"In recent years, there has been a growing interest in diagnosis and prognosis of motors and electrical drives. Effective and accurate diagnosis and prognosis of systems will eventually lead to condition based maintenance, which will decrease maintenance costs and system downtime, improving the reliability of electrical drives. More than 50% of motor failures are due to ball bearings. As such, the area of bearing fault diagnosis and prognosis has attracted a lot of attention in recent years. Although many techniques have been successfully applied for bearing fault diagnosis, prognosis of faults and especially predicting the remaining useful life (RUL) of bearings is a remaining challenge. The main reasons for this are a lack of accurate physical degradation models, limited labeled training data, and the lack of a priori knowledge of the different health states of bearings. There are several factors that contribute to bearing failure, including the mechanical stress of a load and the electrical stress of bearing currents. Due to the intrinsic properties of motors driven by pulse-width modulation (PWM) operation, there are current paths that form from the motor shaft through the races of the bearing and back to ground. These current paths are caused by voltage division interaction with the common mode voltage and stray capacitances within the motor. One type of bearing current, electric discharge machining (EDM) current, causes a significant amount of damage to bearings. The presence of EDM currents causes pitting in the rotating elements of the bearing and ultimately leads to bearing failure. Although this relationship is well known and studied, little work has been done to relate bearing current discharge events to bearing vibrations for failure prognosis. In this work, we propose both computational and experimental approaches for RUL estimation of bearings. In Chapter 2, we present two platforms which were used to accelerate the aging process of bearings. The first, the PRONOSTIA Platform, accelerated bearing degradation via excessive loads, while collecting vibration and temperature data over the course of a run. The second platform is a new test bed we constructed to better understand the relationship between bearing currents, vibrations and failure. This test bed applies an electrical stress on test bearings to induce accelerated aging. Over the course of the experiments, we collect multiple sensor data including current, temperature, and vibration from start to failure in order to correlate current data as well as vibration data to bearing failure. In Chapter 3, we introduce an approach for learning the hidden health states of a bearing from vibration signals. This proposed approach is based on extracting multiple features from sensor signals and identifying change points in the state of the system based on these features. We also propose a framework based on temporal Hidden Markov Model for unsupervised clustering of bearing vibration data in order to identify hidden health states in the data. In Chapter 4, we introduce a data-driven methodology, which relies on both time and time-frequency domain features to track the evolution of bearing faults based on vibration signals. An extended Kalman filter is applied to these features to predict the remaining useful life and to provide a confidence interval to the RUL estimates. Performance of the proposed methods are evaluated on the PRONOSTIA experimental test bed data. In Chapter 5, we propose a computational framework that relates the current discharge events with the evolution of vibration data for a more accurate RUL estimation. We use a current discharge influx event as a trigger to perform RUL estimation on bearings using vibration data, resulting in higher accuracy and efficiency."--Pages ii-iii.
Read
