基于RS-LSTM的滚动轴承故障识别

Fault identification of rolling bearing based on RS-LSTM

针对滚动轴承不同故障位置、不同损伤程度的振动加速度信号的智能分类,提出一种基于随机搜索与长短时记忆(long short term memory,LSTM)神经网络的滚动轴承故障状态识别算法。该算法直接利用原始数据作为非线性输入,避免因人工提取特征值造成的原始信息缺失;使用LSTM与深度神经网络的混合网络提高模型性能;引入随机搜索算法自动优化超参数得到最优的网络配置;使用不同量纲、不同来源、不同损伤结构的两类数据集对模型进行试验验证。试验结果表明,在两类单一数据集及随机混合数据集均可达到99.8%以上的诊断准确度,表明本算法具有较高的泛化能力和鲁棒性。与BP、支持向量机、粒子群算法最小二乘支持向量机、LSSVM、浅层LSTM等方法在同等试验条件下的诊断结果进行比较,本文算法具有更高的识别准确度。

To intelligently classify the vibration acceleration signals of different fault positions and different damage degree of rolling bearings.An algorithm for fault recognition of rolling bearing is proposed.It is based on random search and long short term memory（LSTM）neural network.The algorithm directly uses the original data as a nonlinear input to avoid the loss of original information caused by the artificial extraction of feature values.A mixed network of LSTM and deep neural network is used to improve the performance of the model.Meanwhile,a random search algorithm is used to optimize the hyperparameter automatically and obtain the best network configuration.Finally,two types of data sets with different dimensions,sources,and damage structures were used to verify the model.Experimental results show that over 99.8% diagnostic accuracy can be achieved on both single data sets and random mixed data sets.It is included that the algorithm has high generalization ability and robustness.Compared with the diagnostic results of BP,support vector machine,particle swarm optimization-least squares support vector machine,LSSVM,and shallow LSTM methods under the same experimental conditions,the proposed algorithm has higher recognition accuracy.