Adaptive Warranty Prediction for Highly Reliable Products

Field return rate prediction is important for manufacturers to assess the product reliability and develop effective warranty management. To get timely predictions, lab reliability tests have been widely used in assessing field performance before the product is introduced to the market. This work concerns warranty prediction for highly reliable products. But, due to the high reliability associated with modern electronic devices, the failure data in lab tests are typically insufficient for each individual product, resulting in less accurate prediction for the field return rate. To overcome this issue, a hierarchical reliability model is suggested to efficiently integrate the information from multiple devices of a similar type in the historical database. Under a Bayesian framework, the warranty prediction for a new product can be inferred and updated as the data collection progresses. The proposed methodology is applied to a case study in the information and communication technology industry for illustration. Bayesian prediction is demonstrated to be very effective compared to other alternatives via a cross-validation study. In particular, the prediction error rate based on our updating prediction scheme is significantly improved as more field data are collected, and achieves a prediction error rate lower than 20% after launching the product for 3 months.