基于贝叶斯优化CNN的风电轴承故障诊断策略

Fault Diagnosis Strategy of Wind Turbine Bearings Based on Bayesian Optimized CNN

针对风电机组滚动轴承故障特征微弱、提取困难、诊断效率低下等问题,提出一种基于贝叶斯优化改进卷积神经网络(Convolutional Neural Network,CNN)的风电轴承故障诊断策略。在改进的CNN模型中,在全连接层前新增一卷积层,用于挖掘信号的深层特征;同时对目标函数进行改进,新增L2正则项;对卷积层数据进行批归一化处理并采用带有动量的随机梯度下降(Stochastic Gradient Descent with Momentum,SGDM)算法训练网络,以提高训练速度。使用贝叶斯优化算法对该改进的CNN模型的超参数进行优化。详细介绍改进的CNN模型和贝叶斯优化算法的工作原理,给出了基于贝叶斯优化的CNN超参数优化算法流程。最后利用凯斯西储大学滚动轴承数据库的数据对算法进行性能测试,证明经超参数优化后的CNN模型泛化能力强,诊断率高。

Aiming at the problems of weak fault feature,difficulty in extraction and low diagnosis efficiency for rolling bearings of wind turbines,a new fault diagnosis strategy of wind turbine bearings based on Bayesian optimization and improved convolution neural network(CNN)is proposed.In the improved CNN model,a convolution layer is inserted in front of the full connection layer to extract the deep features of the signal.At the same time,the objective function is improved by adding an L2 normalized term.The convolution layer data is processed by batch normalization(BN)and the stochastic gradient descent algorithm with momentum(SGDM)is used to train the network so as to raise the training speed.Bayesian optimization algorithm is used to optimize the hyperparameters of the improved CNN model.The working principles of the improved CNN model and Bayesian optimization algorithm are introduced in detail.And the flow chart of CNN hyperparameter optimization algorithm based on Bayesian optimization is given.Finally,the performance of the algorithm is tested by using the data of the rolling bearing database of Case Western Reserve University,which proves that the CNN model after hyperparameter optimization has strong generalization ability and high diagnostic rate.