A heuristic method for analysing lifetime data from screening experiments

Designed experiments have been extensively employed to enhance product reliability, often through a two-stage approach. In the initial stage, a two-level screening experiment is conducted to eliminate the factors that are likely to be insignificant, thereby reducing the total number of factors under consideration. The second stage utilizes experiments involving multiple factor levels to optimize the settings of the identified significant factors. The screening stage typically yields datasets that are both limited in size and heavily censored, making standard analysis methods difficult to apply effectively. To overcome this challenge, a heuristic methodology is proposed for analyzing such datasets when formal statistical techniques are either inapplicable or probably yield unreliable results. This method first identifies the best-performing and worst-performing experimental runs based on the response data, and then compares their respective factor levels. A factor is considered potentially significant if the best and second-best runs share the same level for that factor, the worst and second-worst runs share the same level, or if the factor levels of the top-performing and poorly performing runs are different. Advantages of the proposed approach include logical consistency and ease of implementation. The effectiveness of the method is illustrated using two real-world examples.