Robust parameter design for constrained randomization lifetime improvement experiments

Several process parameters affect product reliability. Traditional reliability improvement methods primarily focus on maximizing product lifetime, often overlooking the variation in product lifetime. Manufacturers, however, aim to produce products with minimal variations in their performance. Robust parameter design offers an effective strategy to help manufacturers enhance product reliability while reducing process variations. In practice, reliability experiments frequently involve constrained randomization due to the selection of specific experimental protocols. This study proposes a framework to achieve robust product reliability under a constrained randomization experiment. To consider random effects, we develop a Bayesian method-based Weibull non-linear mixed model for the lifetime response. Important factors are identified according to Bayesian posterior credible intervals. Subsequently, we propose an integrated multi-objective optimization model to determine the optimal factor levels. This model simultaneously considers minimizing the total costs of the manufacturer, maximizing the product lifetime, and minimizing the lifetime variance. An industrial thermostat experiment is conducted to validate the proposed method. Compared to existing approaches, the proposed method demonstrates superior performance in reducing variance and total cost, particularly for long warranty periods. Finally, we discuss the practical implications of the optimal solutions for manufacturers, finding that variations remain tolerable within a certain range.