Early Detection of Failure Precursors Using Symbolic Dynamics, Neural Networks, and Principal Component Analysis

For early detection and monitoring of failure precursors in mechanical transmission couplings, we propose to develop signal processing capabilities that can map patterns in accelerometer data to an anomaly measure. Toward this end, Professor Asok Ray at Penn-State University has pioneered an elaborate mathematical theory based on symbolic time series analysis (STSA), statistical mechanics, and information theory. An anomaly detection algorithm is formulated by applying this novel STSA theory to create a robust statistical pattern recognition technique. For anomaly detection, this STSA technique has been shown to be superior to conventional pattern recognition techniques, such as artificial neural networks (ANN) and principal component analysis (PCA) because it exploits a common physical fact underling most anomalies which conventional techniques do not. This superiority has recently been demonstrated on electrical circuits, fatigue testing machines, and mechanical components undergoing fatigue due to vibrations. The research objectives are: (i) to develop a coupling model where gradually evolving damage phenomena can be introduced, (ii) to formulate and compare real-time algorithms for early detection and monitoring of failure precursors in model simulations based on three principal techniques and their variants - STSA, ANN, and PCA, and (iii) to demonstrate these algorithms on fatigue damage accumulating parts of a vibrating machine experiment.