Reliability and Clustering Techniques for Inspection Optimization of Large Populations

Abstract

This dissertation proposes a methodology for optimizing inspection schedules of large heterogeneous populations, by combining clustering analysis, reliability analysis, and nonlinear optimization techniques. Due to limitation of resources, only a small proportion of the population can be inspected. The proposed methodology first identifies the critical samples with the highest likelihood of failing through clustering analysis; then those critical samples’ inspection schedules are optimized with the purpose of maintaining or exceeding the minimum target reliability level while minimizing inspection costs. The clustering analysis is able to handle both numeric and nominal features. A detailed illustrative example is presented to demonstrate the method’s practical application to inspecting railroad wheels. A general methodology for rolling contact fatigue life prediction under a stochastic loading process is used to calculate the reliability of the critical samples. Then a reliability-based inspection schedule optimization technique is developed for the critical samples, based on various costs and scenarios. The return on investment is also calculated for the proposed methodology.