Abstract
A methodology using artificial intelligence for diagnosing structures and predicting failures in mechanical systems is presented. An Artificial Immune System (AIS) able to identify and locate faults with good predictability was developed, having its op- eration based on the Negative Selection Algorithm (NSA). The Negative Selection Algorithm is divided into two steps: Censor phase and Monitoring. In the first step, the algorithm can learn about the normal operation of the system and created a baseline. In the second step, the algorithm evaluates the system data and becomes able to identify patterns different from what has been learned, in other words, a possible failure. The algorithm developed was optimized with the Clonal Selection Algorithm (ClonalG), aiming at fewer data in training. The results obtained suggest that the AIS can learn about the normal system operation using 5% of the available data, being able to diagnose with an excellent safety margin the predictability of a failure and inform where it is located. With the optimization by ClonalG the need for training data is reduced by 50% and the deviation adopted by 70%, without jeop- ardizing the algorithm hit rate. Thus, the algorithm optimized by ClonalG proved to be an excellent tool in the prevention of failures and accidents, presenting general hit rates above 99.90%. The differentials of this work are the high hit rate presented by the AIS, which performs few misclassifications, and the fact that modeling is not necessary. The system is able to learn by itself about the behavior of the data.
Citations
Related Research
