UC San Diego

The purpose of this study is to use vibration response data to inspect a certain safety mechanism owned by the Department of Energy. These mechanisms are required to have a long accurately predicted service life. However, these mechanisms are permanently enclosed in metal and installed in a higher level assembly hindering inspection. This thesis considers observing the vibrations output from the mechanism by applying the structural health monitoring process. Features that correlated with the change in health of the machine (basic statistics and model parameters) were extracted from the measured vibrations and compared for detection performance. Specifically, failure modes were introduced into a prototype of the device, and supervised learning (involving training data) was applied to assess the features' ability for detection and localization. Performance results were quantified by using the Mahalanobis distance measure to determine how much the features have changed. This distance was used in a receiver operating characteristic curve to compare damage detection results. It was discovered that the use of this structural health monitoring process enabled accurate discrimination of the mechanism as being healthy or unhealthy. A side study of sensor location vs. damage detection indicated which locations sensors were most effective in measuring damage. The principal conclusion was that the structural health monitoring process should be implemented by the department of energy on their mechanism. The benefits include detecting damage before it is severe enough to cause a failure, the ability to locate damage, and a possible reduction of unnecessary inspection labor