Study on fault diagnosis of EMC-CBR (Iv.)

1. System composition

This system is a hybrid system composed of CBR and RBR. The system structure is shown in the following figure. The core of the system is inference machine, reasoning machine according to the user input fault description, can not only transform the fault description into a new case, so as to carry out the operation of the case library, but also can transform the fault description into the fact, the rule base operation. However, the status of CBR and RBR is not equal, the CBR plays the leading role, the RBR plays a supervisory role on CBR, and the inference results are synthesized by the comprehensive mechanism. This system uses the friendly Man-machine interface, after the ordinary user input breakdown description, the system after the inference gives the diagnosis suggestion, and has explained to the suggestion. In addition, in order to facilitate the management of system knowledge (case library, Knowledge Base and fact base), the system can be modified by a knowledge engineer.

2. Case representation and storage

See Bowen "EMC-CBR Fault diagnosis Research (one or two)"

3. Case retrieval

See Bowen "Research on EMC-CBR Fault Diagnosis (III.)"

4. Case Reuse

The main task of case reuse is to revise the conclusions of the retrieved cases. The basic idea is to compare the conclusions of cases in the case library with those of the case that are the most similar to the target cases, if the two are just different parameters of the predicate function, it can be found that there is no need to modify the best match case; If both parameters are different and there is a predicate discrepancy, you need to enable rule-based inference. Finally, the comprehensive mechanism of CBR and RBR is used to revise the conclusion of the best match case by using RBR's conclusion.

5, the preservation of the case

Pattern generalization (schema induction) is performed when the case is saved. The meaning of a pattern generalization is that if the analogy leads to a solution, the generalization of the two similar bodies is performed in order to form an abstract case pattern. In short, it is to look for similarities in two similar bodies and then to abstract and generalize them.

The process of case preservation can be regarded as a process of inductive learning, and many inductive learning methods can be applied to case preservation, such as producing a pre test method (INDUCE1.2).

6, the maintenance of the case

When psychologists study the mechanism of memory, they put forward the famous theory of forgetting curve, that is, information that is not used in the long run will be forgotten gradually. In case-based reasoning, the maintenance of case libraries is important, and the cases that have not been used for a long time must be deleted, because such cases are likely to be noise cases, such as misdiagnosed cases, misjudged cases, and so on. The AHA recommends that you create a matching record for each case in the case library, and you can delete those noise or unwanted cases to maintain the validity of the case library. AHA presents the case refinement algorithm IB3, expressed as follows [51]:

Input: Case library s and test case set T

Output: Refined case Library S '

Process:

(1) Choose a test case T from the test case set T;

(2) Try to find a case s in the case library S, which has the best match with T and its similarity exceeds a certain threshold value;

(3) If no such s is found, go to (8);

(4) If it is confirmed that T and S belong to the same class, turn (9), otherwise the T plus the mark of the Class C to which it belongs is stored in the library s;

(5) From the library s to find all such cases si, they and T similar degree is not more than s and T similar degree, all these si matching records to deduct points;

(6) Remove all cases where the record score is less than a specified value from the library s;

(7) If the test set case has been used up, then the algorithm is over, otherwise turn (1);

(8) Choose a case from the library s, go to (4);

(9) From the library s to find all of these cases si, and their similarity to T is no less than the similarity between S and T;

(10) Go to (7).

7. Rule-based Reasoning

rule-based Inference (RBR) subsystem is a simplified expert system. The RBR subsystem consists of fact base and rule base, in which the expert rules extracted from expert experience are stored in the rule base to produce representation, and the fact base is the fact that the inference machine extracts from the user's input through the Man-machine interface, which is represented by propositional logic.

For example, the rules related to the case6 discussed above are "the use of conductive rubber, gaskets and other shielding materials, not only to ensure good electrical conductivity on the contact surface (contact surface to remove all paint), but also to ensure a certain amount of compression." Can be expressed as follows by a production rule:

Shielding (product, Conductive_tape) ∨shielding (product, Condutive_gasket) => Wipe_off (interface, lacquer) ∧ ( Compress (Conductive_tape) ∨compress (conductive_ gasket))

RBR reasoning uses uncertain reasoning. Uncertain reasoning includes uncertainty of precondition, uncertainty of conclusion and uncertainty of inference process. According to the characteristics of EMC knowledge, it is necessary to comprehensively utilize these three uncertainties in order to simulate the inference process of electromagnetic compatibility experts more accurately. The general representation method for uncertain inference is [20]:

IF [(P1, F1, T1) and (P2, F2, T2) and ... and (PM, FM, TM)]

THEN [(Q1, G1, S1), (Q2, G2, S2) ...)] With CF R

Among them, Pi represents the fuzzy prerequisite, QI, the fuzzy conclusion and action,CF(R) for the credibility of, FI is the expression of the probability distribution of the former pieces, TI is the former true letter degree, GI is the probability distribution of the conclusion, S1 is the conclusion of the degree of certainty.

For the front part, we can adopt the expression scheme: P is fuzzy proposition, X is object name, a is the attribute name of X, D is the deterministic state expression, is the corresponding deterministic proposition of P, and is the uncertainty measure of proposition.

For example, a vague proposition is:

If the inductive coupling (confidence 0.9) and the circuit frequency is high (fuzzy number 0.8), then the coupling is capacitive coupling (confidence 0.7).

Then the proposition can be expressed as:

IF [(p1,0.9) and (p2,0.8)] THEN [(Q1, 0.7)] With CF R

of which,,,

, R is the credibility of the whole proposition calculated on the basis of uncertain inference theory.

8. The comprehensive mechanism of CBR and RBR

Enable RBR when the similarity of source cases in the target case and CBR case Library is lower than the threshold value. If the result can be deduced by using RBR, the conclusion of RBR is taken as the final diagnosis result, and the conclusion of CBR is submitted to the user together as a suggestion; if due to incomplete regular RB