Research On Intelligent Fault Diagnosis Of Rotating Machinery Based On Incomplete Supervision Learning

Rotating machinery plays an important role in modern manufacturing,and for modern manufacturing industry production,the reliable operation of rotating parts is very important.However,with the continuous upgrading of the modern manufacturing industry,the structure of machinery and equipment is more complex and sophisticated,and researchers in equipment maintenance and diagnosis are looking for breakthroughs from the perspective of artificial intelligence and big data technologies to address this situation.However,under real circumstances,most of the fault data of modern manufacturing are incomplete or even missing.How to use these incompletely labeled data for effective diagnosis poses a challenge to diagnose the faults.In order to solve this problem,this paper takes weakly supervised learning technology as the core,and builds a new general intelligent fault diagnosis model.It studies the intelligent fault diagnosis technology of mechanical equipment.Only a small amount of relevant data can be used to accurately diagnose mechanical equipment.The fault categories have greatly improved the efficiency of mechanical equipment diagnosis.In weakly supervised learning,the problems of overfitting,negative transfer,underfitting,and under-adaption are negative-transfer.There are overfitting and under-fitting problems when fitting the unknown probability distribution of rotating equipment fault data;under-adaption refers to the problem that the probability distribution mismatch of fault data obtained under different working conditions cannot be corrected,and negative-transfer It means that the assisted domain tasks have a negative impact on the target tasks,and these have become the research pain points of intelligent diagnosis of rotating equipment under incomplete labeled data.This article focuses on under-fitting and under-adaption fault diagnosis research.The main innovations are:(1)Aiming at the problem of under-fitting,an intelligent fault diagnosis model based on non-parametric transfer learning is proposed from the perspective of domain adaptation to build cross-domain feature migration and reduce the feature distance between different distributions,so only a small number of targets are required The data can get higher classification accuracy.At the same time,the proposed method is applied to the rolling body fault data set,and the effect of no less than any existing transfer learning method can be obtained without parameter adjustment and model selection,making the fault diagnosis technology more practical.(2)For the problem of under-adaptation,it is noted that there are some common features between the target domain and the source domain.For this,the shared features can be directly selected to build the model.Based on this,an intelligent fault diagnosis method based on optimal transmission of unsupervised migration feature selection is proposed and applied to the rolling bearing fault data set.Second extraction based on feature extraction reduces the problem of migration learning domain offset,further improves classification accuracy,and proves the effectiveness of the method.(3)Aiming at the problem of under-fitting,a new graph inference semi-supervised learning algorithm based on Fourier’s law is proposed from the perspective of thermodynamics,which can be more close to the real mechanical vibration characteristics and provide a unique idea for the fault diagnosis of rotating machinery.Then apply it to the rolling bearing data set.Through experimental verification,we can find that this method is the closest to the real state of mechanical operation in the existing semi-supervised learning method.At the same time,the comparison with other methods also shows that this method is the most One of the advanced methods provides theoretical support for intelligent fault diagnosis technology in the field of machinery.(4)Aiming at the problems of under-fitting and under-adaption,an intelligent fault diagnosis model based on transfer semi-supervised learning is proposed,which combines the advantages of domain adaptation and semi-supervised learning from the perspective of subspace learning instead of simple method superposition.The effectiveness and feasibility of this method are verified from the two bearing data sets.Solve the pain points faced by the Intelligent Fault Diagnosis Research Institute under the extreme situation of uneven data distribution and incomplete data labels in the target domain.