Title:Warranty Failure Monitoring and Diagnostics: Product and Process Improvement Based on Field Measurement and Distributed Manufacturing Measurements

Sponsor :National Science Foundation (NSF)

Collaborators: Motorola

Customer feedback in the form of warranty/field performance is an important and direct indicator of quality and robustness of a product. Linking warranty information to manufacturing measurements can identify key design parameters and process variables (DPs and PVs) that are related to warranty failures. Warranty data has been traditionally used in reliability studies to determine failure distributions and warranty cost. This paper proposes a novel Fault Region Localization (FRL) methodology to map warranty failures to manufacturing measurements (hence to DPs/PVs) to diagnose warranty failures and perform tolerance revaluation. The FRL methodology consists of two parts:

1. Identifying relations between warranty failures and DPs and PVs using the Generalized Rough Set (GRS) method. GRS is a supervised learning technique to identify specific DPs and PVs related to the given warranty failures and then determining the corresponding Warranty Fault Regions (WFR), Normal Region (NR) and Boundary region (BND). GRS expands traditional Rough Set method by allowing inclusion of noise and uncertainty of warranty data classes. The method shows better performance than traditional classification methods for analysis of data with high noise, non-normal distributions and small failure sample size that are essential features of field/warranty data.

2. Revaluating the original tolerances of DPs/PVs based on the WFR and BND region identified.

• Java Applet for FRL Methodology

It searches the best subset with minimum number of parameters and their range for studying any particular type of failure the description about the format of training data for using FRL is as follows:

 1. The data should be in stored in a .txt file

 2. All the different values should be separated by spaces

 3. The sequence in which the values are to be arranged is as follows:

• Total number of manufacturing parameters studied and number of samples collected

• The matrix of the measured values corresponding to manufacturing parameters

• The uncertainty associated with customer feed back for each sample

• Lower and Upper Tolerance for each manufacturing parameter

• Sample 1

• Sample 2

• Mannar, K., Ceglarek, D., Niu, F., Abifaraj, B., , 2004, "Fault Region Localization (FRL): Product and Process Improvement Based on Field Performance and Distributed Manufacturing Measurements," submitted to IEEE Trans. on Automation Science and Engineering