Compressor Fault Diagnosis Employing Machine Learning Techniques

Compressors are a potent and crucial component in the industry as they make the work environment fresh, secure, and convenient. Heavy use of this equipment tends to make them faulty. Effective fault identification is a major need for the incessant use of compressors in the industry. Various compressor faults like Bearing faults, Flywheel faults, NRV faults, and LOV faults are considered in the work. Healthy and faulty samples (1024 samples) are taken and preprocessed using Wavelet Transform, such as Daubechies4 and Daubechies10. Statistical measures, namely Error, Skewness, Standard Deviation, Variance, and Kurtosis were computed. Machine learning methods namely KNN, SVM, and Random forest were applied to the statistical measures. A comparison of the three machine learning-based methods was carried out and three other performance metrics such as Precision, Recall, and F1 score have been evaluated. It was proved that KNN had a maximum accuracy (85%) over the other methods. An attempt to improve the accuracy is being carried out by increasing the number of samples and by applying specific preprocessing techniques.