State recognition of motor pump based on multimodal homologous features and XGBoost

Nowadays, state monitoring on industrial equipment attract extensive attention of machinery field. While the sufficient study of the intelligent state recognition of the motor pump operating state in complex operation states is still in need. A motor pump state recognition algorithm based on multimodal homologous features combined with the eXtreme Gradient Boosting (XGBoost) model is proposed. For a single feature characterization method cannot completely reflect the real operating states compared to other characterization methods with failure modes. As known as, the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) is more applicable to Empirical Mode Decomposition (EMD) and Ensemble Empirical Mode Decomposition (EEMD) which may result in the modal aliasing and lacking decomposition process integrity. In this study, the state recognition procedure extracts signal to construct multimodal homologous features in four modals in time domain, frequency domain, wavelet packet and improved Hilbert-Huang transform (HHT). These procedure established the fault feature set. In order to enhance the precision of the motor pump state recognition, the XGBoost algorithm can be optimized by using grid search to improve model hyperparameters effecting the importance calculation of the multimodal homologous features. For the more, to validate the effectiveness of the above methods in bearing data, a public data set of Western Reserve University is selected to learn 10 fault states under 4 load states. The experimental results show that the multimodal homologous feature can effectively characterize the different features of signals in several states. The deep analysis of the XGBoost model, improve the average accuracy of the fault diagnosis under various load states to over 99.77%, which has great significance to the motor pump state recognition.