Fault Detection of Electric Motors via Symmetrized Dot Pattern-Based Features

This study proposes a novel symmetrized dot pattern (SDP) approach using designed SDP-based features, extracted from transformed vibration signals, for fault detection. These features describe the compactness, inclination, and shape of the “snowflake” diagram distributions. They were employed as inputs to a feedforward neural network for the automatic fault detection process. The effectiveness of the new technique was demonstrated on two experimental datasets of electric motors. The use of electric motors dataset is an application of the new technique that includes different fault types, fault severity levels, and operating conditions. The proposed method was also compared with the conventional one consisting of fault detection using the SDP derived images and convolutional neural network. The results demonstrate that the proposed SDP-features and feedforward neural network outperforms both the SDP, images, and convolutional neural network method in terms of classification accuracy and low false positive rates. Moreover, the novel method does not necessitate imaging, achieves higher accuracy and a lower false-positive rate than the conventional method. The new approach demonstrates a reduction in complexity and an enhancement in efficiency in comparison with the conventional one, and an automatic fault detection process.