An availability of MEMS-based accelerometers and current sensors in machinery fault diagnosis

In recent years, micro-electromechanical systems (MEMS)-based sensors have shown huge attraction in machinery fault diagnosis due to their low power consumption, low cost, small size, mobility, and flexibility. Hence, this paper presents a comprehensive fault diagnosis scheme using MEMS-based accelerometers and current sensors to identify several induction motor failures. In this paper, we first verify the reliability of these MEMS-based sensors via frequency analysis for vibration and current signals captured by them. Likewise, this paper validates their suitability for machinery fault diagnosis. To do this, we configure a 147-dimensional feature vector using statistical values (i.e., 21 statistical values × 7 MEMS-based accelerometers and current sensors), analyze fault signatures by employing a kernel principal component analysis, and pinpoint types of induction motor failures with one-against-all multi-class support vector machines (OAA MCSVMs), a random forest (RF), and a fuzzy k-nearest neighbor (Fk-NN). Experimental results indicate that the presented fault diagnosis approach using MEMS-based accelerometers and current sensors yields 100%, 86%, and 80% of classification accuracy with OAA MCSVMs, the RF, and the Fk-NN, respectively. Accordingly, MEMS-based sensors are enough for substituting commercial accelerometers and current sensors that are used for fault diagnosis. Specifically, MEMS-based accelerometers are far more effective for preserving intrinsic information about various induction motor failures than MEMS-based current sensors, offering at least 38% performance improvement in classification accuracy.