Embedded sensing for on -line *bearing condition monitoring and diagnosis
The object of this project is to develop a new methodology to improve bearing condition monitoring and diagnosis, based on embedded sensing approach. The presented study involves work in four areas.
First is investigation of the embedded sensing approach. Studies of the bearing structure dynamics and defect-related vibrations reveal the important characteristics of defect signals. The sensor placement strategy of embedded sensing is verified through theoretical analysis, numerical simulation, and experiment test. The embedded sensing enables the acquisition of bearing signals with a high signal-to-noise ratio, which is critical to the reliable defect detection.
Second is the feasibility investigation of the embedded sensing approach through a wireless sensing module design. The miniaturized sensing module, configured as an accelerometer or dynamic load sensor, have an on-board signal conditioner and a wireless transmitter. It has been designed, prototyped, bench-scale tested, and structurally integrated into a customized bearing test bed for on-line bearing monitoring. A combination of surface mount (SMT) and thick-film hybrid technologies has been used to reduce the size of the sensing module.
Third is the defect signal extraction for the embedded sensing approach. A new signal processing technique has been developed in order to improve the effectiveness of the defect feature extraction. It combines the wavelet analysis on the time-scale domain and the analysis on the frequency domain. Experimental studies have shown that it is more effective in detecting the existence of a small bearing defect of 0.25mm in diameter, as compared to the frequency-domain analysis or the wavelet analysis alone.
Fourth is the defect evaluation. To enable automated bearing defect evaluation, a bearing health index is introduced from the quantization of defect severity. Under different bearing operation conditions (e.g. speed, radial and axial loads), bearing defects are evaluated through a backpropagation neural network. Its input features are constructed based on the defect signatures, which are extracted by the combined wavelet and frequency-domain analysis.
In addition to on-line bearing condition monitoring, the techniques developed for the embedded sensing can be also adapted to the condition monitoring of many other types of manufacturing equipment.