Pressure test control method and system for universal pressure testing machine

Abstract

The invention provides a pressure test control method and system for a universal pressure testing machine, which relates to the technical field of material testing machines, and comprises the following steps: obtaining hydraulic cylinder parameters and servo valve parameters, establishing a pressure control mathematical model, calculating a transfer function, an initial feedback controller, determining the number of hidden layer nodes and setting an activation function and a network weight, and obtaining an adaptive feedback controller; generating a closed-loop control system, constructing an objective function based on an integral absolute error and an adjustment time, determining a Pareto optimal solution set in combination with a multi-objective genetic algorithm, selecting a neural network training sample and generating a training sample set, performing offline pre-training, and updating the network weight to obtain a pre-trained neural network; embedding into a pressure control system, outputting feedback controller parameters, setting an anti-saturation strategy for the feedback controller and determining whether saturation is present, limiting the integral term amplitude according to a relationship between an error signal and a saturation direction and determining whether to stop integration, and generating a press control strategy under a current working condition.

Description

Technical Field

[0001] The present invention relates to the technical field of material testing machines, and in particular to a pressure test control method and system for a universal pressure testing machine. Background Art

[0002] Universal pressure testing machine is a general-purpose equipment used for testing the mechanical properties of materials. It can perform various mechanical tests such as tension, compression, bending, shearing, etc. on materials, parts and structures. It is widely used in industrial production, scientific research and quality inspection. The pressure test control system is one of the core components of the universal testing machine. Its performance directly affects the accuracy, efficiency and reliability of the test.

[0003] Traditional universal testing machines mostly achieve automatic control of the test process by setting parameters such as pressure and displacement. However, due to the nonlinear characteristics of the test object and the time-varying nature of the parameters, traditional control methods are difficult to meet the control requirements under different working conditions, and are prone to overshoot, oscillation and other problems, affecting the accuracy of the test results.

[0004] Therefore, a solution is urgently needed to solve the problems existing in the prior art. Summary of the invention

[0005] The embodiments of the present invention provide a pressure test control method and system for a universal pressure testing machine, which can at least solve some of the problems existing in the prior art.

[0006] A first aspect of an embodiment of the present invention provides a pressure test control method for a universal pressure testing machine, comprising:

[0007] Acquire the hydraulic cylinder parameters and servo valve parameters of the universal pressure testing machine, establish a pressure control mathematical model corresponding to the pressure control system in the universal pressure testing machine according to the hydraulic cylinder flow balance equation and the force balance equation, calculate the transfer function of the dynamic relationship between the voltage input and the load pressure output based on the pressure control mathematical model, stack three layers of back propagation neural network as an initial feedback controller based on the pressure control mathematical model, determine the number of hidden layer nodes based on the input layer nodes and output layer nodes of the initial feedback controller, set the activation function and network weight, and obtain an adaptive feedback controller;

[0008] Integrate the pressure control mathematical model and the adaptive feedback controller to generate a closed-loop control system, construct an objective function based on the integral absolute error and the adjustment time, use the overshoot as a constraint, and determine the Pareto optimal solution set corresponding to the feedback controller parameters in combination with a multi-objective genetic algorithm, select neural network training samples from the Pareto optimal solution set, and generate a training sample set covering different pressure instruction types based on common working conditions of the press, use the training sample set to perform offline pre-training on the back-propagation neural network, and update the network weights of each layer of the back-propagation neural network through the LM algorithm to obtain a pre-trained neural network;

[0009] The pre-trained neural network is embedded in the pressure control system. According to the error signal between the pressure feedback and the command, the network weight is adjusted in combination with the gradient descent algorithm and the feedback controller parameters are output. The anti-saturation strategy of the feedback controller is set based on the integral separator and the integral limiter, and it is determined whether the actuator is saturated under the current control parameters. According to the relationship between the error signal and the saturation direction, the amplitude of the integral term is limited and it is determined whether to stop the integration, so as to generate the press control strategy under the current working conditions.

[0010] In an optional embodiment,

[0011] The hydraulic cylinder parameters and servo valve parameters of the universal pressure testing machine are obtained, and a pressure control mathematical model corresponding to the pressure control system in the universal pressure testing machine is established according to the hydraulic cylinder flow balance equation and the force balance equation. The transfer function of the dynamic relationship between the voltage input and the load pressure output is calculated based on the pressure control mathematical model. Based on the pressure control mathematical model, a three-layer back propagation neural network is stacked as an initial feedback controller. The number of hidden layer nodes is determined based on the input layer nodes and the output layer nodes of the initial feedback controller, and the activation function and the network weight are set. The adaptive feedback controller includes:

[0012] For the hydraulic cylinder of the universal pressure testing machine, the corresponding hydraulic cylinder parameters and servo valve parameters are obtained by querying the equipment manual and actual measurement, wherein the hydraulic cylinder parameters include piston area, dead space volume, equivalent mass and damping coefficient, and the servo valve parameters include flow gain coefficient, bulk modulus of hydraulic oil and leakage coefficient in the hydraulic cylinder;

[0013] A flow continuity equation corresponding to the hydraulic cylinder cavity is constructed based on the hydraulic cylinder parameters and the servo valve parameters, a force balance equation and a corresponding dynamic equation are constructed based on the force balance relationship in the hydraulic cylinder, the flow continuity equation, the force balance equation and the dynamic equation are integrated, and the integrated equations are Laplace transformed to obtain a pressure control mathematical model corresponding to the pressure control system in the universal pressure testing machine, and the displacement term after Laplace transformation is eliminated based on the pressure control mathematical model to obtain a transfer function of the dynamic relationship between voltage input and load pressure output;

[0014] A three-layer back propagation neural network with both input layer nodes and output layer nodes being 1 is selected as the initial feedback controller, the number of hidden node layers is determined according to an empirical formula, the connection weights are randomly initialized through an S-type activation function, a performance index function is defined, and the weights of the hidden layer and output neurons are corrected based on the load pressure error signal through an error back propagation algorithm combined with the negative gradient descent principle to minimize the performance index function, the error is continuously calculated and the weights of the output neurons are repeatedly corrected to obtain the adaptive feedback controller.

[0015] In an optional embodiment,

[0016] The transfer function is shown in the following formula:

[0017] ;

[0018] in, G ( s ) is the transfer function value, which is used to indicate the response relationship between the output data and the input data. s represents the complex frequency variable in the Laplace transform, k q represents the flow gain coefficient, A p represents the piston area, V t Represents the total volume of the hydraulic cylinder, β e represents the effective bulk modulus of the liquid, K c represents the flow-pressure coefficient, B p represents the viscous damping coefficient, C t Indicates the total leakage coefficient.

[0019] In an optional embodiment,

[0020] The pressure control mathematical model and the adaptive feedback controller are integrated to generate a closed-loop control system, an objective function is constructed based on the integral absolute error and the adjustment time, the overshoot is used as a constraint, and the Pareto optimal solution set corresponding to the feedback controller parameters is determined in combination with a multi-objective genetic algorithm. A neural network training sample is selected from the Pareto optimal solution set and a training sample set covering different pressure instruction types is generated based on common working conditions of the press. The back propagation neural network is pre-trained offline using the training sample set, and the network weights of each layer of the back propagation neural network are updated by the LM algorithm to obtain a pre-trained neural network including:

[0021] Integrate the pressure control mathematical model and the adaptive feedback controller to generate the closed-loop control system, take the pressure command as input, and output the control voltage through the adaptive feedback controller to act on the servo valve to complete the closed-loop control;

[0022] Taking the integral absolute error and the adjustment time as the performance evaluation indicators, constructing the objective function and taking the overshoot as the constraint condition, taking the feedback controller parameters as the optimization variables, wherein the feedback controller parameters include the number of hidden layer nodes and the weights of the output neurons;

[0023] Based on the current feedback controller parameters, an initial population is randomly generated by a multi-objective genetic algorithm, wherein each individual in the initial population represents a set of feedback controller parameters, the environmental pheromone matrix and the initial value of the pheromone concentration at each position are initialized, for each individual, the objective function value corresponding to the current individual is calculated and the individual optimal position and the global optimal position corresponding to each individual are initialized, the external archive is initialized, for each individual, the speed and position corresponding to the current individual are updated based on the environmental pheromone matrix, for the individual after the position is updated, the objective function value is calculated and the environmental pheromone corresponding to the current individual is updated, and the individual optimal position is updated. If the updated If the position dominates the position before the update, the updated position is added to the external archive and the position information before the update is deleted. The update is repeated until the position of each individual is updated. For the solutions in the updated external archive, non-dominated sorting is performed and non-dominated levels are assigned. Based on the non-dominated levels, the solutions in the external archive are sorted in descending order and the top 10% of the solutions are selected as candidate individuals. The candidate individuals are crossover and mutation operations are performed to generate second-generation individuals. The second-generation population is obtained by combination. The iteration is repeated until the preset maximum number of iterations is reached. The candidate individuals selected in the last iteration are used as Pareto optimal solutions, and the Pareto optimal solution set is obtained by combination.

[0024] A solution in the Pareto optimal solution set is selected as a neural network training sample, and a pressure instruction type corresponding to a common working condition of a press is selected from the neural network training samples to obtain the training sample set. The back propagation neural network is pre-trained offline based on the training sample set. In combination with the LM algorithm, a damping factor is added to adjust the search direction and update the network weights of each layer of the back propagation neural network. The update is repeated until the back propagation neural network converges or reaches a preset maximum number of iterations to obtain the pre-trained neural network.

[0025] In an optional embodiment,

[0026] The environmental pheromone is updated as shown in the following formula:

[0027] ;

[0028] in, t j ( t +1) indicates the position j At time step t +1 environmental pheromone value, r represents the volatilization rate of environmental pheromones, t j ( t ) indicates the position j At time step t The environmental pheromone value of A ( t ) indicates that at the time step t The total number of individuals at i represents the individual index, Q represents the pheromone intensity factor, rank i ( t ) represents an individual i At time step t Ranking, α represents the path selection adjustment parameter, CD i ( t ) represents an individual i The crowding degree at time step, e Represents a very small decimal number, I Represents the indicator function, which is used to indicate whether the current individual has reached the position j , if it has been reached, the value is 1, otherwise the value is 0.

[0029] In an optional embodiment,

[0030] The pre-trained neural network is embedded into the pressure control system. According to the error signal between the pressure feedback and the command, the network weight is adjusted in combination with the gradient descent algorithm and the feedback controller parameters are output. The anti-saturation strategy of the feedback controller is set based on the integral separator and the integral limiter and it is determined whether the actuator is saturated under the current control parameters. According to the relationship between the error signal and the saturation direction, the integral term amplitude is limited and it is determined whether to stop the integration. The press control strategy under the current working condition is generated, including:

[0031] The pre-trained neural network acquired in advance is embedded into the pressure control system of the universal pressure testing machine, and the network weights are adaptively updated through a gradient descent algorithm according to the pressure feedback of the pressure control system and the error signal of the control command, and the feedback controller parameters are outputted at the same time;

[0032] The anti-saturation strategy of the feedback controller is set based on integral separation and integral limiting. The integral separator and integral limiter are set. According to the current feedback controller parameters, it is determined whether to use integral control according to the integral separation theory. If integral control is used, the integral limiting theory is used to determine whether the actuator is saturated under the current control parameters. If saturated, it is determined whether the error signal and the saturation direction are consistent. If they are consistent, the integral separator input is 0 and the integration is stopped. If they are inconsistent, the error signal is output and integrated. If the error signal is output and integrated, the saturation function is set by the integral limiter in combination with the value range of the pressure error.

[0033] Based on the saturation function, combined with the integral limiter and the integral separator, an anti-saturation control law is generated and the feedback controller parameters are adjusted based on the anti-saturation control law, optimized control parameters are generated and parsed into control instructions, and a press control strategy under the current working conditions is generated.

[0034] In an optional embodiment,

[0035] Based on the saturation function, combined with the integral limiter and the integral separator, the anti-saturation control law is generated as shown in the following formula:

[0036] ;

[0037] in, u ( t ) represents the output signal value of the anti-saturation control law, K p represents the proportional gain, e ( t ) indicates the current time t The error, sat () represents the saturation function, K i represents the integral gain, e s( i ) indicates the time i The error signal at time u s represents the saturation value of the integral limiter, K d represents the differential gain, Indicates the error change rate.

[0038] A second aspect of an embodiment of the present invention provides a pressure test control system for a universal pressure testing machine, comprising:

[0039] The first unit is used to obtain the hydraulic cylinder parameters and servo valve parameters of the universal pressure testing machine, establish a pressure control mathematical model corresponding to the pressure control system in the universal pressure testing machine according to the hydraulic cylinder flow balance equation and the force balance equation, calculate the transfer function of the dynamic relationship between the voltage input and the load pressure output based on the pressure control mathematical model, stack three layers of back propagation neural network as an initial feedback controller based on the pressure control mathematical model, determine the number of hidden layer nodes based on the input layer nodes and output layer nodes of the initial feedback controller, set the activation function and network weight, and obtain an adaptive feedback controller;

[0040] The second unit is used to integrate the pressure control mathematical model and the adaptive feedback controller to generate a closed-loop control system, construct an objective function based on the integral absolute error and the adjustment time, use the overshoot as a constraint, and determine the Pareto optimal solution set corresponding to the feedback controller parameters in combination with a multi-objective genetic algorithm, select neural network training samples from the Pareto optimal solution set and generate a training sample set covering different pressure instruction types based on common working conditions of the press, use the training sample set to perform offline pre-training on the back-propagation neural network, and update the network weights of each layer of the back-propagation neural network through the LM algorithm to obtain a pre-trained neural network;

[0041] The third unit is used to embed the pre-trained neural network into the pressure control system, adjust the network weights and output feedback controller parameters according to the error signal between the pressure feedback and the command in combination with the gradient descent algorithm, set the feedback controller anti-saturation strategy based on the integral separator and the integral limiter and determine whether the actuator is saturated under the current control parameters, limit the integral term amplitude according to the relationship between the error signal and the saturation direction and determine whether to stop the integration, and generate the press control strategy under the current working conditions.

[0042] According to a third aspect of the embodiments of the present invention,

[0043] An electronic device is provided, comprising:

[0044] processor;

[0045] a memory for storing processor-executable instructions;

[0046] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0047] According to a fourth aspect of the embodiments of the present invention,

[0048] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0049] In the present invention, by selecting appropriate parameters from the Pareto solution set as neural network training samples, the controller parameters can meet the constraints while taking into account different performance indicators, thereby realizing the intelligent and automated parameter setting. The combination of offline pre-training and online learning takes into account the generalization ability and real-time performance of the controller, and improves the adaptability and robustness of the control system. The anti-saturation strategy effectively suppresses the integral saturation phenomenon, and avoids the system oscillation and instability caused by the actuator being in a saturated state for a long time. Reasonable setting of the integral limit value can coordinate the consistency of the controller output and the actuator action, and ensure that the control system operates safely and reliably under the constraints of the actuator. In summary, this embodiment takes into account the dynamic performance, steady-state accuracy, safety and real-time performance of the control system, adapts to the complex nonlinear characteristics and multi-operating conditions of the press, realizes the automation and intelligence of the test process, and provides new ideas and solutions for the intelligent upgrading and technological innovation of high-end equipment such as hydraulic universal testing machines, and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A schematic flow chart of a pressure test control method of a universal pressure testing machine according to an embodiment of the present invention;

[0051] Figure 2 It is a structural schematic diagram of a pressure test control system of a universal pressure testing machine according to an embodiment of the present invention. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0054] Figure 1 FIG. 1 is a flow chart of a pressure test control method of a universal pressure testing machine according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0055] S1. Obtain the hydraulic cylinder parameters and servo valve parameters of the universal pressure testing machine, establish a pressure control mathematical model corresponding to the pressure control system in the universal pressure testing machine according to the hydraulic cylinder flow balance equation and the force balance equation, calculate the transfer function of the dynamic relationship between the voltage input and the load pressure output based on the pressure control mathematical model, stack three layers of back propagation neural network as the initial feedback controller based on the pressure control mathematical model, determine the number of hidden layer nodes based on the input layer nodes and output layer nodes of the initial feedback controller, set the activation function and network weight, and obtain an adaptive feedback controller;

[0056] The universal pressure testing machine is a device widely used in material testing, which can perform various tests such as stretching, compression, and bending. The flow balance equation of the hydraulic cylinder describes the relationship between the flow of liquid into and out of the cylinder body. The force balance equation describes the force balance of the piston. The pressure control system is used to adjust the pressure in the hydraulic system to ensure that the pressure is maintained within a set range. The transfer function is a function that describes the relationship between the input and output of the control system. The back propagation neural network is an algorithm for training artificial neural networks. By calculating the error layer by layer and back propagating it, the weight of the network is adjusted to minimize the prediction error. The adaptive feedback controller is a controller that can self-adjust according to the dynamic characteristics of the system and environmental changes.

[0057] In an optional embodiment,

[0058] The hydraulic cylinder parameters and servo valve parameters of the universal pressure testing machine are obtained, and a pressure control mathematical model corresponding to the pressure control system in the universal pressure testing machine is established according to the hydraulic cylinder flow balance equation and the force balance equation. The transfer function of the dynamic relationship between the voltage input and the load pressure output is calculated based on the pressure control mathematical model. Based on the pressure control mathematical model, a three-layer back propagation neural network is stacked as an initial feedback controller. The number of hidden layer nodes is determined based on the input layer nodes and the output layer nodes of the initial feedback controller, and the activation function and the network weight are set. The adaptive feedback controller includes:

[0059] For the hydraulic cylinder of the universal pressure testing machine, the corresponding hydraulic cylinder parameters and servo valve parameters are obtained by querying the equipment manual and actual measurement, wherein the hydraulic cylinder parameters include piston area, dead space volume, equivalent mass and damping coefficient, and the servo valve parameters include flow gain coefficient, bulk modulus of hydraulic oil and leakage coefficient in the hydraulic cylinder;

[0060] A flow continuity equation corresponding to the hydraulic cylinder cavity is constructed based on the hydraulic cylinder parameters and the servo valve parameters, a force balance equation and a corresponding dynamic equation are constructed based on the force balance relationship in the hydraulic cylinder, the flow continuity equation, the force balance equation and the dynamic equation are integrated, and the integrated equations are Laplace transformed to obtain a pressure control mathematical model corresponding to the pressure control system in the universal pressure testing machine, and the displacement term after Laplace transformation is eliminated based on the pressure control mathematical model to obtain a transfer function of the dynamic relationship between voltage input and load pressure output;

[0061] A three-layer back propagation neural network with both input layer nodes and output layer nodes being 1 is selected as the initial feedback controller, the number of hidden node layers is determined according to an empirical formula, the connection weights are randomly initialized through an S-type activation function, a performance index function is defined, and the weights of the hidden layer and output neurons are corrected based on the load pressure error signal through an error back propagation algorithm combined with the negative gradient descent principle to minimize the performance index function, the error is continuously calculated and the weights of the output neurons are repeatedly corrected to obtain the adaptive feedback controller.

[0062] The piston area is the cross-sectional area of ​​the hydraulic cylinder piston, the dead space volume is the space in the hydraulic system that cannot be effectively utilized, the damping coefficient describes the energy dissipation characteristics of the hydraulic cylinder during movement, the flow gain coefficient represents the gain of the flow response of the servo valve or hydraulic cylinder to the input signal, the bulk modulus is the resistance of the liquid to volume changes, reflecting the compression characteristics of the hydraulic oil, the hydraulic cylinder leakage coefficient describes the leakage of the hydraulic cylinder due to poor sealing or other factors, the flow continuity equation describes the conservation of the hydraulic cylinder flow, ensuring that the flow of the inflowing and outflowing liquids is equal, the displacement term refers to the displacement of the piston in the hydraulic cylinder, affecting the working range and output of the hydraulic cylinder, the S-type activation function is a nonlinear activation function, the empirical formula is a formula summarized based on actual data or experimental results, and is used to estimate or predict certain characteristics or behaviors, and the load pressure error signal is the difference between the actual load pressure and the target load pressure.

[0063] Consult the manual of the press equipment to obtain the parameters of the hydraulic cylinder, such as the piston area, dead space volume, equivalent mass and damping coefficient. Through actual measurement or consulting the servo valve parameter table, obtain the parameters such as the servo valve flow gain coefficient, hydraulic oil bulk modulus and hydraulic cylinder internal leakage coefficient. Based on the flow continuity principle of the hydraulic cylinder cavity, write the flow continuity equation to describe the relationship between the servo valve flow, pressure change and hydraulic cylinder displacement. According to the force balance relationship on both sides of the piston in the hydraulic cylinder, establish the force balance equation to describe the balance between the pressure difference force, load force, damping force and inertia force of the piston. Use Newton's second law to construct the dynamic equation of the hydraulic cylinder to describe the relationship between the piston mass, acceleration and force. Integrate the flow continuity equation, force balance equation and dynamic equation to obtain a complete mathematical model of the hydraulic cylinder.

[0064] Perform Laplace transformation on the integrated hydraulic cylinder mathematical model to convert the time domain equation into a complex frequency domain equation. In the Laplace transformed equation, eliminate the displacement term to obtain the dynamic relationship expression between the servo valve input voltage and the hydraulic cylinder output pressure. Simplify the above expression to obtain the transfer function between the voltage input and the load pressure output as the mathematical model of the pressure control system.

[0065] A three-layer back propagation neural network is selected as the initial feedback controller. The number of nodes in the input layer and the output layer are both set to 1. The number of nodes in the hidden layer is determined according to the empirical formula. The connection weights from the input layer to the hidden layer and from the hidden layer to the output layer are randomly initialized. The S-type function is selected as the activation function. The performance index function of the feedback controller is defined. The error back propagation algorithm is used to calculate the negative gradient of the weights of each layer according to the load pressure error signal. Based on the negative gradient descent principle, the gradient descent formula is used to correct the weights of the hidden layer and the output layer. The weights are updated repeatedly until the performance index function is minimized, and the trained adaptive feedback controller is obtained.

[0066] For example, taking a certain type of press as an example, the hydraulic cylinder parameters and servo valve parameters are determined, the hydraulic cylinder flow continuity equation, force balance equation and dynamic equation are constructed, and Laplace transform is performed to obtain the pressure control system transfer function. A three-layer BP neural network with a 1-5-1 structure is selected as the initial feedback controller, and 5 nodes are taken in the hidden layer. The weights from the input layer to the hidden layer and the weights from the hidden layer to the output layer are randomly initialized, the bias is set and the mean square error performance indicator function is defined. The learning rate is set to 0.01, and the error back propagation algorithm is used to iteratively train the network. In each control cycle, the gradient of the loss function is calculated according to the error signal between the pressure feedback and the command, and the weight is updated using the gradient descent formula to continuously reduce the control error. After multiple iterations, when the loss function value is less than the set threshold, an adaptive feedback controller with optimized performance is obtained.

[0067] In this embodiment, by constructing an accurate mathematical model of the hydraulic cylinder, the nonlinear characteristics of the hydraulic system, such as flow-pressure characteristics, force balance relationship and dynamic characteristics, are taken into account, laying the foundation for designing a high-performance controller. A three-layer BP neural network is used as an adaptive feedback controller, and online learning is performed through the error back propagation algorithm. The weights of the hidden layer and the output layer are continuously optimized according to the pressure error signal, so that the controller can adaptively adjust the control strategy to adapt to changes in hydraulic system parameters and external interference. By learning the dynamic characteristics of the hydraulic system, the optimal control law can be generated to improve the dynamic response performance of the system, such as shortening the pressure adjustment time, reducing the pressure overshoot and steady-state error. In summary, this embodiment simplifies the design process of the controller, is easy to implement in hardware, and is convenient for engineering applications.

[0068] In an optional embodiment,

[0069] The transfer function is shown in the following formula:

[0070] ;

[0071] in, G ( s ) is the transfer function value, which is used to indicate the response relationship between the output data and the input data. s represents the complex frequency variable in the Laplace transform, k q represents the flow gain coefficient, A p represents the piston area, V t Represents the total volume of the hydraulic cylinder, β e represents the effective bulk modulus of the liquid, K c represents the flow-pressure coefficient, B p represents the viscous damping coefficient, C t Indicates the total leakage coefficient.

[0072] In this embodiment, by analyzing the structure and parameters of the transfer function, the inherent characteristics of the hydraulic system, such as response speed, damping characteristics and steady-state gain, can be deeply understood, providing a theoretical basis for controller design. By identifying these parameters, the dynamic characteristics of a specific press can be accurately described. When designing a controller, the controller gain can be calculated based on these parameter values ​​to achieve optimal control performance. Through simulation experiments, the controller parameters can be optimized, and the dynamic response of the system under different working conditions can be predicted, providing guidance for the development and debugging of actual control systems. By estimating the transfer function parameters in real time and comparing them with the nominal values, the faults and performance degradation of the hydraulic system, such as increased leakage and damping changes, can be discovered in time, providing a basis for state monitoring and predictive maintenance. In summary, this embodiment reveals the essential characteristics of the pressure control system, provides an important theoretical basis for controller design, parameter setting, simulation verification, fault diagnosis and system optimization, and is of great significance for improving the control performance and reliability of the press.

[0073] S2. Integrate the pressure control mathematical model and the adaptive feedback controller to generate a closed-loop control system, construct an objective function based on the integral absolute error and the adjustment time, use the overshoot as a constraint, and determine the Pareto optimal solution set corresponding to the feedback controller parameters in combination with a multi-objective genetic algorithm. Select neural network training samples from the Pareto optimal solution set and generate a training sample set covering different pressure instruction types based on common working conditions of the press. Use the training sample set to perform offline pre-training on the back-propagation neural network, and update the network weights of each layer of the back-propagation neural network through the LM algorithm to obtain a pre-trained neural network.

[0074] The closed-loop control system is a system that adjusts the control input by continuously measuring the system output and comparing these measurement results with the set target. The integrated absolute error is a control system performance indicator used to evaluate the quality of the system response. The overshoot refers to the degree to which the system response exceeds the target value, usually expressed as a percentage of the target value. The multi-objective genetic algorithm is a type of algorithm used to solve optimization problems with multiple conflicting objectives. It searches the solution space by simulating the natural selection process. The goal is to find a set of balanced solutions so that the value of each objective function is optimized. The pressure instruction types include constant pressure, gradual pressure, and pulse pressure. The offline pre-training refers to the preliminary training of the model using pre-collected data before formal training. The LM algorithm is an optimization algorithm for nonlinear least squares problems. It combines the advantages of the Gauss-Newton method and the gradient descent method, and minimizes the sum of squared errors by iteratively adjusting parameters.

[0075] In an optional embodiment,

[0076] The pressure control mathematical model and the adaptive feedback controller are integrated to generate a closed-loop control system, an objective function is constructed based on the integral absolute error and the adjustment time, the overshoot is used as a constraint, and the Pareto optimal solution set corresponding to the feedback controller parameters is determined in combination with a multi-objective genetic algorithm. A neural network training sample is selected from the Pareto optimal solution set and a training sample set covering different pressure instruction types is generated based on common working conditions of the press. The back propagation neural network is pre-trained offline using the training sample set, and the network weights of each layer of the back propagation neural network are updated by the LM algorithm to obtain a pre-trained neural network including:

[0077] Integrate the pressure control mathematical model and the adaptive feedback controller to generate the closed-loop control system, take the pressure command as input, and output the control voltage through the adaptive feedback controller to act on the servo valve to complete the closed-loop control;

[0078] Taking the integral absolute error and the adjustment time as the performance evaluation indicators, constructing the objective function and taking the overshoot as the constraint condition, taking the feedback controller parameters as the optimization variables, wherein the feedback controller parameters include the number of hidden layer nodes and the weights of the output neurons;

[0079] Based on the current feedback controller parameters, an initial population is randomly generated by a multi-objective genetic algorithm, wherein each individual in the initial population represents a set of feedback controller parameters, the environmental pheromone matrix and the initial value of the pheromone concentration at each position are initialized, for each individual, the objective function value corresponding to the current individual is calculated and the individual optimal position and the global optimal position corresponding to each individual are initialized, the external archive is initialized, for each individual, the speed and position corresponding to the current individual are updated based on the environmental pheromone matrix, for the individual after the position is updated, the objective function value is calculated and the environmental pheromone corresponding to the current individual is updated, and the individual optimal position is updated. If the updated If the position dominates the position before the update, the updated position is added to the external archive and the position information before the update is deleted. The update is repeated until the position of each individual is updated. For the solutions in the updated external archive, non-dominated sorting is performed and non-dominated levels are assigned. Based on the non-dominated levels, the solutions in the external archive are sorted in descending order and the top 10% of the solutions are selected as candidate individuals. The candidate individuals are crossover and mutation operations are performed to generate second-generation individuals. The second-generation population is obtained by combination. The iteration is repeated until the preset maximum number of iterations is reached. The candidate individuals selected in the last iteration are used as Pareto optimal solutions, and the Pareto optimal solution set is obtained by combination.

[0080] A solution in the Pareto optimal solution set is selected as a neural network training sample, and a pressure instruction type corresponding to a common working condition of a press is selected from the neural network training samples to obtain the training sample set. The back propagation neural network is pre-trained offline based on the training sample set. In combination with the LM algorithm, a damping factor is added to adjust the search direction and update the network weights of each layer of the back propagation neural network. The update is repeated until the back propagation neural network converges or reaches a preset maximum number of iterations to obtain the pre-trained neural network.

[0081] The environmental pheromone matrix is ​​a matrix used to store pheromones in the ant colony optimization algorithm. The pheromone concentration refers to the quality of a path or solution represented by the pheromone concentration in the optimization algorithm. The external archive is a set of non-dominated solutions used to store non-dominated sorting in a multi-objective optimization algorithm. The non-dominated ranking is a sorting method used to evaluate the quality of solutions in multi-objective optimization. The non-dominated rank refers to the position rank of a solution in the non-dominated ranking. The Pareto optimal solution is a type of solution defined in multi-objective optimization.

[0082] The pressure control mathematical model and adaptive feedback controller are integrated to generate a closed-loop control system. The pressure command is used as input, and the adaptive feedback controller outputs a control voltage to act on the servo valve. The servo valve adjusts the hydraulic oil flow according to the control voltage, thereby controlling the pressure output of the hydraulic cylinder. The pressure sensor measures the actual pressure and feeds it back to the adaptive controller to form a closed-loop control.

[0083] Taking the integral absolute error and adjustment time as performance evaluation indicators, a multi-objective function is constructed, the overshoot is used as a constraint, the number of hidden layer nodes and the output neuron weight of the feedback controller are used as optimization variables, and the initial population is randomly generated. Each individual represents a set of controller parameters. The environmental pheromone matrix and the pheromone concentration at each position are initialized. For each individual, the objective function value is calculated, the individual optimal position and the global optimal position are initialized, and the external archive is initialized to save the non-dominated solution.

[0084] For each individual, based on the speed and position of the environmental pheromone matrix, calculate the objective function value and update the environmental pheromone for the individual after the position is updated, update the optimal position of the individual, and if the updated position dominates the position before the update, add it to the external archive, sort the solutions in the external archive by non-dominated, select the top 10% of non-dominated solutions as candidate individuals, perform crossover mutation on the candidate individuals, generate a new generation of population, repeat the iteration until the maximum number of iterations is reached, and take the final candidate individuals as the Pareto optimal solution set;

[0085] Some solutions are selected from the Pareto optimal solution set as neural network training samples. Combined with the pressure instructions corresponding to the common working conditions of the press, a training subset is selected from the training samples. The LM algorithm is used for offline pre-training. The search direction is adjusted by adding a damping factor and the network weights of each layer of the back-propagation neural network are updated until the network converges or the maximum number of iterations is reached. The pre-trained back-propagation neural network is obtained as an adaptive feedback controller.

[0086] For example, a 400kN universal pressure testing machine is considered. The hydraulic cylinder parameters and servo valve parameters are set. According to the transfer function model, the pressure control mathematical model of the press can be obtained. A three-layer BP neural network is designed as an adaptive feedback controller. The number of hidden layer nodes ranges from [5, 15], and the output layer weight ranges from [-1, 1]. The integral absolute error and adjustment time are used as performance evaluation indicators. Constraints are set and the population size is set to 50 based on the multi-objective genetic algorithm. The maximum number of iterations is 100, the crossover probability is 0.8, and the mutation probability is 0.1. After 100 iterations, the Pareto optimal solution set was obtained. A solution was randomly selected from the Pareto optimal solution set and used as the initial weight of the back propagation neural network. 50 groups of step, sine and square wave pressure instructions with different amplitudes, frequencies and durations were selected as training samples. The LM algorithm was used for offline training. The maximum number of iterations was set to 1000, the learning rate was 0.01, and the target error was 0.001. After 872 iterations, the back propagation neural network converged, and the fitting error on the test sample was 0.15MPa. The pre-trained back propagation neural network was obtained as an adaptive feedback controller.

[0087] In this embodiment, by integrating the pressure control mathematical model with the adaptive feedback controller, a closed-loop control system is constructed, and the controller parameters are used as optimization variables, the integral absolute error and the adjustment time are used as objective functions, and the overshoot is used as a constraint condition. A multi-objective genetic algorithm is used for parameter optimization, and a set of Pareto optimal solutions that meet the performance requirements can be obtained. The blindness of the traditional trial and error method is overcome, and the efficiency and reliability of the controller parameter setting are improved. The neural network controller can quickly learn the dynamic characteristics of the press, improve the response speed and steady-state accuracy of the system, and evaluate the generalization ability of the neural network by setting verification samples to ensure that the controller can maintain good control performance under unknown working conditions. In summary, this embodiment has significant advantages in improving control accuracy, adaptability, robustness and development efficiency, and has opened up a new path for the intelligent control of hydraulic servo systems and has broad application prospects.

[0088] In an optional embodiment,

[0089] The environmental pheromone is updated as shown in the following formula:

[0090] ;

[0091] in, t j ( t +1) indicates the position j At time step t +1 environmental pheromone value, r represents the volatilization rate of environmental pheromones, t j ( t ) indicates position j At time step t The environmental pheromone value of A ( t ) indicates that at the time step t The total number of individuals at i represents the individual index, Q represents the pheromone intensity factor, rank i ( t ) represents an individual i At time step t Ranking, α represents the path selection adjustment parameter, CD i ( t ) represents an individual i The crowding degree at time step, e Represents a very small decimal number, I Represents the indicator function, which is used to indicate whether the current individual has reached the position j , if it has been reached, the value is 1, otherwise the value is 0.

[0092] In this embodiment, by updating the environmental pheromones, the search experience of excellent individuals can be passed on to other individuals, guiding the population to evolve in a more optimal direction. By appropriately adjusting the pheromone volatilization rate, the algorithm can avoid falling into a local optimum. When calculating the pheromone increment, the two factors of the individual's non-dominated ranking and crowding are considered at the same time, ensuring the convergence and diversity of the population, so that the algorithm can converge to the Pareto front and be evenly distributed, and obtain a set of balanced non-dominated solutions. By introducing the path selection adjustment parameter, the influence weight of crowding in pheromone updating can be controlled. In summary, this embodiment comprehensively considers the individual ranking, crowding and historical search information, guides the population to evolve in a more optimal direction, accelerates the convergence speed of multi-objective optimization, and maintains the diversity of the population, providing an effective method for solving complex multi-objective optimization problems.

[0093] S3. Embed the pre-trained neural network into the pressure control system, adjust the network weights and output the feedback controller parameters according to the error signal between the pressure feedback and the command in combination with the gradient descent algorithm, set the anti-saturation strategy of the feedback controller based on the integral separator and the integral limiter, and determine whether the actuator is saturated under the current control parameters, limit the amplitude of the integral term according to the relationship between the error signal and the saturation direction, and determine whether to stop the integration, so as to generate the press control strategy under the current working conditions.

[0094] The error signal between the pressure feedback and the command is the difference between the actual pressure and the target pressure (command pressure), which is used in the feedback control system to adjust the control signal so that the output of the system is as close to the target value as possible. The integral separator is a component in a control strategy, which is used to prevent excessive accumulation of the integral term when the error is very small or the system is in a steady state. The integral limiter is used to limit the growth range of the integral term to prevent the integral action from causing excessive deviation of the control signal. The anti-saturation strategy of the feedback controller is intended to prevent the control signal from affecting the system performance due to reaching the saturation limit. The actuator is a key component in the control system, which is used to convert the control signal into an actual physical operation. The integral term amplitude is the amplitude of the integral action, which represents the size of the integral term in the control system.

[0095] In an optional embodiment,

[0096] The pre-trained neural network is embedded into the pressure control system. According to the error signal between the pressure feedback and the command, the network weight is adjusted in combination with the gradient descent algorithm and the feedback controller parameters are output. The anti-saturation strategy of the feedback controller is set based on the integral separator and the integral limiter and it is determined whether the actuator is saturated under the current control parameters. According to the relationship between the error signal and the saturation direction, the integral term amplitude is limited and it is determined whether to stop the integration. The press control strategy under the current working condition is generated, including:

[0097] The pre-trained neural network acquired in advance is embedded into the pressure control system of the universal pressure testing machine, and the network weights are adaptively updated through a gradient descent algorithm according to the pressure feedback of the pressure control system and the error signal of the control command, and the feedback controller parameters are outputted at the same time;

[0098] The anti-saturation strategy of the feedback controller is set based on integral separation and integral limiting. The integral separator and integral limiter are set. According to the current feedback controller parameters, it is determined whether to use integral control according to the integral separation theory. If integral control is used, the integral limiting theory is used to determine whether the actuator is saturated under the current control parameters. If saturated, it is determined whether the error signal and the saturation direction are consistent. If they are consistent, the integral separator input is 0 and the integration is stopped. If they are inconsistent, the error signal is output and integrated. If the error signal is output and integrated, the saturation function is set by the integral limiter in combination with the value range of the pressure error.

[0099] Based on the saturation function, combined with the integral limiter and the integral separator, an anti-saturation control law is generated and the feedback controller parameters are adjusted based on the anti-saturation control law, optimized control parameters are generated and parsed into control instructions, and a press control strategy under the current working conditions is generated.

[0100] The integral separation theory is a strategy used in control systems to prevent the integral action from excessively affecting system performance. The integral limiting theory is intended to limit the amplitude of the integral term to avoid the control signal exceeding the system's capabilities or causing saturation. The saturation direction refers to the direction in which the control signal reaches a saturated state. In a control system, saturation can occur in a positive or negative control signal. The anti-saturation control law is a design method used to deal with the control signal saturation problem to avoid system performance degradation. The control instruction is a signal generated in a control system to adjust the system's behavior or output. The press control strategy refers to the strategy and method used to operate and manage a press.

[0101] The input layer of the neural network is connected to the pressure feedback and control command error signal, and the output layer of the neural network is connected to the feedback controller parameters. During the operation of the press, the error between the network predicted output and the expected output is calculated based on the pressure feedback and control command error signal. The weight of the neural network is adjusted in the opposite direction of the error gradient using the gradient descent algorithm to make the network output close to the expected output. The updated network output is parsed into the parameters of the pressure controller to achieve adaptive adjustment of the controller parameters.

[0102] Integral separator and integral limiter are introduced to prevent integral saturation when the actuator is saturated. According to the current control error and actuator status, it is judged whether integral control is needed. If integral control is needed, it is judged whether the actuator is saturated according to the integral limiter theory. If the actuator is not saturated, the error signal is normally output for integral control. If the actuator is saturated, it is judged whether the error signal is consistent with the saturation direction. If they are consistent, the integration is stopped, that is, the integral separator output is zero. If they are inconsistent, the error signal continues to be output and integral control is performed.

[0103] Combined with the value range of pressure error, a saturation function is designed. When the pressure error is less than the lower limit or greater than the upper limit, the saturation function outputs a constant lower limit value or upper limit value. When the saturation function is within the normal range of pressure error, the error value is normally output. The output signal of the integral controller is limited by the saturation function to prevent oversaturation. The integral separator, integral limiter and saturation function are combined to generate an anti-saturation control law. The anti-saturation control law is applied to the feedback controller to optimize and adjust the control parameters. The optimized control parameters are parsed into control instructions to form a press control strategy under the current working conditions.

[0104] For example, consider a 100kN servo press, whose pressure control system adopts the adaptive control strategy of the present invention, and obtains a 3-layer back propagation neural network through offline training, with 2 neurons in the input layer (pressure error and error change rate), 10 neurons in the hidden layer, and 3 neurons in the output layer. The network is embedded in the press as an adaptive feedback controller, and the press is set to operate under 80kN and 1Hz sinusoidal pressure instructions. Initially, the feedback controller parameters and the controller parameters output by each neuron are 2.5, 0.2 and 0.1 respectively. In each control cycle, the pressure sensor measures the actual pressure and calculates the pressure error and error change rate. The input Back propagate the neural network and forward propagate it to obtain the correction value of the control parameter. Use the gradient descent method and the back propagation algorithm to update the network weights. According to the integral separation theory, set a pressure error threshold of 2kN. When the pressure error is greater than the pressure error threshold, stop the integration and set the control parameter to zero. Otherwise, restore the integration and use the corrected control parameter to determine whether the servo valve is saturated. If saturated, limit the integrator output to [-100%, 100%]. Otherwise, output the integral value normally. Combine the integral separator and the integral limiter to obtain an anti-saturation control law, which is parsed into a control voltage command to drive the servo valve to move and achieve precise control of the press.

[0105] In this embodiment, through the self-learning ability of the neural network, the controller parameters can be adaptively optimized, overcoming the shortcomings of the traditional control method of difficult parameter setting and poor adaptability, improving the robustness and intelligence level of the control system, and the adaptive control strategy has better dynamic performance and steady-state accuracy, especially when the press parameters change and the load disturbance is large, it can maintain a good control effect and improve the stability and reliability of the system. The anti-saturation strategy overcomes the integral saturation problem in traditional control, improves the control quality and safety of the system, simplifies the design and debugging process of the press control system, and improves the efficiency and feasibility of engineering applications. In summary, this embodiment can significantly improve the performance and intelligence level of the press control system, and provide effective technical support for the intelligent manufacturing and industrial application of presses.

[0106] In an optional embodiment,

[0107] Based on the saturation function, combined with the integral limiter and the integral separator, the anti-saturation control law is generated as shown in the following formula:

[0108] ;

[0109] in, u ( t ) represents the output signal value of the anti-saturation control law, K p represents the proportional gain, e ( t ) indicates the current time t The error, sat () represents the saturation function, K i represents the integral gain, e s ( i ) indicates the time i The error signal at time u s represents the saturation value of the integral limiter, K d represents the differential gain, Indicates the error change rate.

[0110] In this embodiment, the integral separator decides whether to use integral control through conditional judgment, avoiding the integral saturation problem when the actuator is continuously saturated, combining the advantages of classical control and modern control, overcoming the shortcomings of traditional control methods, improving the robustness and adaptability of the control system, ensuring that the controller output is within the feasible range of the actuator, avoiding the inconsistency between the controller output and the actual action of the actuator, and improving the safety and reliability of the control system. The integral separation mechanism effectively suppresses the integral saturation problem, so that the controller stops accumulating errors when the actuator is saturated, avoiding the infinite growth of the integral term, thereby improving the dynamic performance and stability of the system. In summary, this embodiment limits the oversaturation of the actuator through the saturation function, ensures the safety of the system, overcomes the shortcomings of traditional control methods, and has broad application prospects in engineering practice.

[0111] Figure 2 FIG. 1 is a schematic diagram of the structure of a pressure test control system of a universal pressure testing machine according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0112] The first unit is used to obtain the hydraulic cylinder parameters and servo valve parameters of the universal pressure testing machine, establish a pressure control mathematical model corresponding to the pressure control system in the universal pressure testing machine according to the hydraulic cylinder flow balance equation and the force balance equation, calculate the transfer function of the dynamic relationship between the voltage input and the load pressure output based on the pressure control mathematical model, stack three layers of back propagation neural network as an initial feedback controller based on the pressure control mathematical model, determine the number of hidden layer nodes based on the input layer nodes and output layer nodes of the initial feedback controller, set the activation function and network weight, and obtain an adaptive feedback controller;

[0113] The second unit is used to integrate the pressure control mathematical model and the adaptive feedback controller to generate a closed-loop control system, construct an objective function based on the integral absolute error and the adjustment time, use the overshoot as a constraint, and determine the Pareto optimal solution set corresponding to the feedback controller parameters in combination with a multi-objective genetic algorithm, select neural network training samples from the Pareto optimal solution set and generate a training sample set covering different pressure instruction types based on common working conditions of the press, use the training sample set to perform offline pre-training on the back-propagation neural network, and update the network weights of each layer of the back-propagation neural network through the LM algorithm to obtain a pre-trained neural network;

[0114] The third unit is used to embed the pre-trained neural network into the pressure control system, adjust the network weights and output feedback controller parameters according to the error signal between the pressure feedback and the command in combination with the gradient descent algorithm, set the feedback controller anti-saturation strategy based on the integral separator and the integral limiter and determine whether the actuator is saturated under the current control parameters, limit the integral term amplitude according to the relationship between the error signal and the saturation direction and determine whether to stop the integration, and generate the press control strategy under the current working conditions.

[0115] According to a third aspect of the embodiments of the present invention,

[0116] An electronic device is provided, comprising:

[0117] processor;

[0118] a memory for storing processor-executable instructions;

[0119] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0120] According to a fourth aspect of the embodiments of the present invention,

[0121] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0122] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0123] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A pressure test control method for a universal pressure testing machine, characterized in that: include: Acquire the hydraulic cylinder parameters and servo valve parameters of the universal pressure testing machine, establish a pressure control mathematical model corresponding to the pressure control system in the universal pressure testing machine according to the hydraulic cylinder flow balance equation and the force balance equation, calculate the transfer function of the dynamic relationship between the voltage input and the load pressure output based on the pressure control mathematical model, stack three layers of back propagation neural network as an initial feedback controller based on the pressure control mathematical model, determine the number of hidden layer nodes based on the input layer nodes and output layer nodes of the initial feedback controller, set the activation function and network weight, and obtain an adaptive feedback controller; Integrate the pressure control mathematical model and the adaptive feedback controller to generate a closed-loop control system, construct an objective function based on the integral absolute error and the adjustment time, use the overshoot as a constraint, and determine the Pareto optimal solution set corresponding to the feedback controller parameters in combination with a multi-objective genetic algorithm, select neural network training samples from the Pareto optimal solution set, and generate a training sample set covering different pressure instruction types based on common working conditions of the press, use the training sample set to perform offline pre-training on the back-propagation neural network, and update the network weights of each layer of the back-propagation neural network through the LM algorithm to obtain a pre-trained neural network; The pre-trained neural network is embedded into the pressure control system, and according to the error signal between the pressure feedback and the command, the network weight is adjusted in combination with the gradient descent algorithm and the feedback controller parameters are output, the anti-saturation strategy of the feedback controller is set based on the integral separator and the integral limiter, and it is determined whether the actuator is saturated under the current control parameters, and the integral term amplitude is limited according to the relationship between the error signal and the saturation direction and it is determined whether to stop the integration, so as to generate the press control strategy under the current working condition; The pressure control mathematical model and the adaptive feedback controller are integrated to generate a closed-loop control system, an objective function is constructed based on the integral absolute error and the adjustment time, the overshoot is used as a constraint, and the Pareto optimal solution set corresponding to the feedback controller parameters is determined in combination with a multi-objective genetic algorithm. A neural network training sample is selected from the Pareto optimal solution set and a training sample set covering different pressure instruction types is generated based on common working conditions of the press. The back propagation neural network is pre-trained offline using the training sample set, and the network weights of each layer of the back propagation neural network are updated by the LM algorithm to obtain a pre-trained neural network including: Integrate the pressure control mathematical model and the adaptive feedback controller to generate the closed-loop control system, take the pressure command as input, and output the control voltage through the adaptive feedback controller to act on the servo valve to complete the closed-loop control; Taking the integral absolute error and the adjustment time as the performance evaluation indicators, constructing the objective function and taking the overshoot as the constraint condition, taking the feedback controller parameters as the optimization variables, wherein the feedback controller parameters include the number of hidden layer nodes and the weights of the output neurons; Based on the current feedback controller parameters, an initial population is randomly generated by a multi-objective genetic algorithm, wherein each individual in the initial population represents a set of feedback controller parameters, the environmental pheromone matrix and the initial value of the pheromone concentration at each position are initialized, for each individual, the objective function value corresponding to the current individual is calculated and the individual optimal position and the global optimal position corresponding to each individual are initialized, the external archive is initialized, for each individual, the speed and position corresponding to the current individual are updated based on the environmental pheromone matrix, for the individual after the position is updated, the objective function value is calculated and the environmental pheromone corresponding to the current individual is updated, and the individual optimal position is updated. If the updated If the position dominates the position before the update, the updated position is added to the external archive and the position information before the update is deleted. The update is repeated until the position of each individual is updated. For the solutions in the updated external archive, non-dominated sorting is performed and non-dominated levels are assigned. Based on the non-dominated levels, the solutions in the external archive are sorted in descending order and the top 10% of the solutions are selected as candidate individuals. The candidate individuals are crossover and mutation operations are performed to generate second-generation individuals. The second-generation population is obtained by combination. The iteration is repeated until the preset maximum number of iterations is reached. The candidate individuals selected in the last iteration are used as Pareto optimal solutions, and the Pareto optimal solution set is obtained by combination. Select solutions from the Pareto optimal solution set as neural network training samples, select from the neural network training samples in combination with the pressure instruction types corresponding to common working conditions of the press, and obtain the training sample set; perform offline pre-training on the back propagation neural network based on the training sample set; and adjust the search direction by adding a damping factor and update the network weights of each layer of the back propagation neural network in combination with the LM algorithm; and repeat the update until the back propagation neural network converges or reaches a preset maximum number of iterations, and obtain the pre-trained neural network; The environmental pheromone is updated as shown in the following formula: ； in, τ j ( t +1) indicates the position j At time step t +1 environmental pheromone value, ρ represents the volatilization rate of environmental pheromones, τ j ( t ) indicates position j At time step t The environmental pheromone value of A ( t ) indicates that at the time step t The total number of individuals at i represents the individual index, Q represents the pheromone intensity factor, rank i ( t ) represents an individual i At time step t Ranking, α represents the path selection adjustment parameter, cd i ( t ) represents an individual i The crowding degree at time step, ε Represents a very small decimal number, I Represents the indicator function, which is used to indicate whether the current individual has reached the position j , if it has been reached, the value is 1, otherwise the value is 0.

2. The method according to claim 1, characterized in that The hydraulic cylinder parameters and servo valve parameters of the universal pressure testing machine are obtained, and a pressure control mathematical model corresponding to the pressure control system in the universal pressure testing machine is established according to the hydraulic cylinder flow balance equation and the force balance equation. The transfer function of the dynamic relationship between the voltage input and the load pressure output is calculated based on the pressure control mathematical model. Based on the pressure control mathematical model, a three-layer back propagation neural network is stacked as an initial feedback controller. The number of hidden layer nodes is determined based on the input layer nodes and the output layer nodes of the initial feedback controller, and the activation function and the network weight are set. The adaptive feedback controller includes: For the hydraulic cylinder of the universal pressure testing machine, the corresponding hydraulic cylinder parameters and servo valve parameters are obtained by querying the equipment manual and actual measurement, wherein the hydraulic cylinder parameters include piston area, dead space volume, equivalent mass and damping coefficient, and the servo valve parameters include flow gain coefficient, bulk modulus of hydraulic oil and leakage coefficient in the hydraulic cylinder; A flow continuity equation corresponding to the hydraulic cylinder cavity is constructed based on the hydraulic cylinder parameters and the servo valve parameters, a force balance equation and a corresponding dynamic equation are constructed based on the force balance relationship in the hydraulic cylinder, the flow continuity equation, the force balance equation and the dynamic equation are integrated, and the integrated equations are Laplace transformed to obtain a pressure control mathematical model corresponding to the pressure control system in the universal pressure testing machine, and the displacement term after Laplace transformation is eliminated based on the pressure control mathematical model to obtain a transfer function of the dynamic relationship between voltage input and load pressure output; A three-layer back propagation neural network with both input layer nodes and output layer nodes being 1 is selected as the initial feedback controller, the number of hidden node layers is determined according to an empirical formula, the connection weights are randomly initialized through an S-type activation function, a performance index function is defined, and the weights of the hidden layer and output neurons are corrected based on the load pressure error signal through an error back propagation algorithm combined with the negative gradient descent principle to minimize the performance index function, the error is continuously calculated and the weights of the output neurons are repeatedly corrected to obtain the adaptive feedback controller.

3. The method according to claim 2, characterized in that The transfer function is shown in the following formula: ； in, G ( s ) is the transfer function value, which is used to indicate the response relationship between the output data and the input data. s represents the complex frequency variable in the Laplace transform, k q represents the flow gain coefficient, A p represents the piston area, V t Represents the total volume of the hydraulic cylinder, β e represents the effective bulk modulus of the liquid, K c represents the flow-pressure coefficient, B p represents the viscous damping coefficient, C t Indicates the total leakage coefficient.

4. The method according to claim 1, characterized in that The pre-trained neural network is embedded into the pressure control system. According to the error signal between the pressure feedback and the command, the network weight is adjusted in combination with the gradient descent algorithm and the feedback controller parameters are output. The anti-saturation strategy of the feedback controller is set based on the integral separator and the integral limiter and it is determined whether the actuator is saturated under the current control parameters. According to the relationship between the error signal and the saturation direction, the integral term amplitude is limited and it is determined whether to stop the integration. The press control strategy under the current working condition is generated, including: The pre-trained neural network acquired in advance is embedded into the pressure control system of the universal pressure testing machine, and the network weights are adaptively updated through a gradient descent algorithm according to the pressure feedback of the pressure control system and the error signal of the control command, and the feedback controller parameters are outputted at the same time; The anti-saturation strategy of the feedback controller is set based on integral separation and integral limiting. The integral separator and integral limiter are set. According to the current feedback controller parameters, it is determined whether to use integral control according to the integral separation theory. If integral control is used, the integral limiting theory is used to determine whether the actuator is saturated under the current control parameters. If saturated, it is determined whether the error signal and the saturation direction are consistent. If they are consistent, the integral separator input is 0 and the integration is stopped. If they are inconsistent, the error signal is output and integrated. If the error signal is output and integrated, the saturation function is set by the integral limiter in combination with the value range of the pressure error. Based on the saturation function, combined with the integral limiter and the integral separator, an anti-saturation control law is generated and the feedback controller parameters are adjusted based on the anti-saturation control law, optimized control parameters are generated and parsed into control instructions, and a press control strategy under the current working conditions is generated.

5. The method according to claim 4, characterized in that Based on the saturation function, combined with the integral limiter and the integral separator, the anti-saturation control law is generated as shown in the following formula: ； in, u ( t ) represents the output signal value of the anti-saturation control law, K p represents the proportional gain, e ( t ) indicates the current time t The error, sat () represents the saturation function, K i represents the integral gain, e s ( θ ) indicates the time θ The error signal at time u s represents the saturation value of the integral limiter, K d represents the differential gain and represents the rate of change of error.

6. A pressure test control system for a universal pressure testing machine, used to implement the method described in any one of claims 1 to 5, characterized in that: include: The first unit is used to obtain the hydraulic cylinder parameters and servo valve parameters of the universal pressure testing machine, establish a pressure control mathematical model corresponding to the pressure control system in the universal pressure testing machine according to the hydraulic cylinder flow balance equation and the force balance equation, calculate the transfer function of the dynamic relationship between the voltage input and the load pressure output based on the pressure control mathematical model, stack three layers of back propagation neural network as an initial feedback controller based on the pressure control mathematical model, determine the number of hidden layer nodes based on the input layer nodes and output layer nodes of the initial feedback controller, set the activation function and network weight, and obtain an adaptive feedback controller; The second unit is used to integrate the pressure control mathematical model and the adaptive feedback controller to generate a closed-loop control system, construct an objective function based on the integral absolute error and the adjustment time, use the overshoot as a constraint, and determine the Pareto optimal solution set corresponding to the feedback controller parameters in combination with a multi-objective genetic algorithm, select neural network training samples from the Pareto optimal solution set and generate a training sample set covering different pressure instruction types based on common working conditions of the press, use the training sample set to perform offline pre-training on the back-propagation neural network, and update the network weights of each layer of the back-propagation neural network through the LM algorithm to obtain a pre-trained neural network; The third unit is used to embed the pre-trained neural network into the pressure control system, adjust the network weights and output feedback controller parameters according to the error signal between the pressure feedback and the command in combination with the gradient descent algorithm, set the feedback controller anti-saturation strategy based on the integral separator and the integral limiter and determine whether the actuator is saturated under the current control parameters, limit the integral term amplitude according to the relationship between the error signal and the saturation direction and determine whether to stop the integration, and generate the press control strategy under the current working conditions.

7. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 5.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images