| Rolling bearings are widely used in various rotating machinery and play a vital role.The accurate prediction of the remaining useful life of the bearing can timely know when to replace it and ensure the safe operation of the equipment,which can not only improve the production efficiency,but also eliminate potential safety hazards and prevent the occurrence of casualties and property losses.Therefore,it is of great significance to establish a prediction model for the remaining service life of rolling bearings in key parts to guide the formulation of equipment maintenance strategies and ensure the safe operation of equipment.In recent years,the rise of technologies such as artificial intelligence and big data has set off a frenzy of industrial transformation in the manufacturing industry.In this context,artificial intelligence methods relying on massive monitoring data have gradually occupied the dominant position of the remaining life prediction methods of equipment.This paper uses a data-driven deep learning method to conduct RUL prediction research on rolling bearings.The details are as follows:(1)Taking the bearing vibration signal collected by the sensor as the input,a new deep learning model CNN-Conv LSTM is constructed by combining the convolutional neural network and the convolutional long and short-term memory network to predict the remaining useful life of the bearing.The model uses CNN for feature extraction on the input layer and Conv LSTM for RUL prediction.The validity of the model in RUL prediction of rolling bearings is verified through a bearing life cycle dataset.Its dynamic activation function can dynamically adjust its own slope according to different characteristics,so as to adapt to the characteristics extracted from the vibration signal in the whole life cycle of the bearing.(2)In the whole life cycle of bearing,with the degradation of bearing performance,the vibration signal will contain more and more degradation information.Most of the traditional convolutional neural networks use static activation functions,which cannot perform corresponding nonlinear transformations on the features extracted from the signals of different bearing degradation stages,and have a negative impact on life prediction.Considering the characteristic differences of vibration signals in different degradation stages of rolling bearings,a dynamic activation function is introduced into the convolutional neural network,and a dynamically activated convolutional layer is constructed.(3)Combined with the Dynamically Activated Convolutional Network and Conv LSTM,a rolling bearing RUL prediction method based on the DACN-Conv LSTM model is proposed.DACN is formed by stacking multiple dynamically activated convolutional layers,which can perform corresponding feature extraction on the vibration signals of rolling bearings in different degradation stages.The model first performs adaptive feature extraction on the input layer through DCN,and then inputs the extracted features into Conv LSTM to mine the temporal relationship between adjacent samples.The output of the last Conv LSTM unit is downsampled by the global max pooling layer and then input to a fully connected layer to obtain the RUL prediction result of the bearing.The validity of the model is verified through two bearing life cycle data sets.The verification results also show that the model has superior RUL prediction ability of rolling bearings,and can maintain good prediction performance in different data sets. |
