CN-114492511-B - Fault diagnosis method based on digital twinning

CN114492511BCN 114492511 BCN114492511 BCN 114492511BCN-114492511-B

Abstract

The invention discloses a fault diagnosis method based on digital twin, which comprises the following steps of 1) constructing a corresponding virtual model based on physical characteristics of a physical model, 2) respectively operating the physical model and the virtual model to obtain actual measurement values and virtual values of a plurality of characteristics, 3) selecting parameters to be optimized based on a distance measurement method, then updating the digital twin virtual model by a particle swarm optimization algorithm, 4) operating the updated virtual model, constructing a training set based on data generated by the updated virtual model, training a convolutional neural network, 5) constructing a testing set based on data obtained by the physical model, analyzing the testing set by using the trained convolutional neural network, and outputting a diagnosis effect on equipment faults. According to the invention, through correlation analysis of virtual data and real data, a virtual model is improved, and the problem that the data volume required by an algorithm is not considered in the existing data-driven fault diagnosis method can be solved.

Inventors

LIU JIANG

Assignees

中云开源数据技术（上海）有限公司
中云开源数据技术（上海）有限公司

Dates

Publication Date: 20260421
Application Date: 20211231
Priority Date: 20211231

Claims (2)

1. The fault diagnosis method based on digital twinning is characterized by comprising the following steps: 1) Constructing a corresponding virtual model based on the physical characteristics of the physical model; 2) Selecting a plurality of characteristics, and respectively operating a physical model and a virtual model to obtain actual measurement values and virtual values of the characteristics; 3) Selecting parameters to be optimized of the virtual model from the actual measurement value and the virtual value based on a distance measurement method, and then updating the virtual model by using a particle swarm optimization algorithm; The specific steps of the step 3) are as follows: Extracting root mean square, kurtosis, peak, crest factor and skewness directly from a time domain, extracting average frequency, frequency center, root mean square of frequency distribution, standard deviation of frequency distribution and power envelope spectrum from a frequency domain, extracting effective value of frequency distribution and average envelope spectrum of frequency distribution from a time scale domain, analyzing similarity between a characteristic value obtained from actual measurement in a physical system and a characteristic value output by a dynamic model based on a distance measurement method, selecting parameters with the ranking of 6, and updating a virtual model by a particle swarm optimization algorithm; 4) Running the updated virtual model, constructing a training set based on data generated by the virtual model, and training a convolutional neural network by using the training set; 5) Analyzing the test set by using a trained convolutional neural network, and outputting a diagnosis effect on equipment faults; the process of constructing the virtual model is as follows: 1) Determining equations of motion for physical entities Regarding the rotor as a common multi-node Timoshenko beam, selecting 2 points to represent the dynamic characteristics of a physical system, wherein the degree of freedom adopted by each point is limited to 3; Wherein, the And Respectively represent each Direction and direction Is used for the buckling of the steel sheet, Representative of The torsion of the direction and reflects the section change of the shaft by considering the shearing force effect; the control equations of the physical system taking into account the inertial, restoring and damping forces and the constant vertical force acting on the bearing inner race are shown in equations (1) and (2); + + (1) (2) Wherein, the Representing the mass of the rotor and the mass of the inner ring supported by the bearing, Representing the equivalent viscous damping coefficient of the mass, Is the number of balls to be used, Represents the elastic deformation constant of the Hertz contact, Which means that the inner radial play is given, Represent the first The amplitude of the surface waves at the individual ball locations, Representing the radial controllable load of the device, Representing the forces generated by the imbalance of the rotor, Is the first The angular position of the individual rolling bodies, Is the main displacement of the center of the inner ring, The angular velocity of the cage bars/outer/inner rings, Representing time; subscript + number in formulas (1) and (2), When (when) Indicating loading angular position Is used for generating restoring force; When (when) When it indicates the unloaded angular position The restoring force of the rolling elements is 0; solving the equation by adopting a Newmark method or an implicit Newmark method to obtain the displacement, the speed and the acceleration of each point in the system; 2) Bearing failure modeling The bearing inner and outer ring faults are modeled as small segments with sinusoidal half-wave shape with angular width of Depth of When each ball passes through the defect area, introducing the ball into a virtual model of a motor fault test platform by increasing radial clearance, wherein the instantaneous restoring force when the bearing breaks down is calculated by formulas (3) and (4); (3) (4) Wherein, the The additional clearance is expressed by formula (5); (5) Wherein, the The angular position of the defect is indicated, The angular width of the defect is indicated, The relative angle representing the ball i and the defect location is a function of the cage angular velocity of the bearing represented by equation (6); 。
2. The fault diagnosis method based on digital twin according to claim 1, wherein the physical model is a three-phase asynchronous motor fault test platform, and comprises a controllable load motor, a rotor connected with a driving end of the controllable load motor, and two ball bearings arranged on the rotor, wherein a fault point is respectively arranged on an outer ring and an inner ring of each ball bearing, and sensors for acquiring data are respectively arranged in the horizontal, vertical and axial directions of the driving end of the controllable load motor.

Description

Fault diagnosis method based on digital twinning Technical Field The invention relates to the field of intelligent manufacturing, in particular to a fault diagnosis method based on digital twinning. Background Conventional machine learning algorithms or deep learning algorithms implement preventive maintenance of industrial equipment using historical data, which requires a large number of instruments to obtain the data, and are expensive, resulting in an insufficient amount of data to be collected for model training. In addition, training the machine learning model using the history data cannot guarantee the reliability of the failure diagnosis of the device whose life can be prolonged originally. In order to solve the problems, a digital twin theory technical system is introduced in the intelligent manufacturing field. The digital twin is to fully utilize data such as a physical model, sensor update, operation history and the like, integrate a multi-disciplinary, multi-physical quantity, multi-scale and multi-probability virtual process, and complete mapping in a virtual space, thereby reflecting the full life cycle process of corresponding entity equipment. Digital twinning is a concept beyond reality, and can be regarded as a digital mapping system of one or more important and mutually dependent equipment systems, a virtual model reflecting a physical system is constructed by utilizing the digital twinning idea, and sufficiently diversified data is generated in real time through the virtual model, so that the defect that an existing fault diagnosis method can only train a machine learning model by using historical data can be overcome. Disclosure of Invention The invention aims to provide a fault diagnosis method based on digital twinning, which continuously improves a virtual model through correlation analysis of virtual data and real data, thereby simulating enough diversified data for a machine learning algorithm, solving the defect that the traditional fault diagnosis method only can train the machine learning model by using historical data, and better providing service for preventive maintenance of industrial equipment. The technical aim of the invention is realized by the following technical scheme: a fault diagnosis method based on digital twinning comprises the following steps: 1) Constructing a corresponding virtual model based on the physical characteristics of the physical model; 2) Selecting a plurality of characteristics, and respectively operating a physical model and a virtual model to obtain actual measurement values and virtual values of the characteristics; 3) Selecting parameters to be optimized of the virtual model from the actual measurement value and the virtual value based on a distance measurement method, and then updating the virtual model by using a particle swarm optimization algorithm; 4) Running the updated virtual model, constructing a training set based on data generated by the virtual model, and training a convolutional neural network by using the training set; 5) And analyzing the test set by using a trained convolutional neural network, and outputting the diagnosis effect on equipment faults. The physical model is a three-phase asynchronous motor fault test platform and comprises a controllable load motor, a rotor connected with the driving end of the controllable load motor and two ball bearings arranged on the rotor, wherein the outer ring and the inner ring of each ball bearing are respectively provided with a fault point, and the horizontal, vertical and axial directions of the driving end of the controllable load motor are respectively provided with a sensor for acquiring data. Further, the process of constructing the virtual digital twin virtual model is as follows: 1) Determining equations of motion for physical entities Regarding the rotor as a common multi-node Timoshenko beam, selecting 2 points to represent the dynamic characteristics of a physical system, wherein the degree of freedom adopted by each point is limited to 3; Wherein, the AndRespectively represent eachDirection and directionIs used for the buckling of the steel sheet,Representative ofThe control equations of the physical system taking into account inertial, restoring and damping forces and constant vertical forces acting on the bearing shaft inner race are shown in equations (1) and (2); ++ (1) (2) Wherein, the Representing the mass of the rotor and the mass of the inner ring supported by the bearing,Representing the equivalent viscous damping coefficient of the mass,Is the number of balls to be used,Represents the elastic deformation constant of the Hertz contact,Which means that the inner radial play is given,Represent the firstThe amplitude of the surface waves at the individual ball locations,Representing the radial controllable load of the device,Representing the forces generated by the imbalance of the rotor,Is the firstThe angular position of the individual rolling bodies,Is the main di