation apparatus 31 Model management computer 32 Calculation processing device 33 Storage device 41 Operation computer 42 Calculation processing device 43 Storage device 100 Maintenance part lifetime Prediction system 110 Model creation simulator unit 111 Model creation trial unit 112 Life prediction model creation trial unit 113 Life prediction calculation trial unit 115 Probabilistic parameter sample generation unit 116 Model accuracy convergence determination unit 121 Parameter sample value storage unit 122 Life prediction model storage unit DESCRIPTION OF SYMBOLS 130 Life calculation part 132 Life prediction model preparation part 133 Life prediction calculation part 134 Maintenance part replacement cycle notification part 135 Maintenance management apparatus 140 Reliability data storage part 200 Electromechanical device 200a X-ray CT apparatus 300 Operation center 31 Maintenance part life prediction system screen 314 Data section specification subscreen 315 Life probability distribution parameter calculation subscreen 316 Target specification subscreen 317 Calculation method selection calculation result subscreen 400 Site 410 Maintenance management screen 411 Maintenance part specification subscreen 412 Maintenance plan subscreen 500 communication network

Claims (10)

A maintenance part life prediction system for predicting the life of maintenance parts of electromechanical devices whose life varies depending on the method of use and use environment,
The first data representing the elapsed time or the operation time from the start of use of the maintenance part to the failure and the second data representing the usage method and usage environment of the maintenance part are associated with each other. A reliability data storage unit in which sex data is stored;
The value of the meta parameter included in the parameter prediction model for predicting the value of the parameter included in the life prediction model for predicting the life of the maintenance part from the second data, the first data, and the second data. A model creation simulator unit determined by simulation using a random number according to a probability distribution set in advance for the meta parameter, and
The lifetime that determines the value of the parameter included in the lifetime prediction model from the parameter prediction model having the value of the meta parameter determined by the model creation simulator unit and the value of the second data for the maintenance component that is the lifetime prediction target A prediction model creation unit;
Based on a life prediction model having a parameter value determined by the life prediction model creation unit, a life prediction calculation unit that predicts a predicted life of the maintenance component that is the life prediction target,
A maintenance parts life prediction system characterized by comprising: A maintenance part replacement cycle notifying unit for determining a replacement cycle of the maintenance part based on the predicted life of the maintenance part calculated by the life prediction calculating unit and notifying a maintenance worker of the determined replacement cycle; The service life prediction system according to claim 1 , wherein: The service life prediction system according to claim 1 , wherein the life prediction model is a Weibull model having at least one of a shape parameter, a scale parameter, and a position parameter. The maintenance part lifetime prediction system according to claim 1 , wherein the parameter prediction model is a regression model having data representing the usage method and usage environment as explanatory variables. The simulation that the model creation simulator unit performs to determine the meta parameter is:
The posterior probability distribution for the meta parameter is defined using the probability distribution set in advance for the meta parameter and the likelihood obtained from the first data and the life prediction model, and the defined posterior The maintenance part life prediction system according to claim 1 , wherein the system is a Markov chain Monte Carlo simulation for generating a random sample sequence for the meta parameter according to a probability distribution. A service life prediction method for predicting, by a computer, the service life of an electromechanical device whose service life varies depending on the method of use and use environment,
The calculator is
The first data representing the elapsed time or the operation time from the start of use of the maintenance part to the failure and the second data representing the usage method and usage environment of the maintenance part are associated with each other. A reliability data storage unit for storing sex data,
The value of the meta parameter included in the parameter prediction model for predicting the value of the parameter included in the life prediction model for predicting the life of the maintenance part from the second data, the first data, and the second data. And a model creation simulation process defined by simulation using a random number according to a probability distribution set in advance for the meta parameter, and
The lifetime that determines the value of the parameter included in the lifetime prediction model from the parameter prediction model having the value of the metaparameter determined by the model creation simulation process and the value of the second data for the maintenance component that is the lifetime prediction target Predictive model creation process,
A life prediction calculation process for predicting a predicted life of the maintenance part to be predicted based on a life prediction model having a parameter value determined by the life prediction model creation process;
This is a method for predicting the service life of maintenance parts. The calculator is
Based on the predicted life of the maintenance part predicted by the life prediction calculation process, a maintenance part replacement cycle notification process for determining a maintenance part replacement cycle and notifying the maintenance worker of the determined replacement cycle is further executed. The maintenance part life prediction method according to claim 6 . The service life prediction method according to claim 6 , wherein the life prediction model is a Weibull model having at least one of a shape parameter, a scale parameter, and a position parameter. The maintenance parameter life prediction method according to claim 6 , wherein the parameter prediction model is a regression model having data representing the usage method and usage environment as explanatory variables. The simulation performed to determine the meta parameter in the model creation simulator process is:
The posterior probability distribution for the meta parameter is defined using the probability distribution set in advance for the meta parameter and the likelihood obtained from the first data and the life prediction model, and the defined posterior The service life prediction method according to claim 6 , wherein the service life prediction method is a Markov chain Monte Carlo simulation for generating a random sample sequence for the meta parameter according to a probability distribution.

Images