Pilot valve multi-parameter real-time monitoring control method and system

Abstract

This invention discloses a real-time monitoring and control method and system for multiple parameters of a pilot valve, relating to the field of pilot valve control technology. The method includes: acquiring the electrical parameter feature matrix and fluid pressure gradient array of the pilot valve in the current control cycle; extracting a dynamic back electromotive force (EMF) feature tensor characterizing the mechanical motion state of the valve core; aligning the dynamic EMF feature tensor with the fluid pressure gradient array in feature dimensions, inputting it into a pre-trained physically constrained dynamic hysteresis compensation network, and outputting a desired control current sequence; calculating a dynamic sliding surface threshold based on the fluid pressure gradient array, generating an asymmetric chattering pulse width modulation (PWM) drive signal; and sending the asymmetric chattering PWM drive signal to the underlying drive bridge of the electromagnetic coil to control the valve core opening of the pilot valve. This invention improves the control accuracy and safety of the pilot valve under complex operating conditions and avoids the failure risk caused by reliance on mechanical sensors.

Description

Technical Field

[0001] This invention relates to the field of pilot valve control technology, specifically to a method and system for real-time monitoring and control of multiple parameters of a pilot valve. Background Technology

[0002] As the core pre-actuated element of electro-hydraulic servo systems and precision fluid control systems, the pilot valve's operating state directly determines the response speed of the main valve and the stability of the flow path system. In aerospace, heavy engineering machinery, and high-precision industrial automation scenarios, the pilot valve typically receives a weak electrical signal from the controller and generates electromagnetic attraction through an internal electromagnetic coil to drive the armature and valve core to undergo micro-displacement. This changes the conduction cross-sectional area of ​​the pilot fluid oil or gas path, relying on amplified fluid hydraulic force to drive the main valve core, which has a large mass inertia, to move.

[0003] The control schemes of related technologies typically rely on closed-loop regulation using feedback logic from terminal sensors. Specifically, the controller acquires the actual operating pressure of the system through a pressure sensor installed on the actuator cylinder or main valve outlet side, calculates the difference between the actual operating pressure and the target set pressure, and then processes this difference using a proportional-integral-derivative (PID) control algorithm to calculate the required correction duty cycle signal, which is then amplified and output to the solenoid coil of the pilot valve. The valve opening is changed by the change in coil current until the terminal fluid pressure reaches the target set value. However, in highly dynamic engineering applications, the fluid medium has non-negligible compressibility, and the pilot valve... The movement of the valve core is constantly subjected to nonlinear interference from static friction and dynamic hydraulic forces. This coupling effect of mechanical friction and fluid dynamics will generate a severe nonlinear hysteresis zone when the direction of valve core movement is reversed. The feedback data of the terminal pressure sensor is naturally lagging behind the moment the electromagnetic coil acts in the time link. This delay in the physical link causes the compensation current output by the control system based on the terminal pressure to always lag behind the sudden change in the transient fluid pressure in phase. This error is continuously amplified in the closed-loop system, eventually causing the system to generate continuous and violent oscillations in fluid pressure under high-frequency dynamic loads, and even leading to structural fatigue failure of the hydraulic circuit. Summary of the Invention

[0004] To solve the above technical problems, a method and system for real-time monitoring and control of multiple parameters of a pilot valve are provided. This technical solution solves the above problems.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for real-time monitoring and control of multiple parameters of a pilot valve, comprising: S101, Obtain the electrical parameter feature matrix and fluid pressure gradient array of the pilot valve in the current control cycle. The electrical parameter feature matrix includes the transient voltage sequence and drive current sequence across the electromagnetic coil. S102, calculate the discrete-time derivatives of the transient voltage sequence and the driving current sequence in the electrical parameter feature matrix, and extract the dynamic back electromotive force feature tensor characterizing the mechanical motion state of the valve core; S103, Align the dynamic back electromotive force feature tensor with the fluid pressure gradient array in terms of feature dimensions, input it into a pre-trained physical constraint dynamic hysteresis compensation network, and output the desired control current sequence. S104, calculate the dynamic sliding surface threshold based on the fluid pressure gradient array, and generate an asymmetric chattering pulse width modulation drive signal based on the desired control current sequence and the dynamic sliding surface threshold; S105, the asymmetric dithering pulse width modulation drive signal is sent to the bottom drive bridge of the electromagnetic coil to control the valve core opening of the pilot valve.

[0006] Furthermore, the electrical parameter characteristic matrix and fluid pressure gradient array of the pilot valve in the current control cycle are obtained, including: According to the preset high-frequency sampling clock, the terminal voltage data and loop current data of the electromagnetic coil are collected synchronously. The terminal voltage data and the loop current data are converted into continuous discrete time series to construct a two-dimensional electrical parameter feature matrix; Acquire pressure sensing data located on the fluid outlet side of the pilot valve; The pressure sensing data is subjected to a first-order difference operation to generate the fluid pressure gradient array that reflects the transient fluctuation rate of pressure.

[0007] Furthermore, the dynamic back electromotive force characteristic tensor characterizing the mechanical motion state of the valve core is extracted, including: Obtain the static equivalent DC resistance and inductance parameters of the electromagnetic coil; Substituting the transient voltage sequence and the driving current sequence into the discretized variant of the Kirchhoff voltage equation, and eliminating the ohmic voltage drop component and the inductive voltage drop component, the discretized variant of the Kirchhoff voltage equation is as follows: ; In the formula, The high-frequency sampling period is a preset sampling time interval, measured in milliseconds. The ohmic voltage drop component, in volts (V), is the voltage drop caused by current flowing through the coil resistance. This is the inductive voltage drop component, measured in volts (V), which is the induced voltage drop caused by changes in coil current. It is the transient back electromotive force, with the unit V. It is the reverse induced voltage generated by the valve core's movement cutting the magnetic field and is directly related to the valve core's motion state. for The voltage across the coil at any given moment, in volts (V). for The coil drive current at any given moment, in amperes (A). for The coil drive current at any given moment, in amperes (A). The remaining voltage component is extracted as the transient back electromotive force value; The transient back electromotive force values ​​corresponding to multiple consecutive sampling points are combined into a one-dimensional vector and mapped to the dynamic back electromotive force feature tensor.

[0008] Furthermore, the physically constrained dynamic hysteresis compensation network includes an input mapping layer, a fluid-electromagnetic cross-attention layer, and a compensation decoding layer. The dynamic back electromotive force feature tensor and the fluid pressure gradient array are input into the pre-trained physically constrained dynamic hysteresis compensation network, and the desired control current sequence is output, including: The physically constrained dynamic hysteresis compensation network includes an input mapping layer, a fluid-electromagnetic cross-attention layer, and a compensation decoding layer. It inputs the dynamic back electromotive force feature tensor and the fluid pressure gradient array into a pre-trained physically constrained dynamic hysteresis compensation network and outputs the desired control current sequence, including: The input mapping layer maps the dynamic back electromotive force feature tensor to a priori valve core velocity features and the fluid pressure gradient array to hydrodynamic disturbance features, wherein the mapping formula is: ; In the formula, The input mapping layer nonlinear transformation function is a preset neural network mapping function used to convert input features into a feature form that meets the requirements of subsequent processing. The valve core velocity prior feature is the feature vector characterizing the valve core motion velocity, which is obtained by mapping the dynamic back electromotive force feature tensor. The hydraulic dynamic disturbance characteristic is a feature vector characterizing the hydraulic dynamic disturbance of a hydraulic system, which is obtained by mapping the fluid pressure gradient array. The characteristic tensor of the dynamic back electromotive force; for The fluid pressure gradient at time t; In the fluid-electromagnetic cross-attention layer, the dot product similarity matrix between the valve core velocity prior feature and the hydrodynamic disturbance feature is calculated, and dynamic weighting coefficients characterizing the hysteresis nonlinearity intensity are generated based on the dot product similarity matrix. The dynamic weighting coefficients are fused with the prior features of the valve core velocity and then input into the compensation decoding layer. The compensation decoding layer outputs the desired control current sequence to offset transient hysteresis errors based on the fused features.

[0009] Furthermore, the physical constraint-based dynamic hysteresis compensation network constructs a customized physical constraint loss function during the training phase. The execution logic of the customized physical constraint loss function includes: The mean square error between the predicted current sequence output by the calculation model and the reference current sequence is used as the basic fitting loss. Calculate the overshoot between the theoretically derived fluid pressure value and the target fluid pressure value under the action of the predicted current sequence; When the overshoot exceeds a preset steady-state error threshold, a fluid pressure overshoot penalty is triggered. The total loss value is calculated by weighting and summing the basic fitting loss with the fluid pressure overshoot penalty term. Based on the total loss value, the backpropagation algorithm is executed to update the network node weights of the input mapping layer, the fluid-electromagnetic cross-attention layer, and the compensation decoding layer.

[0010] Furthermore, the dynamic sliding surface threshold is calculated based on the fluid pressure gradient array, including: Extract the peak gradient data and fluctuation frequency data from the fluid pressure gradient array; Multiply the peak gradient data by the basic friction compensation coefficient to obtain the steady-state damping term; Substitute the fluctuation frequency data into a preset exponential decay function to obtain the transient disturbance term; The steady-state damping term is added to the transient disturbance term to generate the dynamic sliding surface threshold.

[0011] Furthermore, based on the desired control current sequence and the dynamic sliding surface threshold, an asymmetric jitter pulse width modulation driving signal is generated, including: The desired control current sequence is converted into a base duty cycle sequence; Determine whether the absolute value of the fluid pressure gradient array is greater than the dynamic sliding surface threshold; When the absolute value of the fluid pressure gradient array is greater than the dynamic sliding surface threshold, the static frictional mechanical parameters between the pilot valve core and the valve sleeve are obtained. The critical excitation frequency that can disrupt the surface tension of the hydraulic oil film is calculated based on the aforementioned static tribological parameters. Using the critical excitation frequency as the reference frequency, a sinusoidal or triangular perturbation wave with a preset amplitude is generated as the high-frequency perturbation waveform. The high-frequency perturbation waveform is injected into the rising edge region of the basic duty cycle sequence in an asymmetric superposition manner to generate the asymmetric dithering pulse width modulation drive signal.

[0012] Furthermore, the high-frequency perturbation waveform is injected into the rising edge region of the basic duty cycle sequence in an asymmetric superposition manner, including: Extract the leading rising phase of each pulse period in the basic duty cycle sequence; The amplitude of the high-frequency perturbation waveform is algebraically added to the basic duty cycle only during the leading rise phase; During the peak and fall phases of the pulse cycle, the injection of the high-frequency perturbation waveform is shielded.

[0013] Furthermore, after sending the asymmetric jitter pulse width modulation drive signal to the bottom drive bridge of the electromagnetic coil, the method further includes: Real-time monitoring of the actual output current waveform of the underlying drive bridge; Calculate the phase delay time between the actual output current waveform and the asymmetric jitter pulse width modulation drive signal; The phase delay time is fed back to the physically constrained dynamic hysteresis compensation network for advance compensation of the phase of the desired control current sequence in the next control cycle.

[0014] A pilot valve multi-parameter real-time monitoring and control system includes: The multi-source parameter acquisition module acquires the electrical parameter feature matrix and fluid pressure gradient array of the pilot valve in the current control cycle. The electrical parameter feature matrix includes the transient voltage sequence and drive current sequence across the electromagnetic coil. The dynamic feature extraction module calculates the discrete-time derivatives of the transient voltage sequence and the driving current sequence in the electrical parameter feature matrix, and extracts the dynamic back electromotive force feature tensor characterizing the mechanical motion state of the valve core. The hysteresis compensation and desired current generation module aligns the dynamic back electromotive force feature tensor with the fluid pressure gradient array in terms of feature dimensions, inputs it into a pre-trained physical constraint dynamic hysteresis compensation network, and outputs the desired control current sequence. The drive signal generation module calculates the dynamic sliding surface threshold based on the fluid pressure gradient array, and generates an asymmetric chattering pulse width modulation drive signal based on the desired control current sequence and the dynamic sliding surface threshold. The valve core opening control module sends the asymmetric dithering pulse width modulation drive signal to the bottom drive bridge of the electromagnetic coil to control the valve core opening of the pilot valve.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention uses a multi-source parameter acquisition module to simultaneously acquire the electrical parameter feature matrix and fluid pressure gradient array of the pilot valve. Combined with a preset high-frequency sampling clock and first-order differential operation, it accurately captures the transient changes in the electrical characteristics of the electromagnetic coil and the internal fluid pressure. This solves the problem that existing control systems only monitor electrical parameters or fluid parameters and cannot take into account the synergistic effect of electromagnetic and fluid. At the same time, by using a dynamic feature extraction module to remove redundant information, it accurately extracts the dynamic back electromotive force feature tensor that characterizes the mechanical motion state of the valve core, realizing indirect and accurate monitoring of the valve core's motion state. This provides reliable feature support for subsequent compensation control and greatly improves the comprehensiveness of the control system's perception of the pilot valve's operating state and the accuracy of its control. 2. This invention innovatively designs a physically constrained dynamic hysteresis compensation network, integrating an input mapping layer, a fluid-electromagnetic cross-attention layer, and a compensation decoding layer. Through the cross-attention mechanism, it accurately captures the correlation between the prior features of valve core velocity and the features of hydrodynamic disturbance, generating dynamic weight coefficients to adapt to changes in hysteresis nonlinearity. Simultaneously, a customized physically constrained loss function is constructed during the network training phase, combining the basic fitting loss and the fluid pressure overshoot penalty term to ensure that the desired control current sequence output by the network can effectively offset transient hysteresis errors. This solves the problems of control deviation and motion jamming caused by hysteresis nonlinearity in the pilot valve core movement, significantly improving the stability and accuracy of valve core opening adjustment, and avoiding insufficient fluid control accuracy caused by hysteresis errors. 3. This invention uses a drive signal generation module to calculate the dynamic sliding surface threshold based on a fluid pressure gradient array, and combines it with the desired control current sequence to generate an asymmetric chattering pulse width modulation drive signal. It adopts a high-frequency micro-perturbation waveform asymmetric superposition mode, injecting high-frequency micro-perturbation waves only at the rising edge of the basic duty cycle sequence. The frequency of the high-frequency micro-perturbation waves is determined based on the static friction mechanical parameters of the valve core and valve sleeve. This can accurately destroy the surface tension of the hydraulic oil film and suppress chattering, while avoiding the impact of chattering on the stability of the drive signal. This design effectively solves the problems of valve core wear, seal damage, and excessive noise caused by chattering in existing pilot valve control, reduces equipment operating losses, and extends the service life of the pilot valve and the underlying drive bridge. 4. On the one hand, the present invention dynamically calculates the threshold of the sliding surface based on the peak gradient and fluctuation frequency of the fluid pressure gradient array, which can adaptively adapt to the changes in different operating conditions of the pilot valve. On the other hand, after the valve core opening is controlled, the actual output current waveform of the underlying drive bridge is monitored in real time, the phase delay time is calculated and fed back to the hysteresis compensation network, and the phase advance compensation of the next control cycle is realized. This solves the defects of the existing control system that is lagging feedback and cannot adapt to changes in operating conditions, greatly improves the control response speed and adaptability, and ensures that the pilot valve can maintain stable operation under different loads and different operating conditions. 5. This invention decomposes the complex pilot valve control process into five functionally independent and seamlessly connected modules. Each module has a clear division of labor, and the entire process from parameter acquisition to valve core control is completed automatically without manual intervention. At the same time, each technical step is based on mature physical principles and engineering practice design. The training logic of the physically constrained dynamic hysteresis compensation network is closely aligned with the electromagnetic-fluid coupling characteristics of the pilot valve, avoiding system redundancy and failure risks caused by complex algorithms. In addition, the various technical features of the system can be directly adapted to the underlying drive architecture of existing pilot valves without the need for large-scale modification of the pilot valve itself, reducing engineering application costs and improving the system's reliability, portability, and engineering practicality. Attached Figure Description

[0016] Figure 1 This is a flowchart of the steps of the present invention. Figure 2 This is a system architecture diagram of the present invention. Detailed Implementation

[0017] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0018] Example 1: Refer to Figure 1 As shown, a method for real-time monitoring and control of multiple parameters of a pilot valve includes: S101, Obtain the electrical parameter feature matrix and fluid pressure gradient array of the pilot valve in the current control cycle. The electrical parameter feature matrix includes the transient voltage sequence and drive current sequence across the electromagnetic coil. S102, calculate the discrete-time derivatives of the transient voltage sequence and the driving current sequence in the electrical parameter feature matrix, and extract the dynamic back electromotive force feature tensor characterizing the mechanical motion state of the valve core; S103, Align the dynamic back electromotive force feature tensor with the fluid pressure gradient array in terms of feature dimensions, input it into a pre-trained physical constraint dynamic hysteresis compensation network, and output the desired control current sequence. S104, calculate the dynamic sliding surface threshold based on the fluid pressure gradient array, and generate an asymmetric chattering pulse width modulation drive signal based on the desired control current sequence and the dynamic sliding surface threshold; S105, the asymmetric dithering pulse width modulation drive signal is sent to the bottom drive bridge of the electromagnetic coil to control the valve core opening of the pilot valve.

[0019] In some embodiments, the technical solution of this application operates within a hardware architecture comprising an embedded microcontroller, a high-frequency analog-to-digital converter array, a full-bridge drive circuit, and an electro-hydraulic pilot valve. This architecture operates in a fluid control environment characterized by high temperature, high pressure, and drastic load fluctuations. Due to inherent constraints of space volume and extreme physical environment, high-precision miniature displacement sensors cannot be installed inside the pilot valve. The controller must rely solely on externally measurable electrical signals and macroscopic physical quantities at the fluid outlet to complete the conversion from the underlying data stream to the logic instruction stream within a sub-millisecond control cycle.

[0020] Specifically, the control cycle is set to 0.5ms to 2ms, the high-frequency analog-to-digital conversion sampling rate is not less than 1MHz, the full-bridge drive circuit supports rapid adjustment of the duty cycle from 0 to 100%, and the rated displacement of the pilot valve core is 0.1mm to 0.5mm.

[0021] In some embodiments, acquiring the electrical parameter feature matrix and fluid pressure gradient array of the pilot valve in the current control cycle includes: synchronously acquiring terminal voltage data and loop current data of the electromagnetic coil according to a preset high-frequency sampling clock; converting the terminal voltage data and loop current data into a continuous discrete time series to construct a two-dimensional electrical parameter feature matrix; acquiring pressure sensing data arranged on the fluid outlet side of the pilot valve; performing a first-order difference operation on the pressure sensing data to generate a fluid pressure gradient array reflecting the transient fluctuation rate of pressure; wherein, the frequency of the high-frequency sampling clock is set to a value much greater than the current response frequency of the electromagnetic coil to ensure that small transient distortions in the current waveform can be captured.

[0022] Specifically, the time axis and physical characteristic axis of the two-dimensional electrical parameter characteristic matrix are orthogonal to each other, accurately mapping the dynamic establishment process of electromagnetic field energy inside the coil; the first-order differential operation filters out the system-level back pressure background interference in the fluid pipeline and directly purifies the high-frequency component of fluid pulsation caused by the valve core movement.

[0023] Specifically: the high-frequency sampling clock frequency is 10kHz to 100kHz; the dimension of the electrical parameter feature matrix is ​​2×N, where N is the number of sampling points per cycle; the fluid pressure gradient array is calculated by the ratio of the pressure difference between adjacent sampling points to the time difference.

[0024] This step achieves forced alignment of heterogeneous signals in the time dimension from the underlying data structure through high-frequency electrical feature extraction and differential preprocessing of fluid pressure.

[0025] In some embodiments, extracting the dynamic back electromotive force characteristic tensor characterizing the mechanical motion state of the valve core includes: obtaining the static equivalent DC resistance parameter and inductance parameter of the electromagnetic coil; substituting the transient voltage sequence and driving current sequence into the discretized variant of Kirchhoff's voltage equation, and removing the ohmic voltage drop component and the inductive voltage drop component; extracting the remaining voltage component as the transient back electromotive force value; combining the transient back electromotive force values ​​corresponding to multiple consecutive sampling points into a one-dimensional vector, and mapping it to the dynamic back electromotive force characteristic tensor; wherein, the ohmic voltage drop component characterizes the pure resistive loss of coil heating, and the inductive voltage drop component characterizes the energy storage during the magnetic field establishment process.

[0026] The discretized variant of Kirchhoff's voltage equation is as follows: ; In the formula, The high-frequency sampling period is a preset sampling time interval, measured in milliseconds. The ohmic voltage drop component, in volts (V), is the voltage drop caused by current flowing through the coil resistance. This is the inductive voltage drop component, measured in volts (V), which is the induced voltage drop caused by changes in coil current. It is the transient back electromotive force, with the unit V. It is the reverse induced voltage generated by the valve core's movement cutting the magnetic field and is directly related to the valve core's motion state. for The voltage across the coil at any given moment, in volts (V). for The coil drive current at any given moment, in amperes (A). for The coil drive current at any given moment, in amperes (A). Specifically, according to Faraday's law of electromagnetic induction, after eliminating the two electromagnetic energy losses mentioned above, the remaining transient back electromotive force is strictly proportional to the relative velocity of the armature cutting the magnetic field lines at the physical level; the dynamic back electromotive force characteristic tensor is the lossless mirror projection of the valve core's microscopic velocity in the electrical dimension.

[0027] Specifically, the discretized Kirchhoff voltage equation is as follows:

[0028]

[0029] In the formula, The back electromotive force is a sequence of back electromotive forces combined into a 1×N tensor, which serves as a characteristic of the valve core's motion velocity. The voltage across the coil terminals is the voltage across the electromagnetic coil at the current sampling point. The driving current at the current sampling point; The resistance of the coil is the static DC resistance of the electromagnetic coil. For inductance, the static inductance of the electromagnetic coil; The rate of change of current is the ratio of the difference in current between adjacent sampling points to the time interval.

[0030] This step utilizes the differential state mapping mechanism of voltage and current to directly characterize the movement speed of the valve core using the back electromotive force value. This design breaks the inherent coupling contradiction of traditional control that must rely on physical displacement sensors, and avoids the hysteresis and failure risk of mechanical sensors.

[0031] In some embodiments, the dynamic back-EMF feature tensor and the fluid pressure gradient array are input into a pre-trained physically constrained dynamic hysteresis compensation network to output a desired control current sequence. This includes: mapping the dynamic back-EMF feature tensor to a valve core velocity prior feature and the fluid pressure gradient array to a hydrodynamic disturbance feature through an input mapping layer; calculating the dot product similarity matrix between the valve core velocity prior feature and the hydrodynamic disturbance feature in a fluid-electromagnetic cross-attention layer, and generating dynamic weight coefficients characterizing the hysteresis nonlinearity based on the dot product similarity matrix; fusing the dynamic weight coefficients with the valve core velocity prior feature and inputting them into a compensation decoding layer; and outputting a desired control current sequence to offset transient hysteresis errors based on the fused features. The valve core velocity prior feature and the hydrodynamic disturbance feature are distributed in different latent feature spaces, and the dot product similarity matrix is ​​used to quantify the actual interference degree of fluid pressure fluctuations on the valve core motion resistance within microseconds.

[0032] The mapping formula is as follows: ; In the formula, The input mapping layer nonlinear transformation function is a preset neural network mapping function used to convert input features into a feature form that meets the requirements of subsequent processing. The valve core velocity prior feature is the feature vector characterizing the valve core motion velocity, which is obtained by mapping the dynamic back electromotive force feature tensor. The hydraulic dynamic disturbance characteristic is a feature vector characterizing the hydraulic dynamic disturbance of a hydraulic system, which is obtained by mapping the fluid pressure gradient array. The characteristic tensor of the dynamic back electromotive force; for The fluid pressure gradient at time t.

[0033] Specifically, when the fluid pressure changes abruptly, the hydrodynamic disturbance characteristics increase dramatically, and the fluid-electromagnetic cross-attention layer adaptively amplifies the dynamic weight coefficient; after receiving the amplified weight instruction, the compensation decoding layer immediately feeds forward to output the distorted compensation current waveform, thus offsetting the upcoming peak value of viscous friction in advance.

[0034] This step utilizes a cross-attention mechanism and adaptive weight adjustment to enable hysteresis compensation to accurately adapt to dynamic changes in fluid pressure, proactively offsetting hysteresis errors. This significantly improves the accuracy of the control current, effectively suppresses hysteresis nonlinear interference in valve core movement, and ensures the stability of valve core movement. In some embodiments, the physically constrained dynamic hysteresis compensation network constructs a customized physically constrained loss function during the training phase. The execution logic includes: calculating the mean square error between the predicted current sequence output by the model and the reference current sequence as the basic fitting loss; calculating the overshoot between the theoretically derived fluid pressure value and the target fluid pressure value under the action of the predicted current sequence; triggering a fluid pressure overshoot penalty term when the overshoot is greater than a preset steady-state error threshold; weighting and summing the basic fitting loss and the fluid pressure overshoot penalty term to calculate the total loss value; and executing the backpropagation algorithm based on the total loss value to update the network node weights of the input mapping layer, the fluid-electromagnetic cross-attention layer, and the compensation decoding layer.

[0035] Specifically, conventional deep learning models only focus on the mathematical fit between input data and output labels, which can easily lead to dangerous mutations that violate the laws of fluid mechanics. This application forcibly incorporates specific physical boundary conditions into the construction logic of the network hierarchy and loss function. The fluid pressure overshoot penalty term prompts the model to prioritize reducing transient surge currents that may cause pipeline rupture during gradient updates, thus constructing a deep mapping relationship between the deep learning algorithm structure and specific physical data and hardware constraints.

[0036] This step integrates fluid dynamics principles and hardware safety constraints into network training by using a customized physical constraint loss function. This ensures that the control commands output by the model meet both mathematical fitting accuracy and actual physical operating principles, thereby avoiding the generation of dangerous mutation commands and improving the safety and reliability of system operation.

[0037] In some embodiments, calculating the dynamic sliding surface threshold based on the fluid pressure gradient array includes: extracting peak gradient data and fluctuation frequency data from the fluid pressure gradient array; multiplying the peak gradient data by the basic friction compensation coefficient to obtain a steady-state damping term; substituting the fluctuation frequency data into a preset exponential decay function to obtain a transient disturbance term; adding the steady-state damping term and the transient disturbance term to generate the dynamic sliding surface threshold; and generating an asymmetric chattering pulse width modulation drive signal based on the desired control current sequence and the dynamic sliding surface threshold, including: converting the desired control current sequence into a basic duty cycle sequence; determining whether the absolute value of the fluid pressure gradient array is greater than the dynamic sliding surface threshold; generating a high-frequency perturbation waveform when the absolute value of the fluid pressure gradient array is greater than the dynamic sliding surface threshold; and injecting the high-frequency perturbation waveform into the rising edge region of the basic duty cycle sequence in an asymmetric superposition manner to generate the asymmetric chattering pulse width modulation drive signal; wherein the basic duty cycle sequence is used to provide macroscopic electromagnetic thrust to maintain the opening of the main valve core, and the high-frequency perturbation waveform is an additional high-frequency low-amplitude energy wave.

[0038] Specifically, the asymmetric superposition method strictly limits the intervention of the high-frequency perturbation waveform to the instant of pulse signal excitation, and to its withdrawal when the electromagnetic thrust remains in a flat-top state or in a discharge state. This selective injection strategy avoids meaningless surges in coil heating and power consumption.

[0039] This step uses dynamic sliding surface threshold calculation and asymmetric perturbation superposition to enable the drive signal to adaptively adjust according to fluid pressure fluctuations. While ensuring the stability of the valve core's macroscopic opening, it avoids wasting coil power consumption, thus balancing control effect and energy consumption control, and improving the system's energy efficiency and stability.

[0040] In some embodiments, the process of generating high-frequency perturbation waveforms can be specifically achieved by obtaining the static frictional mechanical parameters between the pilot valve core and the valve sleeve, calculating the critical excitation frequency that can disrupt the surface tension of the hydraulic oil film based on the static frictional mechanical parameters, and generating a sinusoidal or triangular perturbation wave with a preset amplitude as the high-frequency perturbation waveform using the critical excitation frequency as the reference frequency; wherein the waveform slope of the sinusoidal or triangular perturbation wave is precisely adjusted to match the shear yield limit of the fluid oil film. Specifically, when the fluid pressure changes drastically and triggers the dynamic sliding surface threshold, the system determines that the valve core is trapped in the hysteresis dead zone. At this time, the injected high-frequency micro-perturbation waveform will excite high-frequency magnetic field chatter during the current rise period of the bottom drive bridge. The micro-electromagnetic excitation force generated by the high-frequency magnetic field chatter directly destroys the static friction lock state of the valve core surface.

[0041] In this step, the asymmetric chattering strategy does not change the macroscopic opening of the pilot valve, but rather substantially alters the underlying hardware state. This technique forcibly converts the dead zone friction of the mechanical actuator from static friction to dynamic friction, eliminating the start-up dead zone of the physical mechanism. This design, which deeply binds the underlying signal-driven reconstruction with the abrupt change in the hardware physical state, is not a conventional method used in this field for routine parameter tuning or ordinary pulse width modulation waveform generation.

[0042] In some embodiments, after the asymmetric jitter pulse width modulation drive signal is sent to the bottom drive bridge of the electromagnetic coil, the method further includes: real-time monitoring of the actual output current waveform of the bottom drive bridge; calculating the phase delay time between the actual output current waveform and the asymmetric jitter pulse width modulation drive signal; and feeding the phase delay time back to the physical constraint type dynamic hysteresis compensation network for advance compensation of the phase of the desired control current sequence in the next control cycle; wherein, the phase delay time includes the combined physical delay of the circuit board wiring distributed capacitance and the switching time of the power switch.

[0043] Specifically, by advancing the execution of the next predicted current command by the phase delay time on the time axis, the execution dead zone at the underlying hardware level is completely eliminated.

[0044] In summary, the technical solution provided in this application goes beyond simply adding macroscopic control loops or applying simple algorithms. This application constructs a refined physical control system across the entire chain, from voltage and current back electromotive force analysis to physical constraint cross-attention network prediction, and then to critical excitation frequency asymmetric waveform injection. There are strong interactions between the various technical features, and they support each other functionally. The algorithm execution substantially changes the electromagnetic drive state of the coil's underlying hardware and the force and friction mode of the valve core, possessing specific technical correlations. It significantly overcomes the pressure oscillation problem caused by the nonlinear coupling of fluid hydrodynamics and mechanical friction in pilot valves under complex working conditions, achieving unexpected stable control effects.

[0045] Example 2: Refer to Figure 2 As shown, a pilot valve multi-parameter real-time monitoring and control system includes: The multi-source parameter acquisition module acquires the electrical parameter feature matrix and fluid pressure gradient array of the pilot valve in the current control cycle. The electrical parameter feature matrix includes the transient voltage sequence and drive current sequence across the electromagnetic coil. The dynamic feature extraction module calculates the discrete-time derivatives of the transient voltage sequence and the driving current sequence in the electrical parameter feature matrix, and extracts the dynamic back electromotive force feature tensor characterizing the mechanical motion state of the valve core. The hysteresis compensation and desired current generation module aligns the dynamic back electromotive force feature tensor with the fluid pressure gradient array in terms of feature dimensions, inputs it into a pre-trained physical constraint dynamic hysteresis compensation network, and outputs the desired control current sequence. The drive signal generation module calculates the dynamic sliding surface threshold based on the fluid pressure gradient array, and generates an asymmetric chattering pulse width modulation drive signal based on the desired control current sequence and the dynamic sliding surface threshold. The valve core opening control module sends the asymmetric dithering pulse width modulation drive signal to the bottom drive bridge of the electromagnetic coil to control the valve core opening of the pilot valve.

[0046] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for real-time monitoring and control of multiple parameters of a pilot valve, characterized in that, include: S101, Obtain the electrical parameter feature matrix and fluid pressure gradient array of the pilot valve in the current control cycle. The electrical parameter feature matrix includes the transient voltage sequence and drive current sequence across the electromagnetic coil. S102, calculate the discrete-time derivatives of the transient voltage sequence and the driving current sequence in the electrical parameter feature matrix, and extract the dynamic back electromotive force feature tensor characterizing the mechanical motion state of the valve core; S103, Align the dynamic back electromotive force feature tensor with the fluid pressure gradient array in terms of feature dimensions, input it into a pre-trained physical constraint dynamic hysteresis compensation network, and output the desired control current sequence. S104, calculate the dynamic sliding surface threshold based on the fluid pressure gradient array, and generate an asymmetric chattering pulse width modulation drive signal based on the desired control current sequence and the dynamic sliding surface threshold; S105, the asymmetric dithering pulse width modulation drive signal is sent to the bottom drive bridge of the electromagnetic coil to control the valve core opening of the pilot valve.

2. The pilot valve multi-parameter real-time monitoring and control method according to claim 1, characterized in that, Obtain the electrical parameter characteristic matrix and fluid pressure gradient array of the pilot valve in the current control cycle, including: According to the preset high-frequency sampling clock, the terminal voltage data and loop current data of the electromagnetic coil are collected synchronously. The terminal voltage data and the loop current data are converted into continuous discrete time series to construct a two-dimensional electrical parameter feature matrix; Acquire pressure sensing data located on the fluid outlet side of the pilot valve; The pressure sensing data is subjected to a first-order difference operation to generate the fluid pressure gradient array that reflects the transient fluctuation rate of pressure.

3. The pilot valve multi-parameter real-time monitoring and control method according to claim 1, characterized in that, Extract the dynamic back electromotive force characteristic tensor characterizing the mechanical motion state of the valve core, including: Obtain the static equivalent DC resistance and inductance parameters of the electromagnetic coil; Substituting the transient voltage sequence and the driving current sequence into the discretized variant of the Kirchhoff voltage equation, and eliminating the ohmic voltage drop component and the inductive voltage drop component, the discretized variant of the Kirchhoff voltage equation is as follows: ； In the formula, The high-frequency sampling period is a preset sampling time interval, measured in milliseconds. The ohmic voltage drop component, in volts (V), is the voltage drop caused by current flowing through the coil resistance. This is the inductive voltage drop component, measured in volts (V), which is the induced voltage drop caused by changes in coil current. It is the transient back electromotive force, with the unit V. It is the reverse induced voltage generated by the valve core's movement cutting the magnetic field and is directly related to the valve core's motion state. for The voltage across the coil at any given moment, in volts (V). for The coil drive current at any given moment, in amperes (A). for The coil drive current at any given moment, in amperes (A). The remaining voltage component is extracted as the transient back electromotive force value; The transient back electromotive force values ​​corresponding to multiple consecutive sampling points are combined into a one-dimensional vector and mapped to the dynamic back electromotive force feature tensor.

4. The pilot valve multi-parameter real-time monitoring and control method according to claim 1, characterized in that, The physically constrained dynamic hysteresis compensation network includes an input mapping layer, a fluid-electromagnetic cross-attention layer, and a compensation decoding layer. It inputs the dynamic back electromotive force feature tensor and the fluid pressure gradient array into a pre-trained physically constrained dynamic hysteresis compensation network and outputs the desired control current sequence, including: The physically constrained dynamic hysteresis compensation network includes an input mapping layer, a fluid-electromagnetic cross-attention layer, and a compensation decoding layer. It inputs the dynamic back electromotive force feature tensor and the fluid pressure gradient array into a pre-trained physically constrained dynamic hysteresis compensation network and outputs the desired control current sequence, including: The input mapping layer maps the dynamic back electromotive force feature tensor to a priori valve core velocity features and the fluid pressure gradient array to hydrodynamic disturbance features, wherein the mapping formula is: ； In the formula, The input mapping layer nonlinear transformation function is a preset neural network mapping function used to convert input features into a feature form that meets the requirements of subsequent processing. The valve core velocity prior feature is the feature vector characterizing the valve core motion velocity, which is obtained by mapping the dynamic back electromotive force feature tensor. The hydraulic dynamic disturbance characteristic is a feature vector characterizing the hydraulic dynamic disturbance of a hydraulic system, which is obtained by mapping the fluid pressure gradient array. The characteristic tensor of the dynamic back electromotive force; for The fluid pressure gradient at time t; In the fluid-electromagnetic cross-attention layer, the dot product similarity matrix between the valve core velocity prior feature and the hydrodynamic disturbance feature is calculated, and dynamic weighting coefficients characterizing the hysteresis nonlinearity intensity are generated based on the dot product similarity matrix. The dynamic weighting coefficients are fused with the prior features of the valve core velocity and then input into the compensation decoding layer. The compensation decoding layer outputs the desired control current sequence to offset transient hysteresis errors based on the fused features.

5. The pilot valve multi-parameter real-time monitoring and control method according to claim 4, characterized in that, The physical constraint-based dynamic hysteresis compensation network constructs a customized physical constraint loss function during the training phase. The execution logic of the customized physical constraint loss function includes: The mean square error between the predicted current sequence output by the calculation model and the reference current sequence is used as the basic fitting loss. Calculate the overshoot between the theoretically derived fluid pressure value and the target fluid pressure value under the action of the predicted current sequence; When the overshoot exceeds a preset steady-state error threshold, a fluid pressure overshoot penalty is triggered. The total loss value is calculated by weighting and summing the basic fitting loss with the fluid pressure overshoot penalty term. Based on the total loss value, the backpropagation algorithm is executed to update the network node weights of the input mapping layer, the fluid-electromagnetic cross-attention layer, and the compensation decoding layer.

6. The pilot valve multi-parameter real-time monitoring and control method according to claim 1, characterized in that, The dynamic sliding surface threshold is calculated based on the fluid pressure gradient array, including: Extract the peak gradient data and fluctuation frequency data from the fluid pressure gradient array; Multiply the peak gradient data by the basic friction compensation coefficient to obtain the steady-state damping term; Substitute the fluctuation frequency data into a preset exponential decay function to obtain the transient disturbance term; The steady-state damping term is added to the transient disturbance term to generate the dynamic sliding surface threshold.

7. The pilot valve multi-parameter real-time monitoring and control method according to claim 1, characterized in that, Based on the desired control current sequence and the dynamic sliding surface threshold, an asymmetric jitter pulse width modulation driving signal is generated, including: The desired control current sequence is converted into a base duty cycle sequence; Determine whether the absolute value of the fluid pressure gradient array is greater than the dynamic sliding surface threshold; When the absolute value of the fluid pressure gradient array is greater than the dynamic sliding surface threshold, the static frictional mechanical parameters between the pilot valve core and the valve sleeve are obtained. The critical excitation frequency that can disrupt the surface tension of the hydraulic oil film is calculated based on the aforementioned static tribological parameters. Using the critical excitation frequency as the reference frequency, a sinusoidal micro-perturbation wave or a triangular micro-perturbation wave with a preset amplitude is generated as the high-frequency micro-perturbation waveform; The high-frequency perturbation waveform is injected into the rising edge region of the basic duty cycle sequence in an asymmetric superposition manner to generate the asymmetric dithering pulse width modulation drive signal.

8. The pilot valve multi-parameter real-time monitoring and control method according to claim 7, characterized in that, Injecting the high-frequency perturbation waveform into the rising edge region of the base duty cycle sequence in an asymmetric superposition manner includes: Extract the leading rising phase of each pulse period in the basic duty cycle sequence; The amplitude of the high-frequency perturbation waveform is algebraically added to the basic duty cycle only during the leading rise phase; During the peak and fall phases of the pulse cycle, the injection of the high-frequency perturbation waveform is shielded.

9. The pilot valve multi-parameter real-time monitoring and control method according to claim 1, characterized in that, After the asymmetric jitter pulse width modulation drive signal is sent to the bottom drive bridge of the electromagnetic coil, the method further includes: Real-time monitoring of the actual output current waveform of the underlying drive bridge; Calculate the phase delay time between the actual output current waveform and the asymmetric jitter pulse width modulation drive signal; The phase delay time is fed back to the physically constrained dynamic hysteresis compensation network for advance compensation of the phase of the desired control current sequence in the next control cycle.

10. A pilot valve multi-parameter real-time monitoring and control system, characterized in that, include: The multi-source parameter acquisition module acquires the electrical parameter feature matrix and fluid pressure gradient array of the pilot valve in the current control cycle. The electrical parameter feature matrix includes the transient voltage sequence and drive current sequence across the electromagnetic coil. The dynamic feature extraction module calculates the discrete-time derivatives of the transient voltage sequence and the driving current sequence in the electrical parameter feature matrix, and extracts the dynamic back electromotive force feature tensor characterizing the mechanical motion state of the valve core. The hysteresis compensation and desired current generation module aligns the dynamic back electromotive force feature tensor with the fluid pressure gradient array in terms of feature dimensions, inputs it into a pre-trained physical constraint dynamic hysteresis compensation network, and outputs the desired control current sequence. The drive signal generation module calculates the dynamic sliding surface threshold based on the fluid pressure gradient array, and generates an asymmetric chattering pulse width modulation drive signal based on the desired control current sequence and the dynamic sliding surface threshold. The valve core opening control module sends the asymmetric dithering pulse width modulation drive signal to the bottom drive bridge of the electromagnetic coil to control the valve core opening of the pilot valve.

Images