| In this research, we develop an innovative hierarchal approach to maintenance planning in complex, automated manufacturing systems. Strategic, tactical, and in-process planning are addressed.;At the strategic planning level, we develop an analytical methodology, based on Mean Value Analysis, that provides the maintenance planner with a capability to rapidly and accurately determine optimal preventive maintenance intervals considering production costs and interrelationships associated with machine breakdown.;The algorithm produced consistent estimates within 5% of true system performance over a wide range of Weibull nonhomogeneous failure distributions. In addition to development and validation of the algorithm, an important contribution of the current work has been the quantification of model performance through response surface methods, permitting users of the model to estimate expected error levels for particular applications.;At the tactical planning level, we develop a predictive model which utilizes proportional hazard modeling to analyze machine condition. The predictive maintenance system determines whether a machine failure appears to be pending, and if so, schedules a repair and determines the optimal priority level to assign.;The predictive model demonstrated a remarkably good capability to identify repair requirements in order to prevent failures. In the majority of datasets tested, 100% of failures were prevented. Robust performance was demonstrated across a variety of failure characteristics.;At the in-process level, we develop an innovative neural network system which actively performs machine diagnostics and discovers trends in sensory data. Upon confirmation or correction of a diagnosis, the network can adapt to incorporate this new knowledge.;We have performed an in-depth analysis and testing of the performance of the neural network as a diagnostic tool for machining equipment. This analysis led to the development of a number of preprocessing techniques, network modifications, and a knowledge extraction procedure which improve the network's diagnostic capabilities. The network is robust to noise, can generalize from training examples, has excellent classification accuracy, and rapid, real-time operation.;The three levels have been integrated in a hierarchal planning system with adaptive capability. The system is capable of performing consistent optimized automated maintenance planning in complex manufacturing environments. |
