Research On Motor Bearing Fault Diagnosis Based On Industrial Internet Of Things

With the increasing size,complexity,speed,integration,and intelligence of modern industrial equipment in China,it is necessary to monitor the health and diagnose the faults of key components in mechanical equipment in order to ensure their safe and long-term operation,which can prevent major safety incidents.Currently,the diagnosis and maintenance of the health status of motor bearings are still accomplished using regular inspections carried out by experienced workers.However,this method,which heavily relies on experiential knowledge,has significant limitations in determining the health status of bearings,including the long time required to accumulate expertise and the low accuracy of identification.With the continuous development of artificial intelligence technology,deep learning-based intelligent fault diagnosis is gradually being applied to the diagnosis of faults in mechanical equipment,including electric motors.This technique automatically extracts fault feature information that can help with classification from data using deep learning models,thereby reducing the influence of manual intervention and improving the accuracy of fault identification and the efficiency of system operation.Addressing the problem that operating conditions of bearings often change during practical operation,deep learning methods require that the training and testing data have the same distribution.This has led to certain limitations when deep learning methods are used for diagnosing motor bearing faults under different operating conditions.Transfer learning,as a new learning paradigm,can adaptively apply knowledge learned in one domain to another domain based on data or task similarity,enabling the model to have a transferable ability.Therefore,this thesis focuses on deep transfer learning and industrial Io T application scenarios,and studies some key issues in the field of motor bearing fault diagnosis,such as feature extraction and variable operating condition fault diagnosis.The main contents of this thesis are as follows:(1)Combining the Fast Fourier Transform with the Maximum Mean Discrepancy-based FRNMM model.In real-world industrial equipment operation,it is often difficult to ensure that machines run under consistent working conditions,leading to the collection of non-stationary data.Fault samples collected from real-world scenarios predominantly lack labeled data due to their real-world nature,potentially reducing the accuracy of diagnosis due to significant distribution variations.In order to address this issue of variable working conditions in bearing fault diagnosis,we propose a method that combines fast Fourier transform with feature transfer learning.While model training requires labeled data,data collected from actual industrial internet of things scenarios usually lack such labels,presenting a challenge for fault diagnosis tasks.To address this challenge,we propose a FRNMM model which integrates fast Fourier transform with maximum mean discrepancy.The model is first trained on labeled data under different working conditions,and then utilized to diagnose target data without labels.The experimental results demonstrate that the proposed method performs well in bearing fault diagnosis tasks with variable working conditions.(2)A device data acquisition system based on the Industrial Internet of Things was built in response to the actual operating scenarios of industrial equipment.Opensource components such as EMQ X,TDengine,and Grafana were used to create the system,which has the functions of data collection,storage,visualization,and query.This system provides intuitive observation of the entire data acquisition system’s operational status.To verify the data acquisition system’s usability,Industrial Internet of Things data acquisition system testing was conducted using Node.js backend technology.Overall,modification improved the paragraph’s readability,clarity,and concision while also adhering to academic style.(3)Design and implementation of an electric motor bearing fault diagnosis platform.After analyzing the functional requirements of the motor bearing fault diagnosis system,the present study designed a new motor bearing fault diagnosis system that incorporates the signal feature extraction and diagnosis methods researched herein.The system is capable of performing various functions including health status diagnosis,signal feature extraction,data management,user management,and equipment management for bearings.The system has passed functional test and facilitates diagnosing motor bearings by workers.