Research On Sensor Fault Diagnosis Based On Multi-algorithm Integration Method

With the development of automation technology and the increase of large-scaleautomation projects, sensors used for state monitor and parameter measurement havebecome more and more important. And the sensor faults are inevitable, because thesensor usually works in bad environment. Once the sensor faults occur, which will leadto the degradation of automation system’s performance, and even disastrousconsequences. In short, sensor fault diagnosis is significant and practical.In practice, when carrying on sensor fault diagnosis, the number of the sensorsneed to be diagnosed is usually larger, which brings considerable difficulty to theanalysis of the data. Principal component analysis (PCA) can reduce the dimensions ofthe variables, and the new variables are unrelated, while they can carry the maximumuseful information in original variables. Therefore, PCA is widely used in faultdiagnosis of multi-sensor system. However, PCA can only be used to detect whetherthere is a faulty sensor or not, and it can not determine which sensor has appeared fault,so this paper has proposed the wavelet analysis method based on principal componentanalysis for sensor fault diagnosis.Wavelet analysis with its unique features is very suitable for analyzingnon-stationary signal, and it can also be used as an ideal tool for signal processing infault diagnosis. Both the features of the fault and the necessary information for faultdiagnosis can be constructed and extracted by wavelet analysis respectively. In recentyears, wavelet analysis has attracted widespread attention in the field of fault diagnosis.Although wavelet analysis is specialized in feature extraction, it can not determine thefault type. Signal’s energy is an important physical feature which mirrors the signal’schanges. The distribution of energy can show the characteristics of a signal. So thispaper proposed the energy analysis method based on wavelet transform.In this paper, the general idea of sensor fault diagnosis is shown as followings. First, we use PCA to analyze the data collected, in order to determine whether there isa faulty sensor in multi-sensor system. If the results show that there is a defectivesensor in system, then carry on wavelet analysis to each sensor signal, in order to findout the faulty sensor. Finally, carrying energy analysis on low-frequency signals andhigh-frequency signals come from wavelet analysis, in order to determine the type ofthe fault. After principal component analysis, wavelet analysis and energy analysis, wecan not only detect the fault sensor in system, but also determine the type of fault. Thispaper test and verify the sensor fault diagnosis method based on multi-algorithmintegration method. The result of the experiment on the High-grade NC Machine Toolsand Basic Manufacturing Equipment shows the method is very effective for sensorfault diagnosis.