Estimating Competing Failure Modes in Warranty Data when Modes for Most Failures are Unknown

Warranty data for large populations of commercial products often does not include information about the failure mode by which components may have failed. However, there are times when the failure mode level of detail is needed in order to facilitate engineering problem solving, support warranty cost recovery claims against a supplier, predict future failures during the remaining warranty liability period, properly fund warranty reserve accounts, and so forth. While it is often impractical or impossible to classify every warranty claim according to its failure mode, it is sometimes possible to identify the failure modes associated with a small subset of the claims. When this can be done, the method provided in this paper can be used to estimate the parameters of the failure time distributions for a collection of competing failure modes acting on a component within a complex system. To demonstrate and validate the method, it is first applied to a simulated data set for which the correct answers are known. The method is then applied to a real warranty data case study.