托辊非接触式故障识别方法研究

Study on Non-contact Identification Method of Idler Faults

因托辊故障引发的远程带式输送机事故越来越多,而传统的人工巡检已不能满足需求,且现有接触式加速度信号检测方式存在传感器需求量大及数据收集难的问题,所以有必要通过智能巡检机器人搭载拾音器进行非接触式巡检。托辊运行环境嘈杂,为剔除信号中的噪声,提出基于完全噪声辅助集合经验模态分解(CEEMDAN)、主成分分析(PCA)和鲁棒性独立分量分析(RobustICA)的单通道盲源分离(SCBSS)去噪方法;托辊信号具有非平稳、非线性的特点,仅用梅尔倒谱系数(MFCC)不能完美刻画信号特征参数,提出基于CEEMDAN、PCA、MFCC、MFCC的1阶差分系数和Delta值的自适应特征参数提取方法;最后采用支持向量机(SVM)作为分类器进行故障识别,识别率达到97.2%。

There are more and more accidents of the long-distance belt conveyors caused by idler failure,however,the traditional manual inspection cannot meet the demand,and a great number of sensors are demanded in the existing contact measuring method which has a difficulty of data collection,so it is necessary to carry out non-contact inspection by using an intelligent inspection robot equipped with a sound picker.In order to eliminate the noise in the test signal,the single channel blind source separation(SCBSS)denoising method based on the complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN),the principal component analysis(PCA)and the robust independent component analysis(RobustICA)is proposed.Since the idler signal has the characteristics of non-smooth and nonlinear,and the Mel-frequency cepstral coefficient(MFCC)alone cannot perfectly depict the signal feature,the adaptive feature parameters extraction method based on CEEMDAN,PCA,MFCC,MFCC'first order difference coefficients and Delta value is proposed.Finally,the fault identification is carried out by supporting vector machine as classifier and its recognition rate reaches up to 97.2%.