JP3198793B2 - Failure diagnosis method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for diagnosing a failure in a system constituted by combining elements, parts, units, devices or the like as a single component. .

[0002]

2. Description of the Related Art When a failure occurs in a system composed of a large number of components, or in a complex plant that is continuously processed through many processing steps, it is often difficult to identify the location of the failure. . As a conventional technique of such a failure diagnosis method, a diagnosis expert system is used as a system in which diagnosis knowledge possessed by a diagnosis specialist is stored in a computer in advance, and the diagnosis is automated using the failure diagnosis knowledge. Generally known. in this case,
The knowledge possessed by the expert is knowledge representing the causal relationship between the cause of the failure and the result, and if-then rules are generally widely used as a method for storing this knowledge in a computer. However, in order to acquire such knowledge from an expert, a knowledge engineer (expert system developer) must interview the expert, which involves considerable difficulty.

Conventionally, a model-based diagnosis for diagnosing using a predicted value of an operation by a model to be diagnosed has been devised. As one method of model-based diagnosis, there is, for example, the following document 1. Reference 1: Johan de Kleer and Br
ian C.I. Williams, "Diagnostics
ng Multiple Faults ", Artif
icial Intelligence Vol. 3
2, p. 97-130

FIG. 8 is a diagram for explaining the GDE algorithm shown in the above-mentioned reference 1, and includes M1, M2, M
3 is a multiplier, and A1 and A2 are adders, respectively. Input signals A to A to the arithmetic circuit configured as shown in FIG.
Assuming that A = B = E = 3 and C = D = 2 are input as the signal values of E, the signal values of the output signals F 1 and G are both 1
Should be 2. However, when actually measured, F
= 10 and G = 12, somewhere in the circuit
It can be seen that there is a failure. In such a case, a method of identifying which component in the circuit has failed is GDE.
(General Diagnostic Engineering
e) The algorithm. In this specification, foreign documents
GDE terminology is used to make correspondence with
However, in order to make its meaning clear, a general-purpose diagnostic engine
Also called. Therefore, GDE and general-purpose diagnostic engine are synonymous.
The taste.

In the GDE algorithm, a variable notation is represented by a variable = value {part number used in calculation}. As for the measured values, the inside of the square is an empty list. An example of a diagnostic step by the GDE algorithm in the circuit of FIG. 8 is shown below. [Step 1] A = 3 [Step 2] B = 3 [Step 3] C = 2 [Step 4] D = 2 [Step 5] E = 3

The multiplication results X, Y, and Z can be obtained by using a multiplier model. Then, put the name of the part used for obtaining the signal value in the parentheses. [Step 6] X = 6 ｛M1｝ [Step 7] X = 6 ｛M2｝ [Step 8] X = 6 ｛M3｝ For example, taking step 6 as an example, this means that “M1 has not failed. Then, X = 6. "

For the addition results F and G, [Step 9] F = 12 {A1, M1, M2} [Step 10] G = 12 {A2, M2, M3} Assuming that all of M2 have not failed, F = 12. " On the other hand, since the measured value of F is 10, [Step 11] F = 10 ｛｝ The results of Step 10 and Step 11 are inconsistent. That is, if all of A1, M1, and M2 are considered to be normal, it is strange. On the algorithm, as soon as a contradiction is found, the following conflict is generated. C1 = <A1, M1, M2> This indicates that at least one failed component exists in C1.

Hereinafter, the signal value of each point is obtained by any combination. [Step 12] X = 4 (X = F−Y) ｛A1, M2｝ [Step 13] Y = 4 (Y = F−X) ｛A1, M1｝ [Step 14] G = 10 (G = Y + Z) ｛A1, A
2, M1, M3｝ The measured value of G is [Step 15] G = 12 ｛｝, which contradicts Step 14 and generates the following conflict C2 = <A1, A2, M1, M3>.

The steps of the next combination are: [Step 16] Z = 6 (Z = G−Y) {A2, M2} [Step 17] Y = 6 (Y = G−Z) {A2, M3} [ Step 18] Z = 8 (Z = G−F + X) ｛A1, A
2, M1｝ [Step 19] X = 4 (X = FG + Z) ｛A1, A
2, M3｝, and there is no contradiction any more.

Next, a candidate for a failed component is determined from the obtained conflict. The obtained conflict is as follows: C1 = <A1, M1, M2> C2 = <A1, A2, M1, M3> Although the GDE algorithm supports a plurality of faults, assuming that only one fault occurs at the same time, the fault component that can explain both conflicts is the intersection of C1 and C2. [A1] Or [M1] alone, and these two are candidates for the failed component.

Intuitively, the GDE algorithm stores the parts used in determining the value of a variable, and when there is a discrepancy between the obtained observed value and the measured value, obtains the value. This means that some of the components used at that time have a failure. In this way, by using the GDE algorithm, if only the model to be diagnosed can be defined, it is possible to narrow down the faulty parts by repeating comparison of the observed value and the value calculated by the model.

On the other hand, in the case of a multivariate, non-linear complex plant, it is generally difficult to construct a quantitative model of the object, and instead, as a technique for simply predicting only the qualitative behavior of the object, A method called qualitative simulation has been considered. As an example of the qualitative simulation method, there is the following document 2. Reference 2: Johan De Kleer and Jo
hn Seely Brown, "A Qualita
five Physics Based on Conf
luences ", Artificial Intel
license, Vol. 24, p. 7-83, 198
4

In the qualitative simulation, variables are converted into ternary values as follows instead of dealing with quantitative numerical values. x = [+]; x> normal value x = [0]; x = normal value x = [−]; x <normal value FIG. 9 is a diagram for explaining a qualitative simulation model of a valve, and FIG. , Pin is the inlet pressure, Pout
Is the outlet pressure, Fin is the inlet flow, Fout is the outlet flow, dP
Is the differential pressure between input and output. The qualitative simulation model of the valve 11 in FIG. 9 is generally expressed by the following equations (1) to (3). dP = Pin−Pout (1) Fin = dP (2) Fin = Fout (3)

The physical meaning in the equation (2) means that the input flow rate Fin is proportional to the differential pressure dP between the input and output. However, the equation of the qualitative simulation model is simply connected by an equal sign. In the qualitative simulation model of FIG. 9, the results of the qualitative simulations when the inlet pressure increases and the outlet pressure increases are shown in the following qualitative simulation results a and b, respectively.

Qualitative simulation results a, Pin = [+]: Inlet pressure increase dp = [+]: Differential pressure increase Fin = [+]: Inlet flow rate increase Fout = [+]: Outlet flow rate increase Pout = [+]: Outlet pressure increase The above qualitative simulation result a shows that when the inlet pressure increases, the differential pressure increases, the flow rate from the inlet increases,
The flow rate at the outlet also increases, indicating that the pressure at the outlet increases.

Qualitative simulation result b, Pout = [+]: increase in outlet pressure dp = [-]: decrease in differential pressure Fin = [-]: decrease in inlet flow Fout = [-]: decrease in outlet flow Pin = [+]: Inlet pressure increase The above qualitative simulation result b shows that when the outlet pressure increases, the differential pressure decreases, the flow rate from the inlet decreases,
This indicates that the flow rate exiting the outlet also decreases, and that the pressure at the outlet further increases.

[0017]

However, the aforementioned G
In the DE algorithm, as a result of the diagnosis, the number of failed component candidates is narrowed down to a certain extent. However, parts that are not considered to have actually failed may be included in the failure candidates. Was insufficient. In the qualitative simulation algorithm, when a system including a large number of components or a complex plant in which many processing steps are to be diagnosed is to be diagnosed as a failure candidate for all the components or the processing steps, However, there has been a problem that the diagnosis process generally requires a long time.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problems, and an element, a component, a unit, an apparatus, or the like is considered as one component, and a system configured by combining a plurality of such components is an object to be diagnosed. In this case, it is an object of the present invention to obtain a failure diagnosis method in which the failure component candidates and the failure cause candidates are sufficiently narrowed down and the diagnosis processing time can be reduced as compared with the related art.

[0019]

According to the failure diagnosis method of the present invention, an element, a component, a unit, an apparatus, or the like is used as one component, and a system configured by combining a plurality of the components is used as a diagnosis target. In the failure diagnosis method, the normal operation of the diagnosis target is simulated by qualitative simulation, and the simulation result is compared with the actual diagnosis target symptom using a general-purpose diagnosis engine algorithm. A first step of narrowing down element candidates, and sequentially selecting a qualitative model at the time of failure for some predetermined failure causes for each of the failure component candidates narrowed down in the first step; A qualitative simulation of the diagnosis target is performed using the selected qualitative model at the time of failure, and the simulation results and Is intended to include a second step of identifying the cause of failure by detecting a compare match with the symptoms diagnosed when.

[0020]

According to the present invention, there is provided a failure diagnosis method in which an element, a part, a unit, an apparatus, or the like is regarded as one component and a system constituted by combining a plurality of the components is a diagnosis target. Simulates the normal operation of the diagnosis target by qualitative simulation, and uses the simulation result and the actual symptoms of the diagnosis target in general.
By comparing using the diagnosis engine algorithm, the failure component candidates are narrowed down from the diagnosis target. In the second step, a predetermined number of failure component candidates narrowed down in the first step are determined. A qualitative model at the time of failure for the cause of the failure is sequentially selected, a qualitative simulation of the diagnosis target is performed using the selected qualitative model at the time of the failure, and the simulation result is compared with the actual diagnosis target symptom. The cause of the failure is identified by detecting.

[0021]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS First, a failure diagnosis method according to the present invention comprises connecting one of an element (element), a part (part), a unit, an apparatus, or the like as a constituent element and connecting a plurality of the constituent elements. Applicable to all configured systems. In other words, it can be applied to a small system to a large system.

FIG. 1 is a flowchart showing a schematic process of a failure diagnosis method according to the present invention. The flowchart of FIG. 1 will be described. FIG.
In step S31, when a failure occurs in the diagnosis target, all the components to be diagnosed are considered as failure component candidates. In step S32, the normal operation of the diagnosis target is simulated by qualitative simulation, and the simulation result is compared with the actual symptoms of the diagnosis target using the GDE algorithm, thereby narrowing down failure component candidates from the diagnosis target. Perform That is, the number of failure component candidates is reduced.

In step S33, the previous step S32
For each failure component candidate narrowed down in the above, a qualitative model at the time of failure for several preset failure causes is sequentially selected, and a qualitative simulation of the diagnosis target is performed using the selected qualitative model at the time of failure. The cause of the failure is identified by detecting a comparison match between the simulation result and the symptom of the actual diagnosis target. The more detailed processing in step S33 will be described with reference to FIG.

Consider a case in which the failure diagnosis method of the present invention is applied to the hydraulic circuit of FIG. FIG. 2 is a diagram showing a hydraulic circuit which is an example of a failure diagnosis target of the present invention.
3 and 5 are hydraulic motors, 7 is a solenoid valve, 4 is a pilot check valve, 2 and 6 are other parts, and 8 is an output terminal. The pilot check valve 4 allows oil to flow from the high pressure side to the low pressure side only when the solenoid valve 7 is in the state shown in the figure. In this hydraulic circuit, it is assumed that components other than the hydraulic motors 3 and 5 and the solenoid valve 7 are modeled by the qualitative valve model described in the related art.

In the case of FIG. 2, the following two problems (A) and (B) are considered as possible failure symptoms. (A): The hydraulic motor 3 operates, and the hydraulic motor 5 does not operate. (B): Both hydraulic motors 3 and 5 do not operate. In the case of the symptom (a), assuming that only one failure point is detected, the diagnostician determines that oil leakage has occurred at the pilot check valve 4 between the hydraulic motors 3 and 5 or the hydraulic motor 5 has failed. Judge.

In the case of the condition (b), if any part (2, 3, 4, 5, 6) is clogged, or
It is determined that oil has leaked from the component 2 on the higher pressure side than the two hydraulic motors 3 and 5, or that the solenoid check valve 7 has failed and the pilot check valve 4 is closed. It suffices if the proposed failure diagnosis method makes a conclusion equivalent to the above judgment.

When the pressures at the hydraulic source (input terminal) 1 and the output terminal 8 in FIG. 2 are both normal and the operations of the hydraulic motors 3 and 5 become the above-mentioned symptoms (a) and (b), FIGS. 3 and 4 show the results obtained by performing the processes in steps S31 and S32 of FIG. 1 and narrowing down the failed component candidates. FIG.
In FIG. 4 and FIG. 4, the parts surrounded by the broken line curves are candidates for the narrowed down failed parts. This result includes the failed part as determined by the diagnostician.

Although it is possible to narrow down the number of failed component candidates to some extent only by the diagnostic result using the GDE algorithm in S32, there is a problem. For example, in FIG. 3 showing the diagnosis result of the symptom (a), the solenoid valve 7 is also included in the failure candidate. When the solenoid valve 7 is broken, the pilot check valve 4 which should have been opened is closed, the path is cut off, and both hydraulic motors 3 and 5 are stopped. This is inconsistent with the actual symptoms. That is, the failure of the solenoid valve 7 is not suitable as a failure candidate in the case of the symptom (a). Thus, there is no doubt that contain faulty component in the fault candidates found in the GDE algorithm, Ru <br/> see that sometimes, also comprise extra ones.

Also, for example, looking at the diagnosis result of the symptom (a), it is known that there may be a failure of the pilot check valve 4, but the actual cause of the failure is not known. If the cause of the failure of the pilot check valve 4 is "leakage", the symptom as shown in (a) will occur, but if the cause of the failure of the pilot check valve 4 is "clogging", both hydraulic motors 3 and 5 will have It stops. That is, GDE
Although the diagnosis result of the algorithm indicates which component has a possibility of failure, it is not known what the potential cause of failure is when there are a plurality of possible causes.

In order to solve these problems, FIG.
At 33, a qualitative model at the time of failure is defined for each component, and a qualitative simulation at the time of failure is performed to determine the actual cause of the failure. The simulation at the time of failure refers to estimating the operation of the two hydraulic motors when, for example, the parts 2 in front of the two hydraulic motors 3 and 5 in FIG. 2 leak.

FIG. 5 is a flowchart showing the detailed processing of step S33 in FIG. 1, and FIG. 5 shows the order in which the cause of a failure that can occur in the failure model according to the present invention is specified. FIG.
Will be described. First, in step S41, it is checked whether or not there is a failed component candidate which is not yet assumed to be in failure in the diagnosis target. If not, the process is terminated. It is assumed that one of the new failed component candidates is a failure.

In step S43, the previous step S42
Then, it is determined whether or not there is an unassumed failure cause candidate among several possible failure causes for the component assumed to have failed. As a specific example, assuming that the valve has failed in S42, two possible causes of the valve failure are "leakage" and "clogging". Of the qualitative simulation assuming that
It is to determine whether there is anything that has not been done yet. When the determined result is YES, the process shifts to step S44.
If NO, the process returns to step S41.

In step S44, one of the unpredicted failure cause candidates is assumed to be the cause of the failure. In step S45, the symptom caused by the assumed failure cause is simulated by qualitative simulation. It is confirmed whether the symptom of the diagnosis target matches the simulation result of S45. If the actual symptom and the simulation result match, in step S47, the matching failure cause candidate is displayed as being the cause of the failure, and if not, step S47 is performed.
Returning to step 43, the steps from S43 are repeated again.

When performing a qualitative simulation at the time of failure in step S45 in FIG. 5, if the assumed failed component is a bubble, the following two failure models of "leakage" and "clogging" are set in advance. . Failure model of valve "leak" Pin = [-]: Input pressure decrease Pout = [-]: Output pressure decrease Failure model of valve "clogging" Pin = [+]: Inlet pressure increase Pout = [-]: Output pressure decrease

Here, in the failure model, the failure cause is “leakage”.
Or, in the case of "clogging", it describes what happens to a variable for a part. The model of "leakage" described above indicates that when a valve leaks, both the pressure at the inlet and the output of the valve decrease, and the model of "clogging" indicates that when the valve becomes clogged, the inlet of the valve becomes The pressure increases and the outlet indicates that the pressure decreases. In step S45, a qualitative simulation is performed on a model selected in order among several (two in this example) failure models set in advance as described above.

In the case of the solenoid valve 7 connected to the pilot check valve 4, a failure model is set in which, when a failure occurs, the outlet pressure is originally low when the pressure is originally high, and is high when the outlet pressure is low. For example, in the case of the failure symptom (a) of the hydraulic circuit in FIG. 2, in the diagnosis step S42 in FIG.
It is assumed that the pilot check valve 4 in FIG. 2 has failed. It is assumed that "leakage" is assumed as the cause of the failure in step S44. When a qualitative simulation is performed in step S45 starting from a variable value in the failure model of the valve “leakage”, a symptom that the hydraulic motor 3 operates and the hydraulic motor 5 does not operate can be simulated. The "leakage" of the check valve 4 is displayed as a failure cause candidate in step S47.

Next, it is assumed that "clogging" is assumed as the cause of the failure in step S44. When a qualitative simulation is performed starting from a variable value in the failure model of the valve “clogging”, it is simulated that both hydraulic motors 3 and 5 do not operate. Since this is different from the symptoms that are actually occurring, the assumption of this cause of failure is rejected. Hereinafter, the same verification is repeated for all the faulty component candidates, and the fault causes that cannot simulate the symptoms are rejected, thereby narrowing down the possible fault causes.

With the hydraulic circuit of FIG. 2 as a diagnosis target, first, in S32 of FIG. 1, failed component candidates are narrowed down and determined by the GDE algorithm. Then, in S33 of FIG. 1, a qualitative simulation at the time of failure is performed for each failed component candidate. 6 and 7 show the results of the determination of the cause of the failure for each of the symptoms (a) and (b). 6 and 7
In the above, the failed component indicated by the broken line curve and the cause of the failure completely match the judgments of the experts, respectively, so that the diagnosis was successful.

The purpose of using the GDE algorithm in the present invention is to reduce the number of failed component candidates in advance because it takes a long time to apply qualitative simulation at the time of failure to all components in the diagnosis target. GDE
An advantage of the algorithm is that even if a fault model is not used, a faulty component candidate can be narrowed down to some extent only with a normal model. In general, the more complicated the model of a single component, the higher the effect of using the GDE algorithm.

[0040]

As described above, according to the present invention, a failure diagnosis is performed on a system configured by combining elements, parts, units, devices, or the like as one constituent element and combining the constituent elements. In the method, first in a first step:
The qualitative simulation simulates the normal operation of the diagnosis target, and compares the simulation results with the actual diagnosis target symptoms using a general-purpose diagnosis engine algorithm, thereby narrowing down failure component candidates from the diagnosis target. Then, in the second step, for each of the narrowed-down failure component candidates, a qualitative model at the time of failure for some preset failure causes is sequentially selected, and the qualitative model at the time of the selected failure is selected. The qualitative simulation of the diagnosis target is performed using, and the cause of the failure is identified by detecting a comparison match between the simulation result and the actual symptom of the diagnosis target. Is sufficiently narrowed down, and the total processing time of the first and second steps is Is shorter than the processing time of the law, it has become possible to perform the recovery process of the fault quickly. In addition, since such a failure diagnosis method does not use the knowledge of an expert, interview work with the expert is not required when creating a failure diagnosis system, and development is facilitated.

[Brief description of the drawings]

FIG. 1 is a flowchart showing a schematic process of a failure diagnosis method according to the present invention.

FIG. 2 is a diagram showing a hydraulic circuit which is an example of a failure diagnosis target of the present invention.

3 is a diagram showing a diagnosis result by GDE in the case of the symptom (a) of FIG. 2;

FIG. 4 is a diagram showing a diagnosis result by GDE in the case of the symptom (b) of FIG. 2;

FIG. 5 is a flowchart showing a detailed process of step S33 in FIG. 1;

FIG. 6 is a diagram showing a failure cause determination result according to the present invention in the case of the symptom (a) of FIG. 2;

FIG. 7 is a diagram showing a failure cause determination result according to the present invention in the case of the symptom (b) of FIG. 2;

FIG. 8 is a diagram for explaining a conventional GDE algorithm.

FIG. 9 is a diagram illustrating a qualitative simulation model of a valve.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Hydraulic power source 2, 6 Parts 3, 5 Hydraulic motor 4 Pilot check valve 7 Solenoid valve 8 Output terminal 11 Valve

──────────────────────────────────────────────────続 き Continuation of front page (56) References JP-A-4-264458 (JP, A) JP-A-5-273807 (JP, A) JP-A-4-55776 (JP, A) JP-A-2- 151000 (JP, A) Japanese Patent Application Laid-Open No. 5-99800 (JP, A) Japanese Translation of PCT Application No. 5-502102 (JP, A) "Application of qualitative inference to diagnosis of large-scale systems", Information Processing, February 1991 , Vol. 32, No. 2, p. 137-144, "Problem Solving Using Qualitative Reasoning-Fault Diagnosis-", Proc. Of the 35th Annual Conference of IPSJ (late 1987), September 1987, p. 1575-1576, "Biological model of qualitative inference and its application to medical diagnosis", Information Processing, February 1991, Vol. 32, No. 2, p. 118-125 "Qualitative reasoning and its application to solving intellectual problems", Information Processing, February 1991, Vol. 32, No. 2, p. 105- 117 (58) Fields investigated (Int. Cl. ⁷ , DB name) G06F 11/22-11/277 G06F 9/44 G05B 23/00-23/02 CSDB (Japan Patent Office) Patent file (PATOLIS ) Practical file (PATOLIS) JICST file (JOIS)

Claims (1)

(57) [Claims] 1. A failure diagnosis method in which an element, a component, a unit, an apparatus, or the like is regarded as one component and a system configured by combining a plurality of the components is a diagnostic target. A first step of simulating the operation at the time and comparing the simulation result with the actual symptom of the diagnosis target using a general-purpose diagnosis engine algorithm to narrow down failure component candidates from the diagnosis target; For each of the failure component candidates narrowed down in the first step, a qualitative model at the time of failure for some preset failure causes is sequentially selected, and a diagnostic object is selected using the selected qualitative model at the time of failure. Qualitative simulation of the target, and detects a comparison between the simulation result and the actual symptom of the diagnosis target. Fault diagnosis method characterized by comprising a second step of identifying the cause of failure by.