Method for failure pattern analysis in disk drive manufacturing

This research proposes the design of an algorithm for failure pattern analysis. The actual data is collected from a disk drive manufacturing line. Our work emphasises data from Head Gimbal Assembly (HGA). In our algorithm, the data are first retrieved from the data warehouse and pre-processed. The pre-processed data and critical parameters related to HGA production are then used as inputs. Subsequently, a decision tree is constructed to categorise decision options that indicate problems on the manufacturing environment. The root causes of the yield degradation can then be identified in the form of decision rules. Since only a few selected parameters will be tuned by the engineers, we also apply the hierarchical clustering algorithm in order to identify a group of parameters that should be considered first. The results can be used to automatically create a suggestion for yield improvement. Moreover, our experiments showed that a slow execution time may become problematic. Parallel tree construction is thus designed. The performance of our parallel algorithm is favourable. Data analysts in hard disk drive (HDD) companies can use this tool to automatically summarise the problems on the manufacturing line and yield can be improved as a result.