| The continuously growing demand for improved functionality and reliability results in ever-growing complexity in engineering systems such as manufacturing systems and medical imaging systems. The unprecedented complexity makes system monitoring, diagnosis, and control very challenging engineering problems. On the other side, due to the rapid development of cyber infrastructure and sensing technology, an abundance of data from engineering systems is now readily available. The available data can be roughly put into two categories: (i) the continuous measurement data such as the dimensional measurement of a machined product, system vibration signal, and the tonnage signal from a forging process; and (ii) the discrete event data such as the occurrences of defects on the surface of a product, various logic events that are programmed and triggered in an automation system, and the event logs in computer software systems.; The data rich environment provides great opportunities to develop new fundamental industrial engineering (IE) tools for effective fault management. Targeting on the urgent need and the emerging opportunity, the research has been focusing on the development of rigorous signature-driven statistical tools to model and analyze the data gathered from a vast array of diverse and interrelated sources for fault monitoring, diagnosis, and prediction purposes.; In more details, two aspects have been focused on in this research: (i) a robust signature matching methodology for single and multiple variation sources identification in manufacturing; and (ii) a generic monitoring technique for time between events and a model building methodology with respect to failure events using the Cox Model driven by discrete event-signatures in services and maintenance.; The research seeks to advance fundamental knowledge in monitoring and diagnosis of complex engineering systems by fully exploiting the data-rich environment. The research is interdisciplinary in nature with the integration of advanced statistical modeling methods for both discrete categorical variables and continuous variables, physical knowledge of the system, and computer science, which leads to a new scientific basis for methodology development. The proposed methodology can achieve systematic fault detection, root cause identification, and failure prediction and possesses wide applicability to various engineering systems. |
