SUMMARY

In a large production environment, failing machinery is not only hazardous but can be extremely costly in lost time and resources. Specifically in rotating equipment, shaft misalignment is responsible for over 70% of issues. This thesis investigates the applicability of a convolutional neural network to bearing accelerometer data in the time domain in its ability to preemptively identify and quantify misalignment. Additionally, to determine the effectiveness of the model in a realistic setting, its performance will be further evaluated on noisy data and data with low sampling rates. The same study will be compared with artificial neural net (ANN) and support vector machine (SVM) models. The resulting model will enable users to determine precisely when to fix equipment as to minimize time spent in maintenance.