Machine Learning-Based Sensor Drift Fault Classification using Discrete Cosine Transform

In fourth industrial revolution sensors are playing a crucial role as they provide a real time health and operating conditions of a physical system through continuous real time data. With this real time data in hand, robust and intelligent condition monitoring systems are being implemented through machine learning and deep learning techniques. However, the reliability of the sensor data is critical for good condition monitoring system. Many sensors operate in harsh environment; thus, the sensor signals can be affected by various faults. In this paper we propose a hybrid approach of sensor drift fault classification which involves discrete cosine transform (DCT) for extracting features from time domain sensor signals and in the later stage machine learning algorithms are used to build classification models. Experimental results showed that our proposed DCT-based feature selection method when combined with common machine learning models can exhibit excellent classification accuracy which is above 97% for most common machine learning models.