A digital simulation optimization method for multi-working-condition direction control strategy of air cylinder

Abstract

The application relates to the field of pneumatic control and digital simulation technology, and discloses a digital simulation optimization method for a cylinder multi-working-condition direction control strategy, which comprises the following steps: building a working-condition transition directed graph and configuring local control parameters from multi-working-condition continuous sequence parameters; obtaining a steady-state node ideal boundary constraint vector according to an equation group; separating a preposed constraint and a residual feedforward to obtain an actual injection initial value vector and a virtual truncated residual vector; integrating the initial value to obtain a transient operation curve and a final state and updating the residual feedforward; converting the truncated residual into a penalty energy, combining the equivalent fluid mechanical impact dissipation energy to synthesize total hysteresis dissipation energy measurement to update the control parameters, and over-limit reconstructing the directed graph, which is sent to a bottom programmable logic controller after convergence. The application guarantees integral stability through residual separation and energy conversion, realizes multi-working-condition continuous transition optimization by combining physical dissipation limitation and topological dynamic reconstruction, and avoids calculation divergence and solutionless dead cycle.

Description

Technical Field

[0001] This invention relates to the fields of pneumatic control and digital simulation technology, specifically to a digital simulation optimization method for multi-condition directional control strategies of cylinders. Background Technology

[0002] In the field of pneumatic system control, digital simulation and control parameter optimization of cylinder multi-condition operation are ways to improve system smoothness and control accuracy. Traditional cylinder control strategy simulation methods simplify the complex continuous transient process into a single or discrete steady-state condition for independent calculation. Due to the strong nonlinear gas flow characteristics of pneumatic systems and the transient mechanical shocks that accompany the continuous alternation of multiple conditions, this simplified simulation calculation method severs the intrinsic relationship between multiple conditions. When simulating continuous transient conditions, the simulation model will produce a sudden change in state value at the transient boundary of condition switching. Conventional methods take the operation of directly truncating or ignoring residuals when dealing with the truncation error caused by nonlinear divergence or numerical overshoot, which leads to integral divergence in numerical simulation calculation.

[0003] Furthermore, when dealing with complex constraint spaces containing strong hydromechanical impacts, existing parameter optimization algorithms often fall into local optima due to a lack of accurate quantitative measurement of transient dissipated energy. The output control parameters may exceed the physical limits of the cylinder material's mechanical strength and structural design, resulting in unrealistic control commands. When the control parameter space cannot converge to a safe region under a given hardware structure, the traditional simulation optimization process terminates directly or falls into an infinite loop. It lacks the ability to adaptively adjust and reconstruct the system's operating logic structure and cannot stably output optimized parameters that meet the high-precision continuous control requirements of the underlying programmable logic controller. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a digital simulation optimization method for cylinder multi-condition directional control strategies. This method solves the problems of existing simulation optimization methods, such as computational divergence caused by sudden changes in boundary states during continuous transitions between multiple conditions, parameter optimization getting stuck in local optima and outputting physically infeasible instructions, and computational dead loops caused by the lack of topology reconstruction capability in unsolvable constraint spaces.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a digital simulation optimization method for multi-condition directional control strategies of cylinders, comprising: A directed graph of the working condition transition containing steady-state nodes and directed edges is constructed from the parameters of the continuous operation sequence of multiple working conditions, and local control parameter vectors are assigned to the directed edges; Determine the ideal boundary constraint vector of the steady-state node using the system of equations; The ideal boundary constraint vector of the forward steady-state node and the feedforward vector of the boundary state residual are separated into the actual injected initial value vector and the virtual truncated residual vector. The transient running curve and the final state vector are obtained by integrating the actual injected initial value vector. The final state vector and the ideal boundary constraint vector of the subsequent steady-state node are combined to update the boundary state residual feedforward vector. The virtual truncated residual vector is converted into virtual penalty energy, which is then combined with the equivalent fluid mechanical impact dissipation energy of the transient operating curve to form the total hysteresis dissipation energy measure. The local control parameter vector is updated according to the total hysteresis dissipation energy measure, and the over-limit reconstructed working condition transition directed graph is sent down to the underlying programmable logic controller after convergence.

[0006] This invention is based on the directed graph of working condition transition and the feedforward mechanism of boundary state residuals. It transforms the virtual truncated residual that causes the calculation divergence into penalty energy and adds it to the hysteresis dissipation model. Under the premise of ensuring the integral stability of nonlinear simulation, it realizes dynamic optimization of continuous transition of multiple working conditions and the distribution of low-level parameters, and solves the optimization divergence problem caused by sudden state changes.

[0007] Furthermore, the step of obtaining the ideal boundary constraint vector of the steady-state node according to the system of equations includes: establishing a system of static equilibrium algebraic equations containing mechanical equilibrium equations and thermodynamic state equations as the system of equations; dividing the steady-state node into a charging steady-state node, an exhaust steady-state node, a locking steady-state node, and an end-position steady-state node according to the gas path connection state corresponding to the steady-state node; for the charging steady-state node, constraining the charging chamber pressure to the gas source steady-state supply pressure set value; for the exhaust steady-state node, constraining the exhaust chamber pressure to the external environment back pressure set value; for the locking steady-state node, performing algebraic solution calculation through the mechanical equilibrium equations and the gas chamber closure mass conservation condition; for the end-position steady-state node, introducing a mechanical limit reaction term into the mechanical equilibrium equations and performing algebraic solution calculation.

[0008] This embodiment divides the steady-state nodes according to the internal flow field and mechanical connection state of the cylinder, and sets the limits by mechanical and thermodynamic state equations, discretizing the continuous nonlinear physical process into a static equilibrium solution domain with clear boundary conditions, thus providing a computational boundary basis for transient simulation.

[0009] Furthermore, the step of obtaining the ideal boundary constraint vector of the steady-state node according to the system of equations includes: inputting the macroscopic operating parameters and the state vector corresponding to the previous steady-state node into the steady-state decoupling feature extraction neural network model, and outputting the steady-state prediction vector; when the residual after substituting the steady-state prediction vector into the system of equations is less than or equal to a preset residual threshold, injecting the steady-state prediction vector as a high-confidence initial iteration value into the nonlinear algebra solver; triggering the nonlinear algebra solver to perform multiple approximation root-finding operations on the system of equations starting from the high-confidence initial iteration value, obtaining the analytical solution of the steady-state node at the end of the steady state, and reorganizing the analytical solution of the state into a matrix according to the physical quantity dimension, and outputting the ideal boundary constraint vector of the steady-state node.

[0010] This embodiment utilizes steady-state decoupling features to extract the approximation point of the neural network model output, reducing the invalid root-finding calculations of the nonlinear algebra solver in the multidimensional state space, improving the convergence speed of steady-state ideal boundary constraints, and avoiding numerical divergence caused by improper initial value settings.

[0011] Furthermore, the step of separating the ideal boundary constraint vector of the preceding steady-state node and the feedforward vector of the boundary state residual into the actual injected initial value vector and the virtual truncated residual vector includes: algebraically superimposing the ideal boundary constraint vector of the preceding steady-state node and the feedforward vector of the boundary state residual to obtain the theoretical initial value vector, inputting it into the transient residual nonlinear mapping neural network model, and outputting the smooth mapping steepness gain coefficient matrix. The energy-preserving nonlinear boundary mapping algorithm is invoked, and the theoretical initial value vector is explicitly calculated by combining the set physical lower limit vector and physical upper limit vector of the state variables to obtain the actual injected initial value vector. The difference operation is performed between the theoretical initial value vector and the actual injected initial value vector to output the virtual truncated residual vector that is forcibly stripped outside the physical feasible domain.

[0012] This embodiment uses an energy-preserving nonlinear boundary mapping algorithm to process theoretical initial values, limiting the initial state injected into the transient solver to be within the physically feasible domain, and separately saving the stripped truncated residuals as energy dimension constraints to prevent non-physical boundary overflows from causing numerical simulation divergence.

[0013] Furthermore, the step of updating the boundary state residual feedforward vector by simultaneously combining the final state vector and the ideal boundary constraint vector of the subsequent steady-state node includes: constructing a state transition matrix in the form of a diagonal Boolean mapping matrix based on the degree-of-freedom constraint conditions of the subsequent steady-state node; when the flow field or mechanical dimension corresponding to the subsequent steady-state node is in an unconstrained state with released degrees of freedom, setting the diagonal elements of the state transition matrix to zero; When in a closed and locked state, the diagonal elements of the state transition matrix are set to one; the boundary state residual vector obtained by subtracting the final state vector from the ideal boundary constraint vector of the subsequent steady-state node using the state transition matrix is ​​subjected to matrix multiplication mapping operation to update the boundary state residual feedforward vector.

[0014] This embodiment constructs a diagonal Boolean state transition matrix based on the system's degrees of freedom, adaptively shielding unconstrained dimension residual interference, and maintaining the accuracy of boundary feedforward compensation for multi-degree-of-freedom systems under state locking and releasing transition conditions.

[0015] Furthermore, the step of obtaining the equivalent hydromechanical impact dissipation energy of the transient operating curve includes: extracting the product of the time derivative data of the two-chamber pressure difference and the instantaneous piston velocity data contained in the transient operating curve in the time domain, and performing an absolute value integration operation or a square integration operation on the product; multiplying the integration result with a preset dimension-aligned equivalent coefficient to obtain the equivalent hydromechanical impact dissipation energy of the transient operating curve converted into a work quantity dimension.

[0016] This embodiment combines the time derivative of the pressure difference between the two chambers with the instantaneous velocity of the piston to construct an equivalent energy measurement dimension that reflects the coupling effect of fluid impact and mechanical motion, providing a quantitative indicator for evaluating the smoothness of the control strategy in the transient transition region.

[0017] Furthermore, after obtaining the equivalent hydromechanical impact dissipation energy of the transient operating curve converted into a functional quantity dimension, the process includes: when the calculated equivalent hydromechanical impact dissipation energy of the transient operating curve exceeds a physical dissipation upper limit threshold set based on the yield strength of the cylinder material, the effective cross-sectional area of ​​the piston, and the maximum buffer stroke of the cylinder, a limiting and blocking operation is performed; the physical dissipation upper limit threshold is used as the numerical integration stability output value, and an impact over-limit flag is generated; the impact over-limit flag is used to trigger an over-limit penalty term superposition mechanism in the subsequent optimization process to force the search direction to deviate from the over-limit condition.

[0018] This embodiment combines cylinder material mechanics and structural parameters to set an upper limit for dissipation, introduces hard physical constraints during the optimization process, and uses an impact over-limit flag to guide the optimization algorithm away from extreme damage conditions.

[0019] Furthermore, the step of updating the local control parameter vector based on the total hysteresis dissipation energy measure includes: using the total hysteresis dissipation energy measure and the working condition importance weights corresponding to the directed edges within the working condition transition directed graph to perform a linear weighted accumulation mechanism to construct a global evaluation objective function; and using a hybrid optimization algorithm combining adaptive particle swarm optimization and sequential quadratic programming to concatenate all the local control parameter vectors in the working condition transition directed graph into a global control parameter vector and then performing iterative optimization calculation. After the adaptive particle swarm optimization algorithm satisfies the initial convergence condition, the sequential quadratic programming algorithm is started with the output current global optimal position vector as the initial point of iteration to perform local exact optimization. The quadratic programming subproblem is solved by the Jacobian sensitivity matrix to obtain the control parameter correction direction and correction step size of the current iteration step, and the local control parameter vector is updated in combination with boundary constraint processing.

[0020] This embodiment utilizes a globally weighted objective function combined with adaptive particle swarm optimization and sequential quadratic programming algorithms. It first performs global space exploration, and then uses the Jacobian sensitivity matrix to perform local precise gradient convergence, thereby improving the optimization accuracy of the control parameter set.

[0021] Furthermore, the steps of the over-limit reconstruction working condition transition directed graph include: identifying the target directed edge that causes the over-limit, virtually inserting transient buffer steady-state nodes representing the proposed addition of mechanical buffer damper action area or secondary exhaust throttling area inside the target directed edge; performing a graph theory partitioning operation on the original boundary path to divide the target directed edge into two and reorganize it into a new directed edge sequence with the head and tail connected; assigning independent local control parameter vectors to the new directed edge sequence and outputting a topology dynamic reconstruction command to trigger the re-execution of the secondary loop solution process using the updated working condition transition directed graph.

[0022] In this embodiment, when the control parameter space cannot converge to a safe region, the directed graph topology is split and reorganized. By virtually adding damping or throttling regions in the simulation logic, the dimension of adaptive adjustment from pure parameter optimization to system structure logic is expanded.

[0023] Furthermore, after triggering the step of re-executing the secondary loop solution process using the updated working condition transition directed graph, the process includes: if the total hysteresis dissipation energy measure of the reconstructed system after re-executing the secondary loop solution process is less than or equal to a preset safe hysteresis dissipation threshold, the parameter update deployment is completed; if the number of topology reconstructions reaches the preset maximum hardware reconstruction limit and the total hysteresis dissipation energy measure of the reconstructed system is still greater than the preset safe hysteresis dissipation threshold, the optimization is terminated and a system physical design infeasibility alarm is output.

[0024] This embodiment sets up a recombination convergence determination and exit alarm mechanism to prevent the algorithm from getting stuck in an infinite loop within the unsolvable physical design boundary, and to ensure the reliability of the digital simulation optimization process.

[0025] This invention provides a digital simulation optimization method for multi-condition directional control strategies of cylinders. It has the following beneficial effects: 1. This invention separates the ideal boundary constraint vector of the pre-steady-state node and the feedforward vector of the boundary state residual into the actual injected initial value vector and the virtual truncated residual vector, and converts the virtual truncated residual vector into virtual penalty energy and incorporates it into the hysteresis dissipation model. This avoids nonlinear integral divergence caused by sudden changes in boundary state during continuous transitions under multiple operating conditions. The above residual separation and energy conversion mechanism ensures that the initial state of the transient solver is always within the physical feasible region, thereby improving the numerical simulation stability of the aerodynamic control parameter optimization process.

[0026] 2. This invention extracts the product of the time derivative of the pressure difference between the two chambers and the instantaneous velocity of the piston in the time domain and performs an integral operation to convert it into equivalent fluid mechanical impact dissipation energy. At the same time, it performs a limiting and blocking operation in combination with a set physical dissipation upper limit threshold. The above-mentioned energy extraction and limiting characteristics provide an objective quantitative basis for the smoothness of the control strategy transition. Moreover, it can guide the optimization search direction to deviate and cause equipment structural damage by triggering an over-limit penalty term, thus ensuring the physical security of the execution parameters of the underlying programmable logic controller.

[0027] 3. This invention, when identifying a target directed edge that causes an overlimit, virtually inserts a transient buffer steady-state node inside the target directed edge, performs a graph theory partitioning operation on the original boundary path to reconstruct a new directed edge sequence with the head and tail connected, thereby triggering topological dynamic reconstruction. The above topological reconstruction steps enable the simulation system to simulate the addition of a mechanical buffer damper or a secondary exhaust throttling zone when the existing control parameter space cannot converge, expanding the optimization scope from the pure parameter dimension to the system operation logic structure dimension, and preventing the optimization calculation from entering an infinite loop within the unsolvable physical design boundary. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the system architecture of the present invention; Figure 2 This is a schematic diagram of the method flow of the present invention; Figure 3 This is a schematic diagram illustrating the topology conversion principle of cylinder operating conditions according to the present invention; Figure 4 This is a schematic diagram of the steady-state decoupling calculation process of the present invention; Figure 5 This is a schematic diagram of the boundary mapping and transient integral feedforward principle of the present invention; Figure 6 This is a schematic diagram illustrating the quantification principle of total hysteresis dissipation energy measurement in this invention. Figure 7 This is a schematic diagram of the global iterative optimization process for the control parameters of the present invention; Figure 8 This is a schematic diagram comparing the transient pressure impact states during commutation of the pneumatic control loop of the present invention.

[0029] Among them, 10 is the topology modeling module; 20 is the steady-state decoupling module; 30 is the integral feedforward module; 40 is the dissipative quantization module; and 50 is the optimization and reconstruction module. Detailed Implementation

[0030] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] See attached document Figure 1 This invention provides a digital simulation optimization system for cylinder multi-condition directional control strategy. The system includes: a topology modeling module 10, a steady-state decoupling module 20, an integral feedforward module 30, a dissipative quantization module 40, and an optimization and reconstruction module 50.

[0032] The topology modeling module 10 receives the multi-condition continuous operation sequence parameters of the cylinder physical system. The topology modeling module 10 is configured to generate a directed graph of condition transitions containing steady-state nodes and directed edges. The topology modeling module 10 outputs a directed graph of condition transitions with local control parameter vectors.

[0033] The steady-state decoupling module 20 receives steady-state node data and macroscopic operating parameters contained in the directed graph of the operating condition transition. The steady-state decoupling module 20 is configured with a set of static equilibrium algebraic equations containing mechanical equilibrium equations and thermodynamic state equations. The steady-state decoupling module 20 combines the set of static equilibrium algebraic equations to output the ideal boundary constraint vector of the steady-state node.

[0034] The integral feedforward module 30 obtains the ideal boundary constraint vector and the boundary state residual feedforward vector of the steady-state node. The integral feedforward module 30 has a built-in physical feasible region and smooth mapping function for the system state vector. The integral feedforward module 30 outputs the actual injected initial value vector and the virtual truncated residual vector. The integral feedforward module 30 combines the cylinder differential equation and the actual injected initial value vector to output the transient running curve corresponding to the directed edge. The integral feedforward module 30 combines the built-in state transition matrix to output the mapped boundary state residual vector.

[0035] The dissipative quantization module 40 receives the time derivative data of the two-chamber pressure difference and the instantaneous piston velocity data contained in the transient operating curve. The dissipative quantization module 40 obtains the virtual truncated residual vector output by the integral feedforward module 30. The dissipative quantization module 40 is internally configured with a state energy stiffness matrix. The dissipative quantization module 40 combines the state energy stiffness matrix to output the total hysteresis dissipative energy measure corresponding to the local control parameter vector.

[0036] The optimization and reconstruction module 50 receives the total hysteresis dissipation energy measure. The optimization and reconstruction module 50 is internally configured with a Jacobi sensitivity matrix construction unit, a constraint function construction unit, and a sequential quadratic programming algorithm solution unit. The optimization and reconstruction module 50 outputs the updated local control parameter vector based on the total hysteresis dissipation energy measure, the operating condition importance weight, and the physical upper and lower limits of the local control parameters.

[0037] See attached document Figure 2 This invention provides a digital simulation optimization method for multi-condition directional control strategies of cylinders, the method comprising the following steps: S100, the topology modeling module 10 receives the multi-condition continuous operation sequence parameters of the cylinder physical system. The topology modeling module 10 converts the multi-condition continuous operation sequence parameters into a condition transition directed graph. The topology modeling module 10 extracts the steady-state nodes representing the steady-state execution stage of a single condition and the directed edges representing the transient reversal process between conditions inside the condition transition directed graph. The topology modeling module 10 assigns local control parameter vectors at the positions of the directed edges.

[0038] S200 receives steady-state node data and macroscopic operating parameters contained in the directed graph of the working condition transition through the steady-state decoupling module 20. The steady-state decoupling module 20 substitutes the steady-state node data and macroscopic operating parameters into the static equilibrium algebraic equation system to perform algebraic solution calculation. The steady-state decoupling module 20 obtains the analytical solution of the steady-state node at the end of the steady state and outputs the ideal boundary constraint vector of the steady-state node.

[0039] S300, through the integral feedforward module 30, the ideal boundary constraint vector of the steady-state node and the boundary state residual feedforward vector transmitted by the previous working condition path are algebraically superimposed to obtain the theoretical initial value vector. The integral feedforward module 30 uses the smoothing mapping function to perform nonlinear boundary mapping processing on the theoretical initial value vector. The integral feedforward module 30 separates the actual injected initial value vector inside the physical feasible region of the system state vector and the virtual truncated residual vector outside the physical feasible region of the system state vector. Subsequently, the integral feedforward module 30 uses the actual injected initial value vector as the initial value for calculation, and performs time-domain integration on the local control parameter vector and the cylinder differential equation. The integral feedforward module 30 obtains the transient running curve and final state vector corresponding to the directed edge. The integral feedforward module 30 performs a subtraction operation between the final state vector and the ideal boundary constraint vector of the target node to obtain the boundary state residual vector. The integral feedforward module 30 uses the state transition matrix to perform matrix multiplication mapping operation on the boundary state residual vector. The integral feedforward module 30 outputs the mapped boundary state residual vector.

[0040] S400, the dissipation quantization module 40 uses a preset state energy stiffness matrix to perform quadratic algebraic calculations on the virtual truncated residual vector to extract virtual penalty energy. The dissipation quantization module 40 performs absolute value integration on the product of the two-chamber pressure difference time derivative data and the piston instantaneous velocity data contained in the transient operating curve to obtain the equivalent fluid mechanical impact dissipation energy. The dissipation quantization module 40 performs algebraic summation calculations on the fluid mechanical dissipation extrema and the virtual penalty energy. The dissipation quantization module 40 outputs the total hysteresis dissipation energy measure corresponding to the local control parameter vector.

[0041] S500, through the optimization and reconstruction module 50, uses the total hysteresis dissipation energy measure and the operating condition importance weight to perform linear weighted summation to obtain the global optimization objective function. The optimization and reconstruction module 50 uses the Jacobi sensitivity matrix and the sequential quadratic programming algorithm to perform iterative calculations. The optimization and reconstruction module 50 updates the local control parameter vector corresponding to the inside of the directed edge. Meanwhile, the optimization and reconstruction module 50 extracts the total hysteresis dissipation energy measure and compares it with the preset safe hysteresis dissipation threshold. When the total hysteresis dissipation energy measure is less than or equal to the preset safe hysteresis dissipation threshold, the optimization and reconstruction module 50 outputs the local control parameter vector as the optimal direction control strategy parameter. When the total hysteresis dissipation energy measure is greater than the preset safe hysteresis dissipation threshold, the optimization and reconstruction module 50 generates transient buffer nodes and inserts transient buffer nodes inside the original boundary path of the working condition transition directed graph. The optimization and reconstruction module 50 performs graph theory partitioning and model reorganization operations on the original boundary path. The optimization and reconstruction module 50 outputs topology dynamic reconstruction instructions and feeds them back to the topology modeling module 10 to trigger the system to execute the secondary loop solution process.

[0042] See attached document Figure 3 The process of receiving running sequence parameters and constructing a directed graph for the transition of operating conditions is executed by the topology modeling module 10. For a specific implementation of S100, this can be handled through the following sub-steps: S110, the topology modeling module 10 receives multi-condition continuous operation sequence parameters of the target pneumatic control loop. The multi-condition continuous operation sequence parameters include the predetermined action timing instructions of the cylinder actuator during continuous operation and the matching load boundary conditions. The predetermined action timing instructions include at least the target steady-state node number, target piston displacement, action direction, steady-state holding time, transient reversal duration, and valve action trigger time. The load boundary conditions include at least the external load force, load mass, friction parameters, air supply pressure, exhaust back pressure, and cylinder installation attitude parameters. The parsing program configured inside the topology modeling module 10 converts the hardware geometric parameters including cylinder diameter and piston stroke and the air source pressure data including steady-state air supply pressure input from the target pneumatic control loop into digital format and stores them in the built-in register of the topology modeling module 10.

[0043] S120, the topology modeling module 10 converts the parameters of the continuous multi-condition operation sequence into a directed graph of condition transition. In order to divide the complex continuous aerodynamic working cycle into discrete stages that are easy to simulate in segments, the topology modeling module 10 abstracts the discrete event time sequence characteristics of the target aerodynamic control loop into a mathematical graph theory set form containing nodes and connection paths. The specific algebraic representation formula of the directed graph of condition transition is as follows: ; In the formula, For the generated working condition transition directed graph; It is the mathematical set of all steady-state nodes contained within the directed graph of the operating condition transition; It is a mathematical set of directed edges that form the connection relationships between the steady-state nodes within a directed graph of the working condition transition.

[0044] S130, the topology modeling module 10 extracts the steady-state nodes representing the steady-state execution stage of a single working condition and the directed edges representing the transient reversal process between working conditions within the directed graph of working condition transition. The steady-state nodes correspond to the static execution stage where the cylinder piston displacement remains constant and the fluid pressure in the two chambers reaches equilibrium. The directed edges correspond to the dynamic transition implementation path of fluid state alternation triggered by the action of the electromagnet of the control valve inside the target pneumatic control loop.

[0045] S140, the topology modeling module 10 assigns local control parameter vectors at the directed edge locations. These vectors are used to aggregate controller adjustment setpoints representing the dynamic transition implementation path within the directed edge, influencing the fluid flow direction. The control strategy element variables within the local control parameter vectors include at least one or more of the following: directional valve core dead zone delay time, initial opening of the proportional valve orifice flow cross-section, gradient of the proportional valve orifice flow cross-section change, exhaust throttling opening, and buffer duration. Each control strategy element variable is configured with corresponding physical units, upper hardware setting thresholds, and lower hardware setting thresholds. The topology modeling module 10 establishes a vector matrix containing multi-dimensional parameter elements based on the actuator attributes corresponding to the directed edge. The specific matrix representation formula for the local control parameter vector is as follows: ; In the formula, This is a vector of local control parameters configured at the position of the directed edge pointing from the preceding steady-state node to the target subsequent steady-state node; These are the control strategy element variables encompassed within the local control parameter vector, representing various dimensions. This is the symbol for the matrix transpose operation; For the graph theory location index of the preceding steady-state node; The graph theory position index of the target post-steady-state node; This is the sum of the parameter characteristic dimensions pre-calibrated based on the number of adjustable valves included in the target pneumatic control loop.

[0046] See attached document Figure 4 The steady-state node data reception and ideal boundary constraint acquisition process is performed by the steady-state decoupling module 20. For a specific implementation of S200, this can be processed through the following sub-steps: S210, the steady-state decoupling module 20 receives the steady-state node data and macroscopic operating parameters contained in the directed graph of the operating condition transition. The macroscopic operating parameters include the external load mass and supply back pressure values ​​of the target aerodynamic control loop. The steady-state decoupling module 20 is configured with a set of static equilibrium algebraic equations containing mechanical equilibrium equations and thermodynamic state equations. Considering that the rate of change of various fluid state variables with time in the long-period steady-state execution phase of the target aerodynamic control loop tends to be zero, if the time-domain integration operation is continuously performed by relying on differential equations, it will generate redundant computational burden and introduce numerical truncation error. The steady-state decoupling module 20 establishes the set of static equilibrium algebraic equations by removing the time derivative terms inside the differential equations. The steady-state decoupling module 20 uses macroscopic operating parameters to construct a set of static equilibrium algebraic equations at the static extreme point of the target aerodynamic control loop, and divides the steady-state nodes into inflation steady-state nodes, exhaust steady-state nodes, lock-up steady-state nodes, and end-position steady-state nodes according to the air path connection state corresponding to the steady-state nodes; for inflation steady-state nodes, the inflation chamber pressure constraint is the steady-state supply pressure setting value of the air source; for exhaust steady-state nodes, the exhaust chamber pressure constraint is the external environmental back pressure setting value; For the lock-up steady-state node, the pressure in both chambers of the cylinder is not forced to equal the steady-state supply pressure of the air source or the back pressure of the external environment. Instead, it is solved jointly by the mechanical equilibrium equations and the mass conservation condition of the closed air chamber. For the end-position steady-state node, a mechanical limiting reaction term is introduced into the mechanical equilibrium equations to avoid the static equilibrium algebraic equations becoming unsolvable when the supply pressure, exhaust pressure, and external load cannot simultaneously satisfy mechanical equilibrium. The specific algebraic expression of mechanical equilibrium within the static equilibrium algebraic equations is as follows: ; In the formula, The absolute static pressure of the fluid inside the rodless chamber of the cylinder corresponds to the steady-state node of the directed graph of the working condition transition. This represents the piston action area inside the rodless chamber of the cylinder. The absolute static pressure of the fluid inside the rod chamber of the cylinder corresponds to the steady-state node of the directed graph of the working condition transition. This represents the piston action area inside the rod chamber of the cylinder. The external constant physical load force acting on the end of the cylinder piston rod; This is to overcome the static friction force generated by the deformation of the seal when the cylinder piston is stationary; This is the transient motion trend direction parameter that the cylinder piston will trigger after the steady-state node ends. It takes a positive value when the motion trend is the same as the positive direction set by the external constant physical load force, and a negative value otherwise.

[0047] The calculation of the specific value of static friction can be obtained by performing algebraic multiplication with the preset Coulomb friction empirical constant and the pre-compression of the piston seal. Meanwhile, the specific algebraic expression of the thermodynamic state within the static equilibrium algebraic equations is as follows: ; ; In the formula, The static pressure of the internal fluid in the cylinder's filling chamber under steady-state conditions; The steady-state supply pressure setpoint of the air source associated with the target pneumatic control loop; This represents the internal fluid static pressure in the cylinder exhaust chamber under steady-state conditions. The set value of the external environmental back pressure connected to the exhaust port of the target pneumatic control circuit.

[0048] S220, the steady-state decoupling module 20 substitutes the steady-state node data and macroscopic operating parameters into the static equilibrium algebraic equation system to perform algebraic solution calculation. The steady-state decoupling module 20 is internally configured with a nonlinear algebraic solver containing Jacobian matrix iteration logic. In order to accelerate the root-finding iteration of the nonlinear algebraic solver and avoid the nonlinear algebraic solver diverging, the steady-state decoupling module 20 is configured with a steady-state decoupling feature extraction neural network model.

[0049] The steady-state decoupling feature extraction neural network model adopts a multi-layer perceptron network structure containing an input layer, multiple hidden layers, and an output layer. The input features received by the input layer include at least the steady-state supply pressure setpoint of the air source, the back pressure setpoint of the external environment, the external load force, the effective area of ​​the rodless chamber of the cylinder, the effective area of ​​the rod chamber of the cylinder, the current position of the piston, the type of the target steady-state node, and the state vector of the previous node. For the first steady-state node in a multi-condition continuous operation sequence, since there is no physical preceding condition, the state vector of the preceding node corresponding to the first steady-state node is initialized by the system by default as a zero vector representing the initial state of the hardware device environment or a preset engineering calibration initial value vector. The value range of the elements inside the preset engineering calibration initial value vector is from 0 to 50% of the upper limit value of their respective physical quantities. The steady-state prediction vector output by the output layer includes at least the steady-state pressure of the rodless cavity, the steady-state pressure of the rod cavity, the steady-state displacement of the piston, and the steady-state velocity of the piston.

[0050] The steady-state decoupling feature extraction neural network model is only used to provide initial iteration values ​​for the nonlinear algebra solver. The final analytical solution of the steady-state node is still determined by the algebraic solution of the static equilibrium algebraic equation system. The internal offline training unit of the steady-state decoupling module 20 obtains a large number of input feature samples by performing Latin hypercube random sampling on the macroscopic operating parameters and the state of the preceding system within the physical operating limit range of the target aerodynamic control loop. The internal offline training unit calls high-precision fluid dynamics simulation software to perform steady-state simulation calculations on the large number of input feature samples to obtain the corresponding steady-state convergence analytical solution as a supervised learning label.

[0051] The internal offline training unit uses the mean squared error loss function and the adaptive moment estimation optimization algorithm to perform backpropagation iterative training on the multilayer perceptron network structure until the mean squared error loss function converges to a preset accuracy, which ranges from 10. -6 Up to 10 -4 After training, the steady-state decoupling feature extraction neural network model receives macroscopic operating parameters, steady-state node types, and previous node state vectors and propagates forward to output steady-state prediction vectors. When the residual after substituting the steady-state prediction vector into the static equilibrium algebraic equation system is greater than the preset residual threshold, the preset residual threshold ranges from 0.01MPa to 0.05MPa.

[0052] The steady-state decoupling module 20 discards the steady-state prediction vector and uses the backup initial iteration value generated based on the analytical solution of the previous steady-state node or the initial value of engineering calibration to perform algebraic solution. When the residual is less than or equal to the preset residual threshold, the steady-state decoupling module 20 injects the steady-state prediction vector as a high-confidence initial iteration value into the nonlinear algebra solver. Subsequently, the steady-state decoupling module 20 triggers the nonlinear algebra solver to perform multiple approximation root-finding operations on the static equilibrium algebraic equation system starting from the determined initial iteration value.

[0053] To avoid the risk of nonlinear algebra solver divergence and collapse caused by excessive condition number during the Jacobian matrix iteration process, the steady-state decoupling module 20 has a built-in iterative state monitoring protocol. When the approximation root-finding operation reaches the preset maximum number of iterations and fails to meet the error convergence tolerance, the preset maximum number of iterations ranges from 50 to 100. The steady-state decoupling module 20 triggers the iterative state monitoring protocol to inject a Newton downhill relaxation factor into the nonlinear algebra solver to forcibly reduce the single-step iteration step size and re-execute the root-finding operation. The steady-state decoupling module 20 obtains the analytical solution of the steady-state node at the end of the steady state.

[0054] S230, the steady-state decoupling module 20 reorganizes the analytical solution of the steady-state node at the end of the steady-state state into a matrix according to the physical quantity dimension. The steady-state decoupling module 20 outputs the ideal boundary constraint vector of the steady-state node. The ideal boundary constraint vector represents the macroscopic static physical state boundary of the steady-state node corresponding to the single working condition in the working condition transition directed graph just before the start of the transient commutation process. The specific matrix expression of the ideal boundary constraint vector is as follows: ; In the formula, The single-node ideal boundary constraint vector output by the steady-state decoupling module 20; The steady-state pressure value of the rodless cavity is extracted from the analytical solution of the steady-state node at the end of the steady-state period. The steady-state pressure value of the rod cavity is extracted from the analytical solution of the steady-state node at the end of the steady-state period; The steady-state static displacement of the piston is extracted from the analytical solution of the steady-state node at the end of the steady-state period. The steady-state velocity values ​​of the piston are extracted from the analytical solution of the steady-state node at the end of the steady-state phase. This is the symbol for the matrix transpose operation.

[0055] The selection rules for the error convergence tolerance values ​​involved in the approximation and root-finding operations of nonlinear algebra solvers can be implemented according to the existing technical specifications for numerical calculation methods.

[0056] See attached document Figure 5The transient integral solution and boundary residual feedforward are performed by the integral feedforward module 30. For a specific implementation of S300, this can be achieved through the following sub-steps: S310, the integral feedforward module 30 performs algebraic superposition calculation on the ideal boundary constraint vector of the steady-state node corresponding to the previous single operating condition and the boundary state residual feedforward vector transmitted by the previous operating condition path to obtain the theoretical initial value vector. Specifically, for the first directed edge in a multi-operating condition continuous operation sequence, the boundary state residual feedforward vector is initialized to zero. For a periodic continuous operation sequence, the boundary state residual feedforward vector can be initialized using the boundary state residual feedforward vector output at the end of the previous operating cycle. Considering the fluid compressibility and mechanical hysteresis effects in the actual continuous operation of the target pneumatic control loop, the target pneumatic control loop often cannot accurately reach the ideal equilibrium point of the previous operating condition when switching to the operating condition corresponding to the subsequent steady-state node. At the microscopic level, transient characteristics such as incompletely released cavity back pressure or incompletely reset piston displacement remain. To restore the cross-node transmission law of the physical hysteresis effect of the previous operating condition in the digital simulation model, the integral feedforward module 30 retains the accumulated residual state through a linear superposition mechanism. The specific algebraic expression for the algebraic superposition calculation is as follows: ; In the formula, The theoretical initial value vector is calculated and obtained for the integral feedforward module 30; This is the ideal boundary constraint vector for the steady-state node corresponding to the single operating condition in the preceding step. It is the boundary state residual feedforward vector transmitted from the end of the preceding working condition path.

[0057] S320, the integral feedforward module 30 uses a smoothing mapping function to perform nonlinear boundary mapping processing on the theoretical initial value vector, separating the actual injected initial value vector inside the physical feasible region of the system state vector and the virtual truncated residual vector outside the physical feasible region of the system state vector. Since the residual superposition operation at the pure mathematical level has the risk of negative pressure or exceeding the upper limit of the gas supply pressure of the pressure state variable, directly injecting the out-of-bounds initial value into the differential equation will cause the numerical integral solver to generate a singular matrix or even diverge and collapse.

[0058] The integral feedforward module 30 incorporates a transient residual nonlinear mapping neural network model. This model consists of a cascaded structure of one-dimensional convolutional layers and fully connected layers. To construct this model, the synchronous offline training unit of the integral feedforward module 30 injects Gaussian random noise into the physical neighborhood of the theoretical initial value vector to generate a massive number of initial value feature samples. The synchronous offline training unit utilizes a physical experimental platform or calls a three-dimensional fluid dynamics high-fidelity simulation software to perform offline flow field calculations, capturing the real physical mapping state at the moment of transient reversal as the real injected initial value label.

[0059] The synchronous offline training unit constructs a physical information loss function that includes mean square error and energy conservation penalty term. The synchronous offline training unit uses the physical information loss function to perform gradient descent training on the cascaded structure of one-dimensional convolutional layer and fully connected layer to obtain a converged transient residual nonlinear mapping neural network model. The integral feedforward module 30 inputs the theoretical initial value vector into the transient residual nonlinear mapping neural network model. The transient residual nonlinear mapping neural network model forward inference outputs the correction amount of the smooth mapping steepness gain coefficient matrix and the physical boundary correction amount. The use of the transient residual nonlinear mapping neural network model makes the initial value truncation process more in line with the nonlinear physical laws of fluid dynamics. The actual injected initial value vector and the virtual truncated residual vector are not directly output by the neural network black box, but are explicitly calculated by a continuously differentiable energy-preserving nonlinear boundary mapping algorithm based on the corrected mapping parameters. The integral feedforward module 30 has a built-in continuously differentiable energy-preserving nonlinear boundary mapping algorithm. The specific algebraic expression of the nonlinear boundary mapping is as follows: ; ; In the formula, This is the actual initial value vector obtained after nonlinear boundary mapping. The physical lower limit vector of state variables set for the target aerodynamic control loop; The physical upper limit vector of state variables set for the target aerodynamic control loop; The smoothing mapping steepness gain coefficient matrix is ​​a pre-defined matrix, and the diagonal elements of the smoothing mapping steepness gain coefficient matrix range from 10 to 100. The theoretical initial value vector is calculated and obtained for the integral feedforward module 30; It is the arithmetic mean vector of the physical upper limit vector and the physical lower limit vector; This is a virtual truncated residual vector that is forcibly stripped outside the physically feasible region.

[0060] S330, the integral feedforward module 30 uses the actual injected initial value vector as the initial value for calculation. It performs time-domain integration on the local control parameter vector and the cylinder differential equation. The integral feedforward module 30 obtains the transient operating curve and final state vector corresponding to the directed edge within the directed graph of the target aerodynamic control loop under the operating condition transition. Since the flow field gradient is relatively large and the control rigidity is significant at the moment the air inlet opens, the integral feedforward module 30 calls a rigid ordinary differential equation solver configured with a backward differential formula algorithm to perform variable-step time-domain integration. Simultaneously, to prevent the numerical solver from entering a dead loop in the rigid region due to continuously decreasing step size, the integral feedforward module 30 is configured with a minimum integration step size monitoring command. When the solution step size falls below a preset lower limit, the integration operation is forcibly terminated, and the system vector corresponding to the current iteration time is output as the final state of integration. The preset lower limit ranges from 10. -6 s to 10 -5 The specific algebraic expression for time-domain integration is as follows: ; In the formula, This is the final state vector of the transient operating curve at the end of the integration time. The duration of the transient commutation process corresponding to the directed edge; The cylinder dynamics differential mechanism function characterizing the rate of change of transient pressure and displacement time derivative within the target pneumatic control loop; The transient state vector of the target aerodynamic control loop; This is a vector of local control parameters configured at the position of the directed edge pointing from the preceding steady-state node to the target subsequent steady-state node; For integration time; The differential step size term for the integration time variable; This is the actual initial value vector obtained after nonlinear boundary mapping.

[0061] S340, the integral feedforward module 30 performs a subtraction operation between the final state vector and the ideal boundary constraint vector of the target post-steady-state node to obtain the boundary state residual vector. The integral feedforward module 30 uses the state transition matrix to perform matrix multiplication mapping operation on the boundary state residual vector. The integral feedforward module 30 outputs the mapped boundary state residual vector for subsequent working condition paths inside the working condition transition directed graph to be iteratively called. Since there are topological changes between different working condition nodes, such as actuator locking or degree of freedom release, directly transmitting the residual will lead to physical quantity dimension mismatch.

[0062] The integral feedforward module 30 constructs a state transition matrix based on the degree-of-freedom constraints of the target downstream steady-state node. The state transition matrix is ​​configured as a diagonal Boolean mapping matrix. When the flow field or mechanical dimension corresponding to the downstream steady-state node is in an unconstrained state with released degrees of freedom, the diagonal elements of the state transition matrix are set to zero to block the propagation of historical residuals. When it is in a closed and locked state, the diagonal elements are set to one to retain the propagation of residuals. The specific algebraic expression of the matrix multiplication mapping operation is as follows: ; In the formula, This is the boundary state residual feedforward vector of the next node after the state transition matrix mapping process. This is the final state vector of the transient operating curve at the end of the integration time. The ideal boundary constraint vector is the target post-steady-state node.

[0063] See attached document Figure 6 The virtual penalty energy conversion and total measurement quantization are performed. This process is executed by the dissipation quantization module 40. For a specific implementation of S400, it can be processed through the following sub-steps: S410, the dissipative quantization module 40 receives the virtual truncated residual vector outside the physically feasible region forcibly stripped by the integral feedforward module 30. When the conventional numerical integration algorithm encounters an out-of-bounds initial value, it directly discards the out-of-bounds residual. However, this method masks the mathematical defects of the conventional numerical integration algorithm in handling extreme conditions, which can lead to the failure of the global energy conservation state of the simulation system. To transform the mathematical truncation defect at non-physical boundaries into a physical energy penalty that can be perceived by the system optimization framework, the dissipative quantization module 40 constructs a state energy stiffness matrix. The dissipative quantization module 40 uses quadratic algebraic calculations to transform the virtual truncated residual vector into virtual penalty energy. The specific algebraic expression for the quadratic algebraic calculation is as follows. ; In the formula, The virtual penalty energy acquired through conversion; For the virtual truncated residual vector that is forcibly stripped outside the physically feasible region; This is the symbol for the matrix transpose operation; The state energy stiffness matrix constructed for the energy measurement quantization module.

[0064] The state energy stiffness matrix is ​​a diagonal matrix, and its internal diagonal elements are configured as mapping coefficients that transform the state residuals of different physical dimensions into a unified energy dimension. In specific implementation, for the residuals of the pressure dimension, the corresponding mapping coefficients are allocated by combining the quotient of the cylinder cavity rated volume divided by the fluid bulk elastic modulus. For the residuals of the velocity dimension, the corresponding mapping coefficients are allocated by combining half of the total mass of the piston and load, thereby unifying and quantifying various boundary state residuals into values ​​with equivalent energy dimensions.

[0065] S420, the dissipation quantization module 40 calculates the equivalent fluid mechanical impact dissipation energy based on the transient operating curve obtained by transient integral solution. Because the pneumatic transmission system experiences rapid pressure changes during reversal, the high-frequency mechanical oscillations caused by these pressure transients on the piston accelerate equipment fatigue. During transient reversal, the coupling of fluid pressure fluctuations within the target pneumatic control loop and piston speed generates irreversible mechanical impact dissipation. The dissipation quantization module 40 extracts the rate of change of pressure difference between the two cylinder chambers and the piston speed from the transient operating curve and performs absolute value integration or square integral operations in the time domain to avoid the rate of change of pressure difference and piston speed canceling each other out due to alternating positive and negative signs during high-frequency oscillation, which would lead to an underestimation of the mechanical impact intensity. The specific algebraic expression for calculating the equivalent fluid mechanical impact dissipation energy is as follows. ; In the formula, To calculate the equivalent hydromechanical impact dissipation energy; The duration of the transient commutation process corresponding to the directed edge; The transient pressure difference between the rodless and rod chambers of the cylinder in the target pneumatic control circuit; The preset dimensional alignment equivalence coefficients range from 0.8 to 1.2. The values ​​are empirically allocated and configured in combination with the effective working area of ​​the cylinder piston and the fluid equivalent compliance parameters, which are used to convert the transient product terms into standard working quantity dimensions. For integration time; This refers to the transient velocity variable of the piston during the transient reversal process; This is the differential step size term for the integration time variable.

[0066] To prevent the dissipation energy from diverging due to local high-frequency numerical oscillations in the flow field during integral calculation, the dissipation quantization module 40 is internally configured with an integral limiting instruction to constrain the calculated equivalent hydromechanical impact dissipation energy within a pre-set physical dissipation upper limit threshold based on the material yield strength. The specific value of the physical dissipation upper limit threshold is determined by the product of the cylinder material yield strength, the effective cross-sectional area of ​​the piston, and the maximum buffer stroke of the cylinder. When the calculated equivalent hydromechanical impact dissipation energy exceeds the physical dissipation upper limit threshold, the dissipation quantization module 40 performs a limiting block and uses the physical dissipation upper limit threshold as the numerical integral stability output value, while generating an impact over-limit flag. The impact over-limit flag is transmitted to the optimization and reconstruction module 50 to trigger the over-limit penalty term superposition or topology dynamic reconstruction instruction, avoiding the serious over-limit condition from being misjudged as a normal boundary condition after limiting.

[0067] S430, the dissipation quantization module 40 linearly weights and fuses the virtual penalty energy and the equivalent hydromechanical impact dissipation energy to generate a total hysteresis dissipation energy measure. This total hysteresis dissipation energy measure serves as a comprehensive metric for evaluating the physical impact state of the target aerodynamic control loop and the robustness of the simulation solution. The safety judgment threshold corresponding to the total hysteresis dissipation energy measure is set as the safe hysteresis dissipation threshold. This safe hysteresis dissipation threshold simultaneously reflects the allowable level of mechanical impact and the allowable level of boundary violation penalty. The specific algebraic expression for the weighted fusion is as follows. ; In the formula, The total hysteresis dissipation energy measure generated by the energy measurement quantification module; This is the weighting coefficient for fluid mechanical impact dissipation; To calculate the equivalent hydromechanical impact dissipation energy; For virtual penalty energy weighting coefficient; The virtual punishment energy acquired through conversion.

[0068] Both the fluid mechanical impact dissipation weight coefficient and the virtual penalty energy weight coefficient are non-negative real numbers, and the sum of the fluid mechanical impact dissipation weight coefficient and the virtual penalty energy weight coefficient is set to equal 1. The specific values ​​of the fluid mechanical impact dissipation weight coefficient and the virtual penalty energy weight coefficient are determined empirically based on the tolerance of the target aerodynamic control loop to physical mechanical impact and the requirement for calculation over-limit accuracy. In specific engineering implementation scenarios, in order to balance the physical optimization performance and the stability of algorithm iteration, the value range of the fluid mechanical impact dissipation weight coefficient is set to 0.4 to 0.8, and the virtual penalty energy weight coefficient is the difference between 1 and the fluid mechanical impact dissipation weight coefficient.

[0069] See attached document Figure 7The process of optimizing control parameters and issuing optimal commands is performed by the optimization and reconstruction module 50. For a specific implementation of S500, this can be achieved through the following sub-steps: S510, in order to comprehensively optimize from a global perspective of multi-condition switching, the optimization and reconstruction module 50 receives the total hysteresis dissipation energy measure output by the dissipation quantization module 40 at each directed edge position within the directed graph of condition transition. The optimization and reconstruction module 50 uses a linear weighted accumulation mechanism to sum all the total hysteresis dissipation energy measures within the directed graph of condition transition to construct a global evaluation objective function. Each directed edge is configured with a corresponding condition importance weight. The condition importance weight is used to characterize the degree of influence of the transient reversal process corresponding to the directed edge on the overall machine operation safety, cycle stability, or product processing quality. The global evaluation objective function is used to characterize the comprehensive energy dissipation level and computational stability state of the target pneumatic control loop within the entire multi-condition continuous operation cycle. The specific algebraic expression of the global evaluation objective function is as follows: ; In the formula, The global evaluation objective function is constructed for the parameter iterative optimization module; This is the traversal index variable for directed edges; This represents the total number of directed edges contained within the directed graph of the working condition transition. Internal index of the directed graph for working condition transition The directed edges at corresponding positions are pre-configured with the importance weights of the operating conditions; For index The total hysteresis dissipation energy measure of the directed edge at the corresponding position.

[0070] Meanwhile, to prevent severe over-limit conditions from being masked by conventional values, when the optimization and reconstruction module 50 receives the impact over-limit flag from the dissipation quantization module 40, it activates the over-limit penalty term superposition mechanism. A preset step penalty constant is superimposed on the total hysteresis dissipation energy measure corresponding to the target directed edge that generates the impact over-limit flag. The preset step penalty constant is set to 10,000 to 1,000,000 times the extreme value of normal energy dissipation. By establishing a numerical repulsion potential field in the global evaluation objective function, the search direction of the optimization algorithm is guided to deviate from the physically infeasible over-limit condition region.

[0071] S520, the optimization and reconstruction module 50, based on the global evaluation objective function, employs a hybrid optimization algorithm combining adaptive particle swarm optimization and sequential quadratic programming to perform iterative optimization calculations on the local control parameter vectors. The optimization and reconstruction module 50 concatenates the local control parameter vectors corresponding to all directed edges in the working condition transition directed graph into a global control parameter vector according to a preset order, and converts the hardware-set upper and lower threshold values ​​of the local control parameter vectors into boundary constraints for the global control parameter vectors.

[0072] First, the optimization and reconstruction module 50 initiates a global large-scale optimization, mapping the global control parameter vector to particle position coordinates within the adaptive particle swarm optimization algorithm. Based on the historical best position of particles within the adaptive particle swarm optimization algorithm and the global historical best position, the optimization and reconstruction module 50 performs dynamic update operations on the particle velocity vector within the adaptive particle swarm optimization algorithm. The specific algebraic expressions for updating the particle velocity vector and particle position coordinates are as follows: ; ; In the formula, This refers to the particle velocity vector updated in the next iteration step within the particle swarm optimization algorithm. This is the index variable for the current iteration step; This will serve as the index variable for the next iteration step. This is the inertia weighting coefficient, and its value ranges from 0.4 to 0.9. This represents the original particle velocity vector at the current iteration step. This refers to the individual cognitive learning factor, which ranges from 1.5 to 2.5. The first independent random number that follows a uniform distribution between zero and one; This represents the individual optimal position vector found by the particle in the iteration history; These are the original particle position coordinates in the current iteration step; The social collaborative learning factor ranges from 1.5 to 2.5. It is a second independent random number that follows a uniform distribution from 0 to 1; This represents the globally optimal position vector found by the entire particle swarm throughout the iteration history. This refers to the particle position coordinates updated in the next iteration step within the particle swarm optimization algorithm.

[0073] Once the adaptive particle swarm optimization algorithm satisfies the initial convergence condition, the optimization and reconstruction module 50 uses the current global optimal position vector output by the particle swarm as the initial point of iteration and starts the sequential quadratic programming algorithm to perform local precise optimization. The optimization and reconstruction module 50 calls the integral feedforward module 30 and the dissipative quantization module 40 to calculate the global evaluation objective function based on the current global control parameter vector, and uses the finite difference method or the adjoint sensitivity method to construct the Jacobian sensitivity matrix of the global evaluation objective function relative to the global control parameter vector. The optimization and reconstruction module 50 constructs a quadratic programming subproblem based on the global evaluation objective function, the Jacobian sensitivity matrix, and the boundary constraints, and solves the quadratic programming subproblem through the sequential quadratic programming algorithm to obtain the control parameter correction direction and correction step size of the current iteration step.

[0074] S530, the optimization and reconstruction module 50 performs boundary constraint processing on the parameter correction amount output by the hybrid optimization algorithm. Since the physical opening degree of the adjustable valve, the response frequency of the electromagnet, the dead zone delay time of the valve core, and the gradient of the flow cross section change of the valve port all have hardware limits in the target pneumatic control loop, the optimization and reconstruction module 50 transforms the above hardware limits into inequality constraints.

[0075] During the execution phase of the adaptive particle swarm optimization algorithm, the optimization and reconstruction module 50 uses hard-limiting logic to truncate and restrict the parameter elements inside the particle velocity vector and particle position coordinates. Specifically, when the parameter elements inside the particle velocity vector and particle position coordinates exceed the hardware-set upper limit threshold, they are forcibly assigned the hardware-set upper limit threshold; when they are below the hardware-set lower limit threshold, they are forcibly assigned the hardware-set lower limit threshold. At the same time, in order to prevent the particle swarm from accumulating velocity at the physical boundary and falling into a stagnant dead zone, when the optimization and reconstruction module 50 detects that the parameter element triggers the boundary, it forcibly multiplies the corresponding velocity component by a reverse damping coefficient of -0.5 to guide the particle back to the effective search domain.

[0076] During the execution phase of the sequential quadratic programming algorithm, when the direction of control parameter correction causes any control parameter element to go out of bounds, the optimization and reconstruction module 50 reduces the current iteration step size through a step size back-off mechanism until all updated control parameter elements are within the physical execution domain. By handling out-of-bounds constraints, it is ensured that the controller adjustment setpoint output by the algorithm is completely within the physical execution domain of the target pneumatic control loop.

[0077] S540, the optimization and reconstruction module 50 monitors the convergence state of the global evaluation objective function and outputs the global optimal control parameter matrix.

[0078] During the adaptive particle swarm optimization algorithm phase, when the absolute value of the change in the global evaluation objective function corresponding to two adjacent iterations is lower than the preset primary convergence tolerance for multiple consecutive iterations, such as 50 consecutive iterations, the preset primary convergence tolerance ranges from 0.001 to 0.0001, or when the number of optimization iterations of the adaptive particle swarm optimization algorithm reaches the preset maximum number of iterations, such as an integer in the range of 500 to 2000, the optimization and reconstruction module 50 determines that the primary optimization process has converged and triggers the aforementioned sequential quadratic programming algorithm.

[0079] During the sequential quadratic programming algorithm phase, the optimization and reconstruction module 50 extracts the numerical change of the global evaluation objective function, the correction norm of the global control parameter vector, and the constraint violation amount corresponding to two adjacent iteration steps. When the numerical change of the global evaluation objective function is lower than the preset final convergence tolerance, the correction norm is lower than the preset parameter tolerance, and the constraint violation amount is lower than the preset constraint tolerance, the preset final convergence tolerance ranges from 10. -6 Up to 10 -5 The preset parameter tolerance range is 10. -5 Up to 10 -4 The preset constraint tolerance range is 10. -6 Up to 10 -5 The optimization and reconstruction module 50 determines whether the sequential quadratic programming algorithm has fully converged. If the number of iterations of the sequential quadratic programming algorithm reaches the preset maximum number of iterations but still does not meet the convergence condition, a non-fully converged flag is generated.

[0080] Whether the convergence condition is met to trigger a normal stop or the maximum number of iterations is reached to trigger a forced stop, the optimization and reconstruction module 50 will output the control parameter vector corresponding to the minimum global evaluation objective function obtained so far as the global optimal control parameter matrix to the cache register for closed-loop safety verification. If there is a non-complete convergence flag, a suboptimal solution warning label will be added to the output global optimal control parameter matrix.

[0081] S550, the optimization and reconstruction module 50 extracts the total hysteresis dissipation energy measure corresponding to the global optimal control parameter matrix, and performs closed-loop verification and comparison with the preset safe hysteresis dissipation threshold.

[0082] If the total hysteresis dissipation energy measurement is less than or equal to the preset safe hysteresis dissipation threshold, it indicates that the parameter optimization under the current topology has met the system safety requirements. At this time, the optimization and reconfiguration module 50 formally sends the control strategy element variables covered in the global optimal control parameter matrix to the bottom programmable logic controller of the target pneumatic control loop through the industrial communication bus to complete the physical deployment and closed-loop control of the parameters. The data packaging and message transmission protocol involved in the industrial communication bus can be implemented according to the existing industrial Ethernet communication technology, and the optimization process is officially over. If the total hysteresis dissipation energy measurement is greater than the preset safe hysteresis dissipation threshold, it indicates that relying solely on controller parameter adjustment cannot eliminate the destructive impact caused by severe commutation. The optimization and reconstruction module 50 generates and issues a topology dynamic reconstruction command. In specific implementation, the optimization and reconstruction module 50 identifies the target directed edge that triggers the impact over-limit flag and outputs a topology reconstruction alarm containing hardware modification prompts to the operation terminal. At the same time, a new transient buffer steady-state node is virtually inserted into the target directed edge of the working condition transition directed graph of the simulation system. The transient buffer steady-state node physically represents the working area of ​​the mechanical buffer damper to be added in the gas path or the secondary exhaust throttling area. When the secondary cycle solution process is triggered, the transient buffer steady-state node is equivalently configured by the steady-state decoupling module 20 as an exhaust steady-state node or a locking steady-state node according to the preset rules. The preset rules are to allocate node categories based on the node boundary pressure difference and introduce the corresponding buffer damping reaction force terms into the static equilibrium algebraic equation system to participate in the algebraic solution calculation. Subsequently, the optimization and reconstruction module 50 performs a graph theory partitioning operation on the original boundary path, splitting the target directed edge in two and recombining it into a new sequence of directed edges connected end to end. The optimization and reconstruction module 50 assigns independent local control parameter vectors to the newly generated directed edges and feeds them back to the topology modeling module 10, triggering the system to re-execute the secondary loop solution process with the updated working condition transition directed graph until the total hysteresis dissipation energy measure of the reconstructed system meets the safety threshold requirement or the number of topology reconstructions reaches the preset maximum hardware reconstruction limit. The preset maximum hardware reconstruction limit ranges from 3 to 5 times. If the safety requirement is not met even after reaching the maximum hardware reconstruction limit, the optimization and reconstruction module 50 terminates the optimization and outputs a system physical design infeasibility alarm to the host computer, prompting that the cylinder selection needs to be re-performed or the macroscopic operating parameters need to be modified.

[0083] To aid in understanding the technical solution of this invention, a specific example of a pneumatic control loop application is provided below.

[0084] The application scenario is set as a pneumatic control circuit for a heavy-duty material sorting line. The core actuator of the pneumatic control circuit is configured as a rodless cylinder with a cylinder diameter of 80mm, a maximum stroke of 500mm, an external load mass of 50.3kg, a steady-state air supply pressure of 0.61MPa, an external environmental back pressure of 0.10MPa, and adjustable proportional directional valves at the air inlet and exhaust ends of the pneumatic control circuit.

[0085] During operation, the topology modeling module 10 receives continuous operation sequence parameters of the pneumatic control loop, including the return to extreme position lock-up standby stage and the full load push-out stage. The topology modeling module 10 generates a working condition transition directed graph, which includes a return to extreme position lock-up standby steady-state node, a push-out extreme position unloading waiting steady-state node, and a full load push-out transient directed edge. The topology modeling module 10 allocates local control parameter vectors at the full load push-out transient directed edge positions. The local control parameter vectors include the initial opening of the intake proportional valve, the gradient of the flow cross section change of the exhaust proportional valve, and the reversing delay time.

[0086] The steady-state decoupling module 20 performs algebraic calculations for the push-out extreme position unloading waiting steady-state node. The steady-state decoupling module 20 obtains the analytical solution of the state at the end of the steady state of the push-out extreme position unloading waiting steady-state node. The steady-state decoupling module 20 outputs the ideal boundary constraint vector corresponding to the push-out extreme position unloading waiting steady-state node. At this time, the absolute static pressure of the fluid inside the rodless chamber is equal to the steady-state supply pressure setting value of the gas source, the absolute static pressure of the fluid inside the rod chamber is equal to the back pressure setting value of the external environment, and the steady-state velocity of the piston is equal to zero.

[0087] The integral feedforward module 30 performs calculations for the transient directed edge of the full-load push. Since the gas in the previous operating cycle was not completely released, the integral feedforward module 30 obtains the 0.06MPa boundary state residual feedforward vector passed from the previous operating condition. The integral feedforward module 30 performs algebraic superposition of the ideal boundary constraint vector of the steady-state node and the boundary state residual feedforward vector to obtain the theoretical initial value vector. The integral feedforward module 30 uses the built-in smoothing mapping function to separate the actual injected initial value vector inside the physical feasible region of the system state vector and the virtual truncated residual vector outside the physical feasible region of the system state vector. The integral feedforward module 30 performs time-domain integration operation with the actual injected initial value vector as the initial value for calculation. The integral feedforward module 30 outputs the transient operating curve of the pneumatic control loop.

[0088] The dissipation quantization module 40 receives the transient operating curve, calculates the equivalent hydromechanical impact dissipation energy value, performs quadratic algebraic calculation on the virtual truncated residual vector using the state energy stiffness matrix, extracts the virtual penalty energy value, and performs linear weighted fusion of the equivalent hydromechanical impact dissipation energy and the virtual penalty energy. The dissipation quantization module 40 outputs the total hysteresis dissipation energy measure corresponding to the local control parameter vector.

[0089] The optimization and reconstruction module 50 receives the total hysteresis dissipation energy measure. The preset safe hysteresis dissipation threshold is set to 50.00J. The optimization and reconstruction module 50 uses a hybrid optimization algorithm that combines adaptive particle swarm optimization and sequential quadratic programming to perform iterative optimization calculations on the local control parameter vector. After the first optimization calculation, the total hysteresis dissipation energy measure output by the optimization and reconstruction module 50 is 59.18J. The optimization and reconstruction module 50 determines that the total hysteresis dissipation energy measure is greater than the preset safe hysteresis dissipation threshold.

[0090] The optimization and reconstruction module 50 generates a topology dynamic reconstruction instruction and inserts a transient buffer steady-state node inside the transient directed edge of the full-load push. The transient buffer steady-state node corresponds to the exhaust throttling buffer dead zone region at the physical level. The optimization and reconstruction module 50 divides and reorganizes the transient directed edge of the full-load push into a new directed edge sequence with the beginning and end connected. The optimization and reconstruction module 50 triggers the system to re-execute the secondary loop calculation process with the updated working condition transition directed graph. The optimization and reconstruction module 50 increases the end exhaust throttling degree through the secondary loop calculation process.

[0091] Ultimately, the total hysteresis dissipation energy measure output by the optimization and reconfiguration module 50 is reduced to 31.05J, and the optimization and reconfiguration module 50 outputs the global optimal control parameter matrix to the internal programmable logic controller of the pneumatic control loop.

[0092] Table 1: Comparison of Test Data on the Operating Status of Pneumatic Control Loops under Different Control Strategies Test parameter indicators Traditional engineering experience control methods Optimization methods without topology reconstruction Digital simulation optimization method for multi-condition directional control strategy of cylinder Maximum mechanical impact energy at the end (J) 89.34 59.18 29.47 Maximum pressure overshoot of the inflation chamber (MPa) 0.26 0.13 0.05 Maximum transient speed during full-load propulsion (m / s) 1.34 1.21 1.18 Single cycle execution time (s) 2.47 2.28 2.36 Total hysteresis dissipation energy measure (J) 95.12 61.45 31.05 Number of iterations of the Jacobian matrix diverging (times) 14 0 0 According to Table 1 and Figure 8 It can be seen that the digital simulation optimization method of the cylinder multi-condition directional control strategy surpasses the limitations of pure physical parameter adjustment in terms of mechanical shock suppression. Under the traditional engineering experience control method, the maximum mechanical shock energy at the end reaches 89.34J. The optimization method without topology reconstruction reduces the maximum mechanical shock energy at the end to 59.18J, reaching the limit of valve parameter adjustment. The digital simulation optimization method of the cylinder multi-condition directional control strategy further reduces the maximum mechanical shock energy at the end to 29.47J by inserting transient buffer steady-state nodes to perform topology reconstruction, and reduces the maximum pressure overshoot of the air chamber to 0.05MPa. The topology dynamic reconstruction operation controls the dynamic alternating load borne by the pneumatic control loop.

[0093] In terms of computational robustness, the Jacobian matrix iteration divergence of the digital simulation optimization method for the cylinder multi-condition directional control strategy in Table 1 is 0 times. This is because when the traditional engineering experience control method directly processes the residuals, the out-of-bounds initial value causes the numerical solver to exhibit matrix singularity and divergence 14 times. The digital simulation optimization method for the cylinder multi-condition directional control strategy uses a smooth mapping function to isolate the out-of-bounds negative pressure and overpressure initial values, and uses the dissipative quantization module 40 to transform non-physical truncation defects into optimizable virtual penalty energy. The intervention of the integral feedforward process blocks the mathematical conditions that cause the solver to collapse, thus eliminating the iteration divergence phenomenon.

[0094] At the operational rhythm level, the digital simulation optimization method for cylinder multi-condition directional control strategy, due to the insertion of transient buffer steady-state nodes, has a single-condition cycle execution time of 2.36s, which increases the delay by 0.08s compared to the optimization method without topology reconstruction. At the cost of a 0.08s increase in the action cycle, the system reduces the total hysteresis dissipation energy measure from 61.45J to 31.05J, thereby improving commutation smoothness.

[0095] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A digital simulation optimization method for multi-condition directional control strategy of cylinders, characterized in that, include: A directed graph of the working condition transition containing steady-state nodes and directed edges is constructed from the parameters of the continuous operation sequence of multiple working conditions, and local control parameter vectors are assigned to the directed edges; Determine the ideal boundary constraint vector of the steady-state node using the system of equations; The ideal boundary constraint vector of the forward steady-state node and the feedforward vector of the boundary state residual are separated into the actual injected initial value vector and the virtual truncated residual vector. The transient running curve and the final state vector are obtained by integrating the actual injected initial value vector. The final state vector and the ideal boundary constraint vector of the subsequent steady-state node are combined to update the boundary state residual feedforward vector. The virtual truncated residual vector is converted into virtual penalty energy, which is then combined with the equivalent fluid mechanical impact dissipation energy of the transient operating curve to form the total hysteresis dissipation energy measure. The local control parameter vector is updated according to the total hysteresis dissipation energy measure, and the over-limit reconstructed working condition transition directed graph is sent down to the underlying programmable logic controller after convergence.

2. The digital simulation optimization method for cylinder multi-condition directional control strategy according to claim 1, characterized in that, The steps for finding the ideal boundary constraint vector of the steady-state node according to the system of equations include: A set of static equilibrium algebraic equations, including mechanical equilibrium equations and thermodynamic state equations, is established as the set of equations. Based on the gas path connection state corresponding to the steady-state node, the steady-state node is divided into inflation steady-state node, exhaust steady-state node, lock-up steady-state node and end-position steady-state node; For the inflation steady-state node, the inflation chamber pressure is constrained to the steady-state supply pressure setting value of the air source; for the exhaust steady-state node, the exhaust chamber pressure is constrained to the external environmental back pressure setting value; for the locking steady-state node, algebraic calculation is performed using the mechanical equilibrium equation and the air chamber closure mass conservation condition; for the end-position steady-state node, a mechanical limiting reaction term is introduced into the mechanical equilibrium equation and algebraic calculation is performed.

3. The digital simulation optimization method for cylinder multi-condition directional control strategy according to claim 1, characterized in that, The steps for finding the ideal boundary constraint vector of the steady-state node according to the system of equations include: The macroscopic operating parameters and the state vectors corresponding to the previous steady-state nodes are input into the steady-state decoupling feature extraction neural network model, and the steady-state prediction vector is output. When the residual after substituting the steady-state prediction vector into the system of equations is less than or equal to a preset residual threshold, the steady-state prediction vector is injected into the nonlinear algebra solver as a high-confidence initial iteration value. The nonlinear algebra solver is triggered to perform multiple approximation and root-finding operations on the system of equations starting from the high-confidence initial iteration value, to obtain the analytical solution of the steady-state node at the end of the steady state, and to reorganize the analytical solution of the steady state into a matrix according to the physical quantity dimension, and output the ideal boundary constraint vector of the steady-state node.

4. The digital simulation optimization method for cylinder multi-condition directional control strategy according to claim 1, characterized in that, The steps for separating the ideal boundary constraint vector of the preceding steady-state node and the feedforward vector of the boundary state residual into the actual injected initial value vector and the virtual truncated residual vector include: The theoretical initial value vector is obtained by algebraically superimposing the ideal boundary constraint vector of the pre-steady-state node with the feedforward vector of the boundary state residual, and then input into the transient residual nonlinear mapping neural network model to output the smooth mapping steepness gain coefficient matrix. The energy-preserving nonlinear boundary mapping algorithm is invoked, and the theoretical initial value vector is explicitly calculated by combining the set physical lower limit vector and physical upper limit vector of the state variables to obtain the actual injected initial value vector. Perform a subtraction operation between the theoretical initial value vector and the actual injected initial value vector to output the virtual truncated residual vector that is forcibly stripped outside the physical feasible domain.

5. The digital simulation optimization method for cylinder multi-condition directional control strategy according to claim 1, characterized in that, The steps for updating the boundary state residual feedforward vector by simultaneously combining the final state vector and the ideal boundary constraint vector of the post-steady-state node include: Construct a state transition matrix in the form of a diagonal Boolean mapping matrix based on the degree-of-freedom constraints of the post-steady-state node; When the flow field or mechanical dimension corresponding to the post-steady-state node is in an unconstrained state with released degrees of freedom, the diagonal elements of the state transition matrix are set to zero; when it is in a closed and locked state, the diagonal elements of the state transition matrix are set to one. The boundary state residual vector, obtained by subtracting the final state vector from the ideal boundary constraint vector of the subsequent steady-state node using the state transition matrix, is subjected to matrix multiplication mapping operation to update the boundary state residual feedforward vector.

6. The digital simulation optimization method for cylinder multi-condition directional control strategy according to claim 1, characterized in that, The steps for obtaining the equivalent hydromechanical impact dissipation energy of the transient operating curve include: In the time domain, the product of the time derivative data of the two-chamber pressure difference and the instantaneous velocity data of the piston contained in the transient operating curve is extracted, and the absolute value integration operation or the square integration operation of the product is performed. Multiply the integral result by a preset dimensional alignment equivalent coefficient to obtain the equivalent fluid mechanical impact dissipation energy of the transient operating curve converted into the functional dimension.

7. The digital simulation optimization method for cylinder multi-condition directional control strategy according to claim 6, characterized in that, After obtaining the equivalent hydromechanical impact dissipation energy of the transient operating curve converted into a functional quantity dimension, the process includes: When the calculated equivalent hydromechanical impact dissipation energy of the transient operating curve exceeds the physical dissipation upper limit threshold set based on the yield strength of the cylinder material, the effective cross-sectional area of ​​the piston, and the maximum buffer stroke of the cylinder, a limiting and blocking operation is performed. The physical dissipation upper limit threshold is used as the numerical integration stability output value, and an impact over-limit flag is generated; the impact over-limit flag is used to trigger the over-limit penalty term superposition mechanism in the subsequent optimization process to force the search direction to deviate from the over-limit working condition.

8. The digital simulation optimization method for cylinder multi-condition directional control strategy according to claim 1, characterized in that, The step of updating the local control parameter vector based on the total hysteresis dissipation energy measure includes: A global evaluation objective function is constructed by performing a linear weighted accumulation mechanism using all the total hysteresis dissipation energy measures and the working condition importance weights corresponding to the directed edges within the working condition transition directed graph; A hybrid optimization algorithm combining adaptive particle swarm optimization and sequential quadratic programming is adopted. After concatenating all the local control parameter vectors in the working condition transition directed graph into a global control parameter vector, iterative optimization calculation is performed. After the adaptive particle swarm optimization algorithm satisfies the initial convergence condition, the sequential quadratic programming algorithm is started with the output current global optimal position vector as the initial point of iteration to perform local exact optimization. The quadratic programming subproblem is solved by the Jacobian sensitivity matrix to obtain the control parameter correction direction and correction step size of the current iteration step, and the local control parameter vector is updated in combination with boundary constraint processing.

9. The digital simulation optimization method for cylinder multi-condition directional control strategy according to claim 1, characterized in that, The steps for the transition directed graph of the over-limit reconfiguration working condition include: Identify the target directed edge that causes the overlimit, and virtually insert a transient buffer steady-state node inside the target directed edge to represent the action area of ​​the proposed mechanical buffer damper or the secondary exhaust throttling area. Perform a graph theory partitioning operation on the original boundary path, splitting the target directed edge in two and recombining it into a new sequence of directed edges that are connected end to end; Assign an independent local control parameter vector to the new directed edge sequence and output a topology dynamic reconstruction instruction to trigger the re-execution of the secondary loop solution process using the updated working condition transition directed graph.

10. The digital simulation optimization method for cylinder multi-condition directional control strategy according to claim 9, characterized in that, After triggering the step of re-executing the secondary loop solution process using the updated working condition transition directed graph, the process includes: If the total hysteresis dissipation energy measure of the reconstructed system after re-executing the secondary loop solution process is less than or equal to the preset safe hysteresis dissipation threshold, the parameter update deployment is completed. If the number of topology reconfigurations reaches the preset maximum hardware reconfiguration limit and the total hysteresis dissipation energy measure of the reconfiguration system is still greater than the preset safe hysteresis dissipation threshold, then the optimization will be terminated and a system physical design infeasibility alarm will be output.

Images