Gear fault detection using multi-scale morphological filters

This work gives a thorough exploration of the capacity of multi-scale morphological filters for gear fault detection. Eight types of multi-scale morphological filters are designed based on the mathematical morphology theory. A characteristic frequency intensity coefficient (CFIC) is defined as a quantity criterion for assessing the effectiveness of the filters. Both simulated signal and practical vibration signal measured from a gearbox are utilized to evaluate the fault detection ability of the eight multi-scale morphological filters. The conventional envelope analysis (EA) and another multi-scale analysis technique means the continuous wavelet transform combined with envelope analysis (CWT-EA) are also employed for a comparison. Experimental results have revealed that the averaged multi-scale morphological dilate-erode gradient (AMMGDE) filter achieves the best performance for detecting gear defects. Furthermore, the computation cost of the AMMGDE is comparable to the EA but much less than the CWT-EA technique. Therefore, the AMMGDE filter is a very attractive technique for on-line condition monitoring of gear and other similar rotating machineries.

► We have designed eight types of multi-scale morphological filters (MMFs) based on the mathematical morphological operators.
► The MMFs can achieve more satisfactory demodulation results with low computation cost than other similar techniques.
► The averaged multi-scale morphological dilate-erode gradient filter achieves the best performance for detecting gear defects.