Research on Data Quality Assurance for Health Condition Monitoring of Machinery

Abstract

Health condition monitoring of machinery has entered into the big data era, which brings new opportunities to machinery fault diagnosis. However, due to the abnormal operating environment, disturbance from human and fault data acquisition devices, condition-monitoring data generally include lots of data with abnormal or missing values, which reduces the quality of data seriously. Wrong diagnosis results are probably obtained from the analysis of the low-quality data, leading to inappropriate strategy of machinery maintenance. To solve this problem, a condition-monitoring vibration data recovery method is proposed based on tensor decomposition. A four-order tensor including rotational speed, time-domain window, multi-scale using wavelet transform, and time is constructed. Tucker decomposition is used to process this four-order tensor for extracting the information of health condition and missing data are recovered by tensor completion. Simulated data and real vibration data are used to verify the effectiveness of the proposed method, respectively. The result shows that the data recovered by the proposed method are more close to the real data, compared with traditional data recovery methods, which demonstrates its effectiveness for data recovery in data quality assurance. The proposed method is applied to improve the quality of the condition-monitoring data collected from wind power equipment.