Research On Rotating Machinery Fault Diagnosis Method Based On Deep Reinforcement Learning

Rotating machinery is the most commonly used type of machinery in modern industrial,civil and military applications.Due to the long-term operation of rotating machinery equipment under high load and harsh working conditions,failure will inevitablely occur.If the fault cannot be diagnosed in time,it will cause the entire system to shut down or even cause casualties.The traditional manual featured extraction method relies on expert experience and knowledge,and fails to directly establish the mapping relationship between the original data domain and the fault pattern recognition,which brings certain complexity and instability to the fault diagnosis pair,which will inevitably affects the diagnosis result.Deep reinforcement learning can directly learn from the original time-domain data without manually extracting features.It is an end-to-end diagnostic model that is more in line with human learning mechanisms.Therefore,this paper introduces deep reinforcement learning technology into the field of fault diagnosis to solve this problem,and studies the application of deep reinforcement learning in the field of fault diagnosis.Aiming at the problem that fault features in traditional rotating machinery fault diagnosis depend on manual extraction,a method for fault diagnosis of rotating machinery based on deep Q learning is proposed.This method can directly achieve fault diagnosis through end-to-end learning without extracting fault features in advance.First,use vibration signals to construct an environment state space for the interaction between the agent and the environment;Secondly,the Q function in Q-learning is fitted with CNN to obtain the deep Q network,and the state returned by the environment is input into the deep Q network to learn the specific state feature representation of the fault data,and the learning strategy is represented accordingly;Finally,through continuous interactive learning between the agent and the environment to maximize the Q function value,the optimal strategy is obtained to achieve fault diagnosis.At the same time,in order to solve the problem of low recognition rate when a single agent diagnoses the planetary gearbox in deep Q learning,multiple agents are used for strategy learning,and fuzzy integrals are used to fuse the decision results of multiple agents.A planetary gearbox faults diagnosis method based on deep Q learning and fuzzy integration is proposed.First,the continuous wavelet transform and S transform of the vibration signal are used to obtain the corresponding time-frequency feature matrix,and then the multi-domain environment state space is constructed using the original timedomain data and the obtained time-frequency feature matrix to interact with multiple agents one by one;Through the deep Q-learning algorithm,the Q function value of each agent is maximized to obtain the optimal strategy.Finally,fuzzy integration is used to fuse the decision results of multiple agents to obtain the final diagnosis result.The verification analysis of the planetary gearbox fault data shows that the diagnosis result of decision fusion is better than that of a single Agent.Finally,according to the fault characteristics and diagnostic requirements of rotating machinery,a deep reinforcement learning fault diagnosis system for rotating machinery integrating signal processing,sample division,model training and other functions was developed.The system not only contains training modules for convolutional neural networks and deep Q networks,but also uses ADO.NET technology to achieve effective database management.The system mainly includes: vibration signal basic analysis module(waveform analysis,FFT,Hilbert envelope spectrum analysis,etc.),vibration signal sample division module,vibration signal diagnosis module(including the construction,training,and evaluation of diagnostic models).Finally,the effectiveness and feasibility of the system are verified by the data of bearings and rotors.