Warranty data contain useful information on the product reliability in the field use and therefore, need to be carefully analyzed for process and product improvement. For some product, failures depend on both age and amount of accumulated usage, which results in two-dimensional warranty data. A typical example of such product is an automobile, the life of which is characterized by time in service and mileage.
This thesis develops a method for analyzing two-dimensional warranty data. The warranty data are generated using simulation. In the proposed two-dimensional analysis, only warranty data are used without follow-up survey data, and the sequential regression analysis is conducted under proper assumptions to accommodate the claims censored due to the mileage warranty limit. In addition, the result from the one-dimensional analysis based on failure times is compared with that of the two-dimensional analysis. It is shown that the number of the predicted claims by the proposed two-dimensional analysis agrees well with the number of the claims generated by simulation, and the two-dimensional analysis mitigates the problem of reliability overestimation in the one-dimensional analysis.
