A comparison of methods for diagnostic decision making

Improving the reliability of heuristic multiple fault diagnosis via the EC-based Genetic Algorithm

“…Our initial test results using their 10 by 10 matrix were very encouraging in terms of reliability for our earlier diagnostic method (in fact, the results from our earlier investigations were good enough to motivate us to pursue this line of research and to incorporate the post-evaluation strategy in the CAP-II prototype [9,19]. We then generated (randomly) additional matrices of the following sizes: 10 by 10, 10 by 12, 10 by 15, and 10 by 20 ( Table 1 shows an example of a 10 x 15 tendency matrix).…”

“…The log is an important tool for reviewing the effectiveness of the diagnostic strategy as well as fine tuning the parameters. To date, the system has exhibited very encouraging results (see [9] and [19] for additional experimental results using our earlier improvement scheme, called post-evaluation). In those instances where unexpected results occurred, typically either corrective action on the highlighted problems (those problems requiring immediate action) caused renewed convergence or the system found an unforeseen relationship in the experiential knowledge.…”

“…In addition, edge recombination [31] and local improvement [32] have been proposed to improve traveling salesman problem results. The post-evaluation scheme (the predecessor of our engineered conditioning operator) has been shown to be effective at improving results of set covering type problems such as multiple fault diagnosis [9,19,29]. However, post-evaluation is a process applied to the final GA produced solution, and has no direct relationship to the internal workings of the GA.…”

“…Of particular importance in using the genetic algorithm is the method used to evaluate each diagnosis. We adapt the relative likelihood of the Probabilistic Causal Model (PCM) developed by Peng and Reggia [3,8] to the genetic algorithm [9,19]. A brief description of the PCM is included here, however, the reader is referred to the work of Peng and Reggia for detailed discussions of the motivation and justification for the PCM.…”

Improvements of Simple Genetic Algorithm in Structural Design

“…[1][2][3][4][5][6][7][8] Genetic algorithms simulate the natural evolution process by three major operations-reproduction, crossover and mutation. A population with speciÿed number of randomly generated individuals constructs the initial designs for GA. Each individual in the population is represented by a binary string which simulates the chromosome of natural species.…”