Heuristic intelligent method for machine maintenance and process optimization

Condition monitoring of rotating machineries is a challenging topic across various industries. Through this studies, various signal processing techniques with parameter optimization algorithm are examined. The complex wavelet and empirical wavelet were first examined to extract fault signature. An alternating parameter tuning scheme is proposed to automatic generate an optimal set of parameters to be used in wavelets. The signal-to-noise is improved by using the proposed algorithm in comparison to the traditional wavelet. Then the dictionary learning algorithm is studied by incorporating different parameter updating algorithm. More specifically, the weighted least square and the unscented Kalman filter are implemented to update the dictionaries in order to accommodate the change in fault signal to achieve the optimal diagnostic accuracy. A tight clustering and deep learning based bearing degradation classification methods are proposed to monitor the health condition of bearings in order to make decision for maintenance. The optimized CEEMD using the bootstrap resampling algorithm successfully extract the degradation trend of bearings. It can be used as a preprocessing tool for different classification algorithm. Significant improvement in classification accuracy and a smaller number of features are achieved. The second part of the study includes using machine learning algorithm with physics imbedded and deep convolutional network. The Bayesian optimization algorithm is implemented to tune the parameters within the neural network without performing an exhaustive search. Future research study focusing on possibly improvement for diagnosis and implementation of the deep convolutional network in machine prognosis are discussed.Ph.D