| With the development of science and technology,mechanical equipment is developing towards automation,integration,complexity and systematization,and the combination between equipment is increasing.Once a component of the equipment fails,not only the equipment itself will have problems,but other equipment related to the equipment will also be affected,resulting in a chain reaction.Therefore,timely and accurate fault diagnosis is very important.Vibration signal not only contains the working state information of the equipment,but also is easy to collect.Through the analysis of vibration signal,we can judge the efficiency and operation accuracy of the equipment.The rolling bearing of mechanical equipment is a very important part,and its reliability is very important for the normal operation of mechanical equipment or the whole system.(1)Aiming at the difficulty of metallurgical fault information detection and resolution,this thesis proposes a method combining residual long short-term memory network and convolutional neural network.This method first uses convolutional neural network to learn high-dimensional features of local features of metallurgical faults,and performs local feature extraction and feature fusion through continuous convolution.Secondly,the global feature extraction and feature learning of metallurgical fault information are carried out by using long-term and short-term memory network combined with residual network structure.The learning ability of long-term and short-term memory network for time series is used to parallel the feature information forgotten by quasi-complementary convolutional neural network in the process of feature learning for metallurgical fault information.Finally,the two feature learning modules are fused through the fault detection fusion module,and the experimental verification is carried out on the PHM challenge data set,which proves the effectiveness of the proposed method in the field of metallurgical fault detection.(2)Mechanical failure can cause a certain degree of economic loss,and even cause casualties.Timely and accurate fault analysis is a necessary condition to ensure the safety of industrial production.With the increase of mechanical manufacturing,fault prediction strategies mainly based on deep learning have received additional attention.However,the diagnostic accuracy of existing diagnostic strategies still needs to be improved.To this end,a fault prediction method WDCNN-LSTM combining a wide first layer deep convolutional neural network with a long short-term memory network is proposed.Convolutional neural network is used to extract feature information from one-dimensional real vibration index adaptively.In addition,the fault features can be completely obtained by LSTM extraction.Experiments are performed on the CWRU dataset to confirm the proposed method.Through the test of the experimental results,it is found that the average accuracy of the WDCNN-LSTM model proposed in this thesis is higher than that of the baseline model. |
