Vibration data modeling and design of multivariate EWMA chart for CBM

This study focuses on the cross study of multivariate control chart and condition-based maintenance, which have been extensively studied in isolation but limitedly in an integrated way. Specifically, the multivariate statistical process control charts method is applied to condition based maintenance. My research efforts are divided between analyzing vibration datasets for failure diagnosis and developing a new stochastic model for optimization of maintenance policies.;The main tasks of the failure diagnosis in the rotating machinery are to detect the incipient failure and identify the failure mode or pattern. A novel failure diagnosis scheme for gearboxes was proposed. I used a combination of multivariate time series modeling, dynamic principal component analysis method, and multivariate control chart to implement failure diagnosis. The research results are very appealing in three aspects: First, it provides the whole picture of teeth health condition in one single analysis. Second, it not only reduces the probability of false alarms but also improves the reliability by distinguishing the real alarm pattern from the false alarm pattern. Third, the failure mode of adjacent teeth fracture can be identified by visual inspection from the graph.;A stochastic model was developed determining the optimal policy for monitoring and planned preventive maintenance in a manufacturing process. Specifically, this model integrates the multivariate Exponentially Weighted Moving Average (EWMA) chart and preventative maintenance to minimize the total costs associated with monitoring and maintenance by jointly optimizing the inspection and maintenance policies. The objective is to determine the interval between samples, the control limit, and the multivariate EWMA exponential weight minimizing the expected average cost per unit time. This model can be applied to the other situation when there is a typical warning state.