基于PSO-BP-Adaboost的旋转机械故障诊断方法研究

Research on Fault Diagnosis Method of Rotating Machinery Based on PSO-BP-AdaBoost

针对旋转机械振动信号具有非线性、非平稳的特点,对振动信号进行了多域分析并将PSO-BP-AdaBoost强分类器模型引入到旋转机械故障诊断领域。首先,以粒子群算法优化的BP神经网络为弱分类器,以AdaBoost算法组合构造强分类器建立了分析模型;其次,以某设备振动信号为例,利用集合经验模态分解(EEMD)得到的模态分量计算了其能量特征值,并对时域信号进行分析得到了时域特征值;最后,通过对某转子振动数据进行分析,对比了PSO-BPAdaBoost算法与4种传统算法在以时域、能量域、时域+能量域特征值3种不同输入条件下的诊断准确率及效率。结果表明,在采用PSO-BP-AdaBoost算法及同时考虑时域与能量域特征时,其最高诊断准确率达100%,平均诊断准确率达98%,其诊断精度与效率比传统方法具有优势。

Dealing with the features of nonlinearity and nonstationarity of the vibration signal of rotating machinery,the vibration signal has been analyzed in multi domain,and the PSO-BP-AdaBoost strong classifier model is introduced into the field of rotating machinery fault diagnosis.Firstly,the BP neural network optimized by particle swarm optimization algorithm is used as the weak classifier,and the AdaBoost algorithm is combined to construct the strong classifier,thus the analysis model is established;secondly,taking the vibration signal of an equipment as an example,the energy eigenvalues are calculated by using the modal components obtained by Ensemble Empirical Mode Decomposition(EEMD),and the time-domain eigenvalues are obtained by analyzing the time-domain signal;finally,through analyzing the vibration data of a rotor,the diagnosis accuracy and efficiency of PSO-BP-AdaBoost algorithm under three different input conditions of time domain,energy domain and“time domain+energy domain”eigenvalues are compared with four traditional algorithms.The results show that when the characteristics of time domain and energy domain are taken into consideration with using the PSOBP-AdaBoost algorithm at the same time,the highest diagnostic accuracy is 100%and the average diagnostic accuracy is 98%,whose diagnostic accuracy and efficiency are superior to the traditional methods.