Research On Equipment Condition Monitoring And Intelligent Diagnosis Based On Multi-source Data Fusion

Bearings and gears are the most typical transmission components of rotating machinery.They work in high-speed,high-load conditions for a long time,and they are most prone to failures.To avoid significant economic loss due to machine faults,equipment condition needs to be monitored and diagnosed to ensure safe and reliable operation.A traditional diagnostic measure for equipment fault diagnosis is using the characteristics of vibration signals with a single source and model.In high-temperature and high-corrosion environments,the contact measurement of vibration signals cannot meet the needs of industrial production.In this case,acoustic signals can be collected to reflect the running state of equipment via noncontact measurement with microphones,which can provide complementary information to vibration signals to improve fault diagnosis.With the support of Ministry of Industry and Intelligent Manufacturing Demonstration Project,this dissertation explores the monitoring and intelligent diagnosis techniques of mechanical equipment,taking bearings and gears as the research objects.We applied the latest AI techniques such as deep learning,data fusion,neural network to diagnose the state of devices from two aspects including single-source multi-sensor integration and multi-source multisensor data fusion.Intelligent classifiers are used to obtain the primary diagnosis results of a single source and model.Combined with the improved evidence theory,a reliable device state can be obtained by further fusion of primary results.The main contents of this dissertation are as follows:(1)Evidence theory based conflict resolution for multi-source data fusionTo address the problem that evidence theory cannot effectively deal with evidence conflicts,we proposed an improved fusion algorithm.First,the “one-vote veto” is avoided by using the neighborhood values.Second,the distance function and Delphi method are employed to correct the basic probability assignment of evidence.Third,the corresponding fusion rule is selected to complete data fusion according to the relationship between conflict factors and thresholds.Finally,the improved evidence theory is applied to the fault diagnosis of rolling bearing,which achieved higher accuracy.(2)Integrated model fusion for fault diagnosis based on single-source multi-sensorTo address the problem that a nonintegrated model is constrained by a single structure and has difficulty reflecting the full state of device,an integrated diagnostic model based on improved DS evidence theory is proposed.This proposed model combines learning vector quantization(LVQ)neural network and decision trees.First,the statistical characteristics are extracted from the experimental data of bearing faults published by Case Western Reserve University.Second,principal component analysis is used to reduce the dimension of statistical features.The characteristics of the fan and driver ends are then sent to the LVQ classifier and decision tree model,respectively,for preliminary diagnosis.Finally,to realize the diagnosis of bearing fault,the fault recognition rate of the single model is taken as evidences for further fusion by the improved fusion algorithm.(3)Fault diagnosis based on multi-source multi-sensor data fusionDue to the limitation of single source and sensor performance,it is difficult.to reflect the full state of equipment under different working conditions.To address this issue,we proposed a multi-sensor data fusion method based on convolution neural network(CNN),which combines vibration and acoustic signals.First,a gear box fault platform is built in a semi-anechoic room according to experimental requirements.Vibration and acoustic signals are then collected under different working conditions.Second,vibration signals are preprocessed into time–frequency maps and sent to the adaptive stacked CNN for primary diagnosis.Acoustic signals are sliced directly into the end-to-end stacked CNN model.Finally,the improved fusion algorithm is used to integrate further the primary results of vibration and acoustic signals to obtain accurate and reliable gear state.(4)Development of the integrated system based on multi-source data fusion and big data analysisWe developed an integrated fault diagnosis system based on our theoretical research of multi-source data fusion and big data analysis.First,we analyze the generation,processing and the technical architecture of multi-source manufacturing of big data in a workshop and propose the functional requirements of the system.Then we construct the system's perceptual model and technical framework.Next,we expound the choice of databases and data structure design and the realization of functional modules.Finally,we applied the system in an electrical appliance company in Guizhou Province and demonstrated its effectiveness.By implementing the system,the maintenance cost of the enterprise is reduced along with improved efficiency.