Machine Learning Methods for Abnormality Detection in Hard Disk Drive Assembly Process: Bi-LSTM, Wavelet-CNN and SVM

The research proposes methods to detect abnormality in assembly of a hard disk drive. Three machine learning techniques are employed to classify the drives from assembly process into good and bad class. A good class represent a drive that all components are properly installed while a bad class represent an abnormal drive that some components are missing or improperly installed. The voice coil motor current motor is measured and collected from physical drives in assembly line for using as training and testing data set. Since the amount of bad drives in hard disk drive assembly process are much smaller than the good drive which introduce the imbalance problem dining training process, this paper also set the experiment of varying amount of training data set that can satisfy the training in practical hard disk drive assembly process. Bidirectional Long Short-Term Memory, Wavelet transform with Convolutional Neural Network and Support Vector Machine are chosen as proposed machine learning models to classify this task. The comparison between each technique is discussed.