Research On Fault Diagnosis Dethod Of Motor Based On Improved Deep Reinforcement Learning

As a kind of equipment that converts electric energy into mechanical energy,motor is widely used in the field of ship,accounting for a large proportion of all electrical equipment on board.It is the key equipment of ship power system and plays an important role in the normal operation of the ship.Due to the precise components and parts of the motor and the harsh operating environment on the water,the motor is prone to failure under the influence of vibration,humidity and other factors,which will not only have adverse consequences on the motor itself,but also affect the ship’s operating state,cause unpredictable harm to the ship,and even bring huge economic losses and hidden dangers of personal safety.Therefore,monitoring and diagnosing the fault of the motor is of great significance for the fault diagnosis of the ship’s power system.While using traditional methods for fault diagnosis,it is necessary to carry out complex and cumbersome feature extraction steps on the original data,and professionally distinguish the extracted features through expert knowledge.Therefore,the intelligence and autonomy of fault diagnosis are insufficient.In order to solve this problem,this paper takes the motor bearing as the main research object,takes the most common bearing fault in the motor fault as the main research type,takes the vibration signal collected by the vibration sensor as the data source,and uses the method of deep reinforcement learning to realize the end-to-end autonomous intelligent fault diagnosis of the motor bearing.The main research contents of this paper include the following aspects:(1)Research on the mechanism of motor fault diagnosis.The structure and working principle of the marine motor are expounded,and the mechanism and causes of the mechanical and electrical faults that are easy to occur in the marine motor are analyzed.The bearing fault with the largest proportion is taken as the main research object to focus on in-depth research,to analyze its failure types and causes,and to analyze the vibration mechanism of rolling bearings;the fast Fourier transform method is used to visualize the vibration signal of the motor bearing.(2)Research on GAP-CNN Fault Diagnosis Method.In view of the complex fault characteristics of the collected signals,the manual feature extraction process is cumbersome and needs to rely on the knowledge of professionals.Based on the twodimensional convolutional neural network structure,a fault diagnosis method of motor bearings based on GAP-CNN is proposed for one-dimensional vibration signals.This method directly extracts the features of the time-domain signal,and outputs the fault category through the classifier.Replacing the fully connected layer with a global average pooling layer reduces the computational complexity of the model and reduces overfitting.The experimental results show that the GAP-CNN fault diagnosis model has a better diagnosis effect on the balanced sample set.(3)Research on fault diagnosis method of improved reward DQN.Aiming at the unbalanced distribution of the normal and the faulty bearing sample data volume in the actual environment,the GAP-CNN fault diagnosis model performs poorly,and a method based on the deep Q-learning network model is proposed for fault diagnosis of motor bearings.The method uses the GAP-CNN network structure as the neural network of approximation value function,constructs the personalized reward of the imbalanced data,and uses the interaction process between the agent and the environment to diagnose the fault of the imbalanced sample of the motor bearing.The experimental results show that the model proposed in this paper has a high diagnostic ability in the state of imbalanced data.(4)Research on Fault Diagnosis Method Based on SED-DQN.Aiming at the problems of slow convergence speed and many interactions of the agents in the deep Q network model,the exploration-exploitation dilemma in the reinforcement learning process are deeply analyzed,and a greedy exploration strategy based on stretching exponential decay is proposed.This strategy can control the agent’s exploration and exploitation behavior in the process of interaction with the environment,and control the agent’s exploration transition timing,exploration transition smoothness and decay rate.The improved model is applied to the motor bearing fault diagnosis process and experiments are carried out.The experimental results show that the agent based on SED-DQN speeds up the convergence speed,reduces the number of interactions,and still maintains a high accuracy.Finally,comparative experiments under different loads and different radial forces show that the model has good generalization performance.