Anomaly detection in electromechanical systems using symbolic dynamics

The anomaly detection methodology, investigated in this dissertation, consists of two parts: (1) Forward problem - The objective in the forward problem is to learn how the grammar of the underlying system dynamics evolves as the system parameters gradually change. The forward problem is that of learning where the value of a parameter is associated with an anomaly measure. (2) Inverse problem - The inverse problem is that of inferring the system parameters based on the observed asymptotic behavior.; This dissertation deals with both the forward and inverse problems. The performance of this anomaly detection method is compared with that of other existing pattern recognition techniques from the perspectives of early detection of fatigue damage in polycrystalline alloy structures. The experimental apparatus, on which the anomaly detection method is tested, is a multi-degree of freedom mass-beam structure excited by oscillatory motion of two electromagnetic shakers. The evolution of fatigue crack damage at one of the failure sites is detected from symbolic time series analysis of displacement sensor signal. Also, the dissertation quantifies the progression of damage using other mechanical sensors like accelerometers and load cells and damage sensing devices like ultrasonic flaw detectors. Another mechanical motion apparatus is designed and the accelerometer time series data, generated from this apparatus, has been utilized to detect the change in effective mass/moment of inertia of the system components. The dissertation also exemplifies usage of these techniques to an industrial application as described below.; Industrial Application - Critical components such as bearings, seals, and couplings are subjected to unbalanced axial/radial loads and excessive machine vibrations due to shaft misalignment in rotating machinery. The dissertation presents Symbolic Time Series Analysis ( STSA) of bearing acceleration for detection and estimation of parametric changes in flexible disc/diaphragm couplings due to angular misalignment between shafts. The concept is validated on a simulation model, where the dynamic model of a flexible mechanical coupling is subjected to angular misalignment. Patterns of damage evolution are identified from symbolic time series analysis of data sets, generated for multiple torque inputs. Small changes in the coupling stiffness parameter(s) are detected and estimated for different torque inputs via inverse mapping based on ensemble of the statistical patterns, generated by STSA. (Abstract shortened by UMI.)...