ETNeXt: Integrated feature engineering and classification framework for BLDC motor fault detection

Read the full article See related articles

Discuss this preprint

Start a discussion What are Sciety discussions?

Listed in

This article is not in any list yet, why not save it to one of your lists.
Log in to save this article

Abstract

Motor fault detection is critical for industrial reliability, with acoustic signals serving as a key diagnostic tool. This study employs the ETNeXt framework for robust motor acoustic condition classification using a dataset of 2,021 wav files categorized as "good," "broken," or "heavyload." The methodology includes: (i) applying a 7-level Multilevel Discrete Wavelet Transform (MDWT) with the 'sym4' wavelet to extract approximation coefficients; (ii) using the Triadic Signal Kernel-Local Histogram Pattern to generate feature histograms based on signum, upper ternary, and lower ternary functions; (iii) selecting optimal features via Neighborhood Component Analysis (NCA) and Chi-square (Chi2) methods, with 35 features per level over 5 iterations; and (iv) classifying feature vectors using Fine k-NN and Cubic SVM with 10-fold cross-validation. The ETNeXt framework achieved 99.80% accuracy with Fine k-NN and 100% with Cubic SVM, demonstrating exceptional performance. On a new dataset, it maintained high generalizability, achieving 99.95% accuracy with Cubic SVM and 99.16% with Fine k-NN. Additionally, integrating a cochlear-inspired transformation and ResNet50 model yielded 99.95% accuracy with Fine k-NN and 97.97% with Cubic SVM, further enhancing diagnostic capability. These results highlight the ETNeXt framework's effectiveness and robustness for motor fault detection and classification.

Article activity feed