Distribution of fault currents.
<div><p>This study explores the design of an effective fault classification algorithm for 3 phase induction motor, an integral unit in many industrial systems. It is found that traditional fault detection methods and deep learning approaches are both effective; however, current technique...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | , , , , |
| Published: |
2025
|
| Subjects: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | <div><p>This study explores the design of an effective fault classification algorithm for 3 phase induction motor, an integral unit in many industrial systems. It is found that traditional fault detection methods and deep learning approaches are both effective; however, current techniques can either be computationally exhaustive, or suffer from low accuracy, thus making them inapplicable in many real-world settings. To address these limitations, this study evaluates different machine learning algorithms for accurate and efficient fault detection using a dataset of triaxial vibrational data converted into current variables. A dataset of triaxial vibrational current data at 0.7 mm bearing and rotor faults at various loads (100W, 200W, and 300W) were considered. For the data preprocessing, we handled with the missing values by interpolation and handle data imbalance fault types with Synthetic Minority Over-sampling Technique (SMOTE). Through Fast Fourier Transform (FFT) techniques, the frequency domain information were extracted, which is key for current signals, adding to the feature set. In addition, dimensionality reduction with Principal Component Analysis (PCA) and feature selection was done with SelectKBest. Then, the different machine learning models such as Random Forest (RF), Decision Tree (DT), k-nearest neighbors (KNN), and eXtreme Gradient Boosting (XGBoost) was trained to optimize the hyperparameters and make them perform to its best possible. The results shows the accuracy and performance of all models, DT and RF show good performance, with 99.95% accuracy, while KNN performs well, but at a higher computational cost in testing. Generally known for its capability to handle all the complex dataset, XGBoost wasn’t able to perform in this scenario as it got an accuracy of 87.13%, indicating potentially more optimization is required for the model. This work serves as the groundwork for future work with a multiplicity of fault types, motor specifications, and the incorporation of additional feature-engineering techniques to develop a more robust and intelligent framework for fault detection.</p></div> |
|---|