A slow time-varying weak fault diagnosis method for RV reducer

Abstract

The application discloses a slow time-varying weak fault diagnosis method of RV reducer, comprising the following steps: establishing a slow time-varying micro-damage fault evolution model of the RV reducer according to a fast time-varying power transmission operation process and a slow time-varying damage evolution process of the RV reducer; optimizing and solving parameters of the slow time-varying micro-damage fault evolution model of the RV reducer and performing simulation analysis to obtain a slow time-varying micro-damage fault rule of the RV reducer and verify the rule; establishing a weak fault noise suppression criterion of the RV reducer under different conditions according to fault symptom knowledge; and establishing a deep neural network model based on a swarm intelligence algorithm to identify a weak fault type and quantitatively calculate a damage degree. The application can obtain a filtering and noise reduction method by analyzing a slow fault mechanism of the RV reducer, retain a detail signal and accurately extract a fault feature. Meanwhile, a pattern recognition model fusing a swarm intelligence optimization algorithm and a deep learning network is introduced to solve the problem of fast and accurate identification of the weak fault of the RV reducer.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of RV reducer fault diagnosis, and particularly relates to a slow time-varying weak fault diagnosis method for an RV reducer. BACKGROUND

[0002] The RV reducer is one of the key core components of intelligent manufacturing equipment such as precision CNC machine tools, robots and high-end medical equipment, is an important support for cultivating and developing strategic emerging industries, and is an important symbol of measuring the level of intelligent manufacturing equipment of a country. However, the RV reducer is prone to weak faults, which can cause various shaft systems, gears and bearings to fail, and can even cause major economic losses and even casualties. Therefore, the accurate diagnosis of weak faults is concerned in the field of intelligent manufacturing equipment and high-end medical equipment. State monitoring of the RV reducer and timely identification of weak faults are key issues to ensure safe and reliable operation of equipment.

[0003] Due to the high precision of the RV reducer, the corresponding fault diagnosis identification has the problem of low accuracy. Because the running process of the RV reducer is a fast-changing dynamic process, compared with the time scale of system dynamics, the damage evolution process is a slow-changing process, and the fast-changing dynamic process and the slow-changing damage evolution process have significant differences in time scale. Therefore, theoretically, the fault development of the RV reducer is a nonlinear layered evolution of fast-changing dynamics-slow-changing damage coupling development process. Therefore, the difficulties in solving this problem mainly include: ① the input excitation signal has the characteristics of strong nonlinearity and non-stationarity, the structure parameters are unknown (mass, stiffness, damping), the load working condition is random, the environment is time-varying, and the like are coupled, which makes the fault mechanism time-varying complex, the fault mechanism unclear, and the fault feature unclear. ② The fault signal is weak and the noise interference is strong, especially the test signal is limited, which makes it extremely difficult to separate the fault information from the complex environment. ③ The weak fault feature is not clear and slow, which causes the fault diagnosis identification model to have low precision and poor generalization ability.

[0004] At present, the research methods of weak fault diagnosis and identification mainly include signal processing methods and deep learning methods. The signal processing methods mainly include wavelet transform, empirical mode decomposition, fast spectral kurtosis and other methods, which are used to suppress signal noise interference and enhance the usability of frequency domain spectral analysis methods. However, the effect of wavelet transform is affected by the selection of basis wavelet, and the adaptability is poor. The EMD method is prone to end effect and modal aliasing. The FSK method is affected by random impact, and cannot accurately find the optimal resonance frequency band. At the same time, the weak fault diagnosis and identification based on signal processing method can only achieve simple data-driven fault mechanism constant fault signal noise interference suppression, and when the fault mechanism is time-varying, the fault signal noise interference suppression still lacks in-depth research. The convolutional neural network method is a typical deep learning method, which directly takes the original vibration signal as the input of the network model, and can avoid losing important time domain features. The disadvantage is that the fault feature extraction is subject to human experience, and the model is established on the basis of the laboratory assumption condition that "typical fault information is rich and health mark information is sufficient", so it is difficult to extend and migrate the ideal diagnosis model to engineering practice. Therefore, in view of the problem of extracting weak fault features from the RV reducer state monitoring information under the slow time-varying fault mechanism, more effective methods need to be explored. SUMMARY

[0005] The purpose of the present application overcomes the deficiencies in the prior art, and to achieve the above purpose, a RV reducer slow time-varying weak fault diagnosis method is adopted to solve the problems in the background technology.

[0006] A RV reducer slow time-varying weak fault diagnosis method, the specific steps include:

[0007] According to the fast time-varying power transmission operation process and slow time-varying damage evolution process of the RV reducer, a RV reducer slow time-varying weak fault evolution model is established;

[0008] The RV reducer slow time-varying weak fault evolution model parameters are optimized and simulated to obtain the RV reducer slow time-varying weak fault law and verify it;

[0009] According to the fault symptom knowledge under different conditions, a RV reducer weak fault noise suppression criterion is established;

[0010] A deep neural network model is established based on swarm intelligence algorithm to identify the weak fault type and quantitatively calculate the damage degree.

[0011] As a further scheme of the present application: the specific steps of establishing the RV reducer slow time-varying weak fault evolution model according to the fast time-varying power transmission operation process and slow time-varying damage evolution process of the RV reducer include:

[0012] The dynamic response differential equation of the RV reducer for fast variable power response process is as follows:

[0013]

[0014] Wherein M, C and K are mass matrix, damping matrix and stiffness matrix respectively, F is external excitation force, and q is displacement matrix.

[0015] Let The dynamic response differential equation is converted into the expression in the form of state space as follows:

[0016]

[0017] Wherein, B=[M -1 F 0] T ;

[0018] The differential equation of the RV reducer for damage evolution process is as follows:

[0019] Φ=g(Φ,Y);

[0020] Wherein Φ is a slow variable damage variable.

[0021] The expression of A in the form of state space is as follows:

[0022] A=μ(Φ);

[0023] Based on the significant difference in time scale between the fast variable power response process and the slow variable damage evolution process, a numerical small time scale constant v is introduced to represent the separation of the two processes in time scale.

[0024] According to the above formula and definition, the slow variable micro-damage failure evolution model of the RV reducer under the coupling of dynamics and damage is as follows:

[0025]

[0026]

[0027] Wherein, is a fast variable subsystem for describing the power behavior of the component, is a slow variable subsystem for describing the damage evolution behavior of the component; the two subsystems are coupled through the dynamic response variable Y and the damage variable Φ.

[0028] As a further scheme of the application: the specific steps of optimizing and solving the RV reducer slow variable micro-damage failure evolution model parameters and performing simulation analysis, obtaining the RV reducer slow variable micro-damage failure law and verifying the RV reducer slow variable micro-damage failure law include:

[0029] The damage variable Φ is set within n cycles T of the rotating component transmission of the RV reducer, and the time nT is specified. M-1 It remains unchanged; after n cycles, the damage variable evolves into Φ. M ;

[0030] The solution steps for the slowly varying micro-damage fault evolution model are as follows:

[0031] μ(Φ) M-1 Substituting the equations of the fast variable subsystem, the vibration response Y of the system over n periods is solved numerically. M-1 ;

[0032] Then Y M-1 Substituting into the equations of the slowly varying subsystem, the damage variable Φ after n cycles is solved numerically. M ;

[0033] Repeat the above two steps until the system experiences performance failure;

[0034] Finally, using failure analysis equipment and physical failure analysis methods, the fault law of the RV reducer with gradual micro-damage was verified.

[0035] As a further aspect of the present invention: the specific steps for establishing a weak fault noise suppression criterion for the RV reducer under different conditions based on fault symptom knowledge include:

[0036] If the fault symptoms are known, the specific steps include:

[0037] By combining wavelet filtering, Kalman filtering and principal component analysis, and using feature selection techniques to obtain evaluation functions, fault noise features of key components of the RV reducer are extracted.

[0038] By utilizing the feature selection method of deep neural networks, and combining wavelet filtering, Kalman filtering and time-varying damage model analysis, a weak fault noise suppression criterion for key components of RV reducers is established.

[0039] The feasibility of the algorithms for each part was verified through numerical simulation;

[0040] If the knowledge of the fault symptoms is unknown, the specific steps include:

[0041] By combining multi-source information fusion technology with time-varying damage model analysis, an evaluation function for feature selection is obtained, and fault noise features are optimized and extracted.

[0042] By using the deep neural network fault extraction method, a direct mapping relationship between the high-dimensional fault features extracted by the deep neural network and noise suppression is established, and the weak fault noise suppression criterion of the key components of the RV reducer is determined.

[0043] The feasibility of the algorithm of each part is verified by numerical simulation.

[0044] As a further scheme of the application: the specific steps of the deep neural network model based on the swarm intelligence algorithm, the identification of the weak fault type, and the fault quantitative calculation of the damage degree include:

[0045] Based on the phase space curve variation method, a damage evolution residual model is constructed.

[0046] The field adaptation constraint is introduced, the transfer learning and the deep learning network are fused, the fault features in the RV reducer monitoring data from engineering practice are transferred, and a RV reducer time-varying weak fault deep migration diagnosis method is formed.

[0047] Based on the deep learning network model, the traditional training algorithm is combined with the swarm intelligence algorithm to establish a deep neural network model based on the swarm intelligence algorithm, and the model parameter optimization estimation and the training algorithm improvement are performed.

[0048] Through numerical simulation, the mode recognition of the slowly varying weak fault type of the RV reducer and the fault quantitative calculation of the damage degree are performed.

[0049] Compared with the prior art, the application has the following technical effects:

[0050] By adopting the above technical scheme, the slowly varying fault mechanism of the RV reducer is analyzed, a new filter denoising method based on artificial intelligence is obtained, the detail signal reflecting the essence of the fault is retained, and the fault features are accurately extracted. At the same time, by introducing the pattern recognition model with stronger generalization ability which fuses the swarm intelligence optimization algorithm and the deep learning network, the problem of quickly and accurately identifying the weak fault of the RV reducer is solved. DETAILED DESCRIPTION

[0051] The specific embodiments of the application will be described in detail below with reference to the accompanying drawings:

[0052] Figure 1 The steps of the fault diagnosis method of some embodiments disclosed in the application are shown in the schematic diagram.

[0053] Figure 2 The RV reducer fault diagnosis migration schematic diagram of some embodiments disclosed in the application is shown. DETAILED DESCRIPTION

[0054] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0055] In view of the fact that the slow-varying weak fault diagnosis of the RV reducer is greatly affected by human factors, the calculation efficiency, accuracy and generalization ability of the model are low, and the traditional method of isolated fault mechanism analysis, noise suppression, feature extraction and fault diagnosis is overcome. First, the slow-varying weak damage evolution model of the fault mechanism is established, and the time-varying coupling relationship between different parts, the evolution process and the influence law of the whole machine performance under the conditions of unknown structure parameters of the RV reducer, random working conditions and time-varying environment are explored. Secondly, the hybrid model of slow-varying weak fault noise suppression, feature extraction and intelligent diagnosis is established by combining filtering denoising and improved swarm intelligent neural network method, which solves the pain points of low accuracy and weak generalization ability of traditional fault recognition model, and lays a foundation for realizing the transformation from "fixed time maintenance" to "intelligent maintenance", so as to ensure the healthy and reliable operation of intelligent manufacturing equipment and high-end medical equipment.

[0056] Please refer to Figure 1 and Figure 2 In the embodiments of the present application, a slow time-varying weak fault diagnosis method of RV reducer is provided, and the specific steps include:

[0057] Step S1, according to the fast-varying power transmission operation process and slow-varying damage evolution process of the RV reducer, a slow-varying weak damage fault evolution model of the RV reducer is established;

[0058] 1. Research on RV reducer time-varying weak damage fault mechanism and law.

[0059] In view of the need for research on the slow-varying weak damage fault mechanism and law of the RV reducer, statistical methods, multi-scale dynamics methods, fault physical analysis methods, finite element analysis methods and fault analysis and inspection methods are introduced to carry out research on the slow-varying damage law of the strength and life of the main parts of the RV reducer, the dynamic transmission relationship of each contact link, the dynamic-damage coupling characterization of the RV reducer, and the slow-varying weak damage law considering the time-varying randomness and individual differences of the material, machining accuracy, assembly accuracy and load of the main parts of the RV reducer. The specific route is as follows:

[0060] (1) Considering the time-varying randomness and individual differences of the material, machining accuracy, assembly accuracy and load of the main parts of the RV reducer, the slow-varying damage law of the strength and life of the main parts is derived through fault physical analysis.

[0061] (2) According to the structural characteristics and transmission principle of RV reducer, based on the dynamic transmission relationship of each contact link, considering the fast variable power transmission process and slow variable damage evolution process of RV reducer, the dynamic-damage coupling of RV reducer is characterized, and the slow variable micro-damage fault evolution model of RV reducer is established.

[0062] For the fast variable dynamic response process of RV reducer components, the dynamic response differential equation is as follows:

[0063]

[0064] Where M, C and K are mass matrix, damping matrix and stiffness matrix respectively. F is the external excitation force, and q is the displacement matrix. Let (Formula 1) can be converted into state space form as follows:

[0065]

[0066] Where,

[0067]

[0068] B=[M -1 F 0] T (Formula 4)

[0069] For the damage evolution process of RV reducer components such as wear and fatigue, the differential equation is as follows:

[0070]

[0071] Where, Φ is a slow variable damage variable, which can be a crack depth, wear degree, spalling area and other damage variables with actual physical meaning, or an abstract performance index that cannot be observed inside the system. That is, (Formula 5) represents the slow evolution process of RV reducer component damage.

[0072] With the slow development of damage evolution process, the parameters M, C and K in fast variable power transmission process may change slowly. Generally, the damage variable most directly affects the stiffness of the system, for example, crack propagation will cause the stiffness of the structure to decrease. Therefore, (Formula 3) can be expressed in the form of Φ function:

[0073] A=μ(Φ) (Formula 6)

[0074] Based on the significant difference in time scale between the fast variable power response process and the slow variable damage evolution process, a small numerical time scale constant ν is introduced to represent the separation of the two processes in time scale.

[0075] Based on the above formulas and related definitions, the mathematical model of slow variable damage evolution of RV reducer under the dynamic-damage coupling of RV reducer is as follows:

[0076]

[0077]

[0078] where (Formula 7) is a hierarchical fast-slow coupled system, (Formula 7-a) is the fast-varying subsystem describing the dynamic behavior of the component, and (Formula 7-b) is the slow-varying subsystem describing the damage evolution behavior of the component, and the two subsystems are coupled through the dynamic response variable Y and the damage variable Φ.

[0079] Step S2, the RV reducer slowly varying micro-damage fault evolution model parameters are optimized and simulated, and the RV reducer slowly varying micro-damage fault law is obtained and verified;

[0080] (3) The model parameters are optimized and simulated to analyze the RV reducer slowly varying micro-damage fault law.

[0081] For the fast-slow dynamic-damage coupling model with slow-varying parameters represented by (Formula 7), when we know the explicit expressions of functions f, μ and g, we can use the approximate analytical method of classical nonlinear vibration to solve it, such as perturbation method, multi-scale method, etc. However, in engineering practice, it is difficult to obtain the explicit expression of these function relationships, so numerical method is considered to solve the model.

[0082] Compared with the natural time scale of the system dynamic response, the damage evolution process is very slow. It is assumed that in the n periods (one vibration period is time T) of the RV reducer rotating component transmission, the damage variable Φ M-1 is constant, and after n periods, the damage variable evolves to Φ M . Based on this assumption, the parameter μ(Φ) in the system vibration differential equation (Formula 7-a) no longer varies slowly with time within the n period time when the damage variable is constant. The solving steps of the coupled model are as follows:

[0083] ① Substitute μ(Φ M-1 ) into (Formula-7a) and solve the vibration response Y M-1 of the system within n periods of time by numerical method (such as Runge-Kutta method, Newmark-β method).

[0084] ② Substitute Y M-1 into equation (Formula 7-b), and similarly, solve the damage variable Φ M after n periods by numerical method.

[0085] ③ Repeat steps ① and ② until the system performance failure occurs.

[0086] It is worth noting that the smaller the value of n, the more accurate the solution of the model. However, due to the slow degradation of system performance, the time scale constant v is usually small, and the selected n value is too small, which will produce huge calculation cost. In theory, on the time scale O(1 / ν), ignoring the change of μ(Φ M ) is quasi-stationary.

[0087] (4) Using existing failure analysis equipment, such as the cooperative unit of the national industrial robot product quality supervision and detection center, adopting the method of fault physical failure analysis, verifying the RV reducer gradual micro-damage failure law.

[0088] Step S3, according to the fault symptom knowledge, establish the RV reducer weak fault noise suppression criterion under different conditions;

[0089] 2. Weak fault noise suppression method and noise suppression research of gradual failure mechanism.

[0090] In view of the need of weak fault noise suppression method and noise suppression research of gradual failure mechanism, the gradual weak fault noise interference suppression problem is studied under the conditions that the fault symptom knowledge is known and the fault symptom knowledge is unknown.

[0091] (1) In view of the need of noise suppression method and noise suppression research of known fault, wavelet filtering method, Kalman filtering method and deep neural network feature selection method are introduced, and the research of gradual weak fault feature extraction method, gradual damage model and noise fusion suppression method is carried out. The specific route is as follows:

[0092] ① In the case of Gaussian observation data, under the condition that the fault symptom knowledge is available, wavelet filtering technology, Kalman filtering technology and principal component analysis are combined, designated component analysis (DCA) and other feature selection technologies are used, and appropriate feature selection evaluation function is proposed. Extract the fault noise features of the key components of RV reducer.

[0093] ② Using deep neural network feature selection method, combining wavelet filtering method, Kalman filtering method and time-varying damage model analysis, establish RV reducer key component weak fault noise suppression criterion.

[0094] ③ The feasibility of the algorithm of each part is verified by numerical simulation.

[0095] (2) In view of the need of noise suppression method and noise suppression research of unknown fault, multi-source information fusion method and deep neural network feature extraction method are introduced, and the research of gradual weak fault feature selection, feature classification, feature extraction and damage model and noise fusion suppression method is carried out. The specific route is as follows:

[0096] ①Under the condition of nonlinear observation data, unknown statistical distribution, and unknown fault symptom knowledge, the multi-source information fusion technology is combined with the time-varying damage model analysis to propose a suitable evaluation function for feature selection. From the data-driven perspective, the stress history of the non-stationary load is analyzed to optimize and extract potential fault noise features from the observation data.

[0097] ②The direct mapping relationship between the high-dimensional fault features extracted by the deep neural network and the noise suppression is established by using the deep neural network fault extraction method to determine the weak fault noise suppression criterion for the key components of the RV reducer.

[0098] ③The feasibility of the algorithms in each part is verified through numerical simulation.

[0099] Step S4, based on the swarm intelligence algorithm, a deep neural network model is established to identify the type of weak fault and perform quantitative fault calculation of the damage degree.

[0100] 3. Slowly varying fault diagnosis method and diagnosis research based on swarm intelligence neural network.

[0101] To meet the needs of the slowly varying fault diagnosis method and diagnosis research based on swarm intelligence neural network, the phase space curvature method, multi-source information fusion method, transfer learning deep neural network feature extraction method, and swarm intelligence optimization algorithm are introduced. The research on the construction of deep residual network sharing in the field, the migration of fault features in the RV reducer monitoring data from engineering practice, the establishment of deep neural network model based on swarm intelligence algorithm, and simulation analysis are carried out. The specific route is as follows:

[0102] (1) Based on the phase space curvature method, a damage evolution residual model is constructed.

[0103] The damage evolution process represented by the damage variable Φ is usually difficult to obtain directly through measurement. A system in a damage state starts from an initial state (t0, x0, φ0) (only considering a single damage variable), and after a very short prediction time t p = t - t0, its response is x = X(t p , t0, x0, μ(φ0); ε). For a system in a reference state (healthy state), after the same short time t p , the response is x R = X(t p , t0, x0, μ(φ R ); ε). By comparing the differences between the two, the short-time reference model prediction error e R is obtained to quantitatively represent the dynamic evolution of the system damage.

[0104] e R = X(t p,t0,x0,μ(φ0);ε)-X(t p ,t0,x0,μ(φ R );ε) (Equation 8)

[0105] When the prediction time t p When the time interval is very short, the damage change of the system can be ignored, and the fast-changing subsystem represented by (Equation 7-a) can be regarded as stationary. At this time, the system response term in Equation (8) can be expanded using canonical perturbation as follows:

[0106]

[0107] Where X n For each power series ε after expansion n The coefficients, and X0 is the solution to the original system degradation problem (ε=0) X0=X(t p ,t0,x0,μ(φ i ); 0). If we use O(·) to represent a higher-order infinitesimal function. φ i For φ0 or φ R Substituting (Equation 9) into (Equation 8) yields:

[0108] e R =X(t) p ,t0,x0,μ(φ0);0)-X(t p ,t0,x0,μ(φ R );0)+O(εt p (Equation 10)

[0109] For the first term on the right side of (Equation 10) at φ = φ R The Taylor series expansion at this point yields the following result:

[0110]

[0111] Ignoring higher-order infinitesimal terms, we obtain the following approximate relation:

[0112] e R ≈C(t p ,t 0, x0,φ R )φ+c(t p ,t0,x0,φ R (Equation 12)

[0113] in, c=-Cφ R When ||φ0-φ R || 2 When ε is sufficiently small, the tracking function e RThe damage variable φ is approximately linearly mapped; in the case of large damage, even if the linear mapping relationship no longer holds, higher-order mapping relationships are also feasible. That is, by measuring the fast-changing response of the system, tracking and identifying the slow-changing damage process of the system, and constructing the RV reducer damage evolution residual model.

[0114] (2) Introduce field adaptation constraints, integrate transfer learning and deep learning networks, transfer fault features from RV reducer monitoring data in engineering practice, and form a RV reducer time-varying weak fault deep migration diagnosis method.

[0115] Assume that in the laboratory environment, the engineering practice and simulation experiment of the RV reducer can obtain monitoring data Contains n s samples. y i s ∈Γ is the health label of sample x i s , Γ={1, 2, …, k} is the label space, containing k health states. Sample x i s belongs to the sample space X i s The data generation conforms to the marginal probability distribution P(X i s ), in engineering practice, the reducer state monitoring data set obtained is contains n t samples to be identified. Because the laboratory simulation and the service environment, operating conditions, and even specifications and models of the engineering practice are different, the generation mechanisms of the monitoring data of the two are quite different, which can be described from the statistical point of view: the monitoring data x i t ∈X t of the engineering practice reducer conforms to the marginal probability distribution Q(X i t ), and Q≠P.

[0116] Through the training of the monitoring data samples of the source domain reducer, the nonlinear mapping relationship from the sample space X s of the monitoring data of the source domain equipment to the health label space Γ s is established , which is the obtained reducer fault diagnosis knowledge. As shown in Figure 2 a, because there is a large distribution difference between the monitoring data of the source domain and target domain reducers, the fault diagnosis knowledge f of the source domain reducer cannot accurately identify the health label-free sample categories of the target domain equipment, resulting in misjudgment of the health state of the mechanical equipment. In view of the above field difference problem, as shown in Figure 2(b) as shown, a deep migration diagnostic model is constructed, and the monitoring data distribution of the source domain and the target domain equipment is adapted, so that the failure diagnosis knowledge f of the source domain reducer can identify the health status of the target domain robot RV reducer.

[0117] (3) Combined with swarm intelligence algorithms (such as hybrid frog leap algorithm, particle swarm algorithm, ant colony algorithm, chicken swarm algorithm, etc.), based on deep learning network (or other shallow network such as SVM, VPMCD) model, replace or improve traditional training algorithms such as gradient descent method and quadratic programming method with swarm intelligence algorithm, establish deep neural network model based on swarm intelligence algorithm, and improve model parameter optimization estimation and training (or learning) algorithm. Through numerical simulation, the mode recognition of the slowly varying weak fault type of the RV reducer and the fault quantitative calculation of the damage degree are carried out.

[0118] (4) Using the existing RV reducer fault diagnosis test bench in the laboratory, taking the RV reducer in engineering application as the measured object, formulating the test scheme, completing the RV reducer time-varying weak fault signal test work. Combined with the actual operation signal of the RV reducer, the accuracy, efficiency and generalization ability of the RV reducer slowly varying weak fault diagnosis method based on swarm intelligence filtering neural network are verified.

[0119] According to the needs, using the established RV reducer failure analysis experimental platform, RV reducer performance and reliability experimental platform, and robot whole machine performance and reliability experimental platform, the experiments to be carried out are:

[0120] (1) Simulation experiment

[0121] ① Use the secondary development function of simulation software such as ANSYS and COMSOL to carry out numerical simulation experiment of RV reducer slowly varying micro-damage fault evolution law.

[0122] ② Use MATLAB numerical simulation software to carry out numerical simulation experiment of weak fault noise suppression of slowly varying fault mechanism.

[0123] ③ Use MATLAB numerical simulation software to carry out numerical simulation experiment of RV reducer slowly varying weak fault diagnosis based on swarm intelligence filtering neural network.

[0124] (2) Semi-physical simulation experiment

[0125] For the typical application scenarios (welding, assembly, surgery) of RV reducer, under the conditions of normal state and different types of early potential fault of RV reducer, different sensors or sensor sets are used to carry out information collection semi-physical simulation experiment, which provides information for numerical simulation of weak fault noise suppression of slowly varying fault mechanism.

[0126] (3) Physical experiment

[0127] ①Adopting high-accelerated life experiment technology, the high-accelerated life experiment of RV reducer is carried out, and the slowly changing weak fault of RV reducer is excited quickly.

[0128] ②Aiming at different failure modes of RV reducer, the slowly changing micro-damage fault failure analysis experiment of RV reducer is carried out, and the evolution law of the slowly changing micro-damage fault of RV reducer is verified.

[0129] ③Aiming at the typical application scenarios (welding, assembly, surgery) of RV reducer, the slowly changing weak fault diagnosis engineering application experiment of RV reducer is carried out, and the accuracy, efficiency and generalization ability of the slowly changing weak fault diagnosis method of RV reducer based on swarm intelligent filtering neural network are verified.

[0130] The application carries out the high-accelerated life experiment technology, the optimization selection technology of sensor network and fault feature based on swarm intelligence, and the engineering application experiment scheme design of the typical application scenarios of RV reducer.

[0131] According to the structure characteristics and transmission principle of RV reducer, the fast-changing power transmission process and slow-changing damage evolution process of RV reducer are considered, the pre-research of RV reducer dynamics-damage coupling characterization and RV reducer slowly changing micro-damage fault evolution model is carried out. Based on the phase space curve variation method, the pre-research of damage evolution residual model construction is carried out. The field adaptation constraint is introduced, the fault features in the monitoring data of RV reducer from engineering practice are migrated and learned, and the pre-research work of RV reducer time-varying weak fault deep migration diagnosis method is carried out.

[0132] By discussing the slowly changing fault mechanism of RV reducer, a new filtering and denoising method based on artificial intelligence is found out, the detail signal reflecting the essence of the fault is reserved, the fault feature is accurately extracted, and at the same time, by introducing the pattern recognition model with stronger generalization ability which fuses swarm intelligence optimization algorithm and deep learning network, the problem of quickly and accurately identifying the weak fault of RV reducer is solved.

[0133] Although the embodiments of the application have been shown and described, it can be understood by those skilled in the art that various changes, modifications, replacements and variations can be made to the embodiments without departing from the principles and spirits of the application, the scope of the application is defined by the appended claims and their equivalents, and all should be included in the protection scope of the application.

Claims

1. A slowly time-varying weak fault diagnosis method for RV reducer, characterized in that, The specific steps include: According to the fast variable power transmission operation process and slow variable damage evolution process of the RV reducer, a slow variable micro-damage fault evolution model of the RV reducer is established, and the specific steps include: For the fast variable dynamic response process of the RV reducer, the dynamic response differential equation is as follows: ； Wherein M, C and K are mass matrix, damping matrix and stiffness matrix respectively, F is external excitation force, and q is displacement matrix; Let The power response differential equation is converted into a state space form expression as follows: ； wherein , ; For the damage evolution process of the RV reducer, the differential equation is as follows: ； wherein is a slowly varying damage variable; The expression in the state space form is is expressed as The function form is ； Based on the significant difference in time scale between the fast dynamic response process and the slow damage evolution process, a small numerical time scale constant is introduced to represent the separation of the two processes in time scale; According to the above formula and definition, the slow variable micro-damage fault evolution model of the RV reducer under the coupling action of dynamics-damage is as follows: ； ； wherein, is a fast-varying subsystem describing the dynamic behavior of the component, is a slow-varying subsystem describing the damage evolution behavior of the component; the two subsystems are coupled through the dynamic response variable and the damage variable ; The RV reducer slow variable micro-damage fault evolution model parameters are optimized and simulated to obtain the RV reducer slow variable micro-damage fault law and verify it; According to the fault symptom knowledge under different conditions, the RV reducer weak fault noise suppression criterion is established; Based on the swarm intelligence algorithm, a deep neural network model is established to identify the weak fault type and quantitatively calculate the damage degree.

2. The slowly time-varying weak fault diagnosis method for an RV reducer according to claim 1, characterized in that, The specific steps of the RV reducer slow variable micro-damage fault evolution model parameter optimization solution and simulation analysis, obtaining the RV reducer slow variable micro-damage fault law and verifying it, include: Setting in the RV reducer rotation component drive one cycle time inside, and damage variable Φ M-1 is not changed, one cycle after damage variable evolution Φ M ; The solution step of the slow variable micro-damage fault evolution model is: Substituting μ ( Φ M-1 ) into the equations of the fast subsystem, the vibration response of the system is solved by numerical method n in one period of time Y M-1 ; Again Y M-1 Substituting into the equation for the slow subsystem, we solve n the damage variable after one period Φ M ; Repeat the above two steps until the system performance failure occurs; Finally, through the failure analysis equipment, the RV reducer slow variable micro-damage fault law is verified by the fault physical failure analysis method.

3. The slowly time-varying weak fault diagnosis method for RV reducer according to claim 1, characterized in that, The specific steps of establishing the RV reducer weak fault noise suppression criterion according to the fault symptom knowledge under different conditions include: If the fault symptom knowledge is known, the specific steps include: Wavelet filtering technology, Kalman filtering technology and principal component analysis are combined to obtain the evaluation function by using feature selection technology, and the fault noise features of the key components of the RV reducer are extracted; The wavelet filtering method, Kalman filtering method and time-varying damage model analysis are combined by using the deep neural network feature selection method to establish the RV reducer key component weak fault noise suppression criterion; The feasibility of the algorithm of each part is verified by numerical simulation; If the fault symptom knowledge is unknown, the specific steps include: Multi-source information fusion technology and time-varying damage model analysis are combined to obtain the evaluation function of feature selection, and the fault noise features are optimized and extracted; The deep neural network fault extraction method is used to establish the direct mapping relationship between the high-dimensional fault features extracted by the deep neural network and the noise suppression, and the RV reducer key component weak fault noise suppression criterion is determined; The feasibility of the algorithm of each part is verified by numerical simulation.

4. The slowly time-varying weak fault diagnosis method for an RV reducer according to claim 1, characterized in that, The specific steps of establishing the deep neural network model based on the swarm intelligence algorithm, identifying the weak fault type, and quantitatively calculating the damage degree include: Based on the phase space curve method, a damage evolution residual model is constructed; The field adaptation constraint is introduced, the transfer learning and deep learning network are fused, the fault features in the RV reducer monitoring data from engineering practice are transferred, and the RV reducer time-varying weak fault deep migration diagnosis method is formed. Combining swarm intelligence algorithm, based on deep learning network model, using traditional training algorithm, the deep neural network model based on swarm intelligence algorithm is established to optimize the model parameters and improve the training algorithm; Through numerical simulation, the mode recognition of the slowly varying weak fault type of RV reducer and the fault quantitative calculation of the damage degree are carried out.

Images