Prediction of Spindle Rotation Error through Vibration Signal based on Bi-LSTM Classification Network

Abstract

Abstract One of the important indexes for high-precision machine tools is spindle rotation error, which is closely related to mechanical processing product quality. However, it’s difficult to directly acquire rotation error in the actual machining process. This literature proposed a novel uncompressed approach to predict spindle rotation error through vibration signal based on Bi-LSTM classification network, and this approach mainly consists of three steps, namely, pretreating original data; training Bi-LSTM classification network; predicting spindle rotation error. This approach adopts an uncompressed vibration signal method, which retains the time characteristic information into the predicted network, to build the relationship between easily collected vibration signal and the spindle rotation error. Finally, the proposed approach is applied in a spindle test rig, which has accomplished over 1700 hours of simulated load abrasion, and collected 170 days’ vibration signal and corresponding rotation error at RPM=1000, 2000, 3000, and 4000 conditions. The results show the proposed approach can effectively predict the spindle rotation error.