基于累加式实时串并联变换算法的机械故障声学监测方法

Acoustic monitoring method for mechanical faults based on accumulative real time series-parallel transformation algorithm

针对基于物联网(IoT)的冲压机床故障监测问题,为了降低冲压机床故障监测的计算复杂度,并提高其低频识别的精度,提出了一种无需机器学习技术的实时性机械故障声学监测方法,即基于累加式实时串并联变换算法的机械故障声学监测方法。首先,研究了物联网场景中冲压机床声学低频分析的必要性,并给出了声学信号的表达式;然后,针对频率轴上多个周期信号重叠导致参数估计较为困难的问题,提出了一种累加式实时串并联变换算法,将输入的采样序列馈入多个具有不同输出端口的串并转换器,从累加的波形中检测出最大绝对值,并进行了比较;最后,通过样本时隙划分,将累加式实时串并联变换算法应用于机械故障监测;通过仿真和冲压机床实机测试,对累加式实时串并联变换算法和实时性机械故障声学监测方法的有效性进行了验证。研究结果表明:在无需大量信号样本的情况下,使用累加式实时串并联变换算法有利于提高低频带的识别精度;在直方图相关性方面,累加式实时串并联变换算法和Morlet小波变换具有相同的性能,且均明显优于短时傅立叶变换;同时,尽管累加式实时串并联变换算法需要的加法总数比Morlet小波变换多2.5倍,但是乘法总数减少了20447%,大幅减少了计算的复杂度。

Aiming at the problem of stamping machine fault monitoring based on the Internet of Things(IoT),in order to reduce the computational complexity of stamping machine fault monitoring and improve the accuracy of low-frequency identification,a real-time mechanical fault acoustic monitoring method without machine learning technology was proposed,which was based on the cumulative real-time series-parallel transformation algorithm for mechanical fault acoustic monitoring.Firstly,the necessity of acoustic low-frequency analysis of stamping machine tool in the scene of Internet of Things was studied,and the expression of acoustic signal was given.Then,aiming at the difficulty of parameter estimation caused by the overlapping of multiple periodic signals on the frequency axis,an accumulative real-time serial-parallel transformation algorithm was proposed.The input sampling sequence was fed into a plurality of serial-parallel converters with different output ports,and the maximum absolute value was detected and compared from the accumulated waveforms.Finally,by dividing the sample time slots,the cumulative real-time serial-parallel transformation algorithm was applied to mechanical fault monitoring.The feasibility of the proposed algorithm in mechanical acoustic monitoring and real-time mechanical fault acoustic monitoring method were verified by simulation and real machine test of stamping machine.The research results show that the cumulative real-time serial-parallel transformation algorithm is beneficial to improve the identification accuracy of low-frequency band without a large number of signal samples.In the aspect of histogram correlation,the cumulative real-time series-parallel transform algorithm and Morlet wavelet transform have the same performance,and both are obviously better than short-time Fourier transform.At the same time,although the cumulative real-time serial-parallel transform algorithm needs 2.5 times more additions than Morlet wavelet transform,the total number of multiplications is reduced by 20447%,which greatly reduces the computational complexity.