Gear Fault Identification Based On Acoustic Emission Technology

As a common part of mechanical equipment used for changing the speed and transmit power,gears are widely used in industrial production and are an indispensable part of mechanical connection and drive.Due to the complicated working state of the gears and the long-term operation under heavy load conditions,gears are extremely vulnerable to damage and failure,resulting in failure of the entire mechanical transmission system and even threatening people's safety.Therefore,the diagnosis and identification of the fault of the gear is of great significance for the fault detection of the rotating equipment.In order to improve the accuracy of gear fault identification,avoid relying on the traditional time-frequency analysis method to diagnose gear faults and solve the problem of difficulty in feature extraction of gear signals.In this paper,a combination of acoustic emission technology and deep learning is used.Acoustic emission equipment is used to collect the fault signal of the gear.A gear fault diagnosis model based on Gated Recurrent Unit network is built and the Deep Belief network is used as the comparison network to achieve intelligent identification of gear fault.In this paper,the core components of rotating machinery are used as the research object of fault diagnosis,and the paper has mainly completed following aspects.Firstly,through the QPZZ-II type rotating machinery vibration and fault simulation experimental platform,the acoustic emission signals of the gear in the normal state,the wear fault,the wear and the groove mixed fault are collected.Then,the collected acoustic emission signal is subjected to pre-processing and extracted acoustic emission characteristic parameters as input of the neural network.Finally,the collected acoustic emission signal data is reasonably divided into training set and test set,and the training set sample data is used to train the gear fault diagnosis model based on the Deep Belief network and the Gated Recurrent Unit,then through the performance of the test set data on the network model to adjust the parameters of the network,build the most suitable gear fault diagnosis model.Firstly,the gear fault diagnosis model based on Deep Belief Network is constructed.The network uses a bottom-up layer-by-layer training of Restricted Boltzmann Machines,which gives the whole network a good initial parameter,and then obtain the optimal solution by inversely finetuning the whole network.The experimental results show that the gear fault diagnosis model based on Deep Belief Network has a recognition accuracy of 98.7217%,which is feasible.Secondly,a gear fault diagnosis model based on Gated Recurrent Unit network is built,because the Recurrent Neural Network can capture the long-term dependence between networks,but the traditional Recurrent Neural Network also has the problem of gradient disappearance,which makes the superficial network weights are not updated.In order to ensure the memory between networks and improve the problem of gradient dispersion,this paper uses the Gated Recurrent Unit to identify and diagnose the fault of the gear,and the recognition accuracy reaches 99.8709%.It is verified that the network has a better classification effect on gear fault diagnosis,which improve the accuracy of gear fault diagnosis.