A Study of Supervised Machine Learning Techniques for Structural Health Monitoring

We report on work that is part of the development of an agentbased structural health monitoring system. The data used are acoustic emission signals, and we classify these signals according to source mechanisms. The agents are proxies for communicationand computation-intensive techniques and respond to the situation at hand by determining an appropriate constellation of techniques. It is critical that the system have a repertoire of classifiers with different characteristics so that a combination appropriate for the situation at hand can generally be found. We use unsupervised learning for identifying the existence and location of damage but supervised learning for identifying the type and severity of damage. This paper reports on results for supervised learning techniques: support vector machines (SVMs), naive Bayes classifiers (NBs), feedforward neural networks (FNNs), and two kinds of ensemble learning, random forests and AdaBoost. We found the SVMs to be the most precise and the techniques that required the least time to classify data points. We were generally disappointed in the performance of AdaBoost.