SUMMARY

Rolling Element Bearings (REB) are critical components of a wide range of rotating machines. Different techniques of condition monitoring have been developed to be able to extract information about the REB via: time domain trend analysis or amplitude modulation technics, or by peaks identification in the frequency domain. Those approaches either provide little insights about the type of defects or are sensitive to noise and require post processing to become valuable. The ever-increasing size of databases is also complicating the fault diagnostics. These difficulties are addressable via approaches that leverage recent developments on microprocessors and System on Chip (SoC) that enable more processing power at the sensor level.
The proposed research addresses above-mentioned limitations by presenting a new approach for bearing defect detection using a SoC network for parallel wavelet transformation. To enable a near real-time processing of the data, the Beagle Bone AI SoC is employed its Analog to Digital Converted and its Graphic Processing Unit makes it an excellent candidate for implementation and experimental validation of wavelet transform and defect classification at the edge.
The expected contributions of this work are as follows: first, the real-time data acquisition with the SoC is developed, second the machine learning algorithm for improving the parallel wavelet transforms and the defect identification is implemented, finally the new approach is benchmarked to current approaches in terms of detection accuracy, speed and sensitivity to defect.