Enhancing testing cell set efficiency: A machine learning approach on hard disk drive data

Abstract

Hard Disk Drive (HDD) products undergo meticulous testing procedures to ensure their functionality prior to customer distribution. Nevertheless, anomalies can arise within the testing environment due to various factors, such as an increased number of media discs, leading to heightened current consumption by the spindle motor, and the frequent insertion and removal of HDDs during testing. These factors can induce malfunctions within the testing cell, which are identified by the tester's program. This study leverages diverse data measurements collected from tester HDDs within the testing cell to predict the status of the testing cell itself. Five distinct algorithms—Linear Discriminant Analysis (LDA), Ridge Classifier CV (RCCV), Extra-Tree Classifier (ETC), Random Forest Classifier (RFC), and Extreme Gradient Boosting (XGBoost)—were assessed. The research underscores that the proposed methodology, particularly utilizing XGBoost, achieves a notable prediction accuracy of 87.9% when applied to real datasets.