A Design Based Approach to Constructing Computational Solutions to Diagnostic Problems

Troubleshooting problems in real manufacturing environments impose constraints on admissible solutions that make the computational solutions offered by “troubleshooting from first principles” and the conventional experience based expert systems approaches infeasible. In this paper we present a computational theory for a solution to these problems that is based on the Principle of Locality and exploits the domain specific weak methods of troubleshooters and debugging knowledge of the designers. The computational theory is evaluated by generating focus of attention heuristics for a moderately complex digital device. 1. Computational Problem We are interested in finding computational solutions to diagnostic problems as they occur in real manufacturing environments [Johnson 891. This class of problems introduces two constraints on any computational solution: 1) the devices are large and complex, which precludes any solution that relies on a complete simulation or on enumeration of the various fault propagation paths through one or more abstractions of the device (e.g., [Genesereth 84],[de Kleer 871, [Reiter 871) and 2) the life-cycle of these devices is short, which makes any solution (e.g., [Freiling 851) that relies on trouble-shooting knowledge specific to a particular device unfeasible. The first constraint suggests that any computational solution to this class of problems should have a uniform mechanism for handling the order of complexity of these devices. The second constraint precludes the use of any troubleshooting knowledge that is compiled for a specific device as an adaptation to the task of troubleshooting that device. 2. Computational Theory The task of troubleshooting complex large-scale devices can be viewed as a two step process: 1) Determine an appropriate search space in which the fault is “local” [Davis 841 and 2) Apply weaktroubleshooting methods on this space to locate the fault. A search space is defined by a specification of appropriate ‘pathways of interactions” that are exploited by a weak troubleshooting method. Although the definition of our computational problem precludes the use of device-specific troubleshooting knowledge, device-independent but domain-specific troubleshooting methods [Reed 881 developed by troubleshooters can be a valid component of a computational solution. However, a solution that relies solely on these methods, given the constraints of the computational problem is not adequate. There are two difficulties with such a solution: 1) Domain specific weak methods are used by troubleshooters only when the device specific heuristics fail; this means that the best performance one can expect from a solution that relies solely on these methods would be that of a troubleshooter’s performance on a new device. 2) The locus of search spaces that can be used by troubleshooters is restricted by device representations available in their task environment. The representations of a device available to troubleshooters in real manufacturing environments, however, often consist of just the physical device itself (which leads to the used of a search space based on physical pathways of interaction) and at most one other level of abstraction (such as schematic for digital designs). This means that any solution that relies solely on weak-methods of troubleshooting will have the restriction that it can only work for faults that are ‘local’ in search spaces that can be derived from these representations. A large class of interesting faults are not ‘local’ in a search space derived from the physical representation of a device [Davis 841. Another source of knowledge that is admitted by the definition of the computational problem is that of designers conducting the task of debugging, which though different from troubleshooting, has components that are similar to the troubleshooting task. Designer’s task environment has a rich variety of representations of the device. In any design process, these consist of various by-products of the design. For example, in the case of digital hardware VOLOVIK ET AL. 5 17 From: AAAI-90 Proceedings. Copyright ©1990, AAAI (www.aaai.org). All rights reserved.