Research On Intelligent Fault Diagnosis Method Of Rotating Machinery In The Tensor Space Of Multi-source Signals

As the key equipment in modern intelligent manufacturing,rotating machinery has been widely used in the fields of energy,transportation and national defense.Due to the complexity and high coupling of the system structure,the normal operation of the whole equipment may be affected once a component fault occurs in rotating machinery.Therefore,to ensure the long-term safe,efficient and reliable operation of rotating machinery,it is of great practical significance to carry out health monitoring and intelligent fault diagnosis of rotating machinery and its key components.At the same time,as the rotating machinery equipment tends to be more integrated and intelligent,the system structure of the equipment is also complicated,and the fault modes of the equipment are also increased and coupled with each other.When the equipment has different types of faults,the vibration,temperature,pressure,torque,speed and other multi-source condition signals are used to fully reflect its working condition and ensure the authenticity and integrity of the sample mode,and it is difficult to identify different faults only by one kind of conditon signal.Therefore,it is necessary to explore the intelligent fault diagnosis method of rotating machinery based on multi-source signals.At present,the diagnosis method based on multi-source signals is mainly focused on signal fusion,but signal fusion is essentially intelligent diagnosis method based on the feature vector of multi-source signals,while the spatial structure and related internal information of multi-source signals are ignored.On the contrary,the feature tensor representation of multi-source signals can contain more abundant data structure information.Therefore,with the support of National Natural Science Foundation of China(No.: 51875183,51975193),the fault detection methods,fault diagnosis methods and feature tensor autonomous learning methods of rotating machinery with multi-source signals in the tensor space are researched in depth,which provides new ideas and methods for the intelligent fault diagnosis of rotating machinery with multi-source signals.The main research work of the paper is as follows:(1)Aiming at the problem of fault detection of rotating machinery with multi-source signals,a novel method in the tensor space called one class convex hull based tensor machine(OCCH-TM)is proposed.Firstly,the convex hull model of sample set in the tensor space is defined.Then,the minimum norm optimization problem is constructed based on origin and convex hull.Finally,combining with the feature tensor based on wavelet time-frequency spctrum of multi-source signals,a fault detection method of rotating machinery based on OCCH-TM is proposed.The effectiveness of this method has been verified by the experimental results of gear fault detection.(2)To solve the problem of unbalanced binary classification caused by the lack of fault samples in intelligent fault diagnosis of rotating machinery,support tensor machine with dynamic penalty factors(DC-STM)is proposed.Firstly,a dynamic parameter has been introduced into DC-STM,and its specific calculation formula is defined.Then,according to the dynamic parameters,the penalty factor of the class is changed,and the optimal hyperplane is obtained by increasing the penalty for a small number of samples in the unbalanced binary classification.Finally,combining with the feature tensor based on wavelet time-frequency spctrum of multi-source signals,an intelligent fault diagnosis method for rotating machinery based on DC-STM is proposed.The effectiveness of the proposed method is verified by the experiments of gears and bearings with unbalanced samples.(3)For the problem of underestimation of rotating machinery health conditions with multi-source signals in DC-STM,a novel tensor classfier called flexible and replaceable convex hull based tensor machine(FDCH-TM)is proposed.Firstly,based on tensor convex hull,flexible factor and displaceable factor are introduced to define a new geometric model of tensor space,namely flexible and displaceable convex hull(FDCH).Then,the optimal hyperplane is constructed based on the nearest neighbor points between flexible and displaceable convex hull,and the decision function of FDCH-TM is obtained.Finally,combining with the feature tensor based on wavelet packet decomposition of multi-source signals,an intelligent fault diagnosis method of gearbox with multi-source signals based on FDCH-TM is proposd.The effectiveness of FDCH-TM in intelligent fault diagnosis is verified by gearbox experiment.(4)As the boundary of the DC-STM and FDCH-TM geometric models is too complex and compact,flexible and displaceable hyperdisk based tensor machine(FDHD-TM)is proposed.In FDHD-TM,the tensor hyperdisk model is defined firstly based on the intersection of affine hull and hypersphere of tensor space.Then the flexible factor and displaceable factor in FDCH-TM are introduced to define the flexible and displaceable hyperdisk.Finally,the Quadratically Constrained Quadratic Program is constructed with the idea of maximum interval and it is solved by MOSEK solver.The gearbox experiment verifies that FDHD-TM can effectively identify and intelligently diagnose different health coditions of gear.(5)To overcome the uncertainty of intelligent fault diagnosis results caused by the feature tensor extraction which relies on artificial experience,a feature tensor autonomous learning and intelligent diagnosis method of multi-source signals based on ensemble convolutional neural network(CNN)based multi-source signal is proposed.Firstly,an improved CNN is proposed.Secondly,an ensemble CNN model is constructed for the feature tensor autonomous learning of multi-source signals.Finally,the feature tensors obtained by ensemble convolutional neural network are intelligently diagnosed through a tensor classifier.The gearbox experiment proves that this method can be used for feature tensor autonomous learning and intelligent diagnosis based on multi-source signal of gearbox.Based on the multi-source signals tensor space of rotating machinery,the one classification model,unbalanced binary classification model,multi classification model and feature tensor autonomous learning method have been researched to solve the problems of fault detection and intelligent diagnosis of rotating machinery and its key components.Meanwhile,the corresponding fault detection and intelligent diagnosis methods of rotating machinery with multi-source signals have been proposed.The effectiveness of the proposed methods to the intelligent diagnosis of rotating machinery with multi-source signals has been further verified by the results.