Modeling, fault diagnosis and prognosis under the Zero Defect Manufacturing paradigm

Abstract

This thesis addresses several problems that may arise during the development of Zero Defect Manufacturing strategies. First, we have proposed variation propagation models for vices and self-centering three-jaw chucks using the Stream-of-Variation methodology. We have developed a methodology to adjust linear input-output models using physical-based models, process data and engineering knowledge. We have also proposed a procedure to detect and isolate faults online in a process with the objective to reduce the number of required measurements, using models from the process plan. We have also addressed the problem of predicting the remaining useful life of cutting tools. First, we have proposed a methodology to estimate the surface roughness of the processed parts when those measurements are unavailable, using indirect measurements and a steady-state Kalman filter observer. Additionally, we have proposed a prognosis methodology to predict the optimal moment to replace the cutting tool using an adaptive recursive least squares algorithm.