The diagnostic step of the error recovery process in manufacturing systems is discussed, and an algorithm for constructing and pruning a failure propagation tree is formalized. The methodology of building of failure propagation tree is proposed. This methodology is based on a hierarchical process plan, a precondition-postcondition model of manufacturing operations, and a classification of the failure reasons. The hierarchical operation model embodies the process plan, and the functional and structural classifications for each failure reason of every operation in the model. Based on the hierarchical operation model, the final failure propagation tree is generated or expanded by linking the operations in the model. The expansion process is accomplished by iteratively applying the expansion rule and priority rule to the operations in the hierarchical operation model.>

The authors focus on the diagnostic step of the error recovery process in manufacturing systems and develop theories for evaluating the diagnosability of failure knowledge in manufacturing systems. The theories are based on the action-error-feature data set, described previously by the authors (1989). The definition of one-error diagnosability and the necessary and sufficient conditions for multiple-error diagnosability are given. A simple matrix operation, called covering analysis, for testing the conditions is also presented. The methodology provides a qualitative measure for the embedded diagnostic knowledge in terms of the ability to diagnose the number of multiple simultaneous errors in the system. This provides the diagnosability of the model and the feasibility of the computation, because the covering analysis is simple and at most two analyses are required to determine multiple-fault diagnosability. A discussion of the results obtained and simple examples to illustrate those results are presented.>

The authors describe an approach using machine learning for developing heuristics to be used in diagnosis for automated error recovery in manufacturing systems. This approach can automatically generate diagnostic heuristics for error hypotheses. The approach is based on the integration of set covering and explanation-based learning. In this way, the diagnostic heuristics can be automatically generated and the competing errors can be narrowed down, if not identified. In addition, the set covering concept is potentially useful for diagnostic problems, since it provides a solution to the problem of multiple simultaneous errors. The set covering theory and the theoretical development based on the approach for diagnosing the error and for automatically generating diagnostic heuristics are given. The authors also propose an approach for future research to extract discriminating knowledge for the competing errors when set covering is incapable of identifying the error. A discussion of results thus far obtained, accompanied by simple examples to illustrate those results, is presented.>