基于WLR和PSO-AFS-SVR的滚动轴承可靠度预测方法

在训练数据缺乏的情况下，为了提高支持向量回归机(SVR)对滚动轴承可靠度的预测精度，提出了一种基于威布尔线性回归(WLR)组合可靠度模型结合粒子群人工鱼群-支持向量回归机(PSO-AFS-SVR)的预测方法。首先，使用威布尔统计模型与线性回归（LR）的组合模型作为可靠度模型，利用测量滚动轴承振动信号的加速度计频谱，依据峰值频率分布的变化，分割其性能衰退的各个阶段，对每个阶段单独建模，以便最大程度地挖掘小样本信息；其次，采用k-折交叉验证(k-fold)的平均绝对误差(MAE)和平均相对误差(MAPE)之和作为适应度函数，利用PSO-AFS优化SVR参数，提高其泛化能力和预测精度；最后，采用滚动轴承全寿命周期试验数据进行了验证试验。试验结果表明，所提方法可以对滚动轴承的可靠度进行更准确的预测。

Reliability Prediction Method of Rolling Bearing Based on WLR and PSO-AFS-SVR

To improve the precision of SVR prediction model on predicting the rolling bearing reliablity with a little traning data,a method based on Weibull linear regression(WLR) combined reliability model and particle swarm optimization artificial fish swarm support vector regression (PSO-AFA-SVR) is proposed.Firstly,the combination model of Weibull statistic model and LR was used as the reliability model,and the accelerometer spectrum of the vibration signal of rolling bearing was analyzed to divide each stage of its performance degratation according to the distribution of peak frequency,then each stage was separately modeled,so that the information mining of small sample can be maximized.Secondly,the sum of the mean absolute error (MAE) and the mean absolute percent error (MAPE) of k fold was used as the fitness function,and the SVR parameters were optimized by PSO-AFS,which improved the generalization ability and prediction accuracy.Finally,the rolling bearing life cycle test data were used to perform verification test.The test results show that this method can predict the reliability of the rolling bearing with higher precision.