Applying Decision Tree in Fault Pattern Analysis for HGA Manufacturing

This research proposes the design of a fault pattern analysis algorithm based on the C4.5 decision tree technique. We study the actual data collected from a disk drive manufacturing company. Our work emphasizes the HGA manufacturing data. However, the data from the Wafer and the Slider processes are also explored as they may affect the yield of the HGA production. In our algorithm, the data is first retrieved from the data warehouse, and then pre-processed using the regular data cleaning techniques. The critical external and internal data from all operations that are related to the HGA production (machine parameters and product attributes) are used as inputs in our algorithm. The data preparation steps are added to improve the raw data quality. Subsequently, our decision tree technique is employed to categorize decision options that indicate problems on the actual manufacturing environment. Finally, the root causes of the yield degradation will be identified in three categories of attributes (machine, material and method). The data analysts in a HDD company can use this tool to automatically summarize the problems on the manufacturing line. Yield can then be improved by adjusting parameters and/or attributes as suggested by the algorithm. In this paper, we also describe the algorithm through a simple example. Further study will be performed and the experiments will be elaborated in the near future.