JPH03111938A - Diagnostic support device - Google Patents

Abstract

PURPOSE:To improve an inference processing speed and to easily execute the maintenance control of a knowledge base by dynamically forming the combination of failures on a model to be diagnosed by using a rule module and a control rule corresponding to the model and a diagnostic process. CONSTITUTION:An inference means 4 forms the combination of failures on the objective model 2 by the control rule of rule module inference control system 3 in a production system 1 and applied data. Elements which can not be cause candidates are removed from the failure combination formed on the model 2 by a rule corresponding to inspection data. The residual cause candidates are diagnosed by a rule corresponding to the input of parts data by using the knowledge of a parts level, and if the candidate is abnormal, the diagnosed result is outputted. When the candidate is normal, the candidate is removed from the failure combination on the model 2.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention This invention relates to a diagnostic support device that automatically identifies faulty parts of machines, devices, etc.

Prior Art and Problems of the Invention In conventional diagnostic support devices of this type, IF-THE
The cause is determined using the so-called production rule of N format.
A knowledge base was created that expressed the causal relationships between results using a tree structure, and the knowledge base was explored using input data to perform diagnosis.

However, with conventional diagnostic support devices using such production systems, the exploration space expands as knowledge increases, and it becomes difficult to maintain logical consistency, resulting in problems with processing speed and maintainability of the knowledge base. and development efficiency will decrease.

SUMMARY OF THE INVENTION An object of the present invention is to provide a diagnostic support device that solves the above problems and improves processing speed, maintainability of a knowledge base, and development efficiency.

Means for Solving the Problems The diagnostic support device according to the present invention is modularized based on a target model in which a diagnostic target is broken down into a function that satisfies a certain performance and a group of parts that constitute that function, and a diagnostic process by an expert. A set of faults is dynamically generated on the target model using the rule modules, control rules, and data provided, and local inferences are generated for each rule module. and means for repeatedly identifying failed parts.

By dynamically generating a fault set on the target model using the rule module and control rule according to the model to be diagnosed and the diagnosis process, local inference based on data is possible, and the speed of inference processing is increased. This allows for easier improvement and maintenance of the knowledge base.

Example Hereinafter, the present invention will be described in detail with reference to the drawings.

FIG. 1 shows the overall schematic configuration of the diagnostic support device. This device uses a computer and includes a production system (1), a target model (2), an inference control system (3), an inference means (4), and an interface (5).
It is composed of

A machine, device, etc. that is the object of diagnosis is decomposed into functions that satisfy a certain performance and parts that constitute those functions, and a target model (2) is created in the computer's memory. That is, knowledge (static knowledge) regarding the structure and function of the object is expressed using the object model (2). The target model (2) is
For example, as shown in FIG. 2, it has a hierarchical structure based on the structure and function of the object. In Figure 2, 0
BJECTS connects the object to LINE-1, LINE-
2.・・・・・・・・・LINE-n is the system, un
its-1, units-2, ......, u
nlts-1 is a unit (function), part-1, p
art-2, . . . , part-n indicate parts. Further, XXX is a slot (variable) to which a cause candidate is assigned when the failure cause is a candidate, and to which cause elimination is assigned when it is not a candidate. In addition,
In this embodiment, the target model (2) is roughly divided into two parts: a superordinate concept and a subordinate concept. A superordinate concept is a rough classification of objects by system and function, and a subordinate concept is a detailed representation of the parts that make up the superordinate concept using an object structure. Furthermore, these classifications need to appropriately express the point of view of experts in abnormality diagnosis.

Rules for diagnosis are modularized for each expert's diagnostic process (thought process), and this modularized rule module is used as a production system (1).
Stored in the computer's memory in the form of a program. In other words, in order to realize the expert's thought process in abnormality diagnosis, the expert's thinking (dynamic knowledge) is modularized using production rules in the 1F-THEN format, and this is converted into an expert's thinking process that represents the expert's dynamic knowledge. Use as a model. The rule module includes a phenomenon input rule module (10) for inputting phenomena, which will be described later, a cause candidate rule module (11) for automatically generating a failure set, a confirmation item rule module (12) for inputting investigation data, and a failure module. Cause candidate exclusion rule module (13) for collective evaluation, cause pursuit rule module (14) for inputting parts data
and a diagnostic rule module (15) for diagnosing a set of failed parts. Each module is composed of OR combinations based on production rules. When modularizing rules, it is important to increase the independence of the modules as much as possible to increase maintenance management and development efficiency of the knowledge base. To this end, it is important to make each module into a module that performs one unique function. It is necessary to minimize the degree of coupling of each module. The former was achieved by modularizing the classification based on the expert's thought process, and the latter was achieved by connecting independent modules via necessary data on the target model (2).

The inference control system (3) is a production system (
1) and control rules for controlling inference by the target model (2), and the control rules are stored in the computer's memory in the form of a program. The rule module of the production system (1) is controlled by control rules in accordance with the expert's diagnostic process. The inference control system (3) is a control function that prioritizes the execution of rules during inference execution and resolves conflicts. It is also a means of directing the focus of reasoning towards the most important rules based on their importance at a particular point in time. The inference control system (3) performs the following two controls. First, each rule module of the production system (1) is controlled according to the expert's diagnosis process. In this device, each independent rule module of the production system (1) is connected via the necessary data on the target model (2), so certain data is input to one module. Otherwise, other modules referencing that data will be activated, and the flow of the diagnostic process will become uncontrollable. The above-mentioned control is performed in order to suppress the activation of other modules and prevent the focus of inference from being diverted. In addition, the production rules are controlled based on the expert's way of thinking within the single rule module of the production system (1).The expert makes a diagnosis by prioritizing related data to verify the diagnosis target. In this device, a single module is configured by OR combination of production rules, so the firing order of rules within a single module is arbitrarily determined.
Not sure. Therefore, as it is, it is impossible to make a diagnosis based on the logical thinking order of experts. Therefore, the order in which this related data is used is controlled as described above.

The inference means (4) uses the rule module of the production system (1), the control rules of the inference control system (3), and the provided data to infer the target model (2).
), and identifies faulty parts by repeating local inference for each rule module, and the program for this purpose is stored in the computer's memory.

The interface (5) is for data input/output and graphic display between the inference means (4) and the user, and includes a keyboard, a graphic display, etc.

Next, the operation of the above diagnostic support device will be briefly explained with reference to the principle explanatory diagram of FIG.

First, the target model (
2) Generate above (generation of fault set). The elements of the cause candidate set at this time are based on a rough classification such as the system or unit that constitutes the object. Next, from the fault set generated on the target model (2), elements that cannot be cause candidates are eliminated (review of the fault set) according to rules according to the investigation data. That is, each element in the fault set generated as described above is examined based on the input investigation data, and those with a low probability of being the cause are deleted from the fault set.

Then, according to the rules according to the part data input, the remaining cause candidates are diagnosed using part-level knowledge, and if it is abnormal, the diagnosis result is output, and if it is normal, it is a set of faults on the target model (2). (diagnosis of failed parts set).

Such reasoning is controlled by reasoning means (4). Further, the rule module of the production system (1) for inference is controlled by the inference control system (3).

Next, the above operation will be explained in more detail with reference to the flowchart shown in FIG.

First, the user manually inputs the phenomenon (step 1)
. When the process starts, the phenomenon input rule module (10) is activated by the control rule, and a question prompting the user to input phenomenon data is displayed on the display, so the necessary data is manually entered.

The input phenomenon data is then written into the target model (2).

When the phenomenon input is completed, the cause candidate rule module (11) is activated by the control rule, and a failure set is automatically generated on the target model (2) based on the input phenomenon data (step 2). The cause candidate rule module (11) determines which elements are to be cause candidates for each phenomenon, and when phenomenon data is input, it creates rules for automatically generating a failure set corresponding to that phenomenon. is selected. Then, according to this rule, a cause candidate is assigned to XXX of a certain element, a cause exclusion is assigned to XXX of the remaining elements, and a failure set is automatically generated on the target model (2). For example, as shown in FIG. 5, unlts-1 and units-2 on the target model (2)
Cause candidate is assigned to XXX of , and cause elimination is assigned to XXX of other untts.

When the automatic generation of the failure set is completed, the user manually inputs investigation data (confirmation items) (step 3). The investigation data is used to narrow down cause candidates. When proceeding to step 3, the control rule activates the confirmation item rule module (12), and questions for inputting investigation data necessary for verification are displayed on the display. Because it is done,
Manually enter the required data. Note that these questions are automatically selected in accordance with the elements of the failure cause candidate set, and unnecessary questions are not asked. Furthermore, all input data is written to the target model (2).

When the input of investigation data is completed, the control rule activates the cause candidate exclusion rule module (13), and the failure set on the target model (2) is reviewed based on the input investigation data (step 4). The cause candidate exclusion rule module (13) specifies the rewriting of XXX in the element for each investigation data, and when investigation data is input, the control rule determines the failure set estimation corresponding to the investigation data. Then, according to this rule, XXX of the necessary element is rewritten and the fault set on the target model (2) is reviewed.For example, as shown in (2
) above in units-2 is rewritten from cause candidate to cause exclusion.

When the review of the fault set is completed, it is checked whether the fault set generated on the target model (2) contains an element (step 5). If there is no element, the process is terminated, and if there is an element, Manual input of component data related to the failed element is performed (step 6). The parts data is for diagnosing the final set of failed parts, and there are rules for inputting parts data for each element. Proceeding to step 6, the cause investigation rule module (■4
) will be activated, only the rules for inputting part data of the element that is the cause candidate will be selected, and a question for inputting the part data necessary for verification will be displayed on the display, so enter the necessary data. Enter manually. For example, p
When data A of art-1 and data B of part-2 are selected, as shown in Fig. 7, the target model (2)
Data A is allocated to part-1 above, and data B is allocated to part-2.

When the input of the component data is completed, the diagnosis rule module (15) is activated according to the control rule, and diagnosis is performed according to the rule for diagnosing the set of failed components based on the input component data (step 7). If normal,
The process is completed, and if there is an abnormality, the diagnosis result is output (displayed) on the display (step 8).

The above device uses a target model in which the diagnostic target is broken down into functions and parts that make up the functions, and a rule module in which the rules for diagnosis are modularized based on the diagnostic process of experts. Base maintenance management is easy. Furthermore, since such rule modules and target models are connected by data, local judgments can be made for each rule module, increasing the inference processing speed.

Effects of the Invention According to the diagnosis support device of the present invention, as described above, it is possible to improve the inference processing speed and facilitate the maintenance and management of the knowledge base.

[Brief explanation of drawings]

FIG. 1 is an overall schematic configuration diagram of a diagnostic support device showing an embodiment of the present invention, FIG. 2 is an explanatory diagram showing the configuration of a target model, FIG. 3 is an explanatory diagram of the principle, and FIG. 4 is an example of operation. The flowchart shown in FIG. 5, FIG. 6, and FIG. 7 are explanatory diagrams showing a part of the target model. (1)...Production system, (2)...Target model, (3)...Inference control system, (4)...
・Inference means, (5)...Interface, (10)
(11) (12) (13> (14) (15)...
・Rule module. that's all

Claims (1)

[Claims] A target model that breaks down the diagnostic target into a function that satisfies a certain performance and a group of parts that make up that function, multiple rule modules that are modularized based on the expert's diagnostic process, and controls that control these rule modules. A diagnostic support comprising a rule, and a means for dynamically generating a fault set on the target model using the rule module, control rule, and provided data, and identifying a faulty part by repeating local inference for each rule module. Device.