When using JMP reliability prediction to predict future returns based on time-series data on production volume and returns, selecting a suitable failure model is key. Failure models can vary across products, models, and versions, and can even change over time. In my experience teaching reliability prediction for industrial machinery, weighing equipment, electronic devices, and automotive parts, partial failures are overwhelmingly common. In the case of total failures due to wear-out failures, the cumulative failure rate and number of returns continue to increase over time. However, in the case of partial failures, the cumulative failure rate plateaus over time and the number of failures converges to zero. The cumulative number of failures also becomes modest. Therefore, determining whether a failure is partial or not is an important issue. While the basis for determining the failure model is the statistics in the model comparison table and the likelihood ratio test, early predictions can make it difficult to determine whether a failure is partial or total. We will also consider the causes of partial failures. The data prepared for reliability prediction can contain missing values or have a small number of failures. There can also be a discrepancy between the time of shipment and the start of operation. However, data with the same time period for shipment and failure must be excluded in order to fit the partial failure model. We analyze the failure period (a univariate of lifespan) through trend analysis of complaint data, and perform reliability predictions for areas of concern.
