JPH08202444A - Method and device for diagnosing abnormality of machine facility - Google Patents

Abstract

PURPOSE: To speedily take measures to cope with an abnormal state by judging the adequacy of a diagnostic result by a physical model. CONSTITUTION: An abnormality diagnostic means 3000 detects the abnormal state according to a diagnostic rule wherein the state and abnormality cause of an object facility 1000 are classified by cases and estimates the cause in case of abnormality. When the abnormal state is judged, the cause 4010 of the abnormality estimated and an alarm for the abnormal state are outputted on the display screen of an input/output device 50. And, a causality analyzing means 6000 analyzes the learning result 5020 of a learning means 5000 and specifies information which is relatively strong in relation with the cause of the abnormality among sensor information and observation information. A physical model generating means 7000 constructs a physical model which represents the physical cause and effect relation between the measurement or observation position of the specified information and the cause of the abnormality as a combination of physical relational expressions. The constructed physical model has the relation between actual measurement or observation data and abnormality cause compared as to its qualitative properties and is evaluated.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention measures or observes the state of mechanical equipment such as power generation equipment, heat supply equipment, communication equipment, air conditioning equipment, and various production plants, and detects abnormalities in the target equipment based on the results. And an abnormality diagnosis method and apparatus for estimating the cause of the abnormality.

[0002]

2. Description of the Related Art As a prior art relating to an abnormality (fault) diagnosis device, there is "a device / equipment state diagnosis system".
There is 326 (hereinafter referred to as "prior art 1"). In this conventional technology, the vibration information generated with a specific operating state of the target device or equipment is learned into a neural circuit model as a set of the operating state and the vibration information, and when some vibration information is input, the learning result Based on the above, the operating state is output as the diagnosis result. By this method, even if it is an inexperienced case, it is possible to make a diagnosis from the next time by learning with a neural circuit model.

The prior art also includes a knowledge base for performing a failure diagnosis, as seen in "Plant equipment failure diagnosis work support device", Japanese Patent Laid-Open No. 2-298892 (hereinafter referred to as "prior art 2"). Some have an inference mechanism by an expert system that incorporates knowledge from this knowledge base to estimate the cause of a failure, and a learning mechanism by a neural network.

According to this method, new knowledge can be learned by the learning mechanism when the cause of failure cannot be investigated by the reasoning mechanism of the expert system, and when a similar state occurs from the next time, the cause of the cause can be investigated. Estimate is possible.

[0005]

In the above-mentioned Prior Art 1, there is a possibility that a plurality of neural circuit model outputs, that is, diagnostic results, can be obtained. If the output values in each case are similar, it is difficult to identify which is the correct operating state.

Further, the above-mentioned prior art 1 and prior art 2
In both cases, it is impossible to judge whether or not the relationship between the learned state and the cause of failure is really correct. For example, even in the case of an abnormality of the sensor whose state has been measured, there is a high possibility that a new case will be learned or registered as knowledge. Therefore, in such a case, it results in erroneous diagnosis in the similar state from the next time, and it is difficult to take appropriate countermeasures. Further, in any of the above-mentioned conventional techniques, only the direct relationship between the state of the equipment or facility and the measured value can be known, and the theoretical causal relationship between the two cannot be known. Therefore, it is difficult for the operator to apply the past cases, and it is difficult to appropriately diagnose even the similar cases.

The present invention has been made in consideration of such problems, and a first object of the present invention is to provide an abnormality diagnosis method and apparatus capable of promptly dealing with an abnormal state by minimizing the number of candidates for the diagnosis result. To provide.

A second object of the present invention is to know the theoretical causal relationship between the abnormal cause of the equipment or facility and the measured value, deepen the understanding of the operator, and apply diagnosis to similar cases. An object of the present invention is to provide an abnormality diagnosing method and device that enable the above.

A third object of the present invention is to provide a highly reliable diagnostic rule adding method and apparatus by adding only truly correct cases among newly experienced cases as new diagnosis knowledge. To provide.

[0010]

In order to achieve the above-mentioned object, the present invention is preset with at least one of a measured value or an observation result indicating the state of one or a plurality of parts of mechanical equipment as an input. It is determined whether the mechanical equipment is in a normal state or an abnormal state according to a determination condition, and when it is determined to be an abnormal state, the ruled knowledge for estimating the cause of the abnormality from the input is stored, and the rule is stored. The abnormality diagnosis means for determining whether or not the mechanical equipment is in a normal state based on the converted knowledge and estimating the cause of the abnormality, and the abnormality diagnosis means, the input does not correspond to any of the ruled knowledge. In the case, or when the estimated cause of abnormality is an error, learning means for learning the relationship between the correct abnormality cause and the input, and the learning result of the learning means are analyzed to obtain the correct difference. The causal relationship analysis means for identifying an input item having a strong relationship with the cause, the measurement or observation site of the input item identified by the causal relationship analysis means, and the relationship between the site that is the cause of the abnormality, the registered physical A physical model constructed by a combination of physical or chemical relational expressions and a causal relationship evaluation means for comparing the relationship between the input item identified in the causal relationship analysis step and the cause of abnormality are provided so that no contradiction occurs between the two. In this case, a display means for displaying the relationship of the physical model on the screen is provided.

Preferably, when there is no contradiction in the relationship between the input item specified in the causal relationship analysis step and the cause of abnormality, the relationship between the input item specified by the causal relationship analyzing means and the cause of abnormality is abnormal. It comprises diagnostic rule adding means for adding rules to the knowledge of the abnormality diagnosing means as rules for diagnostic knowledge.

[0012]

The physical causal relationship between the measurement or observation site of the item specified by the causal relationship analysis unit and the site of the abnormal cause is configured by the causal relationship evaluation unit by a combination of physical or chemical relational expressions. The physical model makes it possible to theoretically evaluate the validity of the learned relationships.

By displaying the theoretical relationship on the screen by the display means, the physical or chemical causal relationship can be known.
The operator's understanding can be deepened. As this effect,
Even if the phenomena are not exactly the same, it becomes possible to infer the cause of the abnormality from the theoretical relationship when similar cases occur.

Further, the relation which has been theoretically confirmed to be valid is a new diagnostic knowledge by the diagnostic rule adding means.
Can be added to diagnostic rules. This makes it possible to newly add only the relationship evaluated to be theoretically correct to the diagnostic method without learning a specific state such as an abnormality of the measurement sensor.

If a neural circuit model is used as the learning means, it is possible to estimate the cause by using the output obtained by inputting the measurement or observation data to the learned neural circuit model as the cause of abnormality. In this case, a plurality of outputs (candidates for the cause) may be obtained. However, the causal relationship evaluation means can determine a theoretically valid relationship to specify the cause of the abnormality.

Further, if the priorities are set for the constituent elements when the physical model is created, the model creation time can be shortened without creating an unnecessarily complicated model.

[0017]

FIG. 1 shows an embodiment of the present invention.

FIG. 1 shows the configuration of an abnormality diagnosis device for the target equipment 1000. The state of the target equipment 1000 is measured by various sensors such as level, temperature, pressure, vibration, and sound, and the measurement data 100 is sent to the data processing means 2000. Similarly, observation information such as inspections and repairs is symbolized data 3
00 is sent to the data processing means 2000.

The data processing means 2000 digitizes various sensor signals according to the type of information, or extracts characteristic quantities. Further, the inspection data 200 such as the observation status of the operator of the target equipment 1000 and the periodic inspection are input from the input / output device 500, and this information is also sent to the data processing means 2000 to be digitized or symbolized.

The information processed by the data processing means 2000 is sent to the abnormality diagnosing means 3000. Abnormality diagnosis means 30
In 00, the status of the target equipment 1000 and the cause of the abnormality are described in "IF T
In addition to detecting abnormal conditions based on the diagnosis rules classified by the "HEN" type of rules, the cause of the abnormal conditions is estimated.

When the abnormality diagnosing means 3000 determines that an abnormality has occurred, the cause of abnormality 401 is estimated to be an alarm of the abnormality.
0 through the diagnostic result evaluation means 4000 to the input / output device 5
00 is displayed on the display screen.

The diagnostic result evaluation means 4000 evaluates the validity of the diagnostic result in view of past experience. That is, if the person has been experienced in the past and the diagnosis result at that time is correct, the degree of certainty is evaluated as high, and the greater the number of times of the experience, the greater the degree of certainty is presented. On the other hand, for inexperienced conditions and conditions that have been misdiagnosed,
Present a small degree of certainty. Confidence factor 402 in both cases
0 is sent to the input / output device 500, displayed on the screen, and used as an operator's judgment index.

If the diagnosis result is incorrect, the operator inputs from the input / output device 500 that the diagnosis is wrong, investigates the correct cause of abnormality, and inputs the correct cause of abnormality 510. The learning unit 5000 learns the relationship between the data 2010 digitized or symbolized by the data processing unit 2000 and the correct abnormality cause 510.

The learning means 5000 also has a function of estimating the cause of abnormality based on the learning result. Therefore, when the certainty of the diagnosis result according to the diagnosis rule of the abnormality diagnosis means 3000 is low, or when the diagnosis result is incorrect, the learning means 5000 displays the estimation result 5010 on the input / output device 500 according to the operator's judgment. can do.

The causal relationship analysis means 6000 analyzes the learning result 5020 of the learning means 5000 to determine the target equipment 100.
Information having a relatively strong relationship with the cause of abnormality is specified from among the plurality of sensor information and the observation information indicating the state of 0.

In the physical model creating means 7000, the physical causal relationship between the measured or observed part of the information specified by the causal relationship analyzing means 6000 and the cause of the abnormality is combined by a physical relational expression such as heat transfer, force or electricity. Build the represented physical model. Although only the physical relational expressions are dealt with in this embodiment, chemical relational expressions such as chemical reaction equations may be combined. Further, the physical model may be prepared in advance.

The constructed physical model is model evaluation means 8
At 000, the relationship between the actual measured or observed data and the cause of the anomaly with respect to its qualitative properties is compared and the constructed model is evaluated.

The evaluation result 8010 is presented to the operator at the input / output device 500 to receive an instruction to complete or change the model. Changes made by the operator in case of model change
520 is input, the model is changed in the physical model creating means 7000 according to the contents, and the model evaluating means 800 is again used.
At 0, the same processing is repeated.

Further, the input 53 from the input / output device 500
When the value is 0, the diagnosis method modification means 9000 newly creates the learned relationship as a diagnosis rule and adds it to the abnormality diagnosis means 3000. In the present embodiment, the operator is asked to add the diagnostic rule just in case, but the additional processing may be automatically performed.

Next, each function of this embodiment will be described in detail.

FIG. 2 shows an outline of the diesel engine power generation equipment which is the target equipment of this embodiment. Main fuel tank 1100
The fuel stored in the
Sent to 1200. By opening the fuel cock 1210 from the fuel dispensing tank 1200, the diesel engine 150
Fuel is supplied to 0.

Excess fuel is stored in the fuel drain tank 1220.
The fuel drain pump 1250 returns the fuel to the fuel dispensing tank 1200.

Air required for combustion of the engine is pressurized by the air compressor 1300 and stored in the air tank 1400.
Diesel engine 150 via air valve 1410
Supplied to zero. In addition, the engine exhaust is silencer 170
After passing 0, it is discharged out of the system.

Cooling water for the engine is a cooling water tank 1900.
Is stored in the cooling water pump 1950, is sent to the diesel engine 1500 by the cooling water pump 1950 to cool the engine, and then circulates to the cooling water tank 1900.

Electricity is obtained by driving the generator 1600 by the rotational force of the diesel engine 1500. The diesel engine 1500 and the generator 1600 have a common mount 18
It is installed on 00.

Next, the function of the data processing means 2000 will be described with reference to FIG.

Various sensors are attached to the power generation equipment shown in FIG. 1 in order to monitor the condition of the equipment.

The acoustic sensor 101 and the vibration sensor 102 are an engine 1500, a generator 1600 and a common mount 180.
It is installed on the 0 side. The rotation speed sensor 103 monitors the rotation speed of the generator 1600. Temperature sensor 10
4 measures the temperatures of cooling water, lubricating oil, and exhaust gas.
Further, the level sensor 106 measures the liquid amount of the main fuel tank 1100, the fuel dispensing tank 1200, the cooling water tank 1900, the fuel drain tank 1220, the oil pan and the like. Furthermore, the current and voltage sensors measure the output current and voltage of the generator.

Measurement data 100 sent from each sensor
Are processed by the data processing means 2000 according to the type. Acoustic and vibration signals are waveform and frequency analysis means
Processed at 2100. In waveform analysis, peak value, average value,
Parameters such as the number of occurrences are extracted and the first in frequency analysis.
Parameters such as the peak frequency, the first peak frequency component, the primary rotation frequency, and the primary rotation frequency component are extracted and output as sensor information 2110.

Numerical conversion processing means 2 is used for other sensor information.
At 200, the electric signal is converted into numerical data and output as sensor information 2210.

Observation information other than sensors, that is, inspection data 20
0 is input from the input / output device 500 by the operator. The observation information includes parts replacement information 201 and inspection information 202.
There is. The part replacement information 201 is information on the number of the replaced part and the replacement date, and the inspection information 202 is the result of a situation survey at the time of regular inspection. The specific input contents of the inspection information 202 are the part number and the inspection date of the one whose condition deteriorates or tends to deteriorate at the time of inspection. The observation information, that is, the symbolized data 300 input by the operator is symbolized by the symbolization processing means 2300 and output as the observation information 2310.

The abnormality diagnosing means 3000 detects an abnormality of the device and estimates the cause of the abnormality based on a preset diagnostic rule. An example of the diagnostic rule is shown below for the cooling water temperature (see FIG. 1).

(1) Measured value of temperature sensor 2010 (sensor number T ₁ ) ... X ° C. (2) Reference value corresponding to sensor number T ₁ ... Y ₁ , Y ₂ , Y ₃ ° C. (3) Diagnostic rule ・ IF "X <Y _1" AND "engine stopped" THEN "warning: coolant freezing could occur in the" · IF "X <Y _1" AND "during engine operation" THEN "abnormal (level 1): temperature measurement failure ( T ₁ ) ”“ Abnormality occurred (level 3): Radiator failure ”· IF“ Y ₁ <X <Y ₂ ”THEN“ Normal ”· IF“ Y ₂ <X <Y ₃ ”THEN“ Warning: Overheating may occur・ IF “Y ₃ <X” THEN “Abnormality occurred (level 3): Cooling water tank water volume decrease” Depending on the magnitude relationship between the measured value X and the set value (reference value) Y ₁ , Y ₂ , Y ₃ , It is a rule to make such a diagnosis.

The relationship between the cause of abnormality and the observed state can be considered in various cases in practice, and is not as simple as the above diagnostic rule. However, if all possible cases are ruled, the number of cases will be enormous and the time required for diagnosis will be long. Moreover, the number of candidates for the cause of abnormality increases, which makes it difficult for the operator to make a decision.

Therefore, it is desirable to lower the priority of diagnosis for events that occur with a low probability of occurrence and to reduce the number of diagnosis rules to be judged as much as possible, thereby making efficient diagnosis.

In the present embodiment, the "IF ... THEN ..." type diagnostic rule is used, but a method of physically modeling the target equipment and simulating its behavior to estimate the cause of the abnormality may be used. Even in this case, if the model is configured in a complicated manner, more detailed behavior can be simulated, but it is the same that the time required for calculation becomes long and the diagnosis becomes slow.

Since the purpose of the abnormality diagnosis is to operate the target equipment in a stable state as much as possible, it is of course necessary to accurately investigate the cause of the abnormality, but to enable quick recovery measures. However, the shortest possible diagnosis is desired.

Therefore, the diagnostic rule of this embodiment is initially
The diagnostic rule is set to consider only events with a high probability of occurrence, and the priority of consideration is low for events with a low probability of occurrence. After that, the priority order is changed according to the occurrence frequency of each event.

Further, in the present invention, the learning means 5000 is provided to learn a new abnormal event so as to be able to deal with the case where an event other than the previously considered event occurs.

When the learning is performed by the learning means 5000, the diagnosis cannot be made by the existing diagnosis rule, or the diagnosis result is erroneous. In these cases, the operator investigates the correct cause of the abnormality and then inputs / outputs the device 500.
Enter the correct error cause from. Then, according to the operator's judgment, the learning means 5000 learns the relationship between the state of the target equipment 1000 at that time and the correct cause of abnormality.

FIG. 4 shows an embodiment of the learning means 5000.
In this embodiment, a neural network (neural circuit model) is used as a learning means. Data necessary for learning and abnormality cause estimation are recorded in the data storage unit 5100. The network is an input layer, as shown in Figure 4.
It has a three-layer structure including an intermediate layer and an output layer. Each neuron in the output layer corresponds to a candidate for an abnormal cause.
The input data to the input layer is the sensor information 211 shown in FIG.
0, 2210 and observation information 2310.

The network learning procedure is as follows.

(1) Creation of teacher data (teacher data creation means 5200) The correct cause of abnormality is read from the data storage means 5100, and the corresponding output neuron number is assigned.
As teacher data, 1.
A data set that gives 0 and 0 to the other output neurons is created.

(2) Creation of input data (input data creation means 5500) Various information shown in FIG. 3 is read from the data storage means 5100, and each value is set to 0 using a predetermined representative value.
The input data 5510 is created by normalizing to the range of 1.0. Non-numerical information (symbol information) is set to 1.0 for the corresponding neuron, and 0 for the other neurons.

(3) Signal transmission in the network (neural network 5400) The input data 5510 is the neural network 5400.
Is input to each neuron of the input layer, weighted by an arbitrarily determined coupling strength between neurons, that is, a weighting coefficient, and output to each neuron of the intermediate layer. The middle-layer neuron transforms the total sum of the input signals by a transformation function such as a sigmoid function and outputs it. The output value is weighted by a weighting factor and input to each neuron in the output layer. The output layer neuron converts the sum of the input signals by a conversion function and outputs an output value 5410, as in the middle layer neuron.

(4) Correction of Weighting Factor (Weighting Factor Modifying Unit 5300) The weighting factor between neurons is modified so that the error between the output value 5410 of the neural network 5400 and the teacher data 5210 becomes small.

The backpropagation method is an example of a specific calculation method for the learning method and the weighting coefficient modification.
For details of the learning method, see `` Learning internal representations
by error propagation ", Parallel Distributed Proces
sing: Explorations in the Microstructures of Cognit
ion, Vol.1, DERumelhart and JLMcClelland (Ed
s.), Cambridge, MA ,: MIT Press, pp318-362
The description is omitted here.

The above-mentioned operations (3) and (4) are repeated until a predetermined number of times, or the output value 5410 and the teacher data 5210.
The learning is repeated until the error between and becomes less than a predetermined value.

(5) Output of learning result When estimating the cause of abnormality based on the learning result, the sensor information obtained is input to the learned network, and the candidate of the cause of abnormality is set in descending order of the output value. .

Since a plurality of outputs from the network are usually obtained, it is difficult to specify which is the true cause of the abnormality. In the present invention, the physical model creating means 70
It is possible to theoretically evaluate the validity of the relationship between the sensor information and the candidate for the cause of abnormality with the physical model created by 00, and select the cause of abnormality that does not conflict as a solution.

As described above, it is possible to learn a new event, but it is necessary to judge whether the learned relationship is a correct relationship or an accidental relationship due to an abnormality of the measuring instrument. It is difficult. In the latter case, an erroneous relationship is learned, so that a similar state will be erroneously diagnosed and appropriate countermeasures cannot be taken.

Therefore, in the present invention, the learning result is analyzed,
By clarifying the physical causal relationship between the measured state of the target equipment and the cause of the abnormality, it is possible to newly register only the correct relationship. Examples will be described below.

First, the causal relationship analyzing means 6000 analyzes the relationship learned by the neural network 5400,
The sensor information most closely related to the cause of the abnormality is specified from the input information.

The neural network holds knowledge as a weighting coefficient matrix between neurons. Therefore, the i-th input value xi of the input layer and the k-th output value y of the output layer
It is known that the strength of the relationship with k can be evaluated by the following formula (“Plant operating rule using neural network”, Transactions of the Institute of Electrical Engineers of Japan, Volume D, Vol. 111, No. 1, 1991,
pp20-28).

C = ΣW _{3k, 2j} · W _{2j, 1i} (sum related to j) (1) where W is a weighting coefficient, subscript 3 is an output layer, 2 is an intermediate layer, and 1 is an input layer. By comparing the values of C using the equation (1), it is possible to specify input information having a strong relationship with the cause of abnormality.

Next, the physical model creating means 7000 constructs a physical model that expresses the relationship between the input information specified by the causal relationship analyzing means 6000 and the cause of abnormality by a combination of physical relational expressions.

FIG. 5 shows a configuration example of the physical model creating means 7000. A device configuration database 7300 in which the device configuration of the target facility 1000, the component configuration of the device, and the types of physical formulas related to the device are registered, a connection relation database 7400 in which the connection relation of the component devices is registered, heat transfer,
A physical formula database 7500 in which physical relational expressions such as force and electricity are registered and a constant database 7600 in which physical property values and constant values used in the physical relational expressions are registered are provided, and a component part searching function 7100 and a model creating function 7200 are provided.
To create a physical model.

The input information specified by the causal relationship analyzing means 6000 is input to the constituent part searching function 7100 as a set 6010 with the cause of abnormality at that time. Component search function 7100
Then, a component that links the input information and the cause of the abnormality is extracted from the device configuration database 7300 and the connection relation database 7400, and the primary model is constructed.

As shown in FIG. 6, the equipment configuration database 7300 classifies the components of each equipment (part) into LEVEL-1 to LEVEL 4 in the descending order of the probability of occurrence of an abnormal state.

When constructing the first-order model, this LE
The constituent elements are extracted according to the order of VEL. Next, when modeling the same components (parts),
Create a model considering the EL components. In this way, by setting the priorities at the time of model creation for each component, the time required for model creation can be shortened without creating an unnecessarily complicated model.

Further, the types of related physical expressions are registered in the device configuration database 7300 (in FIG. 6,
(Displayed as (Eq-F1), (Eq-H1)).

FIG. 7 shows an example of constructing the first-order model in LEVEL-1. As shown in FIG. 6, the components registered in LEVEL-1 are the cooling water tank 1900, the cooling water pump 1950, the water pipe 1510, the radiator 1540,
Oil pan 1530, oil pump 1532, oil feed pipe 1
Since it is 531, these are incorporated as a model. Further, in this case, the temperature sensor 2010 and the liquid level sensor 2020 measure the cooling water temperature and the liquid level in the cooling water tank, respectively.

Next, the model creating function 7200 applies a physical relational expression from the physical expression database 7500 to this primary model to construct a secondary physical model. For example, in a radiator, the following heat balance equation corresponding to the symbol (Eq-H1) is selected. Ta = Q / (c · F) + Tb (2) Here, each variable is defined as follows from the information registered in the connection relation database 7400. Q is the amount of heat given from the radiator to the cooling water per unit time, c is the specific heat of the cooling water, F is the mass flow rate of the cooling water, and Ta and Tb are the outlet temperature and the inlet temperature of the cooling water, respectively.

The cooling water is as follows. The flow rate equation F = ρ · u · A (3) corresponding to (Eq−F1) is selected, and ρ is the cooling water density, u is the average cooling water flow velocity, and A is the water pipe cross-sectional area. In this way, the physical relational expression is applied to the primary physical model to construct the secondary physical model.

The model evaluation means 8000 qualitatively evaluates the constructed relational expression. When no contradiction with the learned relationship occurs, it is evaluated that the physical model is completed, the learning contents and the contents of the model are displayed on the input / output device 500, and confirmation of the operator is requested.

When a contradiction occurs with the learned relationship, or when the operator determines that the model is inappropriate, the model is recreated in consideration of the next LEVEL component. Further, the modification contents of the model may be input from the input / output device 500.

In addition, the model evaluation means 8000 may perform a quantitative evaluation of the model if the required physical property values or constant values are appropriately obtained.

When the model evaluation means 8000 evaluates that the model is completed and the operator confirms it, the diagnosis method correction means 9000 adds the learning content or the constructed physical model as a new diagnosis rule by the operator's judgment. can do.

FIG. 8 shows a display example of learning contents and model contents on the input / output device 500.

Windows 510 and 52 on the upper half of the screen
At 0, the input information having the strongest relation with the cause of abnormality is displayed by a symbol or a number. In the window 530,
Up to three contents of the input information are displayed in order of strong relationship (in order of increasing C value in the equation (1)), and the cause of the abnormality is also displayed.

A schematic diagram 550 of the construction model is displayed on the lower half surface of the screen.
Is displayed, and the physical relational expression used for the model is displayed in the vicinity of each component in the window 540. When the button switch 560 on the screen is clicked using the mouse cursor 570, the screen shown in FIG. 9 is displayed. On this screen, the relationship between the cause of abnormality and the measured sensor information is displayed step by step with reference to the physical relational expression. This makes it easier for the operator to understand the theoretical relationship between the cause of the abnormality and the sensor information.

As described above, in the present invention, the detailed model is created as needed, so that the physical model can be efficiently constructed without taking unnecessary calculation time. Further, since the validity of the learned relationship can be confirmed by the physical model, it is not necessary to learn and register wrong knowledge, and the reliability of diagnosis is improved.

Furthermore, since the physical meaning of the learned new relationship can be grasped, there is an effect that the understanding of the operator is deepened and the applied thinking for other problems can be expected.

[0084]

According to the present invention, since the validity of the diagnosis result is confirmed by the physical model, it is possible to quickly deal with the abnormal state by minimizing the candidates of the abnormal cause obtained from the neural circuit model.

Further, it is possible to know the theoretical causal relationship between the abnormal cause of the equipment or facility and the measured value, the operator's understanding is deepened, and the applied diagnosis to similar cases becomes possible.

Furthermore, according to the present invention, among the newly experienced cases, only truly correct cases can be added as new diagnostic knowledge, and highly reliable diagnostic rules can be added.

[Brief description of drawings]

FIG. 1 is a diagram showing an embodiment of the present invention.

FIG. 2 is a diagram showing a configuration example of target equipment.

FIG. 3 is a diagram showing an embodiment of data processing means.

FIG. 4 is a diagram showing an example of learning means.

FIG. 5 is a diagram showing an embodiment of a physical model creating means.

FIG. 6 is a diagram showing an example of registration of device configuration data.

FIG. 7 is a diagram showing an example of creating a physical model.

FIG. 8 is a diagram showing a display screen example of a learning result and a physical model.

FIG. 9 is a diagram showing an example of a display screen of a causal relationship of learning events.

[Explanation of symbols]

100 ... Measurement data, 200 ... Inspection data, 300 ... Symbolized data, 500 ... Input / output device, 1000 ... Target equipment, 2000 ... Data processing means, 3000 ... Abnormality diagnosis means, 4000 ... Diagnostic result evaluation means, 5000 ... Learning means , 6000 ... Causal relationship analysis means, 7000 ... Physical model creation means, 8000 ... Model evaluation means, 9000 ... Diagnostic method correction means.

Claims (15)

[Claims] 1. An abnormality diagnosing method for detecting an abnormality of a mechanical equipment by inputting at least one of a measured value or an observation result indicating the state of one or a plurality of portions of the mechanical equipment and estimating the cause of the abnormality. Based on the input, it is determined whether the mechanical equipment is in a normal state or an abnormal state according to a preset determination condition, and when it is determined to be an abnormal state, the relationship between the input and the cause of the abnormality An abnormality diagnosis step of estimating the cause of abnormality by knowledge that has been ruled in advance, and in the abnormality diagnosis step, when the input does not correspond to any of the ruled knowledge, or when the estimated abnormality cause is an error A learning step of learning the relationship between the correct abnormality cause and the input, and analyzing the learning result of the learning step to identify an input item having a strong relationship with the correct abnormality cause. The relationship between the causal relationship analysis step, the measurement or observation part of the input item identified in the causal relationship analysis step, and the part that is the cause of the abnormality is constituted by a combination of registered physical or chemical relational expressions. A physical model and a causal relationship evaluation step of comparing the relationship between the input item identified in the causal relationship analysis step and the cause of abnormality are provided, and when there is no contradiction between the two, the physical model is displayed on the display screen. An abnormality diagnosis method characterized by displaying. 2. An abnormality diagnosis method for detecting an abnormality of a mechanical equipment by inputting at least one of a measured value or an observation result indicating the state of one or a plurality of portions of the mechanical equipment and estimating the cause of the abnormality. Therefore, in the diagnostic rule addition method of adding ruled knowledge for estimating the cause of abnormality, based on the input, whether the mechanical equipment is in a normal state or is in an abnormal state according to a preset determination condition. If it is determined to be an abnormal state, the relationship between the input and the cause of the abnormality is preliminarily ruled into an abnormality diagnosis step for estimating the cause of the abnormality, and in the abnormality diagnosis step, the input is ruled. A learning step of learning the relationship between the correct abnormality cause and the input when the knowledge does not correspond to any of the knowledge or the estimated abnormality cause is an error; Analyzing the learning result, a causal relationship analysis step of identifying an input item having a strong relationship with the correct cause of abnormality, a measurement or observation site of the input item identified in the causal relationship analysis process, and a site that is the cause of abnormality. The relationship between the, the physical model configured by a combination of registered physical or chemical relational expression, and a causal relationship evaluation step of comparing the relationship between the input item identified in the causal relationship analysis step and the cause of abnormality. When there is no contradiction in the relationship between the two, the relationship between the input item identified in the causal relationship analysis step and the abnormality cause is ruled as abnormality diagnosis knowledge and added to the knowledge of the abnormality diagnosis step. How to add diagnostic rules. 3. The abnormality diagnosing method according to claim 1, wherein the component equipment information in which the mechanical equipment is classified for each component and registered, and the connection in which the connection order of the component elements registered in the component equipment information is registered. From the relationship information, the measurement or observation site of the input item identified in the causal relationship analysis step, and extract the constituent elements between the site that is the cause of the abnormality, the physical or chemical relational expression corresponding to each constituent element Is selected and the physical model of the causal relationship evaluation step is automatically constructed by a combination of the physical or chemical relational expressions. 4. The diagnostic rule adding method according to claim 2, wherein the component device information in which the mechanical equipment is classified by component and registered, and the connection order of the component components registered in the component device information are registered. From the connection relationship information, the measurement or observation site of the input item identified in the causal relationship analysis step and the constituent elements between the site that is the cause of the abnormality are extracted, and the physical or chemical relationship corresponding to each constituent element is extracted. A diagnostic rule adding method, characterized in that a physical model of the causal relationship evaluation step is automatically constructed by selecting an expression and combining the physical or chemical relationship. 5. The abnormality diagnosing method according to claim 3, wherein a priority is set to a component as component information of the mechanical equipment, and a component having a lower priority is adopted as a component of the physical model. First, the abnormality diagnosis method is characterized in that the physical model is constructed only by the components having the higher priority. 6. The diagnostic rule adding method according to claim 4, wherein priority is set to a component as component information of the mechanical equipment, and the component having a lower priority is adopted as a component of the physical model. The diagnostic rule adding method is characterized in that the physical model is constructed only by the components having the high priority. 7. The abnormality diagnosing method according to claim 1, 3, or 5, wherein a neural network is used to learn the relationship between the input item and the cause of the abnormality in the learning step, and the weight of the network is used. An abnormality diagnosis method, characterized in that an input item having a strong relationship with the cause of abnormality is specified by analyzing a coefficient. 8. The diagnostic rule adding method according to claim 2, 4 or 6, wherein a neural network is used for learning the relationship between the input item and the cause of abnormality in the learning step. A method for adding a diagnostic rule, characterized in that an input item having a strong relationship with the cause of abnormality is specified by analyzing a weighting factor. 9. An abnormality diagnosing device for detecting an abnormality of a mechanical equipment by inputting at least one of a measured value or an observation result indicating the state of one or a plurality of portions of the mechanical equipment and estimating the cause of the abnormality Based on the input, it is determined whether the mechanical equipment is in a normal state or an abnormal state by a preset determination condition, and when it is determined to be an abnormal state, the cause of the abnormality is estimated from the input. Abnormality diagnosis means for storing ruled knowledge, determining whether or not the mechanical equipment is in a normal state by the ruled knowledge, and estimating an abnormality cause; and the abnormality diagnosis means, wherein the input is the Learning means for learning the relationship between the correct abnormality cause and the input when the rule does not correspond to any of the ruled knowledge, or when the estimated abnormality cause is an error; Analyzing the learning result of the stage, a causal relationship analysis unit that identifies an input item having a strong relationship with the correct abnormality cause, a measurement or observation site of the input item identified by the causal relationship analysis unit, and an abnormality cause A causal relationship evaluation means for comparing a relationship between a part and a physical model configured by a combination of registered physical or chemical relational expressions with a relationship between the input item identified in the causal relationship analysis step and the cause of abnormality And a display unit that displays the physical model when no contradiction occurs in the relationship between the two. 10. An abnormality diagnostic device for detecting an abnormality of a mechanical equipment by inputting at least one of a measured value or an observation result indicating the state of one or a plurality of parts of the mechanical equipment and estimating the cause of the abnormality. Then, in the diagnostic rule addition device for adding the ruled knowledge for estimating the cause of abnormality, based on the input, whether the mechanical equipment is in a normal state or an abnormal state according to a preset determination condition, If it is determined that the machine is abnormal, the ruled knowledge for estimating the cause of the abnormality from the input is stored, and the ruled knowledge is used to determine whether the machine equipment is in a normal state or not. In the abnormality diagnosis means for estimating the cause, and in the abnormality diagnosis means, if the input does not correspond to any of the ruled knowledge, or the estimated abnormality cause is incorrect. In the case of, a learning unit that learns the relationship between the correct abnormality cause and the input, and a causal relationship analysis unit that analyzes the learning result of the learning unit and identifies an input item that has a strong relationship with the correct abnormality cause. The measurement or observation site of the input item identified by the causal relationship analysis unit, and the relationship between the site that is the cause of the abnormality, a physical model configured by a combination of registered physical or chemical relationship, Equipped with a causal relationship evaluation means for comparing the relationship between the input item specified in the causal relationship analysis process and the cause of abnormality, if there is no contradiction in the relationship between the two, the input item specified by the causal relationship analysis means and the cause of abnormality And a means for adding to the knowledge of the abnormality diagnosing means by making a rule of the relationship with the above as abnormality diagnosis knowledge. 11. The abnormality diagnosing device according to claim 9, wherein the component equipment information in which the mechanical equipment is classified for each component and registered, and the connection in which the connection order of the component elements registered in the component equipment information is registered. Means for storing relationship information, the constituent device information and the connection relationship information, the measurement or observation site of the input item identified by the causal relationship analysis unit, and the components between the site that is the cause of the abnormality An abnormality diagnosing apparatus comprising: a means for extracting and selecting a physical or chemical relational expression corresponding to each constituent element to automatically construct a physical model of the causal relation evaluating means. 12. The diagnostic rule adding apparatus according to claim 2, wherein the component equipment information in which the mechanical equipment is classified for each component and registered, and the connection order of the component elements registered in the component equipment information are registered. A means for storing connection relationship information, and a component between the measurement or observation site of the input item identified by the causal relationship analysis means from the component device information and the connection relationship information, and the site that is the cause of the abnormality And a means for automatically constructing a physical model of the causality evaluation means by selecting a physical or chemical relational expression corresponding to each constituent element. 13. The abnormality diagnosing apparatus according to claim 9 or 11, wherein the abnormality diagnosing means determines that the input does not correspond to any of the ruled knowledge, or the estimated abnormality cause is erroneous. In some cases, the abnormality diagnosis device is provided with means for inputting the correct state of the mechanical equipment or the correct cause of the abnormality by keyboard input, mouse operation, light pen input, voice input or other input means. 14. The abnormality diagnosing device according to claim 10 or 12, wherein the abnormality diagnosing means determines that the input does not correspond to any of the ruled knowledge, or the estimated abnormality cause is erroneous. In some cases, the abnormality diagnosis device is provided with means for inputting the correct state or correct cause of the mechanical equipment by keyboard input, mouse operation, light pen input, voice input, or other input method. 15. The diagnostic rule adding device according to claim 11 or 13, wherein the abnormality diagnosing means does not correspond to any of the ruled knowledge, or the estimated abnormality cause is erroneous. In this case, the diagnostic rule adding device is provided with means for inputting the correct state of the mechanical equipment or the correct cause of abnormality by keyboard input, mouse operation, light pen input, voice input, or another input method.