Research On Bearing Fault Diagnosis And Remote Monitoring System Based On Constrainted Dynamic Ensemble ELM

As the core components of the transmission of rotating machinery,rolling bearings have been widely used in various types of mechanical equipment since their inception.In the field of national major equipment construction,such as high-speed trains,aerospace,etc.,higher requirements are put forward for bearing quality and performance.The healthy and stable operation of the bearing directly determines the performance of major equipment,and if the bearing fails,it will directly affect its normal service,and even bring serious equipment and personnel safety hazards.This paper mainly takes rolling bearings as the research object,through the study of signal analysis technology and machine learning technology,to achieve intelligent diagnosis of bearing faults,and in order to respond to the requirements of the development of the national intelligent manufacturing network,a monitoring system integrating signal acquisition,remote monitoring,real-time diagnosis and personnel and equipment management is designed and developed.The main work of this article is as follows:(1)Starting from signal pre-processing technology for intelligent diagnosis of bearing faults,the common time domain analysis and frequency domain analysis methods based on vibration signals are introduced.Subsequently,the computational process of the neural network based on the back-propagation(BP)algorithm is derived,and the algorithm principle of the Extreme Learning Machine(ELM)algorithm model that cancels the BP process is further derived.To further increase the performance cap of ELM,the advantages and disadvantages of two commonly used ensemble learning models,Bagging and Boosting,and their algorithm principles are analyzed.(2)Aiming at the problem that the hidden layer weight matrix generated by ELM is not fully utilized,an autocorrelation constrained extreme learning machine(ACELM)algorithm based on the timing vibration signal is proposed,which improves the directionality and discrimination of the hidden layer output.When using ACELM as the base classifier for the AdaBoost integration algorithm,although the model converges faster and the stability and accuracy are improved,it loses the characteristic of fast training.For this purpose,the process of sample acquisition and weight calculation in each round of AdaBoost training is improved,and a dynamic integration algorithm is proposed.ACELM was used as the base classifier to construct the Autocorrelation Constrained Dynamic Ensemble Extreme Learning Machine(ACDE~2LM).Performance comparison and testing using the bearing dataset provided from Case Western Reserve University shows that the classification accuracy of the proposed ACDE~2LM algorithm is 2.7%higher than that of the unintegrated ACELM,12.2%higher than that of the traditional ELM,and the training speed is maintained,the training time is maintained within 1s,and the false negative rate for fault classes is only 0.8%.(3)According to the actual needs,the web-based bearing fault diagnosis and remote monitoring system is designed and built.Through the decomposition of system functions,the system is divided into data acquisition module,Web server module,database module and Web client module,and joint development is completed under different system platforms.Through the cloud server,the intranet penetration service is set up to achieve public network access to intranet Web services.The proposed diagnostic algorithm is embedded in the Web server in the form of a scheduled task,and automatic diagnosis can be performed when the system is running.In order to verify the necessity of online model update,the diagnostic model is updated in multiple batches using self-test data,and the accuracy rate continues to increase as the data is supplemented,and it is always better than the traditional ELM algorithm.