Warranty forecasting of electronic boards using short-term field data

The main goal of our study is precisely predicting the reliability performance of electronic boards throughout the warranty period by using short-term field return data. We have cooperated with one of the Europe's largest manufacturers and use their well-maintained data with over 1000 electronic board failures. Before using the field data for our model of warranty forecasting, we filter it to eliminate improper data, correlated to incomplete and poorly collected data. Our model is based on a two-parameter Weibull distribution, chosen from many other distribution options regarding optimum curve fitting. In the fitting process we use and compare 'Bayesian', 'rank regression', and 'maximum likelihood' fitting techniques. Our method has two steps. In the first step, we investigate how the Weibull parameter β changes by increasing the number of months of field data. For this purpose we use an electronic board with 36 months (full warranty period) of field return data. We develop a mathematical model of β as a function of the field data time interval and board dependent parameters. In the second step, we make a warranty forecasting of a new electronic board using its 3-month field data by using the mathematical model developed in the first step. The proposed method is evaluated by applying it to different electronic boards with 36 months (full warranty period) of field return data. The predicted results from our method and the direct results from the field return data matches well. This demonstrates the accuracy of our model.