Stable Hybrid Feature Selection Method for Compressor Fault Diagnosis

Faulty compressors must be detected in advance to speed up the quality control process of compressor's performance. Machine learning models have recently been used as fault classification distinguish between normal and abnormal compressors, facilitating more sophisticated detection methods than those past. However, very few studies conducted on accurate efficient feature selection, despite its high importance. Therefore, this study proposes a new hybrid method that combines merits existing methods, filter wrapper obtain stable, accurate, model. For this, three types filtering with different characteristics, such chi-square, extra tree classifier, correlation matrix, are derive high-ranked features then create powerful set consisting their union sets. Subsequently, using method, one combination highest accuracy was selected among all combinations set. Using two experimental examples numerical example numbers data, robustness proposed were verified through comparison by combining models: support vector machine, K-nearest neighbor, multi-layer perceptron.