Data-Driven Gantry Health Monitoring and Process Status Identification Based on Texture Extraction

Abstract

Abstract As a critical asset, gantry has wide applications in many fields such as medical image area, infrastructure, and heavy industry. Mostly, the gantry is reliable, however, the loss led by the gantry lockout is inestimable enormous. Moreover, there are limited previous gantry studies concentrated on the statistical quality control to detect the fault not to mention the research that focuses on the algorithms applied to the process status sequence to detect the fault. Time series gantry process status sequence usually consists of categorical values, which makes it hard to obtain features for the task of fault identification. This paper proposes a novel method using texture extraction in image processing to obtain the features of gantry process status sequence. The histogram of oriented gradients (HOG) texture extraction technique is used to the process status sequence. To demonstrate the effectiveness of image-based feature extraction, we applied different machine learning algorithms to the time series gantry process status sequences provided by a leading automobile manufacturer. Results demonstrate that the sequence after the transformation of texture extraction technique has improved the accuracy of machine learning algorithms (i.e., 10% increase on average for KNN, 32% increase for LDA, 26% increase for QDA, and 8% increase for linear SVM).