JPH0916255A - Device failure diagnosis method and device - Google Patents

Abstract

(57) [Summary] [Purpose] To automate the failure diagnosis of equipment and to improve the detection accuracy of failure points. [Structure] A data collection unit 5 for transmitting data from an engine control unit to an off-board fault diagnosis device by serial communication, a statistical processing unit 6, and a general-purpose logic unit 7.
The comparison / determination unit 8, the special logic unit 9, the self-tuning unit 10, and the diagnosis result display unit 11 are provided. Then, based on the collected data, first, a normal logic is used to perform a normal fault code check, the water temperature and the power supply voltage are also checked, and then a fault diagnosis condition is executed using a general-purpose logic. Also, as background processing, self-tuning of the standard value table A is performed using the data of the past failure database. Then, in the general-purpose logic, the probability of each feature amount is calculated for each failure phenomenon using the standard value table A and the map B to obtain the overall probability, the overall probabilities are compared, and the failure phenomenon with the large overall probability is displayed. To do.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a failure diagnosis of equipment capable of transmitting input / output data of a control system to a tester or the like by serial communication.

[0002]

2. Description of the Related Art In connection with failure diagnosis of an automobile engine, various input / output data is output from an engine control unit to a tester through serial communication, and the tester displays characteristic quantities relating to engine control conditions such as engine speed and filling efficiency. Methods have been developed to extract and display.
According to this method, the service person can judge the failure location of the engine by looking at the display of the tester. However, the judgment was based solely on the intuition of a service person and did not immediately lead to automation of failure diagnosis.

As a technique for automating the failure diagnosis, for example, the technique described in Japanese Patent Laid-Open No. 3-134951 has been conventionally known. The technology described in this publication is to extract a feature amount from input / output data of a control system, determine an abnormality by comparing with a preset normal distribution range of the feature amount, and output a determination signal. is there.

[0004]

The development has been advanced to the point of extracting the feature quantity of the engine control state from the input / output data of the engine control system and utilizing it for the failure diagnosis.
Furthermore, there is a need to surpass technological development to take it one step further and realize automation of failure diagnosis. However, with the conventional methods, even if the failure can be detected, it is difficult to specify the failure content and the failure location except for a part that is simple. Also in the case of the technique described in Japanese Patent Laid-Open No. 3-134951, it is not possible to accurately detect a fault having a complicated pattern simply by comparing with the data distribution in the normal state.

An object of the present invention is to automate the failure diagnosis of equipment and to detect the failure location with high accuracy.

[0006]

SUMMARY OF THE INVENTION The present invention is to automate the failure diagnosis of equipment based on the input / output data of the equipment control system, and its method is to determine the equipment control status corresponding to each failure phenomenon of equipment. A standard value table that stores the standard value for each failure phenomenon of a plurality of characteristic quantities is created in advance, and the plurality of characteristic quantities in the control state are extracted by collecting and processing the input / output data in the control state of the device, The failure phenomenon is determined based on the extracted values in the control state of the feature quantities and the standard value of the standard value table for each failure phenomenon, and the failure location corresponding to the failure phenomenon is specified. Here, the standard value is, for example, a standard average value and a standard deviation of each of a plurality of feature amounts for each failure phenomenon, and the standard value for at least one of the plurality of feature amounts is different between each failure phenomenon. I shall.

In the failure diagnosing method of the present invention, the failure phenomenon is determined by determining the probability that a plurality of feature values apply to the extracted value in the control state when a specific failure phenomenon occurs as the extracted value for each failure phenomenon. It is preferable to calculate from the standard value, compare the calculated values of the probabilities for each failure phenomenon, and to judge the failure that has occurred with a failure phenomenon with a large calculated value. Also, the calculation of the probability in that case is It is preferable that a plurality of feature quantities be individually calculated for each failure phenomenon, and the overall probability be calculated from the probability value of each feature quantity. Furthermore, the calculation of the probability for each feature amount is preferably performed using a standardized map.

Further, in the failure diagnosis method of the present invention,
In the process of calculating the individual probabilities of a plurality of feature quantities for a particular failure phenomenon, if the calculated probability value of the particular feature quantity is less than or equal to the threshold value, the subsequent calculation for the failure phenomenon is omitted. , It is better to shift to calculation for other failure phenomena. If the overall probability calculated for a particular failure phenomenon is greater than or equal to the threshold value, it is determined that the failure actually occurred due to that failure phenomenon, and the calculations for other failure phenomena after that are omitted. it can.

Further, in the failure diagnosis method of the present invention,
If the abnormality of a specific sensor value is detected as a pre-check, and as a result, the abnormality of the sensor value is detected and the content of the failure is clarified only by that, the subsequent failure diagnosis procedure can be omitted.

In the fault diagnosing method of the present invention, it is preferable to display the diagnosis result so as to prompt the service person to check. In that case, it is complicated if the number of displayed information is unnecessarily large. It is preferable to select a failure phenomenon having a large calculated value of the probability of and to display the data as a diagnosis result.

In the fault diagnosing method of the present invention, in order to prevent erroneous diagnosis, it is preferable to exclude the data of the fault phenomenon in which the calculated value of the overall probability is less than or equal to the threshold value from the display of the diagnostic result. Also, of the calculated value of the overall probability,
For a failure phenomenon that shows a calculated value of a predetermined ratio or less with respect to the maximum calculated value, it is preferable to exclude the data from the display of the diagnostic result.

In the failure diagnosis method of the present invention,
It is possible to improve the diagnostic accuracy by feeding back the diagnostic result. Therefore, it is better to register the diagnostic result in the past failure database. Then, the past failure database is aggregated to obtain a standard value for each failure phenomenon of a plurality of feature quantities, a temporary standard value table storing the standard values is created, and the temporary standard value table is used. Performs a failure diagnosis for each data in the past failure database, compares the diagnosis result with the diagnosis result by the standard value table up to that point, and performs self-tuning with the higher accuracy rate of failure diagnosis as the new standard value table. Is good.

Further, in order to shorten the time required for the diagnosis, the failure diagnosis method of the present invention divides the failure phenomena into a plurality of groups in advance according to the degree of association with a specific sensor value, and extracts the feature quantity related to the sensor value. You may make it select any group based on a value, and perform a failure diagnosis about the failure phenomenon of the selected group.

Further, the apparatus used for the failure diagnosis of the present invention has a data collecting section for collecting the input / output data of the equipment control system, and a plurality of characteristics for statistically processing the collected input / output data to show the equipment control state. A statistical processing unit that extracts a quantity, a probability calculation unit that calculates the probability that a plurality of feature values will apply to the extracted value of the feature value in the control state when a specific failure phenomenon occurs, and a probability calculation for each failure phenomenon It is composed of a comparison / determination unit that compares the results and makes a failure determination, and a display unit that displays the results of the failure determination.

[0015]

In the standard value table, a plurality of feature quantities (for example, in the case of an engine, engine speed, engine speed fluctuation, filling efficiency, air-fuel ratio feedback value, etc.) correspond to each failure phenomenon in the standard value table. Then, the standard value (for example, the average value and the standard deviation) of the feature quantities is stored for each failure phenomenon by obtaining the value obtained by an experiment etc., and the standard value is for each failure phenomenon. Depending on the phenomenon, at least one of the multiple feature quantities differs, and the failure phenomenon is determined by comparing the value of each feature quantity extracted from the input / output data in the control state of the device control system with the table value of this standard value table. Then, the failure location is specified.
Then, the determination of the failure phenomenon is performed by extracting the probability that each feature quantity is applied to the extracted values when a specific failure phenomenon occurs with respect to the extracted values of the plurality of feature quantities in the control state, for each failure phenomenon. And the standard value. Then, the calculation can be performed using the map. Further, the calculation of the probability is performed for each of a plurality of feature quantities for each failure phenomenon, and the overall probability is calculated for each failure phenomenon from the probability value for each feature quantity. The calculated values of the overall probabilities are compared, and a failure phenomenon having a large calculated value is determined as a failure.

Here, in the process of calculating the probabilities of individual feature quantities for a particular failure phenomenon, if the probability value of the particular feature quantity is less than or equal to the threshold value and the probability of the failure phenomenon is extremely low, With respect to the failure phenomenon, the subsequent calculation can be omitted, whereby the total calculation time can be shortened. In addition, when the overall probability of a particular failure phenomenon is greater than or equal to the threshold value and the possibility of this failure phenomenon is extremely high, it is determined that the failure phenomenon has actually occurred, and other failures thereafter. The calculation of the phenomenon can be omitted, and thus the total calculation time can be shortened.

Further, when a specific sensor value is detected as an abnormality and the details of the failure are clear, the subsequent failure diagnosis procedure can be omitted, thereby preventing useless calculation.

Regarding the display of the diagnosis result, by selecting and displaying a failure phenomenon having a large calculated value of the overall probability, the display is not unnecessarily complicated and the check by the service person becomes easy. .

Further, by omitting from the display, failure phenomena that are unlikely to occur, such as the calculated value of the overall probability being less than or equal to the threshold value or being less than or equal to a predetermined ratio with respect to the maximum calculated value, these data are displayed. It is possible to prevent an erroneous diagnosis that may occur due to the mixture of.

Further, by registering the diagnosis result in the past failure database and feeding it back, the diagnosis accuracy can be improved. This feedback is based on the fact that the past failure database in which the diagnostic results have been registered is aggregated to create a new provisional standard value table, and that table is used to perform a failure diagnosis for each data in the past failure database. Compared with the diagnosis result by the table,
This can be done by self-tuning with a higher standard of accuracy as a new standard value table.

In order to shorten the diagnosis time, it is effective to perform the probability calculation for each group. That is, the failure phenomenon is divided into a plurality of groups in advance according to the degree of association with a specific sensor value, one of the groups is selected based on the extracted value of the feature value related to the sensor value, and the failure phenomenon of the selected group is selected. Fault diagnosis can be performed, which can shorten the diagnosis time.

Further, in the failure diagnosis apparatus of the present invention, the data collection unit collects the input / output data of the device control system, and the statistical processing unit statistically processes the input / output data to display a plurality of device control states. The feature quantity is extracted, and when a specific failure phenomenon occurs in the probability calculation section, the probability that a plurality of feature values apply to the extracted value of the feature quantity in the control state is calculated, and in the comparison and determination section, the probability of each failure phenomenon is calculated. A failure judgment is made by comparing the calculation results, and the result of the failure judgment is displayed on the display unit.

Next, the present invention will be described with reference to FIGS. 1, 2 and 3 for the case of engine failure diagnosis. FIG. 1 is a basic block diagram, FIG. 2 is a flow chart of a diagnosis part, and FIG. 3 is a standard value table for each failure by engine type.

A basic block diagram of engine failure diagnosis is shown in FIG.
And a data collecting section which is an internal RAM information collecting means of the engine control unit, and the RAM.
A table (Table A) storing standard values (average value, standard deviation) of feature quantities related to information, and collected R
It is composed of a diagnosis unit that processes AM information to extract a characteristic amount of an engine control state and diagnoses which failure is based on a probability calculation from Table A. The operation is as shown in the flowchart of FIG. 2. First, the internal RAM information of the engine control unit is collected by the data collecting unit. Next, the diagnostic unit extracts a feature amount from the RAM information and uses Table A to stochastically calculate how much the feature amount applies to each failure. Then, when the calculation is completed for all the failures, the calculated values are written in the table A as shown in FIG.
Display (output) in descending order of calculated value.

The principle of the above-mentioned fault diagnosis logic will be described below.

FIG. 4 shows the normal distribution of each sensor value. Here, nA1, nA2, nA3, and nA4 are average values of the sensor values, and σA1, σA2, σA3, and σA4 are standard deviations of the sensor values. When the sensor values when a specific failure (fault A) occurs are A1, A2, A3, A4, the sensor values are A1, A2, A3, A4, respectively.
The probabilities of taking values far apart from each other are PA1, PA2, and
Assuming that PA4 and PA4 are independent of each other, the probability that four sensors can simultaneously take the values of (A1, A2, A3, A4) when failure A occurs: PA is , PA ≦ PA1 × PA2 × PA3 × PA4. Then, the probability of occurrence of failure A is Q
If A, a failure A occurs and the measurement point A (A1, A2,
Probability of taking A3, A4): P is P = QA × PA = QA × PA1 × PA2 × PA3 × PA4. Therefore, the four sensor values are (A1, A2, A
3, A4), the standard value (mean value,
The probability for each failure is obtained using a table (table A) in which standard deviations are set and a probability value table map in which the probability calculation is standardized, and which failure has occurred is determined according to the magnitude of the probability. Make an analogy.

A specific example of this logic will be described below for the case of engine failure diagnosis.

In the engine (V-type engine) failure diagnosis, FIG. 5 shows the characteristic values corresponding to the above sensor values such as rotation speed, charging efficiency, rotation speed fluctuation, right bank O ₂ sensor value (A / F) and Left bank side O ₂ sensor value (A / F)
The fault diagnosis logic in the five cases described above is explained. The upper row shows the value of each feature value in case of failure A (A1, A2,
A3, A4, A5), and the lower part shows the values (A1, A2, A3, A4, A5) of the respective characteristic amounts in the case of the failure B on the normal distribution. Here, when the extracted value of each feature amount is (A1, A2, A3, A4, A5), each feature amount is
1, A2, A3, A4, A5)
Is PA = PA1 × PA2 × PA3 × PA4 × PA5, and when a failure B occurs, each feature quantity is applied to this (A1, A2, A3, A4, A5) probability:
PB is PB = PB1 × PB2 × PB3 × PB4 × PB5, and in this way, in each failure, the probability that the feature quantity is applied to (A1, A2, A3, A4, A5) is Compute by map with and compare. Then, if the result is: QB · PB> QA · PA> QC · PC> QD · PD ... Assuming that QB = QA = QC = QD, then PB>PA>PC> PD ... It is highly possible that

By the way, the failure phenomenon of the V6 engine is as follows.
For example, (A) No ignition for only one cylinder in the right bank (no ignition for one cylinder R) (B) No ignition for only one cylinder in the left bank (no ignition for one cylinder L) (C) Ignition deviation in all cylinders (all ignition cylinders) (D) Leakage in intake path (Intake leak) (E) Small idle flow rate (small intake volume) (F) Large idle flow rate (large intake volume) (G) Maximum idle flow volume (maximum intake volume) ( H) Clogged in intake path (clogged intake air) (I) Minimum fuel in all cylinders (minimum fuel in all cylinders) (J) Abnormal throttle valve (TVO abnormality) (K) Abnormal air flow meter (AFM abnormality) (L) Right No fuel for only one cylinder in the bank or left bank (no fuel for one cylinder R, L) (M) Large fuel for only one cylinder in the right bank or left bank (large fuel for one cylinder R, L) (N) Right bank or left bank 1 cylinder only small fuel (fuel 1 cylinder small R L) (O) No fuel for right bank or left bank 3 cylinders (no fuel for 3 cylinders R, L) (P) Small fuel for all cylinders (small for all cylinders) (Q) Large fuel for all cylinders (all cylinders for fuel) Large) (R) Maximum fuel of all cylinders (maximum total fuel), etc., and five characteristic quantities (rotation speed, charging efficiency, rotation speed fluctuation, right bank side air-fuel ratio feedback value and The left-bank-side air-fuel ratio feedback value distribution pattern is different Fig. 6 shows an example of a five-feature amount distribution pattern for a normal engine condition and each failure phenomenon in a V6 engine vehicle. (R) corresponds to each failure phenomenon described above, and in (L) to (O), the solid line indicates the case of the right bank (R) and the dotted line indicates the case of the left bank (L).

[0030]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment in the case of performing an engine failure diagnosis of a V6 engine vehicle by an off-board failure diagnosis machine will be described below with reference to FIGS. 7 is an overall connection diagram, FIG. 8 is a block diagram of diagnostic processing, FIG. 9 is an overall flowchart including diagnostic condition determination (precheck) and general logic, FIG. 10 is a self-tuning flowchart, and FIG. 11 is general logic (probability calculation). It is a flowchart of (processing).

In this embodiment, in the V6 engine vehicle 1, the engine control unit 2 passes the failure diagnosis diagnostic connector 3 in the no-load idling state after complete warm-up, and the off-board failure diagnosis machine 4 is serially communicated. Data transmission (see FIG. 7), the processing means includes a data collection unit 5 that controls serial communication and collects data, and statistically processes the data obtained by communication to represent five engine states. Amount (engine speed NE, engine speed fluctuation NEV, filling efficiency CE, processing value CFBR according to air-fuel feedback correction value of the right bank, that is, O ₂ sensor value, air-fuel ratio feedback correction value of the left bank, that is, O ₂ sensor value The statistical processing unit 6 for extracting the processed value CFBL) according to the Then, the normalization process for the calculation by the map B (probability value table map) is performed, and the probability calculation result for each failure is calculated by using the standard value table A and the map B. A comparison value determination unit 8 for comparison, a special logic unit 9 for determining a diagnosis condition as a pre-check based on communication data, a past failure database C storing past failure data (feature amount), and a standard value table A It is composed of a self-tuning unit 10 for performing self-tuning of A and a diagnostic result display unit 11 for displaying a diagnostic result (see FIG. 8).
The processing operation is as follows.

The overall operation is as shown in FIG. 9. When started, data is collected from the engine control unit 2 by serial communication in step S101, various sensor values are received as data for special logic, and in step S102, O ₂ Perform a normal failure code check to see if there are any abnormalities such as disconnection or short circuit in the sensor system such as sensors or rotation sensors, and also check if the water temperature and power supply voltage are within the specified range. Then, by looking at the result, whether or not the condition for performing the failure diagnosis by the general-purpose logic is not satisfied (diagnosis condition is not reached) is determined in step S1.
If it is determined in step 03 that the diagnostic conditions have not been reached, step S
At 104, a warning is displayed, and the process ends. In the following cases, general logic is not performed to prevent misdiagnosis. 1. There is a failure code (for example, a disconnection check code for a sensor), and you cannot expect each feature to be accurate. 2. When the water temperature value is below the threshold value H1 and above H2 (standard value is, for example, water temperature 90 ± 10 degrees). 3. When the power supply voltage is abnormal.

On the other hand, when the diagnostic conditions are satisfied, the process proceeds to step S105 to receive the general logic data,
By statistically processing, five engine state characteristic quantities (engine speed NE, engine speed fluctuation NEV, charging efficiency CE, right bank air-fuel ratio feedback correction value CFB
R, left bank air-fuel feedback correction value CFBL) is extracted. Then, the general-purpose diagnostic logic (the routine of FIG. 11) is executed in step S106, and the diagnostic result is displayed in step S107. Then, in step S108, the serviceman checks whether or not the failure phenomenon shown in the display of the diagnosis result actually occurs, and if the failure phenomenon does not actually occur, the failure phenomenon actually occurring. Is input in step S109. Then, the diagnostic results are used as data for self-tuning in step S110.
Add to the record of past failure database with.

Further, as background processing, processing (self-tuning) for feeding back the data of the past failure database in which new failure data is additionally written to the standard value table A is executed. That is, according to the flowchart of FIG. 10, in step S201, the past failure database is aggregated to obtain the average value and the standard deviation for each failure phenomenon, and the provisional standard value table A ′ is created. Then, in order to calculate the correct answer rate, the data of the past failure database (NE, NE fluctuation, CE, CFBR, CFB) is calculated in step S202.
L) is read for each failure phenomenon, and in step 203, the general-purpose diagnostic logic (routine in FIG. 11) is executed for all data in the past failure database using the standard value table A ′, and in S204, , The correct answer rate QA ′, which is how much the expected failure phenomenon matches, is compared with the diagnosis result when the standard value table A is used up to then, and all past failure data is obtained in step S205. It is determined whether or not the calculation of is completed, and the processes of steps S202 to S204 are repeated until all are completed. When all are completed, the correct answer rate QA in the case of the standard value table A and the correct answer rate QA 'in the case of the standard value table A'are compared in step S206, and the correct answer rate is higher in the case of the standard value table A'. If it's high, the table A '
Is newly defined as a standard value table A.

In the general-purpose logic (probability calculation process) according to the flowchart of FIG. 11, first, in step S301, the average value and standard deviation for each feature amount are read from the table A. Then, in step S302, normalization for using the map B is performed for a specific feature amount (rotational speed NE), and a specific failure phenomenon (fault A) is searched by searching the map B.
The NE probability PA1 is calculated. Then, step S3
In 03, see whether the probability PA1 is smaller than the threshold value K,
If the probability PA1 is smaller than the threshold value K, it means that the possibility of the failure phenomenon is extremely low, and therefore the subsequent calculation for the failure phenomenon is omitted to shorten the calculation time. On the other hand, if the probability PA1 of the NE of the failure A is not smaller than the threshold value K, steps S304 and S305, step S306 and step S307, and step S30 for NE fluctuation, CE, CFBR, and CFBL will be described below.
8 and step S309, step S310 and step S311 perform the same processing, and thereafter, in step S312, the probability values (PA1, PA2, P2) of the five feature quantities are calculated.
The total probability PA of failure A is calculated from A3, PA4, PA5) and written in table A. And step S
At 313, it is determined whether PA is larger than the threshold value L,
If PA is larger than the threshold value L, it is extremely likely that the failure phenomenon (fault A). Therefore, in step S314, the failure A is used as an answer, and calculation is performed for other failure phenomena to shorten the calculation time. Is omitted. On the other hand, PA
Is less than the threshold value L, it is determined in step S315 whether or not the calculation for all failure phenomena has been completed, and for other failure phenomena, steps S301 to S31.
The same process is performed in 3, and when the calculation of all failures is completed, the process proceeds to step S316. Then, in step S316, the total probabilities of the respective faults written in the table A are compared,
The one with a high overall probability is selected and output as the calculation result. In addition, those whose overall probability is a threshold value M or less and those whose predetermined probability (N%) is less than the maximum overall probability are excluded from the output of the calculation result.

Table 1 shows characteristic quantity extraction items for general-purpose logic and for determining diagnostic conditions.

[Table 1]

By the way, the failure phenomenon in the case of the V6 engine can be classified, for example, as shown in Table 2 below. For example, the sum of the absolute values of CFBR and CFBL is used as a dividing variable (G) and the threshold value ( KG)
A group of failure phenomena where the air-fuel ratio is abnormal (Group A) and a group of failure phenomena where it is not (Group B)
Can be divided into two groups. Therefore, the failure variable G is used as a segmentation variable G that represents the characteristics of the group.
It is also possible to employ an embodiment in which the variables G are divided into groups, the variable G is calculated to select one of the groups, and only the failure phenomenon of the selected group is diagnosed. In this case, since the diagnosis time can be shortened, it can be a coping method in the case where there are many feature quantities and failure modes.

[Table 2]

FIG. 12 is a flow chart of the general-purpose logic of the embodiment in this case. In the general-purpose logic of this embodiment, first, in step S401, an isolation variable G for the failure phenomenon group is calculated, and in step S402, the variable G is compared with the threshold value KG to select the failure phenomenon group. Then, the processes of steps S404 to S418 are executed only for the failure phenomenon belonging to the selected group.
The processes of steps S404 to S418 are the same as steps S301 to S316 of FIG. 11 described above. Therefore, detailed description is omitted.

Although the embodiment in which the present invention is applied to the engine failure diagnosis of the V6 engine vehicle has been described above, the present invention is not limited to this, and is also applied to the failure diagnosis of other various devices. It is possible.

[0040]

According to the present invention, the standard value of the characteristic amount for each failure is stored in the table in advance, and the value of the characteristic amount obtained from the received control data is evaluated based on the standard value. In addition, it is possible to automatically perform a device failure diagnosis and detect a failure location with high accuracy.

[Brief description of the drawings]

FIG. 1 is a basic block diagram according to the present invention.

FIG. 2 is a flowchart of a diagnostic part according to the present invention.

FIG. 3 is a standard value table for each failure of each engine type according to the present invention.

FIG. 4 is a distribution diagram of sensor values for explaining a failure diagnosis logic according to the present invention.

FIG. 5 is an explanatory diagram illustrating a failure diagnosis logic according to the present invention in the case of an engine.

FIG. 6 is an explanatory diagram of a characteristic amount distribution pattern for each failure phenomenon of the engine according to the present invention.

FIG. 7 is an overall connection diagram of an embodiment of the present invention.

FIG. 8 is a block diagram of an embodiment of the present invention.

FIG. 9 is an overall flowchart of an embodiment of the present invention.

FIG. 10 is a flowchart of self-tuning according to the embodiment of the present invention.

FIG. 11 is a flowchart of general-purpose logic according to an embodiment of the present invention.

FIG. 12 is a flowchart of general-purpose logic according to another embodiment of the present invention.

[Explanation of symbols]

 1 V6 engine vehicle 2 Engine control unit 4 Off-board fault diagnostic device 5 Data collection unit 6 Statistics processing unit 7 General-purpose logic unit 8 Comparison judgment unit 9 Special logic unit 10 Self-tuning unit 11 Diagnostic result display unit

Claims (15)

[Claims] 1. A method of diagnosing a failure of a device based on input / output data of a device control system, wherein each failure phenomenon of a plurality of feature quantities representing a device control state corresponding to each failure phenomenon of the device. A standard value table storing standard values of the device is created in advance, the input / output data is collected and processed in the control state of the device to extract the plurality of feature amounts in the control state, and the feature amount is extracted in the control state. A method for diagnosing a failure of a device, comprising: determining a failure phenomenon based on a value and a standard value based on a standard value table for each failure phenomenon, and identifying a failure location corresponding to the failure phenomenon. 2. The standard value is a standard mean value and standard deviation of the plurality of feature quantities for each failure phenomenon,
The device failure diagnosis method according to claim 1, wherein the standard value for at least one of the plurality of characteristic amounts is different between the failure phenomena. 3. In the determination of the failure phenomenon, the probability that the plurality of feature values apply to the extracted value in the control state when a specific failure phenomenon occurs is the extracted value and the standard value for each failure phenomenon. 3. The method according to claim 1, wherein the calculated value of the probability for each failure phenomenon is compared, and it is determined that the failure has occurred due to a failure phenomenon having a large calculated value.
Failure diagnosis method for the listed equipment. 4. The calculation of the probability is performed for each of the plurality of feature quantities for each failure phenomenon, and the overall probability is calculated from the probability value for each feature quantity. Equipment failure diagnosis method. 5. The device failure diagnosis method according to claim 4, wherein the calculation of the probability for each feature amount is performed using a standardized map. 6. In the process of calculating the individual probabilities of the plurality of feature quantities for a particular failure phenomenon, if the calculated probability value of the particular feature quantity is less than or equal to a threshold value, 6. The device failure diagnosis method according to claim 4, wherein the subsequent calculation is omitted and the calculation is performed for another failure phenomenon. 7. When the overall probability calculated for a specific failure phenomenon is equal to or greater than a threshold value, it is determined that the failure has actually occurred, and the calculation for other failure phenomena thereafter is omitted. 7. A failure diagnosis method for a device according to claim 4, 5 or 6. 8. A failure diagnosis method for a device according to claim 1, wherein an abnormality of a specific sensor value is detected, and whether a failure diagnosis is possible or not is determined based on a result of the detection. 9. The device failure diagnosis method according to claim 4, wherein a failure phenomenon having a large calculated value of the overall probability is selected and the data thereof is displayed as a diagnosis result. 10. The method of diagnosing a failure of a device according to claim 4, wherein, for a failure phenomenon in which the calculated value of the overall probability is a threshold value or less, the data thereof is excluded from the display of the diagnosis result. 11. The device according to claim 4, wherein data of a failure phenomenon showing a calculated value of a predetermined ratio or less with respect to a maximum calculated value among the calculated values of the overall probability is excluded from the display of the diagnostic result. Fault diagnosis method. 12. The device failure diagnosis method according to claim 1, wherein the diagnosis result is registered in a past failure database. 13. The past failure database is totaled,
A standard value for each failure phenomenon of the plurality of feature quantities is obtained, a temporary standard value table storing the standard values is created, and a failure is made for each data in the past failure database using the temporary standard value table. 13. The device failure diagnosis method according to claim 12, wherein the diagnosis is performed, the diagnosis result is compared with the diagnosis results obtained by the standard value table up to that point, and the one having a higher correct answer rate is used as a new standard value table. 14. The failure phenomenon is divided into a plurality of groups in advance according to the degree of association with a specific sensor value, and one of the groups is selected based on the extracted value of the feature quantity related to the sensor value, and the selected one is selected. 4. The device failure diagnosis method according to claim 3, wherein a failure diagnosis is executed for a failure phenomenon of a group. 15. A data collection unit for collecting input / output data of a device control system, and a statistical processing unit for statistically processing the collected input / output data to extract a plurality of feature quantities representing a device control state. When a failure phenomenon occurs, a probability calculation unit that calculates the probability that the plurality of feature quantities apply to the extracted value of the feature quantity in the control state, and the failure determination by comparing the calculation result of the probability of each failure phenomenon. A failure diagnosis device for a device, comprising: a comparison / determination unit to perform; and a display unit that displays a result of failure determination.