JP2019065713A - State discrimination device for plant - Google Patents

Abstract

To provide a state discrimination device for a plant capable of accurately discriminating a state of the plant during subsequent operation by efficiently and accurately creating a plant model representing a correlation between an internal state of the plant and output.SOLUTION: In a state discrimination device for a plant, operation parameters (an engine rotation number NE, an intake pressure Pin, a fuel injection quantity Q and an EGR valve opening degree φegr) representing an internal state of an internal combustion engine 3 as the plant are detected. Based on a predetermined number of data items with a greater Mahalanobis distance DMA among data of the operation parameters, a plant model representing a correlation between the operation parameters and an exhaust temperature Tex that is an output parameter of the internal combustion engine 3 is created (Fig 2) and a threshold (a threshold D_ABN or the like ) is set (step 40). The state of the internal combustion engine 3 is discriminated based on a relation between data of any operation parameters that are detected during operation of the plant, and the threshold (steps 53-61 or the like).SELECTED DRAWING: Figure 7

Description

  The present invention relates to a plant state determination device that creates a plant model that represents the correlation between an internal state of a plant and an output, and determines the state of the plant.

  DESCRIPTION OF RELATED ART Conventionally, what was described in patent document 1 as a determination apparatus which determines the failure of the exhaust gas temperature sensor of an internal combustion engine, for example is known. This determination device takes into consideration that the exhaust temperature sensor has a response delay characteristic, and estimates the output value of the exhaust temperature sensor including the response delay, and also uses this estimated output value as the actual output value of the exhaust temperature sensor. The failure is determined by comparing the exhaust temperature sensor.

The calculation of the above-mentioned estimated output value is performed as follows.
a. First, according to the detected intake oxygen concentration and the set fuel injection timing, a predetermined model expression (second model expression) defining the relationship between the intake oxygen concentration, the fuel injection timing, and the indicated thermal efficiency change amount is used. The indicated thermal efficiency change amount is calculated.
b. Next, the estimated exhaust temperature is calculated according to the indicated thermal efficiency change amount calculated in a using a predetermined model equation (third model equation) that defines the relationship between the indicated thermal efficiency change amount and the exhaust temperature.
c. Further, the estimated exhaust temperature, time constant, and the like calculated in b using a predetermined model equation (first model equation) defining the relationship between the exhaust temperature, the predetermined time constant of the exhaust temperature sensor, the exhaust flow rate According to the exhaust flow rate, an estimated output value reflecting the response delay of the exhaust temperature sensor is calculated.

  The calculated estimated output value is compared with the actual output value of the exhaust temperature sensor, and for example, when the difference between the actual output value and the estimated output value becomes larger than a predetermined threshold value, the exhaust temperature sensor breaks down. It is determined that

Unexamined-Japanese-Patent No. 2015-31169

  As described above, in this conventional determination device, a plurality of model equations (second, third and first model equations) are set in advance, and these model equations are sequentially used to estimate output of the exhaust temperature sensor The value is calculated. For this reason, the actual correlation between the operating parameters is based on the correlation in the model expression due to the variation of the operating characteristics between the individual devices of the internal combustion engine and the failure or deterioration of the sensor that detects the operating parameters. In the case of deviation, the estimated output value of the exhaust gas temperature sensor can not be calculated accurately. In that case, the accuracy of the failure determination of the exhaust temperature sensor using the estimated output value also decreases.

  The present invention has been made to solve the above-mentioned problems, and can efficiently create a plant model representing the correlation between the internal state of the plant and the output with high accuracy, and at the same time, the state of the plant during operation An object of the present invention is to provide a plant state determination device capable of estimating with high accuracy.

  In order to achieve this object, the invention according to claim 1 is an apparatus for determining a state of a plant, which is an internal state parameter representing an internal state of a plant (in the embodiment (hereinafter the same in this section) internal combustion engine 3). Internal state parameter detection means (crank angle sensor 20, intake pressure sensor 23, EGR valve opening sensor 25) for detecting the engine speed NE, intake pressure Pin, fuel injection amount Q, EGR valve opening φegr) Of the data of a large number of internal state parameters, using the predetermined number (M1 pieces) of data for which the distance (Mahalanobis distance DMA) from the center point of the distribution of the data is large, the internal state parameters and the output of the plant Plant model creation means (model) for creating a plant model (reference model) representing the correlation with an output parameter (exhaust temperature Tex) representing 2) and the threshold value (threshold D_ABN, threshold D_DET, threshold D_OPRN, threshold) for determining the state of the plant based on data of a predetermined number of internal state parameters and FIG. Threshold setting means (ECU 2, step 40 of FIG. 6, step 70 of FIG. 9, step 90 of FIG. 12, step 122 of FIG. 14) for setting D_INI) and operation of the plant after the plant model is created State determination means for determining the state of the plant based on the relationship between the data of a large number of internal state parameters detected by the internal state parameter detection means and the set threshold value (ECU 2, step 53 to FIG. 7 61, steps 83 to 88 in FIG. 10, steps 103 to 105 in FIG. 13, and steps 133 to 138) in FIG. It features.

  In this plant state determination device, an internal state parameter representing the internal state of the plant is detected by the internal state parameter detecting means, and the detected internal state parameter is used to represent the internal state parameter and the output of the plant. A plant model is created that represents the correlation with the output parameter. At this time, a predetermined number of data having a larger distance from the center point of the data distribution among the data of a large number of internal state parameters is selected and used for creating a plant model. The predetermined number of data thus selected covers the internal state of the plant and well reflects the smaller number of data. Therefore, a plant model with reduced modeling error is efficiently and accurately generated under a condition where the number of data is small compared to the case where all data including data whose distance from the central point is small is used. Can. Note that using “a predetermined number of data whose distance from the center point of the data distribution is larger” in claim 1 means using a predetermined number of data with a probability proportional to the size of the distance, or This includes the use of a predetermined number of data in order from the one with the largest distance.

  Further, according to the state determination device of the present invention, the threshold value is set based on the predetermined number of data of the internal state parameter used for creating the plant model. Therefore, the set threshold well reflects the internal state of the plant at the time of creation of the plant model. On the other hand, when the state of the plant changes during the subsequent operation of the plant, a change such as an error or deviation occurs in the detected internal state parameter. Therefore, it is possible to appropriately determine the state of the plant based on the relationship between the data of the internal state parameter detected during operation of the plant and the threshold set as described above.

  The invention according to claim 2 is the plant state determination apparatus according to claim 1, wherein the distance from the central point of the distribution of data is the Mahalanobis distance DMA, and the threshold setting means has a predetermined number of internal states. A threshold is set based on the Mahalanobis distance DMA of parameter data.

  According to this configuration, the Mahalanobis distance that reflects the correlation between data is used as the distance from the central point of the data distribution of multiple internal state parameters. Therefore, by setting the threshold value based on the Mahalanobis distance of the data of the predetermined number of internal state parameters used to create the plant model, the threshold value is reflected while the correlation between the data of the internal state parameters is reflected. The value can be set appropriately, whereby the state determination of the plant based on the comparison with the threshold can be performed more accurately.

  The invention according to claim 3 calculates the Mahalanobis distance DMA of data of a large number of internal condition parameters detected during operation of the plant in the plant state determination device according to claim 2 as the Mahalanobis distance DMA during operation. The driving state Mahalanobis distance calculation means (ECU 2, step 53 of FIG. 7, step 83 of FIG. 10, step 103 of FIG. 13, step 133 of FIG. 15) is further provided. And the threshold value to determine the state of the plant.

  According to this configuration, the Mahalanobis distance of the data of a large number of internal state parameters detected during operation of the plant is calculated as the Mahalanobis distance at the time of operation, and the plant is compared with the threshold at the time of operation. Determine the state of Thereby, based on the change of the internal state parameter including the change of the correlation between data since the time of creation of a plant model, state determination of a plant can be performed more appropriately.

  The invention according to claim 4 is the plant state determination apparatus according to claim 3, wherein the threshold setting means sets a threshold D_ABN for abnormality determination as the threshold (step 40 in FIG. 6). State determination means determines that the internal state parameter detection means is abnormal (steps 55, 56, 59 in FIG. 7) when the Mahalanobis distance DMA at the time of operation is larger than the threshold value D_ABN for abnormality determination. ~ 61).

  When an abnormality occurs in the internal state parameter detection means, an error occurs in the detected internal state parameter, and the correlation between the data decreases, thereby increasing the at-operation Mahalanobis distance. Therefore, according to this configuration, when the driving Mahalanobis distance is larger than the threshold value for abnormality determination, it can be appropriately determined that the internal state parameter detection means is abnormal.

  The invention according to claim 5 is the plant state determination apparatus according to claim 3, wherein the threshold value setting means sets the threshold value D_DET for deterioration determination as the threshold value as the plant operation period (vehicle travel distance). It is set according to DV) (step 70 in FIG. 9), and the state determination means determines that the internal state parameter detection means is deteriorated when the Mahalanobis distance during operation is larger than the threshold value D_DET for deterioration judgment. It is characterized by determining (steps 86 to 88 in FIG. 10).

  As the internal state parameter detection means suffers from deterioration over time, the error of the detected internal state parameter gradually increases, and the correlation between those data decreases, whereby the Mahalanobis distance during operation gradually increases. . Therefore, according to this configuration, as the threshold, the threshold for deterioration determination is set according to the operation period of the plant, and when the Mahalanobis distance during operation is larger than the threshold for deterioration determination, It can be appropriately determined that the internal state parameter detection means has deteriorated over time.

  The invention according to claim 6 is the plant state determination apparatus according to claim 5, wherein the state determination means determines the internal state parameter detection means due to deterioration when it is determined that the internal state parameter detection means is deteriorated. It is characterized by learning the deviation of the detection characteristic of (step 89 in FIG. 10).

  According to this configuration, it is possible to learn the deviation of the detection characteristic due to the deterioration at an appropriate timing according to the determination that the internal state parameter detection means is deteriorated.

  The invention according to claim 7 is the plant state determination apparatus according to claim 3, wherein the threshold setting means sets a threshold D_OPRN for operating area determination as the threshold (step in FIG. 12). 90) The state determining means determines that the operating range of the plant represented by the data is the plant model when the Mahalanobis distance DMA during operation of the data of the internal state parameter is larger than the threshold D_OPRN for determining the operating range. It is characterized in that it judges that it deviates from the operation field learned at the time of creation (Steps 105 to 106 in FIG. 13).

  The operating condition of the plant changes due to the characteristics of the operator of the plant, the operation according to preference or the operation different from the normal operation according to the operating environment or due to carelessness, and the internal condition parameter changes accordingly. As the correlation between data decreases, the driving Mahalanobis distance of the data of the internal state parameter increases. According to this configuration, as the threshold value, a threshold value for determining the driving range is set, and when the Mahalanobis distance during driving of the data of the internal state parameter is larger than the threshold value for determining the driving range, It is determined that the operation region of the plant represented by the data deviates from the operation region (hereinafter referred to as a “learning region”) learned when creating the plant model. Thereby, it is possible to appropriately determine the deviation of the operation area of the plant from the learning area, which is caused by the above-described characteristic driving of the driver, an operation different from the normal operation, or the like.

  The invention according to claim 8 is the plant state determination device according to claim 3, wherein the plant model creating unit is a plant model for a plurality of other plants (plural sample engines) having the same configuration as the plant. Threshold value setting means is set as a threshold value based on the Mahalanobis distance (Mahalanobis distance DMAM at the time of model creation) of data of a predetermined number of internal state parameters used to create a plant model of a plurality of plants. The threshold value for determining the initial state of the plant is set (Step 122 in FIG. 14), and the operating Mahalanobis distance calculation means calculates the operating Mahalanobis distance DMA during the initial operation of the plant (Step 133 in FIG. 15). State determination means, when the driving Mahalanobis distance is greater than the threshold for the initial state determination In the initial state of the plant, it is characterized that the deviation of the detection characteristic of the internal state parameter detection means and / or the deviation of the operation characteristic of the device related to the internal state are occurring (steps 135 to 137 in FIG. 15) I assume.

  According to this configuration, a plant model is created for a plurality of plants having the same configuration as the plant to be determined, and based on the Mahalanobis distance of the data of the predetermined number of internal state parameters used at that time, As a threshold value, a threshold value for determining the initial state of the plant is set. In addition, when the Mahalanobis distance during operation is calculated during the initial operation of the plant, and when the Mahalanobis distance during operation is larger than the threshold for the initial state determination, the internal state parameter detection means It is determined that a shift in detection characteristics and / or a shift in operation characteristics of a device related to an internal state has occurred.

  As a result, during initial operation such as test operation in the shipping factory or break-in operation immediately after shipment, deviations in detection characteristics of the internal state parameter detection means in the new plant and internal states not attributable to abnormality or deterioration. Deviations in the operating characteristics of the associated device can be properly determined.

  The invention according to claim 9 is the state determination device for a plant according to any one of claims 1 to 8, wherein the plant is the internal combustion engine 3, and the internal state parameter of the plant is the rotational speed of the internal combustion engine 3 (engine rotation speed Number NE), intake pressure Pin, fuel injection amount Q and EGR valve opening degree φegr, and the output parameter of the plant is an exhaust parameter (exhaust temperature Tex) representing the physical quantity of the exhaust of the internal combustion engine 3 .

  According to this configuration, the plant is an internal combustion engine, and the plant model representing the correlation between the internal combustion engine and the exhaust parameter with the number of revolutions of the internal combustion engine, intake pressure, fuel injection amount and EGR valve opening as internal state parameters is efficient. Well created with high accuracy. Also, based on the relationship between the data of the internal state parameter detected during operation of the internal combustion engine and the threshold value set based on the data of the internal state parameter detected at the time of creation of the plant model, The state can be properly determined.

BRIEF DESCRIPTION OF THE DRAWINGS It is a figure which shows typically the structure of the internal combustion engine as a plant and its state determination apparatus to which this invention is applied. It is a flow chart which shows reference model creation processing. It is explanatory drawing of Mahalanobis distance. It is the figure which compared the error by the sampling method using a Mahalanobis distance, and the random sampling method. It is a flowchart which shows a learning model creation process. It is a flowchart which shows the threshold value setting process by 1st Embodiment of this invention. It is a flowchart which shows the abnormality determination processing of the sensor by 1st Embodiment. It is a figure for demonstrating the relationship of the Mahalanobis distance at the time of model preparation, and the threshold value set by the process of FIG. It is a flowchart which shows the threshold value setting process by 2nd Embodiment of this invention. It is a flowchart which shows the deterioration determination processing of the sensor by 2nd Embodiment. It is a figure which shows the threshold value map set by the process of FIG. It is a flowchart which shows the threshold value setting process by 3rd Embodiment of this invention. It is a flowchart which shows the driving | operation area | region determination processing of the engine by 3rd Embodiment. It is a flowchart which shows the threshold value setting process by 4th Embodiment of this invention. It is a flow chart which shows initial state judging processing of an engine by a 4th embodiment.

  Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings. The present embodiment is an application of the present invention to an internal combustion engine as a plant. As shown in FIG. 1, the state determination device 1 includes an ECU (Electronic Control Unit) 2, and the ECU 2 executes various control processes described later. In addition, before the factory shipment of the vehicle, when creating a reference model which is a plant model for reference, the modeling device 50 is further used.

  An internal combustion engine (hereinafter referred to as "engine") 3 is a multi-cylinder internal combustion engine that uses gasoline as a fuel, and is mounted on a vehicle (not shown) as a motive power source. The engine 3 is provided with a fuel injection valve 4 and a spark plug 5 (only one of which is shown) provided for each cylinder, a throttle valve mechanism 8 and an EGR device 9 and the like.

  The fuel injection valve 4 and the ignition plug 5 are electrically connected to the ECU 2, and the ECU 2 causes the fuel injection amount Q and fuel injection timing by the fuel injection valve 4 to correspond to the operation state of the engine 3. The ignition timing of the mixture is controlled.

  The throttle valve mechanism 8 includes a throttle valve 8a and a TH actuator 8b for driving the same. The throttle valve 8 a is rotatably provided in the intake passage 6. The TH actuator 8b is driven by a control signal from the ECU 2 to change the opening degree of the throttle valve 8a, whereby the amount of air passing through the throttle valve 8a is controlled.

  Further, the EGR device 9 is configured to recirculate a part of the exhaust gas discharged from the cylinders of the engine 3 from the exhaust passage 7 to the intake passage 6, and is constituted by the EGR passage 9a, the EGR valve 9b, the EGR cooler 9c, and the like. There is. One end of the EGR passage 9 a is connected to the exhaust passage 7, and the other end is connected to the intake passage 6 downstream of the throttle valve 8 a.

  The EGR valve 9 b is of a butterfly valve type and is connected to an EGR valve actuator (not shown). The EGR valve actuator is driven by the control signal from the ECU 2 to change the opening degree of the EGR valve 9 b, whereby the amount of recirculated gas (EGR amount) from the exhaust passage 7 to the intake passage 6 is controlled.

  Further, the EGR cooler 9c is a water-cooling type disposed on the side of the exhaust passage 7 with respect to the EGR valve 9b of the EGR passage 9a, and cools the high temperature reflux gas flowing through the EGR passage 9a using engine cooling water. Do.

A crank angle sensor 20, a water temperature sensor 21, an air flow sensor 22, an intake pressure sensor 23, an EGR pressure sensor 24, an EGR valve opening degree sensor 25, and an accelerator opening degree sensor 26 are electrically connected to the ECU 2. A detection signal is input.

  The crank angle sensor 20 outputs a CRK signal, which is a pulse signal, as the crankshaft (not shown) rotates. The CRK signal is output for each predetermined crank angle (for example, 30 °), and the ECU 2 calculates the rotational speed (hereinafter referred to as "the engine rotational speed") NE of the engine 3 based on the CRK signal.

  The water temperature sensor 21 detects an engine water temperature TW, which is the temperature of cooling water circulating in a cylinder block (not shown) of the engine 3, as an absolute temperature. The air flow sensor 22 detects the amount of air Gth passing through the throttle valve 8a as a mass flow rate. The intake pressure sensor 23 detects a pressure (hereinafter referred to as “intake pressure”) Pin in an intake manifold (not shown) of the intake passage 6 as an absolute pressure.

  The EGR pressure sensor 24 detects, as an absolute pressure, a pressure (hereinafter referred to as “EGR pressure”) Pegr of the reflux gas flowing into the EGR valve 9 b from the EGR cooler 9 c. The EGR valve opening degree sensor 25 detects an opening degree of the EGR valve 9 b (hereinafter referred to as “EGR valve opening degree”) φegr. Further, the accelerator opening sensor 26 detects a depression amount (accelerator opening) AP of an accelerator pedal (not shown) of the vehicle.

  The ECU 2 is configured by a microcomputer including a CPU, a RAM, an E2PROM, a ROM, an I / O interface (all not shown) and the like, and will be described later according to detection signals of the various sensors 20 to 26 described above. Execute various control processes. In the present embodiment, the ECU 2 corresponds to threshold value setting means, state determination means, and driving-time Mahalanobis distance calculation means.

  The modeling device 50 is configured of a microcomputer having a CPU, a ROM, a RAM, and the like, and various electric circuits and the like. In the present embodiment, the modeling device 50 corresponds to a plant model creation unit. As described above, the modeling device 50 is connected to the ECU 2 only when creating the reference model, and the exhaust temperature sensor 51 is connected to the modeling device 50.

  The exhaust temperature sensor 51 is attached to an exhaust manifold (not shown) of the exhaust passage 7, detects an exhaust temperature (hereinafter referred to as "exhaust temperature") Tex, and outputs a detection signal to the modeling device 50. The modeling device 50 and the exhaust temperature sensor 51 are respectively removed from the ECU 2 and the engine 3 when the vehicle is shipped from the factory.

  The modeling device 50 executes the reference model creation process shown in FIG. 2 while exchanging various electrical signals with the ECU 2 when the engine 3 is controlled by the ECU 2 in various operation modes.

  In this reference model creation processing, the engine rotational speed NE, the intake pressure Pin, the fuel injection amount Q, and the EGR valve opening degree φegr, which are operation parameters (internal state parameters) of the engine 3, are input variables, and the exhaust temperature is an exhaust parameter. A reference model, which is a plant model for reference having Tex (output parameter) as an output variable, is created and executed at a predetermined cycle. In the following description, “RAM” means the RAM of the modeling device 50.

  As shown in the figure, in this process, first, data of five parameters (NE, Pin, Q, φegr, Tex) are sampled in step 1 (shown as “S1”. The same applies to the following.) The four input variables (NE, Pin, Q, φegr) and one output variable (Tex) are stored in RAM in the state of maintaining their correspondence, that is, in the state of 1-row 5-column matrix data consisting of 5 elements. Do.

  Next, the process proceeds to step 2 and the first sampling number N1 is incremented. The first number of times of sampling N1 represents the number of times of execution of sampling in step 1, and its initial value is set to zero.

  Next, in step 3, it is determined whether or not the first number of times of sampling N1 has become equal to or greater than a predetermined value K1 (for example, a 4-digit integer). When the result of this determination is NO, this processing ends. On the other hand, if the determination result in step 3 is YES, and the number of samplings in step 1 reaches the predetermined value K1, the process proceeds to step 4 to reset the first number of samplings N1.

Then, the process proceeds to step 5 to perform sampling of model data. Specifically, for four input variables (NE, Pin, Q, φegr) in K1 data sets (K1 rows and 5 columns of matrix data) stored in the RAM, the following equations (1) to (3) Then, the Mahalanobis distance DMA is calculated for each data.

X of formula (1) is a matrix of the input variable defined by equation (2), each of the four elements x ₁ ~x ₄ of the matrix x, NE, Pin, Q, corresponds to Faiegr. Further, X_ave of formula (1) is a matrix of the mean value of the input variable defined by the equation (3), each of the four elements x ₁ _ave~x ₄ _ave of this matrix X_ave is NE, Pin, Q , Φ egr corresponds to the average value of the sampling data.

  After the K1 Mahalanobis distances DMA are calculated as described above, M1 (for example, 2-digit integer) input variable data and the corresponding output variables are calculated from the K1 data with a probability proportional to the Mahalanobis distances DMA. The matrix data whose elements are data is sampled as model data and stored in the RAM. As a result, a total of M1 five-element one-row matrix data (M1 × 5 matrix data) is stored in the RAM. As shown in FIG. 3, this sampling process corresponds to intensively sampling data D2 to D4 having a large Mahalanobis distance DMA with respect to data D1 of the central point.

  The reason for performing sampling using the Mahalanobis distance DMA as described above is as follows. That is, as shown in FIG. 4, when the Mahalanobis distance DMA is used, the error V_RMSE is reduced and the sampling accuracy is improved under the condition that the number of data samples is small compared to the case where the normal random sampling method is used. It is because it can be done. Note that this error V_RMSE is the root mean square error in cross validation.

  As described above, after the model data sampling is performed in step 5, the process proceeds to step 6, and the second sampling number N2 is incremented. The second sampling number N2 represents the number of executions of sampling in step 5 above, and its initial value is set to zero.

  Next, in step 7, it is determined whether or not the second sampling number N2 has become equal to or more than a predetermined value K2 (for example, a 3-digit integer). When the result of this determination is NO, this processing ends. On the other hand, if the determination result in step 7 is YES, the number of samplings in step 5 reaches a predetermined value K2, and a total of K2 × M1 five-row one-row matrix data (matrix data of K2 × M1 × 5) When the sampling is completed, the process proceeds to step 8, where two model parameters λ and σ are determined, and the process ends.

  These model parameters λ and σ are model parameters in regression model equations (4) and (5) for calculating the estimated value Tex_est of exhaust gas described later, and are determined by the following method. First, while calculating the error V_RMSE in matrix data having four input variables and one output variable sampled as described above as elements, this error V_RMSE is used as an output, and a function using parameters λ and σ as an input is defined. Do. Then, two model parameters λ and σ are determined by searching for the minimum value of this function using the Nelder-Mead method (the downhill simplex method).

  By the above-described process, a plant model representing the correlation between four input variables (NE, Pin, Q, φegr) and one output variable (Tex) is created. The created plant model is written in the ROM of the ECU 2 as a reference model, and is referred to when the ECU 2 creates a learning model after factory shipment of the vehicle as described later.

  Next, the learning model creation processing executed by the ECU 2 will be described with reference to FIG. This process is performed in a predetermined cycle (for example, several tens of msec) while the driver drives the vehicle after the vehicle is shipped from the factory.

  In this process, first, at step 20, data of four input variables (NE, Pin, Q, φegr) are sampled and stored in the RAM. Next, the process proceeds to step 21 and the third sampling number N3 is incremented. The third sampling number N3 represents the number of executions of sampling in step 20, and its initial value is set to zero.

  Next, in step 22, it is determined whether the third sampling number N3 has become equal to or greater than a predetermined value K3 (for example, a 3-digit integer). When the result of this determination is NO, this processing ends. On the other hand, if the determination result in step 22 is YES, and the number of samplings in step 20 reaches the predetermined value K3, the process proceeds to step 23, and the third number of samplings N3 is reset.

  Next, the process proceeds to step 24, where sampling of learning data is performed. Specifically, for the K3 sets of data (matrix data of K3 rows and 4 columns) consisting of four input variables (NE, Pin, Q, φegr) stored in the RAM, the above formulas (1) to (3) Matrix data with M2 (for example, 2-digit integer) four input variable data as elements from K3 sets of data after calculating K3 Mahalanobis distances DMA by (M2 × 4 matrix data) is sampled as learning data.

  Next, in step 25, data of the learning model is calculated. Specifically, for each set of matrixes of the M2 sets of matrix data sampled in step 24, a learning value Tex_lrn of the exhaust temperature is calculated by referring to a reference model in the ROM. In this case, an interpolation operation method is used. As a result, M2 sets of one-row matrix data (matrix data of M2 rows and 5 columns) having five parameters (NE, Pin, Q, φegr, Tex_lrn) as elements are calculated as data of a learning model.

  Next, the process proceeds to a step 26, wherein the M2 sets of data calculated as described above are written in the E2PROM as part of data of a learning model, followed by terminating the present process. In this case, M2 sets of data are overwritten on data of the oldest learning model of M2 sets. When the data of the learning model is updated as described above, the two model parameters λ and σ described above are calculated again based on the updated data.

The estimated value Tex_est of the exhaust gas temperature is calculated by the following equations (4) and (5) using the plant model created and learned as described above.

  Equation (4) is a regression model equation derived from a kernel ridge regression method using a Gaussian kernel function as a kernel function, and x on the right side represents four input variables (NE, Pin) actually sampled at this control timing. , Q, φegr), and x (j) on the right-hand side is data of the input variable closest to x sampled at this control timing in the learning model described above. It is a matrix.

Furthermore, α in equation (4) is a coefficient vector defined by equation (5), and is derived such that the residual sum of squares is minimized. In the equation (5), K is a matrix having K _ij = k (x (j), x (i)) as the (i, j) component, and represents the similarity between the sampling data. Also, I _n of formula (5) is an n-order unit matrix, y is the learning value Tex_lrn exhaust temperature corresponding to x (j).

  The estimated value Tex_est of the exhaust temperature calculated as described above is used for control of the engine 3, and is used, for example, to calculate a target EGR valve opening that is a target value of the EGR valve opening φegr in EGR control. .

  As described above, according to the present embodiment, four operation parameters (NE, Pin, Q, φegr) representing the internal state of the engine 3 are detected, and using a large number of detected operation parameters, A plant model (reference model) is created that represents the correlation between the operating parameters of and the exhaust temperature Tex that is the output of the engine 3. At this time, the Mahalanobis distance DMA is calculated for each of the K1 operation parameter data, and the M1 data of the larger one of the calculated Mahalanobis distance DMA is selected and used to create a plant model. As a result, a plant model with reduced modeling error is efficiently created with high accuracy under a condition where the number of data is small compared to the case where all data including data whose distance from the central point is small is used. be able to.

  Next, various determination processes for determining the state of the engine 3 will be described. In this state determination, when an error or deviation occurs in the operation parameter which is an input variable of the reference model for calculating the estimated value Tex_est of the exhaust temperature described above, the correlation between the operation parameter decreases, and the Mahalanobis distance is accompanied accordingly It focuses on the increase of DMA. Specifically, a number of Mahalanobis distances (hereinafter referred to as "Mahalanobis distance DMAM at model creation") calculated for four operation parameters (NE, Pin, Q, φegr) used as input variables at the time of creation of the reference model With reference to FIG. 8), the threshold value is set based on this, and the driving parameter Mahalanobis distance (hereinafter referred to as “driving Mahalanobis distance”) DMA obtained during operation of the engine 3 is By comparing with the threshold value, the state of the engine 3 including abnormality or deterioration of the sensor that causes an error or deviation of the input variable is determined.

  Among these determination processes, FIG. 6 and FIG. 7 show the abnormality determination process of the sensor according to the first embodiment. This determination processing is to determine an abnormality of a sensor that constitutes a detection system of the four operation parameters (NE, Pin, Q, φegr) described above.

  First, in the threshold setting process of FIG. 6, a threshold D_ABN for abnormality determination is set (step 40). This process is performed once after the creation of the reference model by the process of FIG. 2 is completed. The threshold D_ABN defines the lower limit of the Mahalanobis distance DMA considered to be abnormal with respect to the above-mentioned Mahalanobis distance DMAM at model creation, and is based on the maximum value D_MAX of the Mahalanobis distance DMAM at model creation. It is set to a larger value.

  Such setting of the threshold D_ABN is, as shown in FIG. 8, an equal Mahalanobis distance line passing the threshold D_ABN (hereinafter referred to as "threshold line") outside the equal Mahalanobis distance line passing the maximum value D_MAX. It corresponds to setting X, and the smaller the value of the threshold D_ABN, the more severe the abnormality determination. In addition, the threshold D_ABN is set in consideration of legal restrictions on exhaust gas characteristics (emissions), the use environment of the vehicle, and the like, and is set to a smaller value, for example, in response to intensification of legal restrictions.

  The abnormality determination process of FIG. 7 is performed during normal operation of the engine 3. In this process, first, in step 50, data of the same four operation parameters (NE, Pin, Q, φegr) as the input variables at the time of reference model creation are sampled, and in the state of matrix data of 1 row 4 columns Store in RAM.

  Next, at step 51, the number of samplings N4 is incremented, and at step 52, it is determined whether or not the number of samplings N4 has reached a predetermined value K4 (for example, a two digit integer). If the answer to this question is negative (NO), the processing returns to steps 50 and 51 to repeat the sampling of data and the increment of the number N4 of times of sampling.

  If the number of times of sampling N4 reaches the predetermined value K4 (step 52: YES), the process proceeds to step 53, and K4 sets of data (K4 rows and 4 columns of matrix data having four operation parameters stored in the RAM as elements) Mahalanobis distances DMA (1) to DMA (K4) during operation of each data are calculated according to the equations (1) to (3).

  Next, in step 54, after setting the index number i instructing the driving Mahalanobis distance DMA to 1, the process proceeds to step 55, and it is determined whether the driving Mahalanobis distance DMA (i) is larger than the threshold D_ABN. To determine If the result of this determination is YES, and the driving Mahalanobis distance DMA (i) is larger than the threshold D_ABN (when located outside the threshold line X as in point A in FIG. 8), the data is In step 56, after the number of abnormal data N_ABN is incremented, the process proceeds to step 57. On the other hand, when the result of the determination in step 55 is NO, the process directly proceeds to step 57.

  In step 57, it is determined whether the index number i has reached the predetermined value K4. If the determination result is NO, the index number i is incremented in step 58, and then the process returns to the steps 55 and 56, The comparison between the driving Mahalanobis distance DMA (i) and the threshold value D_ABN and the incrementing of the number of abnormal data N_ABN according to the comparison result are repeated.

  If the comparison with the threshold value D_ABN is completed for all the K4 operating Mahalanobis distances DMA (step 57: YES), the process proceeds to step 59 and it is determined whether the number of abnormal data N_ABN is equal to or greater than a predetermined value NREF1. To determine

  When the determination result in step 59 is YES, the number of abnormal data for which the Mahalanobis distance DMA exceeds the threshold D_ABN is large during operation, and the cause is the detection system of the above four operation parameters (NE, Pin, Q, φegr) Is determined to be due to an abnormality in the sensors (crank angle sensor 20, intake pressure sensor 23, EGR valve opening degree sensor 25) constituting the sensor, and after the abnormality flag F_ABN is set to "1" in step 60, this processing is ended. . On the other hand, when the determination result in step 59 is NO, it is determined that no abnormality occurs in the above-described sensor, and the abnormality flag F_ABN is set to “0” in step 61, followed by terminating the present process.

  As described above, when the abnormality flag F_ABN is set to “1”, for example, a warning light notifying that the sensor is abnormal is turned on to prompt identification or replacement of the abnormal sensor.

  Alternatively, in this process, an abnormal sensor may be identified. For example, since a plurality of data indicating an abnormality are already grasped by the above-described process, four operation parameters of each data are compared with four operation parameters (input variables) used for creating the reference model. Compare. As a result, if an operating parameter indicating an abnormal value is specified, it can be determined that the sensor that detects the operating parameter is abnormal.

  As described above, according to the present embodiment, the threshold value D_ABN for abnormality determination is set to a value larger than the maximum value D_MAX of the Mahalanobis distance DMAM at the time of model creation, and is detected during operation of the engine 3 Mahalanobis distances DMA (1) to DMA (K4) during operation of each data are calculated for K4 sets of data consisting of four operation parameters. Then, when the calculated number of abnormal data N_ABN for which each driving Mahalanobis distance DMA (i) exceeds the threshold value D_ABN is equal to or more than a predetermined value NREF1, an error occurs in the sensors constituting the detection system of the operating parameter. Since it is determined that there is an abnormality, it is possible to appropriately determine the abnormality of the sensor.

  9 and 10 show a sensor deterioration determination process according to the second embodiment of the present invention. This determination process is for determining the deterioration over time of the sensor that constitutes the above-described operating parameter detection system.

  First, in the threshold setting process of FIG. 9, a threshold D_DET for deterioration determination is set (step 70). This process is performed once after the creation of the reference model is completed. The threshold value D_DET in this case defines the upper limit of the driving Mahalanobis distance DMA, which increases with the deviation of the detection value due to the secular deterioration of the sensor, and as shown in FIG. It is set in the form of a threshold map using DV as a parameter. More specifically, the threshold value D_DET is set to the average value of the Mahalanobis distance DMAM at the time of model creation, with the initial value D_INI when the vehicle travel distance DV = 0, and based on the experimental results etc. Is set to gradually increase as

  The deterioration determination process of FIG. 10 is performed during normal operation of the engine 3. Steps 80 to 82 of this process are basically the same as steps 50 to 52 in FIG. 7. In step 80, data of the above four operation parameters (NE, Pin, Q, .phi. Egr) are sampled, and the ECU 2 Store in RAM. This data sampling is repeated until the number of times of sampling N5 reaches a predetermined value K5 (for example, a two-digit integer) (steps 81 and 82).

  Next, the process proceeds to step 83, and for each of the K5 sets of data (matrix data of K5 rows and 4 columns) having four operation parameters stored in the RAM, each data according to the equations (1) to (3). The driving Mahalanobis distances DMA (1) to DMA (K5) are calculated, and at step 84, the average value DMA_AVE is calculated. Further, in step 85, the threshold value D_DET is calculated by searching the threshold value map of FIG. 11 according to the vehicle travel distance DV at that time.

  Next, at step 86, it is judged if the average value DMA_AVE is larger than the threshold value D_DET. If the result of this determination is YES, and the average value DMA_AVE of the Mahalanobis distance during operation is larger than the threshold D_DET (for example, point B in FIG. 11), the detection system of four operation parameters (NE, Pin, Q, φegr) It is determined that excessive deterioration is occurring in the sensors (crank angle sensor 20, intake pressure sensor 23, EGR valve opening sensor 25) to be configured, and in step 87, the deterioration flag F_DET is set to "1". Next, in step 89 described later, the amount of deviation of the sensor detection value due to deterioration is learned, and the present process is ended.

  On the other hand, when the determination result in step 86 is NO, it is determined that the above sensor does not have excessive deterioration, and in step 88, the deterioration flag F_DET is set to “0”, and the present process is ended.

  The calculation of the amount of deviation of the detection value of the sensor in step 89 is performed as follows, for example. For example, in the case of the EGR valve opening degree sensor 25, the EGR valve 9b is forcibly controlled to the fully closed opening degree, and in that state, the detection value detected by the EGR valve opening degree sensor 25 is It is learned as the amount of deviation of the detected value (0 point learning). Alternatively, in the case of the intake pressure sensor 23, it is assumed that the atmospheric pressure sensor is used in combination, and it is detected under conditions where the intake pressure is assumed to be equal to the atmospheric pressure, for example, under conditions where the vehicle and the engine 3 are stopped. The difference between the detected values of the atmospheric pressure sensor and the intake pressure sensor 23 is learned as the amount of deviation of the detected value of the intake pressure sensor 23.

  As described above, according to the present embodiment, the threshold value D_DET for deterioration determination is set based on the average value of the Mahalanobis distance DMAM at the time of model creation and the vehicle travel distance DV, and the average of the Mahalanobis distance DMA during driving When the value DMA_AVE exceeds the threshold value D_DET, it is determined that the sensor that detects the operation parameter is experiencing excessive deterioration. Thereby, the deterioration determination of the sensor can be appropriately performed. In addition, when it is determined that excessive deterioration has occurred, the amount of deviation of the sensor detection value due to the deterioration can be learned at an appropriate timing.

  In the above example, learning of the deviation of the sensor detection value is performed on the condition that it is determined that excessive deterioration is occurring, but it is performed regardless of the success or failure of the condition. It is also good.

  12 and 13 show an operation area determination process of the engine 3 according to the third embodiment of the present invention. This determination process is for determining whether or not the operating area of the engine 3 deviates from the operating area (hereinafter referred to as a “learning area”) represented by the operating parameters obtained at the time of creation of the reference model. First, in the threshold setting process of FIG. 12, a threshold D_OPRN for area determination is set (step 90). This process is performed, for example, once after creation of the reference model is completed. The threshold D_OPRN defines the outer edge of the learning region at the time of creation of the reference model, and is set to, for example, the maximum value D_MAX of the Mahalanobis distance DMAM at the time of model creation.

  The operating range determination process of FIG. 13 is performed during normal operation of the engine 3. Steps 100 to 102 of the present process are basically the same as steps 50 to 52 in FIG. 7. In step 100, data of the four operation parameters (NE, Pin, Q, φegr) are sampled, and the ECU 2 is executed. Store in the RAM of. This data sampling is repeated until the number of times of sampling N6 reaches a predetermined value K6 (for example, a two-digit integer) (steps 101 and 102).

  Next, the process proceeds to step 103, and for each of the K6 sets of data (K6 rows and 4 columns of matrix data) having four operation parameters stored in the RAM, each data is calculated according to the above equations (1) to (3). Mahalanobis distances DMA (1) to DMA (K6) are calculated during driving.

  The next steps 104 to 108 are also basically the same as steps 54 to 58 in FIG. 7, and after setting the index number i to 1 in step 104, the process proceeds to step 105 and the Mahalanobis distance DMA at each operation It is determined whether (i) is larger than the threshold D_OPRN. If the result of this determination is YES, and the driving Mahalanobis distance DMA (i) is larger than the threshold D_OPRN, it is determined that the operating region of the engine 3 represented by the data deviates from the learning region, and step 106 After incrementing the out-of-area data number N_OPRN, the process proceeds to step 107. On the other hand, when the result of the determination in step 105 is NO, the process directly proceeds to step 107.

  In this step 107, it is determined whether or not the index number i has reached the predetermined value K6. If the determination result is NO, the index number i is incremented in step 108, and then the driving is performed in the steps 105 and 106. When the Mahalanobis distance DMA (i) and the threshold value D_OPRN are compared, and the out-of-area data number N_OPRN is incremented according to the comparison result.

  When the comparison with the threshold value D_OPRN is completed for all the K6 at-operation Mahalanobis distance DMAs (step 107: YES), the operation of the engine 3 is performed for the operation parameters determined to be out of the learning region at step 109. The state is learned, and this processing ends. The learning of the driving state is performed by, for example, considering the number N_OPRN of out-of-region data and analyzing data determined to be out of the learning region (hereinafter referred to as “out-of-region data”). More specifically, while picking up data indicating the same or similar operation parameters from among the plurality of out-of-region data, the operation parameters are compared with the operation parameters obtained at the time of creation of the reference model, and the singular By grasping the nature, the operating state of the engine 3 is learned. By such learning, it is possible to grasp, for example, a driving condition different from a normal or careless driving condition according to the driver's characteristic driving pattern or driving environment.

  As described above, according to the present embodiment, the threshold value D_OPRN for operating region determination is set to the maximum value D_MAX of the Mahalanobis distance DMAM at the time of model creation, and the Mahalanobis distance DMA (i) at the time of operation is the threshold value. When it is larger than D_OPRN, it is determined that the operating region of the engine 3 represented by the data deviates from the learning region. Thereby, the presence or absence of the deviation of the driving | operation area | region of the engine 3 from a learning area | region can be determined appropriately. Also, by analyzing the out-of-region data, it is possible to learn the operating state of the engine 3 that has deviated from the learning region.

  14 and 15 show an initial state determination process of the engine 3 according to the fourth embodiment of the present invention. This determination process is different from the determination processes of the first to third embodiments which are executed during the normal operation of the engine 3, and for the new engine 3 before or after the factory shipment, the detection characteristics of the sensor It is determined whether or not there is a shift in the operating characteristics of the device.

  In the threshold setting process of FIG. 14, first, in step 120, with respect to each of N sample engines having the same configuration as the engine 3, the reference model is created by the process of FIG. Then, N average values DMAM_AVE (1) to DMAM_AVE (N) are calculated by obtaining an average value DMAM_AVE obtained by averaging the multiple Mahalanobis distances DMAM calculated at the time of model creation. Next, overall average value DMAM_AVE2 is calculated by further averaging these average values DMAM_AVE (1) to (N) (step 121), and set as threshold value D_INI for initial state determination (step 122). ), End this processing.

  The initial state determination process of FIG. 15 is performed during an initial operation of the engine 3 such as a test operation in a factory before shipment or a break-in operation immediately after shipment after the process of FIG. The steps 130 to 132 are basically the same as the steps 50 to 52 in FIG. 7. In the step 130, data of the four operation parameters (NE, Pin, Q, φegr) are sampled and stored in the RAM of the ECU 2. Remember. This data sampling is repeated until the number of times of sampling N7 reaches a predetermined value K7 (for example, a two-digit integer) (steps 131 and 132).

  Next, the process proceeds to step 133, and the K7 sets of data (the K7 rows and 4 columns of matrix data) stored in the RAM are stored in accordance with the equations (1) to (3). The at-operation Mahalanobis distances DMA (1) to DMA (K7) are calculated, and at step 134, the average value DMA_AVE is calculated.

  Next, at step 135, it is judged if the average value DMA_AVE of the driving Mahalanobis distance is larger than the threshold value D_INI. When the determination result is YES and the average value DMA_AVE is larger than the threshold value D_INI, the sensors (crank angle sensor 20, intake pressure sensor) constituting the detection system of the four operation parameters (NE, Pin, Q, φegr) 23, detection characteristics of the EGR valve opening degree sensor 25) or operation characteristics of devices related to four operation parameters (for example, the throttle valve mechanism 8, the fuel injection valve 4, the EGR device 9) are represented by the threshold value D_INI. In step 136, the initial shift flag F_INI is set to "1", and the process ends.

  On the other hand, when the determination result in step 135 is NO, it is determined that the deviation of the detection characteristic of the sensor or the deviation of the operation characteristic of the device does not occur, and the initial deviation flag F_INI is set to "0" in step 137. And the process ends.

  When the initial deviation flag F_INI is set to “1” as described above, for example, a display indicating the initial deviation is displayed, and accordingly, replacement of the sensor or the device is appropriately performed. Alternatively, the detected value of the sensor or the deviation of the operation amount of the device may be learned, and the detected value or the operation amount may be corrected using the learned value.

  As described above, according to the present embodiment, reference models are created for a plurality of other sample engines having the same configuration as the engine 3, and a number of models calculated at that time are used to create the Mahalanobis distance DMAM. Is set as an overall average value DMAM_AVE2 is set as a threshold D_INI for initial state determination. And, when the average value DMA_AVE of the driving Mahalanobis distance DMA calculated during the initial operation of the engine 3 is larger than the threshold D_INI, the detection characteristic of the sensor constituting the detection system of the driving parameter, and / or the driving parameter It is determined that the operation characteristic of the device related to is deviated from the characteristic of the central product. As a result, it is possible to appropriately determine the presence or absence of the deviation of the detection characteristic of the sensor in the brand-new engine 3 and the operation characteristic of the device.

  In addition, this invention can be implemented in various aspects, without being limited to the described embodiment. For example, in the embodiment, the Mahalanobis distance DMA is used as the distance from the central point of the distribution of the operating parameter (internal state parameter), but not limited to this, the distance from the central point such as Euclidean distance What is necessary is just to represent.

  In the embodiment, sampling of model data in step 5 of FIG. 2 and sampling of learning data in step 24 of FIG. 5 are performed with a probability proportional to the Mahalanobis distance DMA, but the invention is not limited to this. The sampling may be performed in order from the one with the largest DMA.

  Further, in the first to fourth embodiments, as the state of the engine 3, the abnormality of the sensor, the deterioration of the sensor, the operating region of the engine 3 and the initial state of the engine 3 are determined. The determination may be performed on other states of the engine 3 as long as the correlation of the data of the three operation parameters (internal state parameters) is affected.

  Furthermore, the setting method of the threshold shown in each of the first to fourth embodiments and the calculation method of the driving Mahalanobis distance DMA compared with the threshold are merely examples, and the configuration of the details thereof is changed. It is possible.

  In the embodiment, the four operating parameters (NE, Pin, Q, φegr) are used as the internal state parameters of the engine 3, and the exhaust temperature Tex is used as the output parameter. The operation parameter is not limited to this, and other appropriate parameters such as the ignition timing IG, the engine coolant temperature TW, the opening degree of the throttle valve 8a, and the like may be further used as long as they have correlation with the exhaust temperature Tex. Also, instead of the exhaust temperature Tex, other exhaust parameters representing the physical quantity of the exhaust, such as the exhaust pressure or the exhaust recirculation amount may be used.

  Furthermore, the present invention is not limited to the input / output system using the exhaust parameter as the output parameter as in the embodiment, but can be applied to other input / output systems. For example, in an input / output system in which an intake parameter (for example, a cylinder intake quantity and a boost pressure) representing a physical quantity of intake of the engine 3 is used as an output parameter, and a plurality of operation parameters having correlation with the intake parameter are used as internal condition parameters, The present invention may be applied. Alternatively, for example, in an automatic driving apparatus for a vehicle, since the detection values detected by a large number of sensors are generally used as input variables to obtain an output such as external world recognition, the present invention relates to such an automatic driving apparatus Can be effectively used in the input / output system of

  The embodiment is an example in which the present invention is applied using an internal combustion engine for a vehicle as a plant, but the present invention is not limited to this, and is a plant including an internal combustion engine for ships and other industrial equipment. It is possible to apply widely.

1 State determination device 2 ECU (Threshold setting means, State determination means, Mahalanobis distance calculation means during driving)
3 Internal combustion engine (plant)
4 Fuel injection valve (device)
8 Throttle valve mechanism (device)
9 EGR device (device)
20 Crank angle sensor (internal condition parameter detection means)
23 Intake pressure sensor (internal condition parameter detection means)
25 EGR valve opening sensor (internal condition parameter detection means)
50 Modeling device (plant model creation means)
NE engine speed (internal condition parameter)
Pin Intake pressure (internal condition parameter)
Q Fuel injection amount (internal condition parameter)
φegr EGR valve opening (internal condition parameter)
DMA Mahalanobis Distance (During Mahalanobis Distance)
Mahalanobis distance at the time of DMAM model creation Tex exhaust temperature (output parameter, exhaust parameter)
Threshold for abnormality determination of D_ABN sensor Threshold for deterioration determination of D_DET sensor
DV Vehicle travel distance (plant operation period)
D_OPRN Threshold value for driving range judgment D_INI Threshold value for initial state judgment

Claims (9)

Internal condition parameter detection means for detecting an internal condition parameter representing an internal condition of a plant;
An output parameter representing the internal condition parameter and the output of the plant using a predetermined number of data with a larger distance from the center point of the distribution of the data among the data of the detected multiple internal condition parameters Plant model creation means for creating a plant model representing the correlation with
Threshold setting means for setting a threshold for determining the state of the plant based on the data of the predetermined number of internal state parameters;
The plant based on the relationship between data of a large number of the internal state parameters detected by the internal state parameter detection means during operation of the plant after the plant model is created, and the set threshold value. State determination means for determining the state of
A plant state determination apparatus comprising:   The distance from the central point of the distribution of the data is a Mahalanobis distance, and the threshold setting means sets the threshold based on the Mahalanobis distance of the data of the predetermined number of internal state parameters. The plant state determination device according to claim 1, characterized in that The system further comprises an operating Mahalanobis distance calculation means for calculating the Mahalanobis distance of data of a number of internal state parameters detected during operation of the plant as the operating Mahalanobis distance,
The plant state determination device according to claim 2, wherein the state determination unit determines the state of the plant by comparing the calculated Mahalanobis distance at the time of operation with the threshold value. The threshold setting means sets a threshold for abnormality determination as the threshold.
4. The apparatus according to claim 3, wherein the state determination means determines that the internal state parameter detection means is abnormal when the driving Mahalanobis distance is larger than the threshold value for abnormality determination. Plant status determination device. The threshold setting means sets a threshold for deterioration determination as the threshold according to the operation period of the plant,
4. The apparatus according to claim 3, wherein said state determination means determines that said internal state parameter detection means is deteriorated when said driving Mahalanobis distance is larger than said threshold value for deterioration judgment. The state determination apparatus of the described plant.   6. The apparatus according to claim 5, wherein, when it is determined that the internal state parameter detection means is deteriorated, the state determination means learns a deviation of detection characteristics of the internal state parameter detection means due to the deterioration. The state determination apparatus of the described plant. The threshold setting means sets a threshold for operating region determination as the threshold.
The state determination means, when the operating Mahalanobis distance of the data of the internal state parameter is larger than the threshold value for the operating region determination, the operating region of the plant represented by the data is the plant model 4. The plant state determination device according to claim 3, wherein it is determined that the vehicle deviates from the operation region learned at the time of creation of. The plant model creation means creates the plant model for a plurality of other plants having the same configuration as the plant,
The threshold setting unit is configured to determine an initial state of the plant as the threshold based on a Mahalanobis distance of data of the predetermined number of internal state parameters used to create a plant model of the plurality of plants. Set the threshold of
The at-operation Mahalanobis distance calculation means calculates the at-operation Mahalanobis distance during initial operation of the plant;
When the operating Mahalanobis distance is greater than the threshold value for determining the initial state, the state determining means may detect the deviation of the detection characteristic of the internal state parameter detecting means or the internal state in the initial state of the plant. The plant state determination apparatus according to claim 3, characterized in that it is determined that a deviation of operation characteristics of a related device has occurred. The plant is an internal combustion engine,
The internal state parameters of the plant are the number of revolutions of the internal combustion engine, the intake pressure, the fuel injection amount, and the EGR valve opening degree.
The plant state determination device according to any one of claims 1 to 8, wherein the output parameter of the plant is an exhaust parameter representing a physical quantity of exhaust of the internal combustion engine.

Images