Fault Diagnosis Method Of Rotating Machinery Based On Improved Geometric Model

Rotating machinery is an important part of all kinds of mechanical equipment,the most common are gears and bearings.In order to reduce the loss caused by the failure of such key parts,it plays an important role in carrying out fault diagnosis to ensure the healthy operation of the whole equipment system.In this thesis,the fault diagnosis model is constructed by combining fault feature extraction and pattern recognition.We study the theoretical basis and practical application of Convex Hull(CH)and Polyhedral Conic(PC)geometric classification models.The two geometric models are improved.The Probability Output Flexible Convex Hull classification model is proposed and Rotated and Extended Polyhedral Conic classification model also have been successfully applied to fault diagnosis of rotating machinery.The major contents of this thesis are as follows:(1)The classification hyperplane of the original convex hull and flexible convex hull(FCH)model classifiers only uses the decision-making contribution of the outer boundary points of the convex hull.Therefore,when the sample distribution is uneven or the outer boundary points are outliers,the classification accuracy of CH and FCH will be reduced.To solve this problem,this thesis proposes a new method Probability output flexible convex hull binary classification model.In this method,we map the distance from data point to its original classification hyperplane to between 0 and 1,and the initial classification hyperplane is further optimized by probability fitting.Finally,the output probability is used for classification decision,which improves the defects of the original classifier.(2)In order to solve the multi classification problem,two multi classification models are proposed based on the probability output flexible convex hull binary classifier,the probability output flexible convex hull decision tree multi classification model based on decision tree and the multi probability coupling output multi classification model based on one-to-one.Experiments on different fault data show that the two multi classification models have good effect and good performance,and can effectively identify the fault.(3)In order to solve the problem that the decision boundary of polyhedral cone classification model is difficult to fit the real positive class region,which leads to low classification accuracy and poor robustness,a rotated and extended polyhedral conic classifier is proposed.The classifier adds rotation factor on the basis of L1 norm vectorization,which can achieve the appropriate scale expansion and boundary domain increase on the classification boundary;the boundary of the classifier can better fit the positive class region;thus the accuracy of the original classifier is improved.Combined with the time-frequency characteristics,the method is applied to the fault diagnosis of rotating machinery.The experimental data analysis results show that the rotating polyhedron cone classifier is effective,and it can accurately identify the mechanical working state and fault type in the case of small samples,and has good robustness and anti-interference ability.