Model based diagnostics and prognostics of practical industrial systems

In this thesis, the diagnosis and prognosis of single and simultaneous multiple incipient faults in practical industrial systems is demonstrated in two papers. For the purpose of fault detection and failure prediction, a model based strategy is adopted, which involves using an online adaptive estimator. The online estimator estimates the system dynamics. Residual is generated based on the physical system model and the estimated system dynamics. This residual is used to detect the fault. Once the fault is detected, a tunable nonlinear neural network online approximator in discrete-time (OLAD) along with a robust adaptive term is initiated. The OLAD utilizes the parameter update law to learn the unknown fault dynamics of the encountered fault. The robust adaptive term guarantees the asymptotic convergence of the residual. Once the fault is detected, fault isolation and time-to-failure (ITF) prediction are performed by another online approximator. In the first paper, a vapor compression system with a scroll compressor is taken and analyzed for mechanical faults which are refrigeration cycle faults like refrigerant leakage and heat exchanger fouling. In the second paper, a 3 phase squirrel cage Induction motor with a variable load like a vapor compressor is considered to study for electrical faults like the insulation degradation and rotor bar breakage faults. Thus the scheme can be employed to a wide variety of industrial systems and can be used to detect and predict different types of faults --Abstract, page iv