基于粒子群优化以及深度胶囊网络的轴承故障诊断

Bearing Fault Diagnosis Based on PSO and CapsNet

针对发动机轴承损坏情况严重以及基于模型方法预测精度不稳定的问题,提出一种基于深度胶囊网络和粒子群优化算法的轴承故障预测方法。通过将观测振动信号自适应降噪后,基于粒子群优化算法进行稀疏盲分离,得到轴承振动信号,通过S变换获取时域图以及轴承振动特征,其次将时域图经由卷积层卷积,输入到胶囊层进行预测。将高低胶囊层之间的算法转化为数学优化问题,提升传输效率,最后得出高层胶囊的预测向量。结合具体轴承监测数据进行实例分析,与基于数据的浅卷积网络以及经验模态分解预测相比,算法体现了更稳定更精确的预测性能。

Aiming at the serious damage of engine bearing and the low prediction accuracy of model-based method,a bearing fault prediction method based on depth capsule network(CapsNet)is proposed.After adaptive noise reduction of observed vibration signals,sparse blind separation based on particle swarm optimization(PSO)algorithm is used to obtain bearing vibration signals,which are transformed into time-frequency images by using time-frequency transform.Finally,time-frequency images are convolu⁃tion via convolution layer.The prediction vector is obtained as input to the lower capsule layer,and the dynamic routing algorithm is used as the input of the high layer capsule.The algorithm between the high and low capsule layers is transformed into a mathematical optimization problem.Finally,the prediction vector of the high layer capsule is obtained.Theoretical analysis and case analysis,compared with Convolutional Neural Network(CNN)and Empirical Mode Decomposition(EMD),this method can improve the fault identification rate reaching up to 99.14%.