Learning error recoveries in automated manufacturing by general explanation of observed program augmentations
P.J. Fielding, Frank DiCesare · 1989
It has been found that when developing an automated system, 80% to 90% of the development effort is dedicated to generating automated error recovery routines. Any technique that eases the effort required to generate recovery routines, will directly facilitate the development of automated systems. Other researchers have realized this potential, and pursued solutions independently. This thesis begins by coalescing their work, yielding a perspective on the error-recovery problem that was previously not possible. Unlike the reviewed works, this thesis formally develops and analyzes the error-recovery problem from basic automata theory concepts. It is found that an analogy exists between ideal versus actual programmed manufacturing processes, and the undecidable nature of the program equivalence problem which is implied by Church's thesis. The consequence of this is that the problem of generating a complete set of automatic error recovery routines is an undecidable problem. A restricted equivalence relationship is developed, which leads to a solvable problem. As perturbations about the restricted equivalence are examined, a knowledge structure evolves that is useful for capturing manufacturing programmed processes in a manner that facilitates programmed augmentation. Having developed this knowledge structure, a relational construct is derived that yields useful, general explanations of observed program augmentations made to programs implemented within the constraints of the developed knowledge structure. This developed relational construct allows the concepts of explanation-based learning to be mapped to this domain. To demonstrate the usefulness of these concepts, a programming shell is created which embodies the developed theoretical constructs. A simple example is implemented in the devised shell, clearly demonstrating how an automation programmer's efforts might be leveraged when creating error recoveries in such an environment.