Test machine phase difference optimization method, system, and rubber product fatigue test machine

Abstract

The present application belongs to the technical field of fatigue testing, and discloses a testing machine phase difference optimization method and system and a rubber product fatigue testing machine. The method comprises the following steps: presetting a loading path signal, collecting sensor data, performing feature extraction on the sensor data to obtain phase difference features; taking the testing specimen parameters and the phase difference features as inputs of a double-channel lightweight model to obtain phase difference prediction values of each channel; based on a multi-objective dynamic parameter optimization algorithm, the control parameters of the testing machine are optimized and adjusted in combination with the phase difference prediction values and the coupling features between channels; the specimen strain distribution is monitored in real time, and when the strain unevenness is greater than a preset strain threshold, the channel load is redistributed and the phase difference is re-optimized. The present application can realize real-time sensing of specimen characteristics and loading path changes, reduce phase difference prediction errors, improve testing efficiency, reduce energy consumption, and significantly improve the authenticity of multi-axis load simulation, phase control accuracy and system comprehensive performance.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fatigue testing, and more particularly, to a test machine phase difference optimization method and system and a rubber product fatigue test machine. BACKGROUND

[0002] In the field of dynamic fatigue testing, for fatigue test machines with multiple channels or complex loading conditions, such as electro-hydraulic servo fatigue test machines, accurate control of the phase difference between different loading channels is a core requirement for simulating the actual stress conditions of structures. Such devices need to adjust the phase difference of each loading signal to reproduce the complex stress path of materials or structures under multi-directional cyclic loading, thereby accurately evaluating their fatigue performance. For example, in the simulation of bridges, aerospace components and other scenarios that bear multi-dimensional dynamic loads, accurate control of the phase difference directly affects the authenticity and reliability of the test results.

[0003] Chinese patent application No. CN115792770A discloses a method and system for obtaining inherent phase reference calibration data between channels of a vector network analyzer. The ports of two target receivers and a third receiver port of the vector network analyzer are connected with a power divider, and forward connection measurement data is obtained by testing. The vector network analyzer includes multiple receivers and the power divider. The ports of the two target receivers are reversely connected, and reverse connection measurement data is obtained by testing. Inherent phase reference calibration data is obtained according to the forward connection measurement data and the reverse connection measurement data. This invention eliminates the phase reference measurement error caused by the inconsistency of the amplitude and phase characteristics of each channel of the vector network analyzer receiver, and significantly improves the phase reference calibration accuracy of using the vector network analyzer for multi-source phase reference signal testing.

[0004] Although the above method can meet most scenarios, research and practical application of the above method and prior art have found that the above method and prior art at least have the following defects:

[0005] The above method mainly relies on fixed calibration data and a preset mathematical model, and cannot adjust the phase difference in real time according to the dynamic changes of the specimen characteristics and the loading path. Moreover, when multi-axis loading is performed, the pressure fluctuations among the hydraulic channels interfere with each other, such as the change of the flow rate of channel A causing the pressure pulsation of channel B, which easily leads to distortion of the preset phase difference and affects the authenticity of multi-axis load simulation.

[0006] In view of this, the present application provides a test machine phase difference optimization method and system and a rubber product fatigue test machine to solve the above problems. SUMMARY

[0007] In order to overcome the above-mentioned defects of the prior art and achieve the above-mentioned purpose, the present application provides the following technical solutions: a test machine phase difference optimization method, comprising the following steps:

[0008] Pre-set N-channel loading path signals, collect sensor data of each channel, the sensor data includes independent data and inter-channel pressure coupling signals, perform feature extraction on the sensor data to obtain phase difference features; the phase difference features include intra-channel phase difference features and inter-channel coupling features;

[0009] Take the test specimen parameters and the phase difference features as inputs of a two-channel lightweight model to obtain phase difference prediction values of each channel;

[0010] Based on a multi-objective dynamic parameter optimization algorithm, the control parameters of the testing machine are optimized and adjusted in combination with the phase difference prediction values and the inter-channel coupling features.

[0011] Real-time monitoring of the strain distribution of the test specimen, when the strain non-uniformity is greater than a preset strain threshold, triggering channel load redistribution and re-optimizing the phase difference.

[0012] Further, the method for optimizing and adjusting the control parameters of the testing machine based on the multi-objective dynamic parameter optimization algorithm comprises:

[0013] Defining a target function, performing normalization processing on the target function to obtain a normalized target;

[0014] Initializing R particles, the particles representing decision variables, the decision variables consisting of PID gains and filter medium frequencies;

[0015] Randomly distributing the particles in a parameter space of the decision variables, initializing particle velocities;

[0016] Taking the phase error and the error change rate as input variables of a fuzzy logic controller, obtaining corresponding output variables consisting of learning factors through a fuzzy rule table predefined by the fuzzy logic controller;

[0017] Dividing the particles into an exploration group and a development group, updating particle positions of the exploration group in a PSO mode, and updating particle positions of the development group in a CPO mode;

[0018] Merging all the particles, updating a Pareto front through non-dominated sorting, and reserving the top Y% elite solutions as global optimal solutions for next iteration;

[0019] Repeating iteration on the particles until a preset termination condition is met;

[0020] From the Pareto front, selecting a solution ranked first, and outputting to a testing machine controller for parameter compensation.

[0021] Further, the target function includes phase synchronization error, energy consumption, dynamic stability; according to the pressure signals between different channels, a channel coupling matrix is constructed through pressure covariance, a single-channel root mean square error is calculated according to the phase difference prediction value and the actual measured phase difference, and the phase synchronization error is calculated according to the single-channel root mean square error, the current phase difference and the channel coupling matrix; the energy consumption is calculated according to the real-time voltage signal and the real-time current signal collected by the sensor; the Lyapunov index is calculated by a preset initial value and an exponential function with an exponential change of a decay coefficient, and the Lyapunov index is taken as the dynamic stability.

[0022] Further, in the multi-objective dynamic parameter optimization algorithm, in the PSO mode, the weight is dynamically updated based on the error standard deviation, and the speed is updated in combination with the dynamically updated weight, the current optimal solution and the global optimal solution;

[0023] In the CPO mode, the speed is updated in combination with the global optimal solution and the distance between the current position and the global optimal solution, simulating the defense behavior of the crown porcupine;

[0024] Based on the updated speed, the particle position is updated, and when the position exceeds the preset range, the mirror reflection method is used for position correction and speed correction.

[0025] Further, the method for obtaining the phase difference prediction value comprises:

[0026] A dual-channel lightweight model including a Transformer-XL channel and a GRU-attention channel is built;

[0027] Test specimen parameters are obtained, the test specimen parameters and long-term phase difference features are taken as inputs of the Transformer-XL channel, long-time phase difference prediction values are obtained, the test specimen parameters and short-term phase difference features are taken as inputs of the GRU-attention channel, and short-time phase difference prediction values are obtained;

[0028] Real-time strain distribution data are collected by M strain gauge arrays attached to the surface of the test specimen, and strain unevenness is calculated according to the strain distribution data;

[0029] When the strain unevenness is lower than a preset uniformity threshold, the dual-channel lightweight model is used to output the phase difference prediction value;

[0030] When the strain unevenness is not lower than the preset uniformity threshold, a strain-phase mapping network is used to output the inter-channel compensation phase;

[0031] The strain distribution data, the material elastic modulus and the current phase difference are taken as inputs of the strain-phase mapping network, inter-channel compensation phases are obtained, the current phase difference is corrected according to the compensation phases, and the phase difference prediction value is obtained.

[0032] Further, the method for outputting the phase difference prediction value by the dual-channel lightweight model comprises:

[0033] The dual-channel weights are dynamically adjusted based on the interactive multiple model filtering, and the dual-channel weights are calculated based on the channel likelihood probability;

[0034] The test specimen parameters, the phase difference characteristics, the long-time phase difference prediction value of the Transformer-XL channel at the last moment, and the short-time phase difference prediction value of the GRU-attention channel at the last moment are taken as the current input, and the mixed state is obtained by combining the dual-channel weights;

[0035] The mixed state is taken as the input of the Transformer-XL channel and the GRU-attention channel respectively, and the predicted long-time phase difference prediction value and the predicted short-time phase difference prediction value are obtained, and the long-time prediction error and the short-time prediction error are calculated;

[0036] The long-time mean and the long-time variance are obtained according to the long-time prediction error statistics, the probability density of the long-time prediction error subject to Gaussian distribution is calculated, and the long-time likelihood probability is calculated by combining the long-time weight at the last moment;

[0037] The short-time mean and the short-time variance are obtained according to the short-time prediction error statistics, the probability density of the short-time prediction error subject to Laplace distribution is calculated, and the short-time likelihood probability is calculated by combining the short-time weight at the last moment; the normalized long-time weight and the normalized short-time weight are calculated according to the long-time likelihood probability and the short-time likelihood probability;

[0038] According to the normalized long-time weight and the normalized short-time weight, the predicted long-time phase difference prediction value and the predicted short-time phase difference prediction value are weighted to obtain the phase difference prediction value.

[0039] Further, the method for obtaining the phase difference characteristics comprises:

[0040] The mechanical vibration signal in the independent data is obtained, the wavelet basis and the decomposition layer number are adaptively selected to decompose the mechanical vibration signal, and two N subband signals are obtained, the normalized energy of the subband signal is calculated, and the energy entropy is calculated according to the normalized energy;

[0041] The correlation weight between the PID gain and the phase difference is calculated by gray correlation degree analysis, and the control modal characteristics are selected and obtained;

[0042] The test specimen parameters in the independent data are spatially wavelet decomposed, and the strain gradient amplitude is extracted and obtained;

[0043] The strain energy density distribution is calculated according to the strain value and the material elastic modulus in the independent data;

[0044] Stack the energy entropy, control modal feature, strain gradient amplitude and strain energy density distribution according to the time step G into a three-dimensional tensor and perform normalization processing to obtain the channel phase difference feature;

[0045] An electrical signal in the inter-channel pressure coupling signal is acquired, the electrical signal is transformed through Hilbert transform to obtain an analytic signal, and an instantaneous phase is extracted according to the analytic signal; a frequency spectrum is obtained by performing Fourier transform on the electrical signal, and a joint distribution matrix is obtained by analyzing the frequency spectrum;

[0046] A nonlinear model of the environmental variable and the phase difference is established through a partial least squares method, the environmental variable includes temperature and oil viscosity; fitting coefficients of the nonlinear model are acquired, and a coupling feature is obtained by splicing;

[0047] The joint distribution matrix and the coupling feature are stacked into a three-dimensional tensor according to the time step G and are normalized to obtain the inter-channel coupling feature.

[0048] Further, the method for obtaining the joint distribution matrix comprises:

[0049] The frequency axis of the frequency spectrum is divided into K frequency intervals, and the instantaneous phase is equally divided into Q phase intervals, each interval representing a discrete range;

[0050] Each frequency interval k is traversed, the frequency components in the frequency interval k are screened, each phase interval q is traversed, the corresponding instantaneous phase is calculated for the screened frequency components, the components whose corresponding instantaneous phase belongs to the phase interval q are screened, and the accumulated energy of the electrical signal belonging to the frequency interval k and the phase interval q is calculated; the total energy is calculated by accumulating the accumulated energy; and the joint distribution matrix is calculated by the accumulated energy and the total energy, wherein k = 1, 2, …, K; q = 1, 2, …, Q.

[0051] Further, the method for obtaining the control modal feature comprises:

[0052] The system state parameter and the phase difference are acquired;

[0053] The system state parameter is normalized;

[0054] The phase difference is acquired as a reference sequence, the fth control parameter sequence is acquired as a comparison sequence, the grey correlation degree is calculated based on the resolution coefficient according to the reference sequence and the comparison sequence, and the system state parameter with the grey correlation degree higher than a preset correlation degree threshold is screened as the control modal feature, wherein f = 1, 2, …, F, and F is the total number of control parameters.

[0055] A test machine phase difference optimization system, which implements the test machine phase difference optimization method, comprises:

[0056] The acquisition analysis module: preset N channel loading path signals, acquire sensor data of each channel, the sensor data includes independent data and inter-channel pressure coupling signals, and the sensor data is subjected to feature extraction to obtain a phase difference feature; the phase difference feature includes an intra-channel phase difference feature and an inter-channel coupling feature;

[0057] The phase difference prediction module: taking the test specimen parameters and the phase difference feature as inputs of a dual-channel lightweight model, phase difference prediction values of each channel are obtained;

[0058] The phase difference optimization module: based on a multi-objective dynamic parameter optimization algorithm, the control parameters of the testing machine are optimized and adjusted in combination with the phase difference prediction values and the inter-channel coupling features;

[0059] The monitoring optimization module: the strain distribution of the test specimen is monitored in real time, when the strain non-uniformity is greater than a preset strain threshold, the channel load is redistributed and the phase difference is re-optimized.

[0060] The testing machine phase difference optimization method is applied to a rubber product fatigue testing machine.

[0061] The testing machine phase difference optimization method, system and rubber product fatigue testing machine of the application have the following technical effects and advantages:

[0062] The application constructs a dual-channel lightweight model containing Transformer-XL and GRU-attention, combines a strain non-uniformity dynamic switching model with an interactive multi-model filtering weight adjustment, realizes accurate modeling of long-time trends, short-time mutations and local deformation of the test specimen, constructs a channel coupling matrix through a pressure covariance, integrates it into a multi-objective function containing phase synchronization error, energy consumption and dynamic stability, quantifies the inter-channel coupling effect and realizes multi-dimensional index collaborative optimization, uses a multi-objective dynamic parameter optimization algorithm of a PSO and CPO hybrid mode, combines a fuzzy logic controller to dynamically adjust a learning factor, balances global search and local development capabilities, and ensures parameter effectiveness through a mirror reflection method; the phase difference features containing multi-source data such as energy entropy, strain field features and joint distribution matrix are extracted to provide rich input dimensions for the model. The application effectively solves the problems of phase difference adjustment lag and insufficient channel interference processing in the prior art, enables the system to real-time perceive test specimen characteristics and loading path changes, reduces phase difference prediction error, improves testing efficiency, reduces energy consumption, and significantly improves the authenticity of multi-axis load simulation, phase control accuracy and system comprehensive performance.

[0063] By collecting sensor data and extracting multi-domain features, a phase difference feature containing specimen characteristics and loading path information is constructed to provide multi-dimensional input for intelligent prediction; a dual-channel lightweight model is also used to capture the long-term trend and short-term dynamics of the phase difference, and through interactive multi-model filtering, the channel weight is dynamically adjusted to realize high-precision output of the phase difference prediction value; based on a multi-objective dynamic parameter optimization algorithm, the phase synchronization error minimization, energy consumption constraint and dynamic stability are taken as optimization objectives, and the test machine control parameters are adaptively adjusted combined with the prediction value to form a data acquisition-feature fusion-intelligent prediction-parameter optimization closed loop; the present application can dynamically adapt the phase difference according to the specimen material properties, geometric topology and preset loading path, reduce the test errors caused by phase mismatch, environmental interference, etc., and at the same time, through multi-objective balance, the system stability and energy efficiency are improved, the repeated calibration loss under the traditional fixed parameter mode is avoided, the test efficiency is improved, and the intelligent upgrading of complex working condition equipment such as electro-hydraulic servo fatigue testing machine can be applied. BRIEF DESCRIPTION OF DRAWINGS

[0064] Figure 1 It is a test machine phase difference optimization method flow diagram of embodiment 1 of the present application.

[0065] Figure 2 It is a data flow diagram of embodiment 1 of the present application.

[0066] Figure 3 It is a data flow diagram of embodiment 2 of the present application.

[0067] Figure 4 It is a test machine phase difference optimization system structure diagram of embodiment 3 of the present application. DETAILED DESCRIPTION

[0068] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0069] Embodiment 1:

[0070] Please refer to Figure 1 、 Figure 2 The present embodiment provides a test machine phase difference optimization method, including the following steps:

[0071] Preset N-channel loading path signals, collect sensor data of each channel, the sensor data includes independent data and inter-channel pressure coupling signals, perform feature extraction on the sensor data to obtain phase difference features; the phase difference features include intra-channel phase difference features and inter-channel coupling features;

[0072] The method for obtaining the phase difference feature comprises:

[0073] The phase difference feature comprises an intra-channel phase difference feature and an inter-channel coupling feature;

[0074] A mechanical vibration signal in independent data is obtained, a wavelet basis and a decomposition layer number are adaptively selected to decompose the mechanical vibration signal, two N subband signals are obtained, a normalized energy of the subband signal is calculated, and an energy entropy is calculated according to the normalized energy;

[0075] The correlation weight of the PID gain and the phase difference is calculated through the grey correlation degree analysis, and the key control variable is screened;

[0076] System state parameters and a phase difference are obtained, the system state parameters comprising a servo valve opening, a PID gain, a discrete command or a mixed data stream;

[0077] The servo valve opening and the PID gain in the system state parameters are normalized;

[0078] The phase difference is obtained as a reference sequence, and the fth control parameter sequence is obtained as a comparison sequence, wherein f = 1, 2, …, F, F is the total number of control parameters, the grey correlation degree is calculated based on the resolution coefficient according to the reference sequence and the comparison sequence, and the system state parameter with a grey correlation degree higher than a preset correlation degree threshold is screened as a control modal feature, preferably, the value range of the resolution coefficient is (0, 1], and 0.5 is usually taken, which can be adjusted according to actual conditions;

[0079] The test specimen parameters in the independent data are subjected to spatial wavelet decomposition, and a strain gradient amplitude is extracted;

[0080] The strain energy density distribution is calculated according to the strain value and the material elastic modulus in the independent data;

[0081] The energy entropy, the control modal feature, the strain gradient amplitude and the strain energy density distribution are stacked into a three-dimensional tensor according to a time step G and subjected to normalization processing, and an intra-channel phase difference feature is obtained;

[0082] An electrical signal in the inter-channel pressure coupling signal is obtained, the electrical signal is transformed through Hilbert transform, an analytical signal is obtained, and an instantaneous phase is extracted according to the analytical signal, wherein the value range of the instantaneous phase is [-π, π];

[0083] The frequency spectrum is obtained by Fourier transform of the electrical signal, the frequency axis of the frequency spectrum is divided into K frequency intervals, the instantaneous phase is equally divided into Q phase intervals, and each interval represents a discrete range;

[0084] Traverse each frequency interval k, screen the frequency components in the frequency interval k, traverse each phase interval q, calculate the corresponding instantaneous phase for the screened frequency components, screen the components whose corresponding instantaneous phase belongs to the phase interval q, and calculate the accumulated energy of the electrical signal belonging to the frequency interval k and the phase interval q; the total energy is obtained by accumulating the accumulated energy; the joint distribution matrix is obtained by calculating the accumulated energy and the total energy, wherein k = 1, 2, …, K; q = 1, 2, …, Q;

[0085] A nonlinear model of the environmental variable and the phase difference is established by the partial least squares method, and the environmental variable includes temperature and oil viscosity; the fitting coefficients of the nonlinear model are obtained, and the coupling features are spliced;

[0086] The joint distribution matrix and the coupling features are stacked into a three-dimensional tensor according to the time step G and normalized to obtain the inter-channel coupling features.

[0087] The above method can capture the key influence of the signal stability in the channel and the control parameter on the phase difference by adaptively decomposing the mechanical vibration signal by using the wavelet to obtain the energy entropy, and screening the control modal features by using the grey correlation degree analysis; the strain field features are extracted by using the spatial wavelet decomposition and the strain energy density calculation, which can reflect the local deformation characteristics of the test piece in real time, so that the model can adapt to the nonlinear change of the test piece material; the joint distribution matrix is constructed by using the Hilbert transform and the Fourier transform on the electrical signal, and the environmental variable coupling features are extracted by using the partial least squares method, which can quantify the frequency-phase coupling relationship between the channels and the environmental interference such as temperature and oil viscosity, and break through the limitations of the traditional method for modeling the dynamic interaction between the channels and the environmental factors. The above feature extraction method fuses the multi-dimensional information of the mechanical vibration, the control parameter, the strain field, the electrical signal and the environmental variable, constructs a complex feature system containing the dynamic characteristics in the channel and the coupling effect between the channels, provides rich input dimensions for the phase difference prediction and multi-objective optimization, and enables the system to perceive the dynamic changes of the test piece characteristics and the loading path in real time, effectively solves the problems of phase difference adjustment lag and insufficient channel interference processing in the prior art, and significantly improves the authenticity of multi-axis load simulation and the phase control precision.

[0088] The test piece parameters and the phase difference features are used as the input of the dual-channel lightweight model to obtain the phase difference prediction value;

[0089] The method for obtaining the phase difference prediction value includes:

[0090] A dual-channel lightweight model including a Transformer-XL channel and a GRU-attention channel is built.

[0091] The test specimen parameters are obtained, and the test specimen parameters and long-term phase difference characteristics are taken as inputs of the Transformer-XL channel; long-time phase difference prediction values are obtained, wherein the long time corresponds to a test period of more than 10 minutes or a time range of more than 1000 loading cycles; the test specimen parameters and short-term phase difference characteristics are taken as inputs of the GRU-attention channel; short-time phase difference prediction values are obtained, wherein the short time corresponds to a time range of 10-100 loading cycles or a data window of 1-99 time steps;

[0092] The real-time strain distribution data are collected by the M strain gauge arrays attached to the surface of the test specimen, and the strain unevenness is obtained according to the strain distribution data;

[0093] When the strain unevenness is lower than the preset uniformity threshold, the phase difference prediction values are output by the dual-channel lightweight model;

[0094] The method for outputting the phase difference prediction values by the dual-channel lightweight model comprises the following steps:

[0095] The dual-channel weights are dynamically adjusted based on the interactive multiple model filtering; and the dual-channel weights are obtained based on the likelihood probability of the channels;

[0096] The test specimen parameters, the phase difference characteristics, the long-time phase difference prediction values of the Transformer-XL channel at the previous moment and the short-time phase difference prediction values of the GRU-attention channel at the previous moment are taken as current inputs, and the mixed state is obtained in combination with the dual-channel weights;

[0097] The mixed state is taken as the input of the Transformer-XL channel and the GRU-attention channel respectively, and the predicted long-time phase difference prediction values and the predicted short-time phase difference prediction values are obtained; and the long-time prediction error and the short-time prediction error are calculated and obtained;

[0098] The long-time mean and the long-time variance are obtained according to the long-time prediction error statistics; the probability density of the long-time prediction error subject to the Gaussian distribution is calculated and obtained; and the long-time likelihood probability is calculated and obtained in combination with the long-time weight at the previous moment;

[0099] The short-time mean and the short-time variance are obtained according to the short-time prediction error statistics; the probability density of the short-time prediction error subject to the Laplace distribution is calculated and obtained; and the short-time likelihood probability is calculated and obtained in combination with the short-time weight at the previous moment; and the normalized long-time weight and the normalized short-time weight are calculated and obtained according to the long-time likelihood probability and the short-time likelihood probability;

[0100] The phase difference prediction values are obtained by weighting the predicted long-time phase difference prediction values and the predicted short-time phase difference prediction values in combination with the normalized long-time weight and the normalized short-time weight.

[0101] When the strain non-uniformity is not lower than the preset uniformity threshold, a strain-phase mapping network is adopted to output inter-channel compensation phases;

[0102] The strain distribution data, the material elastic modulus and the current phase difference are taken as inputs of the strain-phase mapping network to obtain channel compensation phases, and the current phase difference is corrected according to the compensation phases to obtain a phase difference prediction value.

[0103] The above method can realize accurate modeling of long-term trends and short-term mutations of the phase difference by capturing long-term phase difference features through the Transformer-XL channel, extracting short-term dynamic features through the GRU-attention channel, and dynamically adjusting the weights of the two channels based on the Gaussian distribution characteristics of long-term errors and the Laplace distribution characteristics of short-term errors through interactive multi-model filtering. When the strain non-uniformity is low, the phase difference prediction value is obtained through the fusion of mixed state inputs and dynamic weights, effectively balancing the prediction requirements of different time scales. When the strain non-uniformity is high, the strain-phase mapping network is activated to generate inter-channel compensation phases based on the strain distribution, elastic modulus and current phase difference, and to correct the deviation of the phase difference caused by local deformation or damage of the test piece. Through the multi-time scale modeling capability of the dual-channel lightweight model and the adaptive compensation mechanism of the strain feedback, the method breaks through the response limitations of traditional fixed models to dynamic changes in test piece characteristics and loading paths, significantly improves the real-time performance and accuracy of multi-channel phase difference prediction, and solves the problems of insufficient channel interference processing and phase adjustment lag in the prior art, providing key technical support for improving the authenticity and test accuracy of multi-axis load simulation.

[0104] Based on a multi-objective dynamic parameter optimization algorithm, the control parameters of the testing machine are optimized and adjusted in combination with the phase difference prediction value and the inter-channel coupling characteristics.

[0105] The method for optimizing and adjusting the control parameters of the testing machine based on the multi-objective dynamic parameter optimization algorithm comprises the following steps:

[0106] A target function is defined, which includes a phase synchronization error, energy consumption and dynamic stability. A channel coupling matrix is constructed through a pressure covariance based on the pressure signals between different channels. A single-channel root mean square error is calculated based on the phase difference prediction value and the actual measured phase difference. A phase synchronization error is calculated based on the single-channel root mean square error, the current phase difference and the channel coupling matrix.

[0107] The energy consumption is calculated based on real-time voltage signals and real-time current signals collected by the sensor.

[0108] The Lyapunov exponent is calculated from a preset initial value and an exponential function that exponentially changes with a decay coefficient, and the Lyapunov exponent is taken as the dynamic stability.

[0109] The existing method mainly relies on fixed calibration data and preset mathematical model, and cannot adjust the phase difference in real time according to the dynamic changes of the specimen characteristics and the loading path, and when multi-axial loading is performed, the pressure fluctuations among the hydraulic channels interfere with each other, for example, the flow change of channel A causes the pressure pulsation of channel B, which easily leads to the distortion of the preset phase difference and affects the authenticity of the multi-axial load simulation, and the above problems have not been effectively solved in the prior art. The application constructs a target function including phase synchronization error, energy consumption and dynamic stability, specifically, a channel coupling matrix is constructed by the pressure covariance of the pressure signals among different channels, and the phase synchronization error is calculated by combining the single-channel root mean square error and the current phase difference, so that the influence of the pressure fluctuation interference among the channels on the phase difference can be quantified, the target function can reflect the coupling characteristics in multi-axial loading in real time, and the limitation of the fixed model on dynamic phase difference adjustment is broken; the energy consumption is calculated according to the real-time voltage and current signals, so that the system energy efficiency can be considered while the phase difference is optimized; the Lyapunov index is calculated by using a preset initial value and an exponential function with an exponential change of a decay coefficient to represent the dynamic stability, so that the nonlinear dynamic characteristics of the system can be effectively evaluated, and the system instability caused by phase adjustment can be avoided. The target function realizes adaptive response to the dynamic changes of the specimen characteristics and the loading path through multi-dimensional index fusion, solves the problems of phase difference adjustment lag and insufficient channel interference processing in the prior art, and significantly improves the authenticity of multi-axial load simulation and the comprehensive performance of the system.

[0110] The target function is normalized to obtain a normalized target;

[0111] R particles are initialized, the particles represent decision variables, and the decision variables are composed of PID gains and filter medium frequencies;

[0112] The particles are randomly distributed in the parameter space of the decision variables, and the particle velocities are initialized to explore the feasible region;

[0113] The phase error and the error change rate are taken as input variables of a fuzzy logic controller, and the corresponding output variables composed of learning factors are obtained through a fuzzy rule table predefined by the fuzzy logic controller;

[0114] The particles are divided into an exploration group and a development group, the particle positions of the exploration group are updated in a PSO mode, and the particle positions of the development group are updated in a CPO mode;

[0115] In the PSO mode, the weights are dynamically updated based on the error standard deviation, and the velocities are updated in combination with the dynamically updated weights, the current optimal solution and the global optimal solution;

[0116] In the CPO mode, the speed is updated in combination with the global optimal solution, the distance between the current position and the global optimal solution and the defense behavior of the guinea pig;

[0117] Based on the updated speed, the particle position is updated, and when it exceeds the preset range, the mirror reflection method is used for position correction and speed correction. When the distance Kp between the particle position and the optimal particle is greater than 10, it is set to 20−Kp and the speed is reversed.

[0118] The dimension difference of the multi-objective quantity is eliminated by target function normalization processing, and the fair trade-off of phase synchronization error, energy consumption and dynamic stability is ensured; the particle swarm is initialized and randomly distributed in the parameter space, and the fuzzy logic controller is combined to dynamically adjust the learning factor based on the phase error and error change rate, so that the algorithm adapts to the requirements of different optimization stages; the above steps divide the particles into an exploration group (PSO mode) and a development group (CPO mode), the PSO mode dynamically updates the weight based on the error standard deviation, and combines the current optimal and global optimal solutions to improve the global search ability and break through the limitation of traditional PSO easily falling into local optimum; the CPO mode simulates the defense behavior of the guinea pig, and strengthens the local fine search through the distance information between the global optimal solution and the current position, and makes up for the search blind area of the single algorithm; when the particle position is updated, the mirror reflection method is used to correct the out-of-bound parameters, and it is ensured that the control parameters (PID gain, filter frequency) are within the effective range. Through the mixed search strategy and dynamic parameter adjustment, the algorithm realizes the precise adaptation to the multi-channel coupling characteristics and the dynamic changes of the test piece, solves the problems of phase difference adjustment lag and insufficient channel interference processing in the prior art, and significantly improves the authenticity of multi-axis load simulation and the optimization efficiency of control parameters.

[0119] Merge all particles, update the Pareto front by non-dominated sorting, and keep the top Y% (such as 20%) elite solutions as the global optimal solution for the next iteration;

[0120] Iterate the particles repeatedly until the preset termination condition is met (such as the maximum number of iterations (such as 50 times); or the RMSE average of the Pareto front solution is less than 0.05° and the constraint satisfaction rate is 100%; or the optimal solution has no significant change for 10 consecutive iterations, such as a change rate of less than 1%);

[0121] From the Pareto front, select the solution ranked first and output it to the test machine controller for parameter compensation.

[0122] The method for updating the Pareto front by non-dominated sorting includes:

[0123] For each particle, calculate the fitness of the normalized target, and select the non-dominated solution;

[0124] Mark the particles exceeding the preset energy consumption as infeasible solutions, and mark the particles not exceeding the preset energy consumption as feasible solutions;

[0125] Define the dominance relationship, and in particles A and B, if all targets of A are not inferior to B, or at least one target is strictly superior to B, then A dominates B.

[0126] If j is a feasible solution and i is an infeasible solution, j dominates i, mark i as dominated, and exit the loop; if i and j are both feasible solutions or both infeasible solutions, if j < i (i.e., j dominates i), mark i as dominated, and exit the loop; if i is not dominated by any particle, add i to the non-dominated solution set;

[0127] Iterate through all particles to obtain the Pareto front.

[0128] The multi-objective dynamic parameter optimization algorithm defines a multi-objective function including phase error, energy consumption, and stability, and normalizes the processing. The particle swarm composed of PID gains and filter parameters is initialized. The fuzzy logic controller dynamically adjusts the learning factor according to the real-time phase error and the rate of change. The particles are divided into a PSO mode group for global exploration and a CPO mode group for local development. The non-dominated sorting is used to maintain the Pareto front and retain the elite solution for iterative optimization. Finally, the control parameters with the optimal comprehensive performance are selected from the front solution and output to the test machine. According to the material properties of the test specimen and the preset loading path, the control parameters can be adaptively adjusted to optimize the phase difference. The PSO mode covers a wide parameter space to match the long-period fatigue characteristics of different test specimens. The CPO mode finely adjusts to cope with short-term dynamic changes in the loading path. The fuzzy logic responds to phase fluctuations in real time. The Pareto optimization balances the multi-objective conflict, thereby reducing the test errors such as phase loss and overshoot caused by fixed parameters or single optimization, improving the loading frequency tracking efficiency and system stability, avoiding repeated calibration and shutdown adjustment, and significantly improving the test efficiency and reliability of the electro-hydraulic servo fatigue test machine.

[0129] The strain distribution of the test specimen is monitored in real time. When the strain non-uniformity is greater than the preset strain threshold, the channel load is redistributed and the phase difference is re-optimized.

[0130] Embodiment 2:

[0131] Please refer to Figure 3 The embodiment provides a cross-specimen knowledge transfer method, which comprises the following steps:

[0132] According to the test specimen parameters (such as specimen material properties and geometric topology) and historical test data, a specimen characteristic fingerprint is constructed. The specimen characteristic fingerprint and the new specimen characteristic fingerprint are used as inputs of a graph neural network to obtain a specimen similarity. When the specimen similarity is greater than a preset threshold, a historical optimal solution is loaded. The multi-objective dynamic parameter optimization algorithm initializes the particle swarm based on the historical optimal solution, such as 50% of the particles being distributed around the historical optimal solution in a Gaussian distribution and 50% being randomly distributed.

[0133] The test piece characteristic fingerprint is constructed by extracting test piece material properties, geometric topology and other parameters and historical test data, the similarity of the characteristic fingerprint of the new test piece and the historical test piece is obtained by using the graph neural network, when the similarity is higher than the preset threshold, the historical optimal control parameter solution is directly loaded, and the particle swarm of the multi-objective dynamic parameter optimization algorithm is initialized, such as 50% particles around the historical optimal solution in Gaussian distribution to quickly converge, and 50% particles are randomly distributed to explore new characteristics. The above method can reuse the historical optimization experience based on the similarity of the characteristics of the test piece, avoid repeated optimization of similar test pieces, shorten the parameter adjustment time; at the same time, through the particle initialization strategy combining Gaussian distribution and random distribution, the exploration ability of the slight characteristic difference of the new test piece is reserved on the basis of inheriting the historical optimal solution, so that the test machine control parameter quickly adapts to the phase difference demand under the preset loading path, reduces the test error and time loss caused by parameter optimization from the beginning, and significantly improves the test efficiency and phase control precision of the electro-hydraulic servo fatigue test machine for similar test pieces.

[0134] Embodiment 3:

[0135] Referring to Figure 4 The embodiment provides a test machine phase difference optimization system, which comprises:

[0136] The acquisition and analysis module: preset N-channel loading path signals, acquires sensor data of each channel, the sensor data includes independent data and inter-channel pressure coupling signals, performs feature extraction on the sensor data, and obtains phase difference characteristics; the phase difference characteristics include intra-channel phase difference characteristics and inter-channel coupling characteristics;

[0137] The phase difference prediction module: taking the test piece parameters and the phase difference characteristics as inputs of a double-channel lightweight model, the phase difference prediction values of each channel are obtained;

[0138] The phase difference optimization module: based on a multi-objective dynamic parameter optimization algorithm, the control parameters of the test machine are optimized and adjusted in combination with the phase difference prediction values and the inter-channel coupling characteristics;

[0139] The monitoring and optimization module: real-time monitoring of the strain distribution of the test piece, when the strain non-uniformity is greater than a preset strain threshold, triggering channel load redistribution and re-optimizing the phase difference.

[0140] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0141] Finally: the above only for the preferred embodiments of the present application, and not for limiting the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application, should be included in the scope of protection of the present application.

Claims

1. A test machine phase difference optimization method characterized by, The method comprises the following steps: presetting an N-channel loading path signal, collecting sensor data of each channel, the sensor data including independent data and inter-channel pressure coupling signals, performing feature extraction on the sensor data to obtain phase difference features; the phase difference features include intra-channel phase difference features and inter-channel coupling features; taking the test specimen parameters and the phase difference features as inputs of a dual-channel lightweight model to obtain phase difference prediction values of each channel; optimizing and adjusting control parameters of the testing machine based on a multi-objective dynamic parameter optimization algorithm and in combination with the phase difference prediction values and the inter-channel coupling features; monitoring strain distribution of the test specimen in real time, triggering channel load redistribution and re-optimizing the phase difference when the strain non-uniformity is greater than a preset strain threshold; the method for optimizing and adjusting the control parameters of the testing machine based on the multi-objective dynamic parameter optimization algorithm comprises the following steps: defining an objective function, normalizing the objective function to obtain a normalized objective; initializing R particles, the particles representing decision variables composed of PID gains and filter medium frequencies; randomly distributing the particles in a parameter space of the decision variables, initializing particle velocities; taking the phase error and the error change rate as input variables of a fuzzy logic controller, obtaining corresponding output variables composed of learning factors through a fuzzy rule table predefined by the fuzzy logic controller; dividing the particles into an exploration group and a development group, updating particle positions of the exploration group in a PSO mode and updating particle positions of the development group in a CPO mode; merging all the particles, updating a Pareto front through non-dominated sorting, and reserving the top Y% elite solutions as global optimal solutions for next iteration; repeatedly iterating the particles until a preset termination condition is met; selecting a solution ranked first from the Pareto front and outputting the solution to a testing machine controller for parameter compensation.

2. The test machine phase difference optimization method of claim 1, wherein, The objective function includes a phase synchronization error, energy consumption and dynamic stability; a channel coupling matrix is constructed through pressure covariance according to pressure signals between different channels, a single-channel root mean square error is calculated according to the phase difference prediction values and the actual measured phase difference, and the phase synchronization error is calculated according to the single-channel root mean square error, the current phase difference and the channel coupling matrix; the energy consumption is calculated according to real-time voltage signals and real-time current signals collected by the sensors; the Lyapunov index is calculated by a preset initial value and an exponential function with an exponential change of a decay coefficient, and the Lyapunov index is taken as the dynamic stability.

3. The test machine phase difference optimization method of claim 1, wherein, In the multi-objective dynamic parameter optimization algorithm, in the PSO mode, the weights are dynamically updated based on error standard deviations, and the velocities are updated in combination with the dynamically updated weights, the current optimal solution and the global optimal solution; in the CPO mode, the global optimal solution, the current position and the distance between the global optimal solution and the current position are combined to update the velocities, simulating the defense behavior of a guinea pig; based on the updated velocities, the particle positions are updated, and when the positions exceed a preset range, the positions are corrected and the velocities are corrected by using a mirror reflection method.

4. The test machine phase difference optimization method of claim 1, wherein, The method for obtaining the phase difference prediction values comprises the following steps: building a dual-channel lightweight model including a Transformer-XL channel and a GRU-attention channel; The test specimen parameter is acquired, and the test specimen parameter and the long-term phase difference feature are taken as inputs of a Transformer-XL channel; a long-time phase difference prediction value is obtained; the test specimen parameter and the short-term phase difference feature are taken as inputs of a GRU-attention channel; a short-time phase difference prediction value is obtained; A real-time strain distribution data is collected through an array of M strain gauges attached to the surface of the test specimen, and a strain unevenness is calculated according to the strain distribution data; When the strain unevenness is lower than a preset uniformity threshold, a phase difference prediction value is output by the dual-channel lightweight model; When the strain unevenness is not lower than the preset uniformity threshold, a compensation phase between channels is output by the strain-phase mapping network; The strain distribution data, the material elastic modulus and the current phase difference are taken as inputs of the strain-phase mapping network, a compensation phase of each channel is obtained, the current phase difference is corrected according to the compensation phase, and a phase difference prediction value is obtained.

5. The test machine phase difference optimization method of claim 4, wherein, The method for outputting a phase difference prediction value by a dual-channel lightweight model comprises: The dual-channel weights are dynamically adjusted based on interactive multiple model filtering; the dual-channel weights are calculated based on the likelihood probability of the channels; The test specimen parameter, the phase difference feature, the long-time phase difference prediction value of the Transformer-XL channel at the previous moment and the short-time phase difference prediction value of the GRU-attention channel at the previous moment are taken as current inputs, and a mixed state is obtained in combination with the dual-channel weights; The mixed state is taken as inputs of the Transformer-XL channel and the GRU-attention channel respectively, a predicted long-time phase difference prediction value and a predicted short-time phase difference prediction value are obtained, and a long-time prediction error and a short-time prediction error are calculated; The long-time mean and the long-time variance are obtained by statistics based on the long-time prediction error, the probability density of the long-time prediction error subject to Gaussian distribution is calculated, and a long-time likelihood probability is calculated in combination with the long-time weight at the previous moment; The short-time mean and the short-time variance are obtained by statistics based on the short-time prediction error, the probability density of the short-time prediction error subject to Laplace distribution is calculated, and a short-time likelihood probability is calculated in combination with the short-time weight at the previous moment; the normalized long-time weight and the normalized short-time weight are calculated based on the long-time likelihood probability and the short-time likelihood probability; The phase difference prediction value is obtained by weighting the predicted long-time phase difference prediction value and the predicted short-time phase difference prediction value in combination with the normalized long-time weight and the normalized short-time weight.

6. The test machine phase difference optimization method of claim 1, wherein, The method for obtaining a phase difference feature comprises: The mechanical vibration signal in independent data is acquired, a wavelet base and a decomposition layer number are adaptively selected to decompose the mechanical vibration signal, two N subband signals are obtained, the normalized energy of the subband signals is calculated, and the energy entropy is calculated according to the normalized energy. The correlation weight of the PID gain and the phase difference is calculated through grey correlation degree analysis, and a control modal feature is selected and obtained; The test specimen parameter in the independent data is subjected to spatial wavelet decomposition, and a strain gradient amplitude is extracted and obtained; The strain energy density distribution is calculated based on the strain value and the material elastic modulus in the independent data; The energy entropy, the control modal feature, the strain gradient amplitude and the strain energy density distribution are stacked into a three-dimensional tensor according to a time step G and subjected to normalization processing, and an intra-channel phase difference feature is obtained; An electrical signal in the inter-channel pressure coupling signal is acquired, the electrical signal is transformed through a Hilbert transform to obtain an analytic signal, and an instantaneous phase is extracted according to the analytic signal; a Fourier transform is performed on the electrical signal to obtain a frequency spectrum, and a joint distribution matrix is obtained by analyzing the frequency spectrum; A non-linear model of the environmental variables and the phase difference is established through a partial least squares method, the environmental variables including temperature and oil viscosity; fitting coefficients of the non-linear model are acquired, and coupling features are obtained by splicing the fitting coefficients; The joint distribution matrix and the coupling features are stacked into a three-dimensional tensor according to a time step G and are normalized to obtain inter-channel coupling features.

7. The test machine phase difference optimization method of claim 6, wherein, The method for obtaining the joint distribution matrix comprises: a frequency axis of the frequency spectrum is divided into K frequency intervals, and the instantaneous phase is equally divided into Q phase intervals, each interval representing a discrete range; each frequency interval k is traversed, frequency components in the frequency interval k are screened, each phase interval q is traversed, the screened frequency components are calculated to obtain corresponding instantaneous phases, the instantaneous phases belonging to the phase interval q are screened, and the accumulated energy of the electrical signal belonging to the frequency interval k and the phase interval q is calculated; the total energy is calculated by accumulating the accumulated energy; and the joint distribution matrix is calculated by the accumulated energy and the total energy, wherein k = 1, 2, …, K; q = 1, 2, …, Q.

8. The test machine phase difference optimization method of claim 6, wherein, The method for obtaining the control modal features comprises: acquiring system state parameters and a phase difference; normalizing the system state parameters; acquiring the phase difference as a reference sequence, taking an fth control parameter sequence as a comparison sequence, calculating a grey correlation degree based on a resolution coefficient according to the reference sequence and the comparison sequence, and screening system state parameters with a grey correlation degree higher than a preset correlation degree threshold as control modal features, wherein f = 1, 2, …, F, and F is a total number of control parameters.

9. A test machine phase difference optimization system implementing the test machine phase difference optimization method of any one of claims 1-8, characterized by, comprises: an acquisition and analysis module: presetting N-channel loading path signals, acquiring sensor data of each channel, the sensor data including independent data and inter-channel pressure coupling signals, and extracting features from the sensor data to obtain phase difference features; the phase difference features include intra-channel phase difference features and inter-channel coupling features; a phase difference prediction module: taking test specimen parameters and the phase difference features as inputs of a double-channel lightweight model to obtain phase difference prediction values of each channel; a phase difference optimization module: based on a multi-objective dynamic parameter optimization algorithm, combining the phase difference prediction values and the inter-channel coupling features to optimize and adjust control parameters of a testing machine; a monitoring and optimization module: monitoring strain distribution of a test specimen in real time, and triggering channel load redistribution and re-optimizing the phase difference when the strain non-uniformity is greater than a preset strain threshold.

10. Rubber article fatigue testing machine characterized in that, The application is applied to the testing machine phase difference optimization method in any one of claims 1-8.

Images