Machine Learning Method For Fault Diagnosis Of A Class Of Shafting Equipment

The development of production and the continuous progress of science make the structure of modern mechanical equipment more and more complicated.As various functions are becoming more and more comprehensive,the degree of automation of mechanical equipment is also constantly improving.It has made an important contribution to economic development.The shafting equipment is an important part of mechanical equipment.The service life of shafting equipment is limited and prone to failure.These failures may cause problems such as reducing the expected efficiency of the equipment and causing the equipment to stop functioning.It even caused more serious catastrophic accidents.It is important to find faults in time and identify fault types.This will not only help extend its service life but also effectively avoid the occurrence of dangerous accidents.Therefore,it is the necessary to research on the fault diagnosis method of shafting equipment.In recent years,machine learning algorithms have developed rapidly.It has the advantages of high degree of automation and fast learning speed.Machine learning has been widely used in fault diagnosis of various devices.Therefore,this paper mainly focuses on the machine learning based fault diagnosis method of shafting equipment.The main research content includes three major steps: data collection,data preprocessing,classifier training and classification.First,sensors are used to monitor the device status.Then the collected data was pre-processed to extract feature information.Finally,a modal classifier is designed based on the feature information to diagnose the current operating mode of the device.The simulation data is obtained from the ZHS-2 flexible rotor test bench.Theinnovative work is mainly reflected in the data preprocessing stage and the classifier training stage.It mainly includes the following three points:1.In order to obtain machine learning training and test samples.It is necessary to segment the original time-domain data collected by the monitoring sensor.For shafting equipment,when a single spin cycle period is used as the segmentation criterion,the features contained in the obtained single sample are not enough.When segmenting in multiple cycles,the resulting single sample contains a large amount of data.It will affect the training and testing efficiency of most machine learning classifiers.To solve this problem,this paper proposes three data preprocessing methods based on the information fusion theory: weighted fusion preprocessing method,merging fusion preprocessing method and direct fusion preprocessing method.They are used to further pre-process the data after multiple period division processing.In the experimental verification section.The effectiveness of the above method is verified through simulation experiments.2.To solve the problem that the parameters of the traditional machine learning classifier cannot be updated;the problem that the untrained unknown fault data cannot be classified.A fault diagnosis method for shafting equipment based on Kalman Filter based Sequential Kernel Extreme Learning Machine is proposed.It can continuously update the parameters according to the test data in the subsequent test process to achieve the purpose of tracking failure.In addition,a method that based on multi Kalman Filter based Sequential Kernel Extreme Learning Machine networks is proposed.By setting a threshold to determine whether the unknown operating mode data has been trained.And automatically build a new classifier for untrained data to classify it.It can be used to detect and diagnose unknown faults.Finally,through simulation analysis and comparison with other methods,the effectiveness of the proposed method is verified.3.To solve the problem that the main features in the data at the early stage ofthe fault are easily overwhelmed by noise.This paper proposes a fault diagnosis method combining multi-sensor fusion with cyclic bispectrum slice analysis.First use multi-sensor fusion to enhance the main features contained in the data.Then using the cyclic bispectrum estimation to analysis the fused data.By analyzing the slice chart of the frequency of the cyclic bispectrum at the peak of the spectrum,the characteristic frequency of the fault is extracted.By analyzing the cyclic slice of the frequency at the peak of the spectrum to extract the fault characteristic frequency.Eventually,the frequency corresponds to the fault type one by one,and the fault type can be quickly obtained by using such a mapping relationship.Early diagnosis of faults can be achieved using relatively few fault data.And verify it through simulation analysis.