基于极值点概率密度和听觉模型的瞬态信号提取方法研究

A method extracting transient signals based on probability density of extreme points and an auditory model

零件间的碰撞会诱发瞬态振动信号成分,所以发现和提取瞬态信号有助于设备故障的识别。人类听觉系统对于突发声音具有本能的敏感性,因此,基于听觉系统的运行机制提出了一种瞬态信号提取方法。提出了信号极值点幅值概率密度曲线中存在小幅值局部波动这一含瞬态成分信号的重要特征,并结合频带连续性和起始同步性做为瞬态信号提取的线索。基于三种线索,首先对信号进行Gammatone带通滤波、相位调整和极值点提取,继而计算各滤波信号极值点的幅值概率密度,并判断各滤波信号中是否存在瞬态成分,根据判断结果提取可能与瞬态成分有关的极值点,但因背景信号和干扰噪声的影响,所提取到的极值点会有一部分与瞬态成分无关,因此,将无关点分为四类并设计了相应的筛选方法。最终,利用筛选后所得极值点生成瞬态信号。数值仿真和实测数据验证的结果表明,所提方法对于瞬态信号提取具有良好的性能,且在干扰噪声和背景信号较强时也可实现较好的提取效果。

Transient vibration signals are usually induced by impacts between mechanical parts.So,it is important to find and extract transient signals for recognition of mechanical faults.Considering that the human auditory system is instinctively sensitive to a burst of sound,a method of transient signal extraction based on the operation mechanism of an auditory system was proposed.An important feature was found that there are some fluctuations with small amplitude in the probability density curve of the amplitude of a signal extreme point,if the signal contains transient components.Then,this feature,the continuity of frequency band and its onset synchronism were taken as the cues for extracting transient signals. Based on these three cues,the band-pass filtering with Gammatone filters,phase adjustment and extreme points extraction for signals were implemented at first.Based on the extreme points,their amplitude probability densities were calculated to judge if there exist transient components in filtered signals.According to judgment results,those extreme points that might be related to transient components were extracted.However,due to the effects of noise and background signals,parts of the extracted extreme points were not related to transient commponents.Therefore,these points were divided into four kinds and the corresponding screening methods were designed.At last,the transient signals were formed by using the left extreme points after screening.The results of numerical simulation and actual test data showed that the proposed method is effective,especially,it can extract transient signals with stronger noise interferences and background signals.