Endocrine CNN-Based Fault Detection for DC Motors

Abstract

This paper presents a novel method for detecting and classifying faults in dynamic control systems empowered with DC motors, operating under laboratory conditions. The approach employs a convolutional neural network model enhanced with an artificial endocrine influence to evaluate the condition of the rotating motor shaft by analysing information from the vibration sensors mounted on the shaft itself. The trained network effectively classifies the level of unbalance in the system into three categories based on the vibrations: optimal (no unbalance), first and second degree of unbalance. To validate the efficiency of the proposed model, its performance was compared with the performance of deep learning algorithms commonly recommended for time-series classification: default convolutional neural network, fully convolutional neural network, and residual network. The new model was shown to perform classification tasks with the highest accuracy, proving to be an efficient fault diagnosis tool with a viable potential to be applicable in industrial predictive maintenance processes.

Keywords

• convolutional neural network
• time-series classification
• artificial endocrine gland
• fault diagnosis
• vibration analysis