Statistical Approaches for Fault Diagnostics and Root Cause Analysis with Industrial Applications

Abstract

With the advancement of sensors and data storage technology, condition-based maintenance (CBM) in manufacturing industries is becoming an appropriate approach to build a monitoring system. In this thesis, CBM is conducted for two manufacturing systems: multilayer ceramic capacitor (MLCC) stacker and power plant turbine system. A MLCC stacking machine is a core process of defining a quality of products. It is known that unparalleled upper and lower plates in a pressing step might cause MLCC misalignment. A machine health index which can represent status of this unevenness of the plates has been developed. To prove effectiveness of this machine health index, there have been several experiments and its validated algorithm is implemented in a real production system. Since a turbine system in power plants is core components, many diagnosis systems are already installed. Much information related to a power plant maintenance exists in a form of written documents, but these historical records are mostly not computerized. In addition, such information is often electronically stored as a string data format which is not appropriate data type for statistical analysis. Therefore, we propose to develop a knowledge-based expert system for a power plant monitoring system to overcome such limitations of computerization of scattered written information. Furthermore, an algorithm based on the recursive Bayesian estimation is suggested to recommend the most appropriate root cause from multiple observed symptoms of machine fault.