Stationary and non-stationary process condition monitoring and fault diagnosis and its application to drilling processes

In this research, a new multiple fault classification method has been developed. It is based on pairwise linear discriminant function method. Optimal linear discriminant functions have been used. Moreover, a recursive algorithm has been developed for on-line training and updating the optimal linear discriminant function. It makes the fault classification model easier for real application. Both simulation results and experimental results show the superiority of this method over conventional methods for multiple fault classification.; Transient state machine condition condition monitoring has also been addressed. A non-stationary signal analysis approach based on Choi-Williams time-frequency distribution analysis and singular value decomposition has been developed for the transient state condition monitoring. This signal analysis approach is particularly suitable for automatic non-stationary signal change detection. Satisfactory simulation results have been obtained, and the experiments of drilling process, machining chatter, and bearing failure monitoring show that it is promising for the automatic transient state condition monitoring.; As an application, the multiple spindle drilling process condition monitoring and fault diagnosis have been studied. The condition monitoring is to detect normal and abnormal drilling conditions, and the fault diagnosis is to locate a worn drill position. In this research, a low-cost and minimum number of sensor scheme have been adopted. Namely, a vibration sensor has been used for the monitoring and diagnosis. A {dollar}chisp2{dollar} test approach for vibration signal change detection has been applied for the condition monitoring. For the fault diagnosis, features extracted from the vibration signal have been used, and the new multiple fault classification method has been implemented.