A Fault Diagnosis System for Rotary Machinery Supported by Rolling Element Bearings

Abstract

The failure of rolling element bearings is one of the foremost causes of breakdown in rotary machinery. So far, a variety of vibration-based techniques have been developed to monitor the condition of bearings; however, the role of vibration behavior is rarely considered in the proposed techniques. This thesis presents an analytical study of a healthy rotor-bearing system to gain an understanding of the different categories of bearing vibration. In this study, a two degree-of-freedom model is employed, where the contacts between the rolling elements and races are considered to be nonlinear springs. The analytical investigations confirm that the nature of the inner ring oscillation depends on the internal clearance. A fault-free bearing with a small backlash exhibits periodic behavior; however, bearings categorized as having normal clearance oscillate chaotically. The results from the numerical simulations agree with those from the experiments confirming bearing’s chaotic response at various rotational speeds. Bearing faults generate periodic impacts which affect the chaotic behavior. This effect manifests itself in the phase plane, Poincare map, and chaotic quantifiers such as the Lyapunov exponent, correlation dimension, and information entropy. These quantifiers serve as useful indices for detecting bearing defects. To compare the sensitivity and robustness of chaotic indices with those of well-accepted fault detection techniques, a comprehensive investigation is conducted. The test results demonstrate that the Correlation Dimension (CD), Normalized Information Entropy (NIE), and a proposed time-frequency index, the Maximum Approximate Coefficient of Wavelet transform (MACW), are the most reliable fault indicators. A neuro-fuzzy diagnosis system is then developed, where the strength of the aforementioned indices are integrated to provide a more robust assessment of a bearing’s health condition. Moreover, a prognosis scheme, based on the Adaptive Neuro Fuzzy Inference System (ANFIS), in combination with a set of logical rules, is proposed for estimating the next state of a bearing’s condition. Experimental results confirm the viability of forecasting health condition under different speeds and loads.