Flow regulating valve servo force control method and system based on non-force sensor

Abstract

The invention relates to the technical field of intelligent control, provides a flow regulating valve servo force control method and system based on a force sensor, and aims to solve the technical problems of response delay, weak overshoot suppression capability and poor long-term operation stability. The method comprises the following steps: acquiring a servo driving current data set and a valve displacement track data set of a target regulating valve; performing pressure feature mapping processing on the servo driving current data set to generate a pressure fluctuation feature set corresponding to the driving current waveform data; the pressure fluctuation characteristic set and the valve displacement track data set are input into a preset force control decision model for dynamic matching processing, and a servo control instruction set is generated; executing multi-stage dynamic adjustment operation on a servo driving unit of the target adjusting valve according to the servo control instruction set, and generating real-time pressure balance state data; and iteratively updating the dynamic matching processing parameters of the force control decision model based on the deviation value of the real-time pressure balance state data and the preset pressure reference value.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent control, and particularly relates to a servo force control method and system for a flow regulating valve based on a forceless sensor. Background Art

[0002] In the field of industrial fluid control, the accurate perception of the pressure state is the key to realizing the closed-loop control of the flow regulating valve. Traditional methods usually rely on pressure sensors to directly measure the pressure parameters of the oil rail or pipeline, and adjust the valve opening based on the feedback signal. However, such solutions have significant limitations: First, the pressure sensor is prone to signal drift or distortion under high-pressure and high-frequency impact conditions, especially in high-temperature and corrosive environments, and its long-term stability and reliability are difficult to guarantee; Second, the installation of the sensor is limited by the mechanical structure, resulting in a dynamic response delay, which is difficult to meet the transient pressure regulation requirements. In addition, the open-loop compensation strategy based on a fixed parameter model in the prior art lacks the ability to analyze the dynamic coupling relationship between electromagnetic force and fluid resistance, and is prone to overshoot or adjustment oscillation during complex working condition switching.

[0003] Although some studies have attempted to optimize the control algorithm through data-driven methods in recent years, the existing solutions still have the following defects: 1) relying on a single sensor signal and not effectively exploring the correlation of multi-source dynamic parameters of the servo system; 2) the feature extraction dimension is single and cannot accurately map the dynamic balance state during the pressure establishment stage; 3) the control model lacks an adaptive matching mechanism for time-varying working conditions and is easily affected by mechanical wear or environmental interference during long-term operation. These problems make it difficult for the prior art to meet the requirements of complex industrial scenarios in terms of response speed, anti-interference ability, and long-term stability. Summary of the Invention

[0004] The present invention provides a servo force control method and system for a flow regulating valve based on a forceless sensor, so as to overcome the technical problems of response delay, weak overshoot suppression ability, and poor long-term operation stability to a certain extent.

[0005] In a first aspect, an embodiment of the present invention provides a servo force control method for a flow regulating valve based on a force sensor, which is applied to a servo force control system of a flow regulating valve. The method includes: obtaining a servo drive current data set and a valve displacement trajectory data set of a target regulating valve, where the servo drive current data set includes drive current waveform data under different working conditions; performing pressure feature mapping processing on the servo drive current data set to generate a pressure fluctuation feature set corresponding to the drive current waveform data, where the pressure fluctuation feature set includes a current phase offset and a current amplitude attenuation coefficient; inputting the pressure fluctuation feature set and the valve displacement trajectory data set into a preset force control decision model for dynamic matching processing to generate a servo control instruction set, where the servo control instruction set includes a valve opening compensation value and a pressure balance adjustment coefficient; performing a multi-level dynamic adjustment operation on a servo drive unit of the target regulating valve according to the servo control instruction set to generate real-time pressure balance state data; and iteratively updating dynamic matching processing parameters of the force control decision model based on a deviation amount between the real-time pressure balance state data and a preset pressure reference value.

[0006] In a second aspect, an embodiment of the present invention provides a servo force control system for a flow regulating valve, which includes a processor and a memory. Among them, the memory stores a computer program, and when the computer program is executed by the processor, the processor is caused to execute the steps of the above method.

[0007] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium, which includes a computer program. When the computer program runs on a servo force control system of a flow regulating valve, the computer program is used to cause the servo force control system of the flow regulating valve to execute the steps of the above method.

[0008] It can be seen that the core innovation of the embodiment of the present invention is to construct an indirect perception system of the pressure state by deeply mining the current and displacement dynamic characteristics of the servo system itself, thereby significantly reducing the dependence on traditional mechanical sensors.

[0009] First of all, the embodiment of the present invention collects two types of native data of the servo system, namely, the servo drive current waveform and the valve displacement trajectory, with high precision, and uses time-frequency joint analysis technology to analyze the hidden pressure feature information therein. In the processing of the drive current waveform, a dynamic time warping algorithm is used to extract key indicators such as the current phase offset and the amplitude attenuation coefficient. These parameters can effectively reflect the dynamic balance state of electromagnetic force-fluid resistance in the fuel pressure establishment stage. For example, the slope change of the current rising edge can map the pressure establishment rate, and the ripple feature of the steady-state holding current has a strong correlation with the pressure pulsation of the high-pressure fuel rail. This mapping relationship realizes the accurate conversion from the current signal to the pressure value through a pre-trained pressure inversion model.

[0010] Secondly, the embodiment of the present invention realizes the closed-loop optimization of the pressure control strategy by constructing a force control decision-making model with multi-dimensional feature fusion, dynamically matching the current feature with the valve displacement trajectory. The acceleration feature of the valve displacement trajectory and the integral of the position tracking error, combined with parameters such as the current phase offset, jointly generate a servo control instruction with a feed-forward compensation function. This data-driven method not only avoids the signal distortion problem of traditional mechanical sensors under harsh working conditions such as high pressure and high-frequency impact, but also realizes the real-time inversion of the pressure state through in-depth analysis of the dynamic characteristics of the servo system itself. For example, under the condition of rapid acceleration, the risk of pressure overshoot is predicted in advance according to the high-frequency oscillation feature in the current waveform, and the valve opening compensation value is dynamically adjusted, rather than relying on the lag feedback of the pressure sensor, which significantly improves the transient response speed.

[0011] In addition, the embodiment of the present invention continuously optimizes the pressure feature mapping relationship through an iterative learning mechanism, forming an adaptive pressure control system. The deviation between the real-time pressure balance state and the preset reference value is used as a feedback signal to automatically correct the feature matching weight and the generation logic of the compensation coefficient of the force control decision-making model. This mechanism enables the pressure control accuracy not to depend on the long-term stability of the mechanical sensor, but to maintain the control efficiency through continuous monitoring of the dynamic characteristics of the servo system and model update. Under extreme working conditions of high temperature and high pressure, traditional pressure sensors are prone to drift or damage. However, the embodiment of the present invention utilizes the inherent anti-interference characteristics of servo current and displacement data, combined with a feature extraction algorithm based on physical mechanisms, to ensure the reliability and environmental adaptability of pressure control, providing a highly robust solution for complex industrial scenarios. In this way, when facing complex working conditions, the flow regulating valve control technology provided by the embodiment of the present invention can improve the technical problems of response delay, weak overshoot suppression ability, and poor long-term operation stability. Description of the Drawings

[0012] Figure 1 It is a schematic flow chart of a servo force control method for a flow regulating valve without a force sensor provided by an embodiment of the present invention.

[0013] Figure 2 It is a schematic structural diagram of a servo force control system for a flow regulating valve provided by an embodiment of the present invention. Detailed Embodiments

[0014] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, rather than all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments described in this document of the present invention without creative efforts shall fall within the scope of protection of the technical solutions of the present invention.

[0015] See Figure 1 , which is a servo force control method for a flow regulating valve based on a force sensor provided in an embodiment of the present invention. This method can be applied to a servo force control system of a flow regulating valve, and the specific process is as follows: Step 101 - Step 105.

[0016] Step 101: Obtain a servo drive current data set and a valve displacement trajectory data set of a target regulating valve. The servo drive current data set includes drive current waveform data under different working conditions.

[0017] In an exemplary application scenario, the target regulating valve is the core flow control component of a fuel injection system, and its servo drive current data set reflects the dynamic response characteristics of the drive unit under different operating states of the engine. Under the idle condition, the drive current waveform presents a periodic pulse sequence with low amplitude and narrow pulse width, corresponding to the fine adjustment requirement of the small opening of the valve core; while under the rapid acceleration condition, the current waveform shows a composite waveform of a steep rising edge and high-frequency oscillation superimposed, reflecting the dynamic coupling of electromagnetic force and mechanical inertia during the rapid opening of the valve core.

[0018] In addition, the valve displacement trajectory data set can synchronously record the position change of the valve core through a high-precision displacement sensor. For example, in a common rail system, the valve core needs to complete the accurate displacement from fully closed to a specified opening within a very short time, and its trajectory curve shows rapid response while suppressing overshoot.

[0019] In the embodiment of the present invention, the acquisition methods of the servo drive current data set and the valve displacement trajectory data set may include: based on the real-time signal acquisition module of the engine control unit (Electronic Control Unit, ECU), synchronously capture the drive current waveform and the valve core displacement amount with a microsecond-level time resolution, and classify and store the working conditions according to the injection cycle to form a complete data set covering cold start, steady-state operation, and transient load changes.

[0020] It can be understood that during the flow control process of the fuel injection system, the servo drive current data set of the target regulating valve and the valve displacement trajectory data set constitute the core observation indexes of the system dynamic characteristics. The drive current waveform data is captured in real time by the high-precision acquisition module of the engine control unit, which can reflect the energy input characteristics of the electromagnetic drive unit under different operating conditions. For example, in the idle condition, the control unit needs to precisely adjust the valve core to a small opening. At this time, the collected current waveform presents a periodic pulse sequence with a low amplitude and a narrow pulse width. This morphological feature is directly related to the micro-flow control requirements of the fuel system. When the engine enters the rapid acceleration state, the drive current waveform will change significantly, showing a high-frequency oscillation composite waveform with a steep rising edge. This dynamic characteristic reveals the complex coupling effect between the electromagnetic driving force and mechanical inertia. The data acquisition system synchronously records the drive current and the valve core displacement with a microsecond-level time resolution, and combines the working condition classification mechanism of the injection cycle to construct a complete data set covering cold start, steady-state operation and transient load changes, providing a multi-dimensional dynamic information basis for subsequent analysis and processing.

[0021] Step 102: Perform pressure characteristic mapping processing on the servo drive current data set to generate a pressure fluctuation characteristic set corresponding to the drive current waveform data. The pressure fluctuation characteristic set includes a current phase offset and a current amplitude attenuation coefficient.

[0022] In the embodiment of the present invention, the pressure characteristic mapping processing establishes a correlation model between the drive current characteristics and the fuel pressure fluctuation through signal analysis means. Among them, the current phase offset is used to quantify the synchronization deviation of the drive current waveform relative to the ideal injection timing. Exemplarily, when the fuel viscosity increases or there is wear in the valve core moving parts, an observable phase lag phenomenon will occur in the current waveform at the initial stage of fuel injection. This offset is positively correlated with the system pressure establishment delay. The current amplitude attenuation coefficient reflects the energy loss degree of the drive current during the valve core holding stage. For example, if the stiffness of the injector return spring decreases or the electromagnetic coil ages, the amplitude attenuation rate of the current during the closing stage will increase significantly, resulting in incomplete closing of the valve core and fuel pressure leakage.

[0023] For example, the exemplary methods of pressure characteristic mapping processing include: using time-frequency analysis technology to perform multi-scale decomposition on the current waveform to extract the frequency band energy distribution characteristics strongly related to the pressure fluctuation; comparing the morphological differences between the actual current waveform and the reference waveform through the dynamic time warping algorithm to generate a set of characteristic parameters representing pressure anomalies.

[0024] Specifically, when performing pressure characteristic mapping processing on servo drive current data, it is necessary to establish an association model between the current signal and fuel pressure fluctuations. This processing first extracts time-domain and frequency-domain characteristic indicators with physical significance from the current waveform. The time-domain characteristics include the current rising edge slope representing the electromagnetic force establishment speed and the steady-state holding duration reflecting the spool stability. The frequency-domain characteristics focus on analyzing parameters closely related to pressure vibration, such as the proportion of the third harmonic component and the fundamental wave amplitude volatility. These indicators are integrated into a composite current feature vector through a feature fusion algorithm. After inputting it into a pre-trained pressure inversion model, a real-time pressure estimate can be obtained. The training process of this pressure inversion model is based on current samples and pressure sensor data collected in a standard pressure test environment. Feature vectors are extracted through time-frequency joint analysis of the current waveform, and a mapping relationship from current features to pressure values is established using a neural network. In practical applications, by comparing and analyzing the real-time pressure estimate with the historical reference value, the current phase offset reflecting the system response delay and the current amplitude attenuation coefficient representing energy loss can be calculated. These pressure fluctuation characteristics provide a quantitative basis for formulating subsequent control strategies.

[0025] Step 103: Input the pressure fluctuation feature set and the valve displacement trajectory data set into a preset force control decision model for dynamic matching processing to generate a servo control instruction set, which includes a valve opening compensation value and a pressure balance adjustment coefficient.

[0026] In the embodiment of the present invention, the force control decision model is constructed based on the closed-loop control requirements of the fuel injection system, and the dynamic matching processing mechanism realizes the feedforward compensation of pressure fluctuations by coupling and analyzing current characteristics and spool movement states.

[0027] Among them, the generation logic of the valve opening compensation value is as follows: when the current phase offset exceeds a preset threshold, the force control decision model generates a directional opening compensation instruction according to the offset direction and the integral of the spool displacement tracking error. For example, in the scenario of lagging fuel injection pressure establishment, the positive compensation value can increase the initial movement speed of the spool to shorten the response time.

[0028] In addition, the pressure balance adjustment coefficient dynamically adjusts the proportional-integral-derivative control parameters of the drive unit by analyzing the correlation between the current amplitude attenuation rate and the fuel pressure oscillation amplitude. For example, in the case of overshoot of the high-pressure fuel rail pressure, the damping coefficient of the control loop is increased to suppress oscillation, or the integral action is enhanced to eliminate the steady-state error when the pressure response is slow. It is worth mentioning that the force control decision model can continuously optimize the matching rules through an online learning mechanism to ensure the real-time adaptation of control instructions to the dynamic characteristics of the system.

[0029] In specific implementation, during the dynamic matching and processing stage, the pressure fluctuation characteristics and valve displacement trajectory data are jointly analyzed. By analyzing the spool displacement curve, the displacement acceleration characteristics and the integral of the position deviation can be extracted. The former reflects the dynamic response ability of the spool movement, and the latter quantifies the cumulative effect of the position tracking error. These motion characteristics are subjected to multi-level coupling operations with the current phase offset and the amplitude attenuation coefficient to generate a preliminary compensation coefficient and a dynamic correction coefficient. After constructing a multi-dimensional compensation matrix, the strategy selector will search for the optimal control path within the three-dimensional constraint boundary according to environmental parameters such as the fluid medium density, the pipeline vibration amplitude, and the environmental temperature gradient. This search process uses an adaptive step size traversal algorithm to generate a set of candidate paths, and each path contains an axial displacement compensation sequence and a radial pressure adjustment coefficient sequence. By extracting the pressure fluctuation frequency and the amplitude attenuation rate from the real-time collected pipeline pressure transient feedback signal as evaluation indicators, the optimal path with the highest dynamic stability score can be selected, thereby determining the two key control parameters: the valve opening compensation value and the pressure balance adjustment coefficient.

[0030] Step 104: Perform a multi-level dynamic adjustment operation on the servo drive unit of the target regulating valve according to the servo control instruction set to generate real-time pressure balance state data.

[0031] In the embodiment of the present invention, the multi-level dynamic adjustment operation is implemented layer by layer according to the priority and timing requirements of the control instructions: the primary adjustment focuses on fast response. For example, the target position of the spool is corrected by feedforward according to the valve opening compensation value, and the rough adjustment of the fuel injection quantity is completed within milliseconds; the secondary adjustment is based on the pressure balance adjustment coefficient proportional-integral-derivative parameters, and the pressure tracking accuracy is improved by adjusting the gain characteristics of the drive current.

[0032] Among them, the real-time pressure balance state data is jointly collected by the rail pressure sensor and the high-frequency vibration sensor, and includes core indicators such as the pressure mean value, the fluctuation amplitude, and the spectrum characteristics. For example, after the adjustment is executed, the fuel pressure fluctuation amplitude converges from significantly exceeding the allowable range to a strictly defined threshold interval, and at the same time, the pressure build-up time is shortened to an acceptable level, and the system stability level is improved to a preset standard.

[0033] It can be understood that when performing multi-level dynamic adjustment operations, the control instructions are decomposed into axial displacement compensation components and radial pressure compensation components. The primary adjustment generates a dynamic pulse sequence through pulse width modulation technology to quickly correct the valve core position deviation; the secondary adjustment uses frequency-phase collaborative processing to generate a frequency-modulated carrier signal to finely adjust the pressure balance state. The composite servo drive signal is amplified by power and then drives the actuator, causing the valve core to simultaneously generate axial displacement and radial pressure compensation actions. During this process, the current sampling circuit continuously monitors the transient characteristics and steady-state ripple of the coil current. When the detected rising edge slope exceeds the limit or the ripple is distorted, a real-time correction mechanism is triggered to perform amplitude limiting and phase compensation processing on the control signal. The actual offset and fluctuation amplitude feedback by the displacement sensor and pressure transmitter are used to calculate the displacement following error and pressure balance deviation. These real-time status data provide closed-loop feedback for the iterative optimization of the control strategy.

[0034] Step 105: Based on the deviation between the real-time pressure balance state data and the preset pressure reference value, iteratively update the dynamic matching processing parameters of the force control decision model.

[0035] In the embodiment of the present invention, the preset pressure reference value is dynamically set according to the engine working conditions. For example, at idle speed, it corresponds to a low-amplitude and stable pressure curve, while at high load, it is mapped to a target pressure trajectory with a rapid climb. The deviation evaluation uses a multi-dimensional weighted algorithm to generate a quantitative evaluation result by comprehensively considering indicators such as pressure tracking error, overshoot, and oscillation frequency. The iterative update mechanism is exemplified as follows: when the deviation between the real-time pressure data and the reference value continuously remains in an unacceptable range, a model parameter self-correction process is triggered, and the feature matching weight or fuzzy rule confidence is adjusted through the gradient descent method until the pressure balance state returns to the optimal level. For example, the statistical variance of the pressure fluctuation amplitude is gradually reduced within multiple injection cycles, and at the same time, the standard deviation of the pressure build-up time is compressed to a significantly lower level of the original magnitude, ultimately enabling the force control decision model to adapt to the dynamic characteristic changes of the fuel injection system and achieve long-term stable control.

[0036] Specifically, based on the deviation analysis between the real-time pressure balance state and the preset reference value, the iterative update mechanism of the dynamic matching parameters can be started. Among them, the positive and negative polarities of the deviation direction determine whether to use the gradient ascent algorithm or the momentum optimization algorithm to adjust the model weight coefficient and integral threshold. The updated parameters need to be confirmed for the improvement effect through stability evaluation. If the improvement amplitude does not meet the expectation, the historical parameters are rolled back to maintain system stability. Further, the parameter update that successfully passes the evaluation will be synchronized to all running control model instances, and the pressure fluctuation amplitude is monitored in real time during the synchronization process to prevent system instability. The entire above update mechanism enables the force control decision model to adapt to the dynamic characteristic changes of the fuel injection system by continuously optimizing the feature matching rules and fuzzy control confidence, and ultimately achieves a comprehensive improvement in pressure build-up time shortening, fluctuation amplitude convergence, and long-term operation stability.

[0037] In addition, an abnormal fluctuation detection mechanism can be embedded during the actual application process to identify the mutation spikes in the current waveform through time-frequency joint analysis. For example, it can distinguish between the persistent abnormality caused by mechanical jamming and the transient interference caused by fluid impact, and trigger the servo current limiting instruction and the pressure buffering control strategy respectively. The valve self-cleaning program started for the mechanical jamming problem can effectively eliminate the resistance of the moving parts, while the response rate adjustment implemented for the fluid impact suppresses the pressure oscillation by dynamically adjusting the control parameters. The data set after abnormal processing is marked as the purified data set to ensure that subsequent feature extraction and control decisions are based on high-quality data. This innovative design concept combining closed-loop control and self-repair enables the servo force control of the flow regulating valve to maintain high-precision flow control and pressure balance capabilities under complex working conditions.

[0038] It can be seen that the core innovation of the embodiment of the present invention lies in constructing an indirect perception system of the pressure state by deeply mining the current and displacement dynamic characteristics of the servo system itself, thereby significantly reducing the dependence on traditional mechanical sensors.

[0039] First of all, the embodiment of the present invention analyzes the pressure characteristic information hidden therein by using the time-frequency joint analysis technology through high-precision acquisition of two types of native data of the servo system, namely, the servo drive current waveform and the valve displacement trajectory. In the processing of the drive current waveform, the dynamic time warping algorithm is used to extract key indicators such as the current phase offset and the amplitude attenuation coefficient. These parameters can effectively reflect the dynamic balance state of electromagnetic force-fluid resistance during the fuel pressure establishment stage. For example, the slope change of the current rising edge can map the pressure establishment rate, and the ripple characteristic of the steady-state holding current has a strong correlation with the pressure pulsation of the high-pressure fuel rail. This mapping relationship realizes the accurate conversion from the current signal to the pressure value through a pre-trained pressure inversion model.

[0040] Secondly, the embodiment of the present invention realizes the closed-loop optimization of the pressure control strategy by constructing a force control decision model with multi-dimensional feature fusion, dynamically matching the current characteristics with the valve displacement trajectory. The acceleration characteristic of the valve displacement trajectory and the integral of the position tracking error, combined with parameters such as the current phase offset, jointly generate a servo control instruction with a feed-forward compensation function. This data-driven method not only avoids the signal distortion problem of traditional mechanical sensors under harsh working conditions such as high pressure and high-frequency impact, but also realizes the real-time inversion of the pressure state through in-depth analysis of the dynamic characteristics of the servo system itself. For example, in the case of rapid acceleration, the risk of pressure overshoot is predicted in advance according to the high-frequency oscillation characteristics in the current waveform, and the valve opening compensation value is dynamically adjusted, rather than relying on the lag feedback of the pressure sensor, which significantly improves the transient response speed.

[0041] In addition, the embodiment of the present invention continuously optimizes the pressure characteristic mapping relationship through an iterative learning mechanism, forming an adaptive pressure control system. The deviation between the real-time pressure balance state and the preset reference value is used as a feedback signal to automatically correct the characteristic matching weights and compensation coefficient generation logic of the force control decision model. This mechanism enables the pressure control accuracy to no longer depend on the long-term stability of the mechanical sensor, but instead maintains the control efficiency through continuous monitoring of the dynamic characteristics of the servo system and model updates. In extreme working conditions of high temperature and high pressure, traditional pressure sensors are prone to drift or damage. However, the embodiment of the present invention utilizes the inherent anti-interference characteristics of servo current and displacement data, combined with a feature extraction algorithm based on physical mechanisms, to ensure the reliability and environmental adaptability of pressure control, providing a highly robust solution for complex industrial scenarios.

[0042] In a design concept, after obtaining the servo drive current data set and the valve displacement trajectory data set in step 101, it further includes: Step 1011: Perform abnormal fluctuation detection processing on the servo drive current data set to identify the sudden spike region in the current waveform.

[0043] During the operation of the fuel injection system, the servo drive current waveform of the target regulating valve may generate unexpected sudden spikes due to mechanical or fluid abnormalities. The abnormal fluctuation detection processing uses a dynamic threshold algorithm to scan the current signal segment by segment, and identifies the abnormal region by comparing the deviation degree of the sampling point amplitude within the current window with the historical average value. For example, when the amplitude of a certain segment of the current waveform continuously exceeds the preset fluctuation range, this region is marked as a candidate sudden spike region. During the detection process, a sliding window mechanism and a continuous trigger condition are combined to ensure that only spikes that meet the characteristics of time persistence and amplitude exceeding the standard are effectively identified, avoiding misjudgment caused by instantaneous noise interference. The identified spike region will be attached with a timestamp and a working condition label to provide positioning information for subsequent time-frequency analysis.

[0044] Step 1012: Perform a joint time-frequency analysis on the identified sudden spike region to determine whether the identified sudden spike region is caused by mechanical jamming or fluid impact.

[0045] In this step, the time-frequency joint analysis synchronously analyzes the time-domain morphological characteristics and frequency-domain energy distribution characteristics of the spike region by integrating the short-time Fourier transform and the discrete wavelet transform techniques. The spikes caused by mechanical jamming show persistent high-amplitude fluctuations in the time domain, and the frequency-domain energy is concentrated in the high-frequency band and decays slowly; the spikes caused by fluid impact show the time-domain characteristics of rapid rise and exponential decay, and the frequency-domain energy is mainly distributed in the low-frequency band and has a wide-band characteristic. Fault tracing is achieved by calculating the instantaneous frequency slope and the energy decay time constant of the spike region. For example, when it is detected that the proportion of high-frequency energy exceeds the critical threshold and the decay time exceeds the set range, it is determined as an abnormal type of mechanical jamming, otherwise it is classified as the influence of fluid impact.

[0046] Step 1013: When it is determined that the identified mutant spike region is caused by mechanical jamming, generate a servo drive current limiting command and trigger the valve self-cleaning program.

[0047] For example, for mechanical jamming abnormalities, a two-level processing strategy is implemented: First, the instantaneous peak value of the drive current is limited by dynamically adjusting the pulse width modulation parameters to prevent damage to the electromagnetic coil due to overload. The current limiting command calculates the safety threshold in real time according to the spike amplitude. For example, when an abnormal amplitude is detected, the output current is limited within a set proportion range of the rated value of the current working condition. Second, activate the valve self-cleaning program, drive the valve core to perform high-frequency micro-amplitude vibration by injecting a reverse pulse sequence with a specific frequency, and use the mechanical resonance effect to remove foreign objects in the kinematic pair. During the self-cleaning process, the displacement trajectory of the valve core is monitored synchronously, and the program is automatically terminated when the displacement fluctuation amplitude returns to the normal range to ensure the coherence of the control process.

[0048] Step 1014: When it is determined that the identified mutant spike region is caused by fluid impact, generate a pressure buffering control command and adjust the valve response rate.

[0049] Among them, the transient abnormality caused by fluid impact needs to be suppressed through the coordinated control of pressure buffering and dynamic response. The pressure buffering control command reduces the coupling strength between the valve core movement speed and the fuel pressure fluctuation by adjusting the damping coefficient and the integral time parameter of the proportional-integral-derivative control loop. At the same time, the rising edge slope of the drive current is dynamically adjusted to make the valve core opening rate match the fluid pressure establishment process. For example, during the high-pressure injection stage, by reducing the current rising rate, the valve core opening time is extended to reduce the impact effect caused by the change of the fuel pressure gradient. The adjustment amount of the control parameters is dynamically calculated according to the deviation between the real-time pressure estimation value and the target trajectory to ensure the adaptive optimization of the response.

[0050] Step 1015: Mark the servo drive current data set after the abnormal fluctuation detection and processing as the purified data set.

[0051] It can be understood that the current data after anomaly detection and processing needs to be subjected to integrity verification and quality marking. The interpolation algorithm is used to repair the removed anomaly data points, and the residual noise is eliminated through waveform smoothing processing. The data quality identifier is attached to the purified data set, recording the type of anomaly processing and the correction parameters. For example, the mechanical jamming correction area and its corresponding amplitude limit threshold are marked. This data set serves as the basic data source for subsequent feature extraction to ensure that the input data for pressure feature mapping processing meets the quality specification requirements.

[0052] In one implementation, the pressure feature mapping processing of the servo drive current data set in step 102 generates a pressure fluctuation feature set corresponding to the drive current waveform data, including: Step 1021: Extract the time-domain features and frequency-domain features of the drive current waveform data. The time-domain features include the current rising edge slope and the steady-state holding duration, and the frequency-domain features include the proportion of the third harmonic component and the fundamental wave amplitude volatility.

[0053] In this embodiment, the time-domain feature extraction focuses on the dynamic response characteristics of the drive current: the current rising edge slope is obtained by calculating the time differential in a specified amplitude range, reflecting the electromagnetic force establishment speed; the steady-state holding duration measures the duration of the current maintaining the target amplitude, characterizing the stability of the spool position. The frequency-domain feature analysis uses the fast Fourier transform to decompose the current waveform. The proportion of the third harmonic component reveals the nonlinear characteristics of the electromagnetic system, and the fundamental wave amplitude volatility quantifies the periodic stability of the drive energy. The sliding window mechanism is adopted in the feature extraction process to ensure representative feature parameters are obtained at different working condition stages.

[0054] Step 1022: Perform feature fusion processing on the time-domain features and frequency-domain features to generate a composite current feature vector.

[0055] It can be understood that the feature fusion processing realizes the organic integration of multi-dimensional features through normalization and weighted splicing. The time-domain features and frequency-domain features are respectively subjected to Z-score standardization processing to eliminate the influence of dimensional differences. The weight coefficients are assigned according to the correlation between the features and the pressure fluctuation. For example, a higher weight is given to the current rising edge slope to reflect its direct impact on pressure establishment. The fused composite feature vector constructs a multi-dimensional feature space, providing a unified input format containing time-frequency domain characteristics for the pressure inversion model.

[0056] Step 1023: Input the composite current feature vector into the pre-trained pressure inversion model to output the real-time pressure estimation value corresponding to the drive current waveform data.

[0057] In this step, the pre-trained pressure inversion model is constructed based on a neural network architecture, and a non-linear mapping relationship between the composite current features and the fuel pressure is established through offline training. The input layer of the model receives the standardized feature vector, and after non-linear transformation in the hidden layer, the pressure estimation value is output. During the online inference process, the model completes the conversion from the feature vector to the pressure value with a microsecond-level delay. For example, during the fuel injection process, the estimated curve of the rail pressure is output in real time. The weight parameters built into the model are solidified in the control unit memory to ensure the operation efficiency and determinacy.

[0058] Step 1024: Calculate the current phase offset and the current amplitude attenuation coefficient based on the difference between the real-time pressure estimation value and the historical pressure reference value.

[0059] It can be understood that the current phase offset is calculated through the timing deviation of the pressure estimation value relative to the target pressure curve, reflecting the difference in the synchronization established between the drive current and the fuel pressure. For example, when the pressure build-up is delayed, the phase offset shows a positive growth trend. The current amplitude attenuation coefficient is derived from the correlation model between the current decline rate and the pressure loss rate in the steady state stage, quantifying the degree of change in the electromagnetic energy conversion efficiency. The calculation process introduces a moving average filtering process to eliminate the influence of instantaneous fluctuations on the characteristic parameters.

[0060] Step 1025: Generate a pressure fluctuation feature set based on the current phase offset and the current amplitude attenuation coefficient.

[0061] In specific implementation, the pressure fluctuation feature set integrates the time series data of the phase offset and the attenuation coefficient in a structured data format, and synchronously records the corresponding operating condition parameters and environmental variables. The feature set adopts a time-aligned storage method, with each fuel injection cycle corresponding to a set of characteristic parameters, forming a multi-dimensional data set reflecting the dynamic characteristics of the system. This set serves as the core input of the force control decision model, providing a quantitative basis for optimizing the control strategy.

[0062] In a preferred implementation manner, the training process of the pre-trained pressure inversion model includes: Step 201: Collect the servo drive current sample data of the sample control valve in a standard pressure test environment and the measured data of the corresponding pressure sensor.

[0063] Exemplarily, a standard test platform can be built in a controlled experimental environment, and the drive current waveform and the rail pressure change curve are synchronously recorded through a high-precision data acquisition system. The test covers the full operating condition range, including typical scenarios such as cold start, idle speed, acceleration, and high load. Each test case contains a complete control instruction sequence, environmental parameter records, and sensor data, constructing a sample data set with statistical significance.

[0064] Step 202: Perform waveform segmentation processing on the servo drive current sample data to generate multiple current waveform segment data.

[0065] Among them, waveform segmentation divides the current waveform according to the phase characteristics of the fuel injection control signal, and each segment corresponds to a specific stage of the spool movement, such as the opening process, steady-state holding, and closing process. The edge detection algorithm is used in the segmentation process to automatically identify the waveform turning points, ensuring that each segment contains a complete dynamic response process. Phase tags and measured pressure values are added to the segmented waveform segments to form sample units required for supervised learning.

[0066] Step 203: Perform time-frequency joint analysis on the data of each current waveform segment, and extract the sample time-domain features and sample frequency-domain features.

[0067] Specifically, time-domain analysis and frequency-domain analysis are respectively performed on the segmented current waveform segments. The time-domain analysis extracts dynamic parameters such as the rising edge slope and steady-state holding duration, and the frequency-domain analysis calculates spectral features such as the third-harmonic ratio and fundamental-wave volatility. The standardized processing flow is adopted in the analysis process to ensure the comparability of the characteristic parameters among different samples. The extracted characteristic parameters and the corresponding pressure values form characteristic-label pairs for model training.

[0068] Step 204: Perform standardized splicing processing on the sample time-domain features and sample frequency-domain features to generate a sample composite feature vector.

[0069] Optionally, after the feature standardization eliminates the influence of different dimensions, the time-domain and frequency-domain features are spliced into a feature vector of a unified dimension in a preset order. The physical relevance between the features is retained during the splicing process. For example, the time-domain features reflecting the dynamic response and the frequency-domain features characterizing the system stability are arranged in order. The generated composite feature vector is used as the input of the machine learning model, and the pressure sensor data is used as the supervision signal to form a complete training sample.

[0070] Step 205: Using the sample composite feature vector as the input and the measured data of the pressure sensor as the output target, train the preset neural network model using the gradient descent algorithm until convergence.

[0071] In specific implementation, the neural network model iteratively optimizes the weight parameters through the backpropagation algorithm to minimize the mean square error between the predicted pressure and the measured value. The early stopping mechanism is adopted during the training process to prevent overfitting, and the training is terminated when the loss function of the validation set has not improved for multiple consecutive training rounds. The optimized model can accurately capture the nonlinear relationship between the current characteristics and the pressure fluctuation, and achieve high-precision pressure state estimation.

[0072] Step 206: Configure the trained preset neural network model as a pressure inversion model and solidify the model weight parameters.

[0073] As you can understand, the trained model weight parameters are quantized and embedded into the engine control unit's firmware to ensure efficient real-time inference. The model consolidation process includes precision optimization and memory alignment, ensuring that the neural network's forward propagation calculations meet microsecond real-time requirements. The consolidated pressure inversion model is integrated into the fuel injection control software, allowing the model interface to be directly called during online runtime to obtain pressure estimates.

[0074] In an alternative implementation, step 103 inputs the pressure fluctuation feature set and the valve displacement trajectory data set into a preset force control decision model for dynamic matching processing to generate a servo control instruction set, including: Step 1031: Perform motion feature analysis on the valve displacement trajectory data set to generate displacement acceleration features and position deviation integrals.

[0075] During the fuel injection control process, the dynamic characteristics of the valve core displacement trajectory directly affect the accuracy of fuel flow regulation. The motion feature analysis process extracts the second-order derivative of the displacement trajectory through differential operations, generating a displacement acceleration feature that reflects the transient changes in the valve core motion state. For example, under rapid acceleration conditions, the valve core needs to complete the full-close to full-open stroke within milliseconds, and its displacement acceleration feature presents a waveform that rises steeply and then converges rapidly. At the same time, the absolute value of the deviation between the actual valve core displacement and the target displacement curve is accumulated through integral operations to generate the position deviation integral. This parameter quantifies the cumulative effect of the control error. For example, when the valve core is delayed in opening due to mechanical resistance, the position deviation integral increases linearly with the injection cycle, providing a quantitative basis for subsequent compensation control.

[0076] Step 1032: Perform a first-level coupling operation on the displacement acceleration characteristics and the current phase offset to generate a preliminary compensation coefficient.

[0077] In this embodiment, the first-level coupling operation aims to establish a correlation model between the dynamic response of the valve core and the timing deviation of the electromagnetic drive. The operation convolves the time series of the displacement acceleration characteristics with the changing trend of the current phase offset to generate a preliminary compensation coefficient that reflects the degree of dynamic matching between the two. For example, when the current phase lag causes the valve core to open late, the coupling operation generates a positive compensation coefficient by amplifying the compensation weight of the high acceleration interval to improve the initial response rate of the drive current. This coefficient serves as the basic parameter of the feedforward control and directly affects the amplitude adjustment of the drive signal to ensure the timing synchronization of the valve core movement and the pressure buildup process.

[0078] Step 1033: Perform a second-stage coupling operation on the position deviation integral and the current amplitude attenuation coefficient to generate a dynamic correction coefficient.

[0079] In this embodiment, the second-level coupling operation focuses on the collaborative optimization of control error accumulation and energy loss. By multiplying the growth rate of the integral of the position deviation by the change in the current amplitude attenuation coefficient, a dynamic correction coefficient is generated. For example, when the current amplitude attenuation intensifies and the position deviation continues to accumulate, the dynamic correction coefficient shows an exponential growth trend, triggering the integral gain adaptive adjustment mechanism of the control loop. This coefficient acts on the integral term of the proportional-integral-derivative controller through a proportional scaling factor to achieve the gradual elimination of the steady-state error and the active suppression of energy loss.

[0080] Step 1034: Construct a multi-dimensional compensation matrix based on the preliminary compensation coefficient and the dynamic correction coefficient.

[0081] It can be understood that the multi-dimensional compensation matrix takes the time series as the row vector dimension and the compensation parameters as the column vector dimension, integrating the time-varying characteristics of the preliminary compensation coefficient and the dynamic correction coefficient. During the matrix construction process, the compensation coefficients are normalized and smoothed through a sliding window mechanism to eliminate instantaneous fluctuation interference. For example, the compensation coefficients collected within five consecutive fuel injection cycles are filtered by a Kalman filter to form a structured matrix containing timestamps, operating condition codes, and compensation parameter values. This matrix provides a decision basis with spatio-temporal characteristics for subsequent path optimization.

[0082] Step 1035: Perform an optimal path search process on the multi-dimensional compensation matrix through the policy selector in the force control decision model, and output the valve opening compensation value and the pressure balance adjustment coefficient.

[0083] Specifically, the policy selector searches for a control path that satisfies multi-objective constraints in the multi-dimensional compensation matrix based on the dynamic programming algorithm. During the search process, the row vectors of the compensation matrix are mapped to state nodes, and the column vectors are used as state transition weights. The response speed, energy consumption efficiency, and pressure stability indicators of each path are evaluated through a cost function. For example, the branch and bound algorithm is used to screen out the optimized path with the smallest pressure fluctuation amplitude and a controllable increase in energy consumption, and finally, a control instruction set containing the axial displacement compensation sequence and the radial pressure adjustment coefficient is output.

[0084] In another alternative implementation, the step of performing an optimal path search process on the multi-dimensional compensation matrix through the policy selector in the force control decision model in step 1035 and outputting the valve opening compensation value and the pressure balance adjustment coefficient includes: Step 10351: Obtain the fluid medium density, pipeline vibration amplitude, and ambient temperature gradient in the current operating condition environment parameter set; where the fluid medium density, pipeline vibration amplitude, and ambient temperature gradient are independent of each other.

[0085] In this embodiment, the environmental parameter set is synchronously collected by multi-source sensors. Among them, the fluid medium density is measured in real time by a fuel density meter, reflecting the physical property parameter fluctuations caused by changes in fuel temperature and composition; the pipeline vibration amplitude is obtained by an acceleration sensor, characterizing the mechanical vibration intensity caused by fuel pressure pulsation; the environmental temperature gradient is calculated by a distributed temperature sensor group, describing the non-uniformity of the axial temperature distribution of the valve body. The three types of parameters respectively construct control constraint conditions from the dimensions of fluid characteristics, mechanical state, and thermal environment.

[0086] Step 10352: Map the three orthogonal dimensions of the multi-dimensional compensation matrix to the fluid medium density, pipeline vibration amplitude, and environmental temperature gradient respectively based on the space coordinate system to obtain the medium density threshold line, vibration amplitude isosurface, and temperature gradient surface.

[0087] For example, the space coordinate system mapping adopts the parameter normalization method, mapping the fluid medium density to the X-axis coordinate, the vibration amplitude to the Y-axis coordinate, and the temperature gradient to the Z-axis coordinate. The medium density threshold line is defined as the upper density limit for the safe operation of the fuel injection system, forming a vertical boundary on the X-axis; the vibration amplitude isosurface is calculated by frequency domain energy integration, forming a gradient-distributed surface in the Y-Z plane; the temperature gradient surface is generated by polynomial fitting, characterizing the three-dimensional distribution characteristics of the axial heat conduction of the valve body. This mapping transforms the abstract environmental parameters into visual geometric constraint boundaries.

[0088] Step 10353: Divide a three-dimensional constraint boundary composed of the medium density threshold line, vibration amplitude isosurface, and temperature gradient surface within the space coordinate system.

[0089] For example, the three-dimensional constraint boundary is generated by combining and cutting geometric elements through Boolean operations. The medium density threshold line defines the maximum allowable density value along the X-axis, eliminating the over-limit area; the vibration amplitude isosurface divides different vibration intensity level areas along the Y-axis; the temperature gradient surface divides the thermal stable area and the high-temperature risk area along the Z-axis. The closed space formed by the intersection of the three is defined as the system safe operation domain, and the optimal path search is strictly restricted within this boundary to ensure the physical feasibility and system safety of the control strategy.

[0090] Step 10354: Use the adaptive step size traversal algorithm to generate a set of candidate control paths within the three-dimensional constraint boundary. Each candidate control path in the set of candidate control paths contains an axial displacement compensation sequence and a radial pressure adjustment coefficient sequence.

[0091] Specifically, the adaptive step-size traversal algorithm dynamically adjusts the search step size based on the gradient of environmental parameters. Large step sizes are used for rapid coverage in areas of gently varying density, while smaller step sizes are used for refined search in areas of drastic vibration amplitude fluctuations. Each candidate path consists of an axial displacement compensation sequence and a radial pressure adjustment coefficient sequence. The axial sequence defines the magnitude and timing of valve spool opening compensation, while the radial sequence sets the intensity and duration of pressure adjustment. A random perturbation factor is introduced into the path generation process to avoid local optimality traps.

[0092] Step 10355: Collect the transient feedback signal of the pipeline pressure sensor in real time, and extract the pressure fluctuation frequency and amplitude attenuation rate from the transient feedback signal as path evaluation indicators.

[0093] For example, the pressure fluctuation frequency is calculated through Fast Fourier Transform to reveal the periodic characteristics of fuel pressure oscillations. The amplitude decay rate is calculated through envelope fitting to obtain the exponential decay coefficient, characterizing the system's damping characteristics. For example, during high-pressure injection, the ideal pressure fluctuation frequency should match the inverse of the injection pulse width, and the amplitude decay rate must reach a preset critical value to ensure rapid pressure stabilization. These two metrics together constitute the core parameters for path stability assessment.

[0094] Step 10356: Input the pressure fluctuation frequency and amplitude attenuation rate into the path stability evaluation function to calculate the dynamic stability score of each candidate control path.

[0095] Specifically, the evaluation function uses a weighted summation formula to linearly combine the squared error of the pressure fluctuation frequency from the target value with the logarithmic gain of the amplitude decay rate. For example, if the frequency weight is set to 0.7 and the decay rate weight is set to 0.3, when a path has a frequency deviation of 5 Hz and a decay rate increase of 15%, its stability score is: (52 × 0.7) + (0.15 × 0.3) = 17.5 + 0.045 = 17.545. The score reflects the path's overall effectiveness in suppressing pressure oscillations and improving system stability.

[0096] Step 10357: Select the candidate control path with the highest dynamic stability score as the optimal path, and extract the mean value of the axial displacement compensation sequence of the optimal path as the valve opening compensation value.

[0097] The optimal path is selected based on the ranking results. The compensation amounts at each time point in the axial displacement compensation sequence are arithmetic averaged to generate a statistically representative valve opening compensation value. For example, an optimal path includes a compensation sequence of 10 time points [0.2mm, 0.25mm, …, 0.18mm]. The average value of 0.21mm is output as the final compensation command. This value is loaded into the drive circuit via a digital-to-analog converter to achieve precise correction of the valve core position.

[0098] Step 10358: Synchronously extract the peak maintenance coefficient of the radial pressure regulation coefficient sequence in the optimal path as the pressure balance regulation coefficient.

[0099] In this embodiment, the peak maintenance coefficient is calculated by analyzing the extreme value distribution characteristics of the radial pressure regulation coefficient sequence, specifically the product of the proportion of coefficients exceeding the set threshold in the sequence and their duration. For example, if the proportion of a certain sequence with a coefficient value higher than 0.8 within 5 ms reaches 80%, then the peak maintenance coefficient is 0.8×5 = 4.0. This coefficient is input into the proportional-integral-derivative controller as the core parameter for pressure balance regulation to dynamically adjust the gain characteristics of the drive current and ensure that the fuel pressure quickly converges to the target range.

[0100] As an alternative embodiment, the multi-level dynamic adjustment operation on the servo drive unit of the target regulating valve according to the servo control instruction set in step 104 generates real-time pressure balance state data, including: Step 1041: Input the valve opening compensation value in the servo control instruction set into the axial displacement compensator for component decomposition processing to generate an axial displacement compensation component and a radial pressure compensation component.

[0101] In the closed-loop control of the fuel injection system, the valve opening compensation value needs to be decomposed into two spatially orthogonal control components to meet the multi-dimensional adjustment requirements. The axial displacement compensator uses the orthogonal projection algorithm to decompose the compensation value into a displacement compensation component along the spool axis direction and a pressure compensation component perpendicular to the axis according to the geometric relationship between the spool movement direction and the fuel flow direction. For example, when the valve opening compensation value is 0.3 mm, the component decomposition processing calculates the linear displacement increment of the axial displacement compensation component and the equivalent force amplitude of the radial pressure compensation component according to the angle between the current spool position and the fuel injection direction. The decomposed components act on the spool displacement control loop and the pressure balance regulation loop respectively to achieve multi-degree-of-freedom collaborative control.

[0102] Step 1042: Input the axial displacement compensation component into the first-level pulse width modulator for dynamic duty cycle distribution processing to generate an axial drive pulse sequence with a dynamic pulse width.

[0103] For example, the first-level pulse width modulator dynamically adjusts the duty cycle characteristics of the pulse signal according to the amplitude change of the axial displacement compensation component. The modulator uses a duty cycle distribution algorithm based on slope detection to convert the compensation component into a time series signal corresponding to the pulse width. For example, when the axial compensation component increases, the modulator generates a pulse sequence with a gradually widened pulse width to drive the electromagnetic coil to generate a stronger axial driving force. The rising edge slope of the pulse sequence is synchronously adjusted with the change rate of the compensation component to ensure that the spool displacement acceleration matches the target trajectory. The modulated pulse sequence is output as a drive signal with millisecond-level time accuracy through the digital-to-analog conversion circuit.

[0104] Step 1043: Input the radial pressure compensation component into the second-stage carrier generator for frequency-phase collaborative processing to generate a radial frequency-modulated carrier signal that matches the current pressure balance adjustment coefficient.

[0105] For another example, the second-stage carrier generator adopts digital phase-locked loop technology to synchronize the radial pressure compensation component with the system clock reference to generate a modulated carrier signal with adjustable frequency and phase. The carrier frequency is dynamically adjusted according to the amplitude of the pressure balance adjustment coefficient. For example, when the adjustment coefficient increases, the carrier frequency is increased by a preset ratio to enhance the pressure regulation response speed. The phase synchronization mechanism ensures that the negative feedback phase of the carrier signal is 180 degrees different from the pressure oscillation phase by tracking the fuel pressure fluctuation period in real time, achieving active damping control. After the generated frequency-modulated carrier signal eliminates harmonic interference through a band-pass filter, a standardized pressure compensation control signal is formed.

[0106] Step 1044: Input the axial drive pulse sequence and the radial frequency-modulated carrier signal into the signal synthesis module for time-domain superposition processing to generate a composite servo drive control signal.

[0107] Exemplarily, the signal synthesis module adopts a time-domain linear superposition algorithm to align and merge the axial drive pulse sequence and the radial frequency-modulated carrier signal on the time axis. During the superposition process, the module eliminates the mutual interference between the two signals through a phase compensation circuit. For example, a microsecond-level delay is applied to the radial carrier signal to match the rising edge timing of the axial pulse. The synthesized composite signal contains a high-amplitude pulse component for axial displacement control and a continuous carrier component for radial pressure regulation, forming a composite drive waveform with dual-channel control characteristics. This signal is amplified by a differential amplifier circuit to increase the drive level and ensure meeting the power requirements of the servo unit.

[0108] Step 1045: Load the composite servo drive control signal to the actuator of the servo drive unit through a power amplifier to drive the spool to generate axial displacement action and radial pressure compensation action.

[0109] Among them, the power amplifier adopts an H-bridge topology to convert the composite drive control signal into a power output with sufficient drive current and voltage. The pulse signal of the axial displacement control channel is amplified to drive the linear motor to generate an accurate axial thrust to push the spool to achieve the target displacement; the carrier signal of the radial pressure regulation channel generates a periodic radial force through an electromagnetic actuator to offset the vibration of the spool caused by fuel pressure fluctuations. The two-degree-of-freedom movement of the actuator enables the spool to complete the opening adjustment and pressure stabilization functions simultaneously. For example, during the opening process, a radial damping force is superimposed to suppress the oscillation caused by fluid impact.

[0110] Step 1046: Real-time collect the spool displacement sensor data and the post-valve pressure transmitter data of the actuator, and generate a dynamic response data set containing the actual displacement offset and the pressure fluctuation amplitude.

[0111] Among them, the high-precision magnetostrictive displacement sensor measures the actual displacement of the spool in real time with a micron-level resolution, and the data acquisition system records the displacement trajectory curve at a sampling rate of 10 kHz. The piezoresistive pressure transmitter connected synchronously monitors the change of the fuel pressure after the valve and captures the high-frequency characteristics of the pressure fluctuation. After the collected displacement data and pressure data are aligned by the time stamp, a structured data set containing the actual displacement value, the instantaneous pressure value and their corresponding time marks is formed. For example, each fuel injection cycle generates a displacement-pressure data pair containing 500 sampling points, which completely records the dynamic process of the spool movement and the pressure response.

[0112] Step 1047: Perform deviation integral calculation on the actual displacement offset in the dynamic response data set and the valve displacement trajectory data set to generate a displacement following error value.

[0113] Specifically, the displacement following error value quantifies the cumulative deviation between the actual displacement and the target trajectory through integral operation. The trapezoidal integral algorithm is adopted in the calculation process. The absolute value of the actual displacement offset is calculated in each sampling period and accumulated and summed throughout the fuel injection cycle. For example, when the spool opening is delayed due to mechanical friction, the displacement offset is continuously a positive deviation in the initial stage, and the integral result shows a linear growth trend. This error value is input into the adaptive adjustment module as an evaluation index of control accuracy, providing a quantitative basis for the subsequent optimization of control parameters.

[0114] Step 1048: Perform a difference comparison process on the pressure fluctuation amplitude and the preset pressure reference value to generate a pressure balance deviation index.

[0115] It can be understood that the pressure balance deviation index is obtained by calculating the ratio of the real-time pressure fluctuation peak-to-peak value to the allowable fluctuation range of the reference pressure. During the processing, the local extreme points of the pressure data are extracted, and the difference between adjacent wave peaks and wave valleys is calculated as the instantaneous fluctuation amplitude, and then normalized with the reference value. For example, when the allowable fluctuation of the reference is ±0.5 MPa, and the actual fluctuation amplitude of 0.8 MPa is detected at a certain moment, the deviation index is (0.8 - 0.5) / 0.5 = 60%. This index reflects the deviation degree of the current pressure stability state from the ideal target and is used to evaluate the pressure control effect.

[0116] Step 1049: Generate real-time pressure balance state data according to the displacement following error value and the pressure balance deviation index.

[0117] Among them, the real-time pressure balance state data integrates the displacement and pressure control effect information through a data fusion algorithm. A two-dimensional state space is constructed, with the displacement following error value as the abscissa and the pressure balance deviation as the ordinate. The Euclidean distance between the current state point and the ideal origin is calculated as the comprehensive state index. For example, when the displacement error is 0.1 mm and the pressure deviation is 30%, the comprehensive state value is √(0.1² + 0.3²) = 0.316. This data also records the timestamp, working condition code, and environmental parameters simultaneously, forming a state data packet containing multi-dimensional evaluation results, providing complete feedback information for the closed-loop optimization of the control strategy.

[0118] As another alternative embodiment, after inputting the axial drive pulse sequence and the radial frequency-modulated carrier signal into the signal synthesis module for time-domain superposition processing to generate a composite servo drive control signal in step 1044, the method further includes: Step 301: Real-time capture the coil current waveform data of the servo drive unit through a current sampling circuit, and extract the transient rising edge slope and the steady-state holding section ripple amplitude of the coil current waveform data.

[0119] In the real-time monitoring of the servo drive unit, the current sampling circuit captures the transient and steady-state characteristics of the coil current at a high sampling rate. The transient rising edge slope calculates the change rate of the current rising from the initial value to the target value through differential operation, reflecting the dynamic establishment speed of the electromagnetic force; the steady-state holding section ripple amplitude measures the fluctuation amplitude of the current in the stable stage through the peak-to-peak detection algorithm, characterizing the energy conversion efficiency of the drive system. For example, in the rapid opening stage of the spool valve, the transient rising edge slope is directly related to the response speed of the electromagnetic coil, while the steady-state ripple amplitude reveals the filtering performance of the power supply and the coil impedance matching state. The data extraction process adopts a sliding window mechanism to ensure accurate capture of the key characteristic parameters of the current waveform under dynamic working conditions.

[0120] Step 302: Input the transient rising edge slope into the first comparator for overrun judgment processing with a preset transient safety threshold, and generate a transient slope overrun flag signal.

[0121] In the actual application process, the first comparator is built with a dynamic threshold adjustment algorithm, and dynamically sets the transient safety threshold according to the rated current value under the current working condition and historical operation data. When it is detected that the transient rising edge slope exceeds the threshold, the comparator outputs a high-level overrun flag signal, and this signal triggers the protection mechanism. For example, in the rapid acceleration working condition, if the slope exceeds the physical tolerance limit of the electromagnetic coil, the flag signal activates and immediately interrupts the overload risk. The threshold setting comprehensively considers the coil temperature rise characteristics and the withstand voltage level of the insulating material to ensure that the system operates within the safety boundary.

[0122] Step 303: When the transient slope overrun flag signal is in the active state, trigger the current suppression circuit to perform slope limiting processing on the pulse front edge of the composite servo drive control signal, and generate an axially driven pulse sequence with limited amplitude correction.

[0123] Specifically, the current suppression circuit adopts a cooperative control strategy of analog limiting and digital filtering, and dynamically adjusts the rising edge slope of the drive signal through a programmable gain amplifier. The limiting process is specifically manifested as: superimposing a reverse compensation current on the pulse front edge to limit the rising rate of the actual coil current within a preset safe range. The corrected axially driven pulse sequence keeps the target amplitude unchanged and only adjusts its time response characteristics. For example, limiting the original slope of 50 A / ms to 40 A / ms can prevent the coil from overheating and maintain the dynamic performance of the spool movement.

[0124] Step 304: Input the ripple amplitude in the steady-state holding section into a band-pass filter for spectrum separation processing to separate the effective ripple component matching the commutation frequency of the servo motor and the high-frequency noise component of electromagnetic interference.

[0125] Specifically, the passband range of the band-pass filter is set to the fundamental wave and third harmonic intervals of the commutation frequency of the servo motor, effectively retaining the characteristic ripple reflecting the commutation state of the motor. The high-frequency noise component is blocked outside the stopband. For example, filter out the switching noise from the PWM modulator and environmental electromagnetic interference. The separated effective ripple component is used to analyze the commutation loss of the motor, while the noise component is input into the electromagnetic compatibility monitoring module. The filter design uses an elliptic function type, and sets a steep attenuation characteristic in the transition band to ensure the separation accuracy.

[0126] Step 305: Input the effective ripple component and the high-frequency noise component of electromagnetic interference into a waveform matcher to calculate the morphological similarity with a preset ideal steady-state waveform template, and generate a waveform distortion correction coefficient.

[0127] Exemplarily, the waveform matcher adopts a dynamic time warping algorithm to non-linearly align and compare the measured ripple waveform with the ideal template. The similarity calculation evaluates the waveform morphological difference through a correlation coefficient matrix, and the distortion correction coefficient is generated according to the exponential law of the difference amplitude. For example, when there is a local depression distortion in the effective ripple, the correction coefficient linearly increases with the product of the depression depth and duration. This coefficient quantifies the distortion degree of the current waveform and provides a quantitative basis for subsequent phase compensation.

[0128] Step 306: Adjust the phase offset compensation amount of the second-stage carrier generator according to the waveform distortion correction coefficient to generate a phase-compensated radial frequency modulation carrier signal.

[0129] For example, the phase offset compensation amount is achieved by a digital phase shifter, and the compensation step is dynamically adjusted according to the distortion correction coefficient. When a positive waveform distortion is detected, the phase of the carrier signal is advanced to cancel the delay effect; a negative distortion triggers a phase lag compensation. The compensated carrier signal is precisely anti-phase aligned with the pressure fluctuation period. For example, the maximum reverse regulating force is applied at the pressure wave peak stage, significantly enhancing the active damping effect. The phase shift accuracy reaches the level of 0.1 degrees, ensuring effective suppression of high-frequency pressure oscillations.

[0130] Step 307: Input the radially frequency-modulated carrier signal after phase compensation and the axially driven pulse sequence input signal after amplitude limiting correction into the signal synthesis module for timing alignment and superposition processing to generate an optimized composite servo drive control signal.

[0131] In this step, the timing alignment processing adopts a hardware synchronous triggering mechanism to unify the starting phases of the two signals using the system clock reference. During the superposition process, the signal synthesis module eliminates crosstalk between signals through an impedance matching network and finely adjusts the relative positions of the pulse sequence and the carrier signal in the time domain. For example, the trough of the radially carrier signal is made to correspond to the rising edge of the axially pulse to maximize the effective utilization rate of the driving energy. The synthesized optimized signal combines the high dynamic characteristics of axial displacement control and the fine damping ability of radial pressure regulation.

[0132] Step 308: Load the optimized composite servo drive control signal onto a power amplifier to drive the actuator of the servo drive unit to generate corrected axial displacement actions and radial pressure compensation actions.

[0133] Combined with the above content, the power amplifier adopts a cascaded H-bridge topology structure to convert the optimized signal into a power output with sufficient driving ability. The axial drive channel precisely controls the output current amplitude through a current feedback loop to ensure the accurate execution of the pulse sequence after amplitude limiting correction; the radial regulation channel adopts a voltage follower mode to ensure the waveform fidelity of the carrier signal. The linear motor and the piezoelectric actuator of the actuator respectively respond to the axial and radial control components. For example, during the high-pressure injection stage, the axial component achieves rapid opening adjustment, and the radial component synchronously suppresses the oil rail pressure oscillation.

[0134] Step 309: Real-time collect the spool displacement sensor data of the actuator and update the actual displacement offset, and synchronously collect the post-valve pressure transmitter data and update the pressure fluctuation amplitude.

[0135] In this embodiment, the displacement sensor adopts the magnetic grating ruler measurement technology to update the spool position data in real time with a resolution of 0.1 micron. The pressure transmitter captures the transient fluctuations of the pressure behind the valve at a sampling rate of 20 kHz, and the data acquisition system realizes the strict synchronization of the displacement and pressure data through the hardware timestamp. The updated data set covers a complete control cycle. For example, 500 sets of displacement-pressure data pairs are generated within a 5-ms fuel injection cycle, accurately recording the real-time effect of the control strategy.

[0136] Step 310: Input the updated actual displacement offset into the deviation integration calculation module of the displacement following error value to generate the updated displacement following error value.

[0137] For example, the deviation integration module adopts the variable-step trapezoidal integration algorithm to adaptively adjust the integration step according to the change rate of the displacement offset. A small step is used in the acceleration stage of the spool movement to improve the calculation accuracy, and a large step is switched to in the steady state stage to reduce the operation load. The updated error value reflects the cumulative tracking deviation in the latest control cycle. For example, when the integration result increases by 15% compared with the previous cycle, it indicates that the integral action of the control loop needs to be enhanced.

[0138] Step 311: Perform a difference comparison process on the updated pressure fluctuation amplitude and the preset pressure reference value to generate the updated pressure balance deviation index.

[0139] Among them, the difference comparator adopts the window comparison mechanism to perform multiple local extreme value detections and compare with the reference value within a single control cycle. The deviation index is calculated by the weighted sum of the overrun times and the overrun amplitude. For example, when three transient pressure peaks exceeding 50% of the allowable fluctuation range are detected, the deviation index is raised to the warning level. This index dynamically reflects the pressure control effect and provides immediate feedback for online calibration.

[0140] Step 312: Feed back the updated displacement following error value and the updated pressure balance deviation index to the strategy selector of the force control decision model to trigger the online calibration of the dynamically matched processing parameters.

[0141] Furthermore, the feedback data is transmitted to the adaptive learning module of the strategy selector through the high-speed data bus. The online calibration adopts the incremental gradient descent algorithm to adjust the feature matching weight coefficient and the fuzzy rule confidence according to the error value and the deviation index. For example, when the displacement error persists as positive and the pressure deviation remains high, the compensation weight of the current phase offset is automatically increased. The calibration process is completed within milliseconds, ensuring that the control strategy adapts to the changes in the system dynamic characteristics in real time and maintaining the optimal control performance.

[0142] As an alternative embodiment, based on the deviation between the real-time pressure balance state data and the preset pressure reference value in step 105, iteratively update the dynamic matching processing parameters of the force control decision model, including: Step 1051: Extract the positive and negative polarities of the pressure balance deviation index from the real-time pressure balance state data as the basis for determining the deviation direction.

[0143] In the analysis of the pressure balance state, the positive and negative polarities of the pressure balance deviation index characterize the deviation direction between the real-time pressure and the target reference: a positive deviation indicates that the current pressure is higher than the upper limit of the allowable range, and a negative deviation corresponds to the pressure being lower than the lower limit. The polarity determination is achieved through a sign function operation. For example, when the deviation index value is +15%, it is determined as a positive deviation of 15%, and when it is -8%, it is determined as a negative deviation of 8%. This polarity information provides directional guidance for subsequent parameter adjustment to ensure that the correction direction of the control strategy is consistent with the system requirements.

[0144] Step 1052: When a positive pressure deviation is detected, use the gradient ascent algorithm to incrementally adjust the displacement acceleration feature weight coefficient of the force control decision model to generate an updated displacement acceleration feature weight coefficient.

[0145] It can be understood that the gradient ascent algorithm dynamically calculates the weight adjustment amount according to the amplitude of the positive deviation, and its incremental amplitude is positively correlated with the deviation index. The adjustment process follows the principle of the maximum directional derivative. By increasing the weight proportion of the displacement acceleration feature in the composite feature vector, the contribution of the spool dynamic response ability to pressure control is strengthened. For example, when the initial value of the weight coefficient is 0.6, for a +20% positive deviation, the algorithm increases it to 0.68, making the force control decision model pay more attention to the suppression effect of the acceleration feature on pressure overshoot.

[0146] Step 1053: When a negative pressure deviation is detected, use the momentum optimization algorithm to decrementally adjust the threshold of the position deviation integral of the force control decision model to generate an updated threshold of the position deviation integral.

[0147] Among them, the momentum optimization algorithm introduces an exponentially decaying average mechanism of the historical adjustment amount to avoid oscillation phenomena during the threshold adjustment process. For negative deviations, the algorithm gradually reduces the integral threshold to enhance the sensitivity of the error accumulation effect. For example, the threshold is reduced from the initial 0.15 mm·s to 0.12 mm·s. This adjustment prompts the control strategy to trigger the integral compensation mechanism faster when the pressure is insufficient, and accelerates the pressure recovery speed by improving the spool position tracking accuracy.

[0148] Step 1054: Input the updated displacement acceleration feature weight coefficient and the updated threshold of the position deviation integral into the model stability evaluator to calculate the improvement amplitude of the pressure control stability index.

[0149] Exemplarily, the model stability evaluator constructs a multi-dimensional evaluation system: by comparing parameters such as the standard deviation of the pressure fluctuation amplitude, the peak overshoot, and the stabilization time within three consecutive control cycles before and after adjustment, the improvement effect is quantified. For example, the improvement amplitude calculation formula is (original fluctuation amplitude - adjusted fluctuation amplitude) / original fluctuation amplitude × 100%. When the calculation result exceeds the preset threshold, it is determined as an effective improvement. The sliding window mechanism is adopted in the evaluation process to eliminate the influence of instantaneous interference and ensure the reliability of the results.

[0150] Step 1055: When the improvement amplitude is less than the preset convergence threshold, trigger the historical parameter rollback mechanism to restore the displacement acceleration feature weight coefficient and the position deviation integral threshold to the state before adjustment.

[0151] It can be understood that the historical parameter rollback mechanism stores the records of the last five parameter adjustments through a circular buffer. When the improvement amplitude does not reach the convergence threshold (e.g., 0.5%), the most recent valid parameters are automatically called to overwrite the current values. The historical momentum cache of the momentum optimization algorithm is cleared synchronously during the rollback process to prevent residual data from interfering with subsequent adjustments. This mechanism ensures the rapid restoration of system stability when parameter optimization fails and avoids the continuous deterioration of control performance.

[0152] Step 1056: When the improvement amplitude reaches the preset convergence threshold, synchronize the updated displacement acceleration feature weight coefficient and the updated position deviation integral threshold to all online running force control decision model instances.

[0153] For example, parameter synchronization adopts a publish-subscribe mechanism. The master control node encapsulates the verified valid parameters into data packets and broadcasts them to each slave node through the real-time communication bus. The synchronization process follows strict timing constraints to ensure that all model instances complete parameter loading within the specified control cycle. For example, within a 1ms synchronization window period, each node needs to complete parameter verification, memory writing, and effectiveness confirmation operations to ensure the consistency of control strategies for multiple execution units.

[0154] Step 1057: During the synchronization process, continuously monitor the fluctuation amplitude of the pressure balance deviation index. If the fluctuation amplitude exceeds the dynamic stability tolerance, suspend the synchronization and enable the redundant model parameter backup.

[0155] Specifically, the dynamic stability tolerance is dynamically set according to the current working conditions. For example, it is set to ±3% under the idle condition and relaxed to ±5% under the rapid acceleration condition. When it is monitored that the fluctuation amplitude exceeds the limit twice consecutively, the synchronization process is immediately interrupted and switched to the redundant parameter set. The redundant parameters are extracted from the historical optimal parameter library and a quick stability check is performed after loading. This mechanism forms a dual insurance strategy to minimize the risk of system oscillation caused by parameter updates.

[0156] Step 1058: After completing parameter synchronization, re - execute the dynamic matching process, and generate a new set of servo control instructions using the updated displacement - acceleration feature weight coefficients and the position deviation integral threshold.

[0157] For example, the updated force - control decision model recalculates the feature - matching rules based on the new parameters. For instance, after the weight of the displacement - acceleration feature is increased, the model assigns a higher priority to the dynamic response characteristics when generating the opening compensation value. The new set of control instructions is sent to the drive unit after multi - level verification, synchronously triggering the resampling process of the pressure sensor data, forming a complete closed - loop from parameter update to control - effect verification. This process enables the system to continuously approach the optimal control state and achieve self - improvement in long - term operation stability.

[0158] In a non - limiting embodiment, the method further includes: Obtain the valve opening compensation value and the pressure - balance adjustment coefficient in the set of servo control instructions output by the updated force - control decision model; Input the valve opening compensation value into the displacement - acceleration coupler, perform time - domain alignment processing with the acceleration data of the current valve displacement trajectory, and generate a dynamic displacement compensation gain; Input the pressure - balance adjustment coefficient into the phase - compensation controller, and generate a leading - phase compensation angle in combination with the real - time monitoring value of the pipeline pressure fluctuation frequency; Perform frequency - domain superposition processing on the dynamic displacement compensation gain and the leading - phase compensation angle to generate a composite compensation control signal; Adjust the rising - edge time of the current waveform and the duty cycle of the steady - state holding section of the servo drive unit through the composite compensation control signal; Real - time collect the current ripple amplitude of the servo - motor winding and the spool - displacement following - error data, and generate a set of dynamic response deviations; Input the set of dynamic response deviations into the parameter - correction module of the force - control decision model to adjust the dynamic matching weight coefficient between the displacement - acceleration feature and the current - phase offset; Recalculate the orthogonal basis vectors of the multi - dimensional compensation matrix according to the adjusted dynamic matching weight coefficient, generate an optimized set of servo control instructions, and update the drive parameters of the actuator.

[0159] In the control method of the fuel injection system, it further includes a cooperative optimization mechanism for dynamic displacement compensation gain and phase compensation. Specifically, input the valve opening compensation value output by the force - control decision model into the displacement - acceleration coupler, and the coupler aligns the compensation value with the current spool - displacement acceleration data through the time - domain convolution algorithm to generate a dynamic displacement compensation gain. For example, when the spool generates a peak acceleration under a rapid - acceleration condition, the compensation gain is dynamically adjusted based on the slope of the acceleration curve to ensure that the displacement compensation amount is synchronized with the mechanical dynamic response.

[0160] Meanwhile, the pressure balance adjustment coefficient is input into the phase compensation controller, which calculates the leading phase compensation angle in real-time by monitoring the pipeline pressure fluctuation frequency to offset the phase delay effect of the pressure fluctuation. For example, when the detected main frequency of the pressure fluctuation is 200 Hz, the controller generates a leading compensation angle of 5 degrees. The dynamic displacement compensation gain and the leading compensation angle are transformed into frequency-domain signals by Fourier transform and then superimposed to form a composite compensation control signal. This signal realizes the precise matching of electromagnetic force and mechanical motion by adjusting the rising edge time of the drive current waveform (e.g., shortening from 0.5 ms to 0.4 ms) and the steady-state duty cycle (e.g., increasing from 70% to 75%). The current ripple amplitude (such as ±0.2 A) of the servo motor winding and the spool displacement error data (such as 0.05 mm) are collected in real-time to generate a dynamic response deviation set containing the ripple distortion rate and the error integral value. After this deviation set is input into the parameter correction module, the dynamic matching weight coefficient of the displacement acceleration feature and the current phase offset (e.g., increasing the acceleration weight from 0.6 to 0.65) is adjusted by the gradient descent method, and the orthogonal basis vectors of the multi-dimensional compensation matrix are recalculated to generate an optimized control instruction set. The updated instruction drives the actuator to achieve higher-precision displacement tracking and pressure balance.

[0161] In a non-limiting embodiment, the method further includes: extracting the main frequency component of the pressure fluctuation frequency from the real-time pressure balance state data; performing cross-correlation analysis on the main frequency component and the acceleration frequency band of the valve displacement trajectory to identify the mechanical resonance risk frequency band; adjusting the dynamic damping distribution ratio of the pressure balance adjustment coefficient according to the mechanical resonance risk frequency band; inputting the adjusted dynamic damping distribution ratio into a multi-stage adaptive notch filter to generate band-stop filter parameters matching the resonance frequency band; synchronously calculating the inertial delay compensation of the valve opening compensation value according to the fluid medium density to generate a density-adaptive displacement pre-compensation amount; suppressing the energy amplitude of the corresponding frequency band in the servo drive current by the band-stop filter parameters and generating an anti-resonance control signal in combination with the displacement pre-compensation amount; driving the servo drive unit to execute the anti-resonance control signal and collecting the spectral attenuation rate of the pressure sensor and the root mean square value of the displacement following error in real-time; feeding back the spectral attenuation rate and the root mean square value to the dynamic damping distribution module of the force control decision model to iteratively update the cut-off frequency and attenuation depth of the band-stop filter parameters.

[0162] To further suppress the influence of mechanical resonance on fuel injection stability, the risk frequency band is identified through spectral cross-correlation analysis. Specifically, the main frequency component of the pressure fluctuation (e.g., 120 Hz) is extracted from the real-time pressure balance state data, and a cross-correlation operation is performed with the frequency-domain characteristics of the spool displacement acceleration to calculate the correlation coefficient matrix of the two in the 50 - 300 Hz frequency band. When the correlation coefficient of a certain frequency band (e.g., 180 Hz) exceeds the threshold, it is determined as the mechanical resonance risk frequency band. Dynamically adjust the damping distribution ratio in the pressure balance adjustment coefficient (e.g., increase the damping weight of the risk frequency band from 0.3 to 0.5), and input this ratio into the multi-stage adaptive notch filter to generate band-stop filter parameters (center frequency 180 Hz, stopband width 20 Hz, attenuation depth -30 dB).

[0163] Meanwhile, according to the fluid medium density (e.g., 850 kg / m³), an inertial delay compensation calculation is performed on the valve opening compensation value to generate a density-adaptive displacement pre-compensation amount (e.g., apply the compensation 0.1 ms in advance when the density increases). The anti-resonance control signal is generated by suppressing the energy of the risk frequency band in the drive current (e.g., reducing the amplitude of the 180 Hz component by 40%) and combining the pre-compensated displacement amount. After executing this signal, the real-time collected pressure spectrum attenuation rate (e.g., increased from 15 dB / ms to 20 dB / ms) and the root mean square value of the displacement error (e.g., decreased from 0.1 mm to 0.06 mm) are fed back to the dynamic damping distribution module to iteratively update the filter parameters (e.g., adjust the cut-off frequency to 185 Hz and increase the attenuation depth to -35 dB), forming a closed-loop anti-resonance control.

[0164] In a non-limiting embodiment, the method further includes: Based on the pressure fluctuation amplitude in the real-time pressure balance state data, multiple levels of pressure threshold intervals are divided and a hierarchical adjustment instruction set is generated; The high-pressure threshold instruction in the hierarchical adjustment instruction set is input into the pressure balance adjustment coefficient calculation module to generate an increment of the high-pressure compensation coefficient; The increment of the high-pressure compensation coefficient is superimposed on the pressure balance adjustment coefficient to generate a radially frequency-modulated carrier signal adapted to multiple levels of pressure; The radially frequency-modulated carrier signal adapted to multiple levels of pressure is superimposed with the axial drive pulse sequence through the signal synthesis module to generate a composite servo drive control signal for hierarchical pressure control; The data of the post-valve pressure transmitter after the hierarchical adjustment of the actuator is collected in real time, and it is judged whether the pressure fluctuation falls into the target threshold interval; If it does not fall into the target threshold interval, trigger the reallocation of the priority of the hierarchical adjustment instruction set and update the superimposition ratio of the increment of the high-pressure compensation coefficient.

[0165] For the requirements of multi - level pressure conditions, a hierarchical pressure threshold control strategy is implemented. Three - level threshold intervals are divided according to the real - time pressure fluctuation amplitude (low pressure: ±0.3 MPa, medium pressure: ±0.5 MPa, high pressure: ±0.8 MPa), and corresponding hierarchical adjustment instructions are generated. When the pressure enters the high - pressure interval, the pressure balance adjustment coefficient calculation module generates a high - pressure compensation coefficient increment (such as a coefficient increase of 0.15) based on the difference amplitude (for example, exceeding the limit by 0.2 MPa). After this increment is superimposed on the original adjustment coefficient, a radially frequency - modulated carrier signal adapted to multiple levels is generated, and its frequency is dynamically adjusted according to the pressure level (for example, increasing from 150 Hz to 180 Hz under high pressure). The signal synthesis module superimposes this carrier signal on the axial drive pulse sequence to generate a composite control signal to drive the actuator. If the pressure behind the valve does not fall back to the target interval within a specified period (such as 5 ms), priority re - allocation is triggered: for example, increasing the superposition ratio of the high - pressure compensation increment from 100% to 120% and enhancing the integral control effect. This process is verified in real - time through the pressure sensor data until the fluctuation amplitude converges to the allowable range.

[0166] In a non - limiting embodiment, the method further includes: Obtaining the current environmental temperature gradient change rate and the fluid medium temperature compensation coefficient; Calculating the temperature drift compensation amount of the servo drive current according to the deviation direction between the current environmental temperature gradient change rate and the preset temperature threshold interval; Performing a linear superposition process on the fluid medium temperature compensation coefficient and the temperature drift compensation amount to generate a comprehensive temperature compensation instruction; Performing a dynamic gain adjustment process on the drive current waveform data in the servo drive current data set according to the comprehensive temperature compensation instruction to generate temperature - compensated servo drive current waveform data; Performing a pressure feature mapping process on the temperature - compensated servo drive current waveform data to regenerate an updated pressure fluctuation feature set; Iteratively performing a dynamic matching process based on the updated pressure fluctuation feature set to generate a temperature - compensated servo control instruction set; Performing a multi - level dynamic adjustment operation on the servo drive unit according to the temperature - compensated servo control instruction set to generate temperature - corrected real - time pressure balance state data; Feeding back the temperature - corrected real - time pressure balance state data to achieve the current iterative update of the dynamic matching processing parameters of the force control decision model; wherein, the current iterative update includes: triggering the synchronous update of the temperature compensation parameters of the force control decision model.

[0167] To eliminate the influence of temperature changes on control accuracy, a temperature compensation mechanism is embedded. The ambient temperature gradient change rate (such as 2℃ / s) and the fluid medium temperature compensation coefficient (such as -0.05% / ℃) are obtained through a temperature sensor, and the temperature drift compensation amount of the servo drive current is calculated (for example, when the temperature rises by 10℃, the compensation current amplitude is 0.5A). After the comprehensive temperature compensation instructions are linearly superimposed, the gain characteristics of the drive current waveform are dynamically adjusted (for example, the steady-state current is increased from 8A to 8.4A). The compensated current data is re-executed for pressure characteristic mapping processing to update the phase offset (such as reducing by 0.1ms) and amplitude attenuation coefficient (such as reducing by 0.02A / ms) in the pressure fluctuation characteristic set. The temperature compensation control instructions generated based on the new characteristic set drive the actuator to maintain the stability of the fluctuation amplitude within ±0.4MPa in the high-temperature working condition for the pressure balance state data. Finally, the temperature correction data is fed back to the force control decision-making model to trigger the online update of temperature compensation parameters (such as the medium temperature coefficient) to ensure the adaptive control ability of the system within the full temperature range.

[0168] It can be seen that the core innovation of the embodiment of the present invention lies in constructing an indirect perception system of the pressure state by deeply excavating the current and displacement dynamic characteristics of the servo system itself, thereby significantly reducing the dependence on traditional mechanical sensors.

[0169] First, the embodiment of the present invention acquires two types of native data of the servo system, namely the servo drive current waveform and the valve displacement trajectory, with high precision, and uses time-frequency joint analysis technology to analyze the implicit pressure characteristic information therein. In the processing of the drive current waveform, the dynamic time warping algorithm is used to extract key indicators such as the current phase offset and amplitude attenuation coefficient. These parameters can effectively reflect the dynamic balance state of electromagnetic force-fluid resistance in the fuel pressure establishment stage. For example, the slope change of the current rising edge can map the pressure establishment rate, and the ripple characteristic of the steady-state holding current has a strong correlation with the pressure pulsation of the high-pressure oil rail. This mapping relationship realizes the accurate conversion from the current signal to the pressure value through a pre-trained pressure inversion model.

[0170] Secondly, in the embodiments of the present invention, by constructing a force control decision-making model with multi-dimensional feature fusion, the current feature is dynamically matched with the valve displacement trajectory, realizing the closed-loop optimization of the pressure control strategy. The acceleration feature of the valve displacement trajectory and the integral of the position tracking error, combined with parameters such as the current phase offset, jointly generate a servo control instruction with a feed-forward compensation function. This data-driven method not only avoids the signal distortion problem of traditional mechanical sensors under harsh working conditions such as high pressure and high-frequency impact, but also realizes the real-time inversion of the pressure state through in-depth analysis of the dynamic characteristics of the servo system itself. For example, under the condition of rapid acceleration, according to the high-frequency oscillation characteristics in the current waveform, the risk of pressure overshoot is predicted in advance, and the valve opening compensation value is dynamically adjusted, rather than relying on the lag feedback of the pressure sensor, significantly improving the transient response speed.

[0171] In addition, in the embodiments of the present invention, the pressure feature mapping relationship is continuously optimized through an iterative learning mechanism, forming an adaptive pressure control system. The deviation between the real-time pressure balance state and the preset reference value is used as a feedback signal to automatically correct the feature matching weight and the compensation coefficient generation logic of the force control decision-making model. This mechanism enables the pressure control accuracy to no longer depend on the long-term stability of the mechanical sensor, but instead maintains the control efficiency through continuous monitoring of the dynamic characteristics of the servo system and model update. Under extreme working conditions of high temperature and high pressure, traditional pressure sensors are prone to drift or damage. However, in the embodiments of the present invention, by utilizing the inherent anti-interference characteristics of servo current and displacement data, combined with a feature extraction algorithm based on physical mechanisms, the reliability and environmental adaptability of pressure control are ensured, providing a highly robust solution for complex industrial scenarios. In this way, when facing complex working conditions, the flow regulating valve control technology provided by the embodiments of the present invention can improve the technical problems of response delay, weak overshoot suppression ability, and poor long-term operation stability.

[0172] Based on the same inventive concept, the embodiments of the present invention also provide a flow regulating valve servo force control system. Refer to Figure 2 As shown in the figure, it is a schematic structural diagram of a possible flow regulating valve servo force control system provided in the embodiments of the present invention. Figure 2 In the figure, the flow regulating valve servo force control system 200 includes: a processor 210 and a memory 220. Among them, the memory 220 stores a computer program executable by the processor 210. The processor 210 can execute the steps of the above-mentioned flow regulating valve servo force control method without a force sensor by executing the instructions stored in the memory 220.

[0173] Based on the same inventive concept, an embodiment of the present invention provides a computer-readable storage medium, which includes a computer program. When the computer program runs on a flow control valve servo force control system, the computer program is used to cause the flow control valve servo force control system to execute the steps of the above-mentioned flow control valve servo force control method based on a force sensorless. In some possible implementation manners, various aspects of the flow control valve servo force control method provided by the present invention can also be implemented in the form of a program product, which includes a computer program. When the program product runs on the flow control valve servo force control system, the computer program is used to cause the flow control valve servo force control system to execute the steps in the above-mentioned flow control valve servo force control method based on a force sensorless. For example, the flow control valve servo force control system can execute as Figure 1 the steps shown therein.

Claims

1. A servo force control method for a flow regulating valve based on a force sensor, characterized in that, Including: Obtain the servo drive current data set and valve displacement trajectory data set of the target regulating valve, where the servo drive current data set contains drive current waveform data under different working conditions; Perform pressure characteristic mapping processing on the servo drive current data set to generate a pressure fluctuation characteristic set corresponding to the drive current waveform data, where the pressure fluctuation characteristic set contains current phase offset and current amplitude attenuation coefficient; Input the pressure fluctuation characteristic set and the valve displacement trajectory data set into a preset force control decision model for dynamic matching processing to generate a servo control instruction set, where the servo control instruction set contains valve opening compensation value and pressure balance adjustment coefficient; Perform a multi-level dynamic adjustment operation on the servo drive unit of the target regulating valve according to the servo control instruction set to generate real-time pressure balance state data; Based on the deviation between the real-time pressure balance state data and a preset pressure reference value, iteratively update the dynamic matching processing parameters of the force control decision model.

2. The method according to claim 1, characterized in that The performing pressure characteristic mapping processing on the servo drive current data set to generate a pressure fluctuation characteristic set corresponding to the drive current waveform data includes: Extract the time-domain characteristics and frequency-domain characteristics of the drive current waveform data, where the time-domain characteristics include current rising edge slope and steady-state holding duration, and the frequency-domain characteristics include the proportion of the third harmonic component and the fundamental wave amplitude volatility; Perform feature fusion processing on the time-domain characteristics and the frequency-domain characteristics to generate a composite current feature vector; Input the composite current feature vector into a pre-trained pressure inversion model to output a real-time pressure estimate value corresponding to the drive current waveform data; Based on the difference between the real-time pressure estimate value and the historical pressure reference value, calculate the current phase offset and the current amplitude attenuation coefficient; Generate the pressure fluctuation characteristic set according to the current phase offset and the current amplitude attenuation coefficient.

3. The method according to claim 2, wherein The training process of the pre-trained pressure inversion model includes: Collect the servo drive current sample data and the corresponding pressure sensor measured data of the sample regulating valve in a standard pressure test environment; Perform waveform segmentation processing on the servo drive current sample data to generate multiple current waveform segment data; Perform time-frequency joint analysis on each current waveform segment data to extract sample time-domain characteristics and sample frequency-domain characteristics; Perform standardized splicing processing on the sample time-domain characteristics and the sample frequency-domain characteristics to generate a sample composite feature vector; Using the sample composite feature vector as the input and the pressure sensor measured data as the output target, train a preset neural network model using the gradient descent algorithm until convergence; Configure the trained preset neural network model as the pressure inversion model and solidify the model weight parameters.

4. The method according to claim 1, characterized in that, The inputting the pressure fluctuation characteristic set and the valve displacement trajectory data set into a preset force control decision model for dynamic matching processing to generate a servo control instruction set includes: Perform motion feature analysis processing on the valve displacement trajectory data set to generate displacement acceleration characteristics and position deviation integral; Perform a first-level coupling operation on the displacement acceleration feature and the current phase offset to generate a preliminary compensation coefficient; Perform a second-level coupling operation on the position deviation integral and the current amplitude attenuation coefficient to generate a dynamic correction coefficient; Construct a multi-dimensional compensation matrix based on the preliminary compensation coefficient and the dynamic correction coefficient; Perform an optimal path search process on the multi-dimensional compensation matrix through the strategy selector in the force control decision model, and output the valve opening compensation value and the pressure balance adjustment coefficient.

5. The method according to claim 4, characterized in that, The performing an optimal path search process on the multi-dimensional compensation matrix through the strategy selector in the force control decision model and outputting the valve opening compensation value and the pressure balance adjustment coefficient includes: Obtain the fluid medium density, pipeline vibration amplitude, and environmental temperature gradient in the current working condition environment parameter set; wherein, the fluid medium density, the pipeline vibration amplitude, and the environmental temperature gradient are independent of each other; Map the three orthogonal dimensions of the multi-dimensional compensation matrix to the fluid medium density, the pipeline vibration amplitude, and the environmental temperature gradient respectively based on the space coordinate system to obtain the medium density threshold line, the vibration amplitude isosurface, and the temperature gradient surface; Divide a three-dimensional constraint boundary formed by the medium density threshold line, the vibration amplitude isosurface, and the temperature gradient surface in the space coordinate system; Use an adaptive step size traversal algorithm to generate a candidate control path set within the three-dimensional constraint boundary, and each candidate control path in the candidate control path set includes an axial displacement compensation sequence and a radial pressure adjustment coefficient sequence; Real-time collect the transient feedback signal of the pipeline pressure sensor, and extract the pressure fluctuation frequency and amplitude attenuation rate from the transient feedback signal as path evaluation indicators; Input the pressure fluctuation frequency and the amplitude attenuation rate into the path stability evaluation function to calculate the dynamic stability score of each candidate control path; Select the candidate control path with the highest dynamic stability score as the optimal path, and extract the mean value of the axial displacement compensation sequence of the optimal path as the valve opening compensation value; Synchronously extract the peak maintenance coefficient of the radial pressure adjustment coefficient sequence in the optimal path as the pressure balance adjustment coefficient.

6. The method according to claim 1, wherein The performing a multi-level dynamic adjustment operation on the servo drive unit of the target regulating valve according to the servo control instruction set to generate real-time pressure balance state data includes: Input the valve opening compensation value in the servo control instruction set into the axial displacement compensator for component decomposition processing to generate an axial displacement compensation component and a radial pressure compensation component; Input the axial displacement compensation component into the first-level pulse width modulator for dynamic duty ratio allocation processing to generate an axial drive pulse sequence with a dynamic pulse width; Input the radial pressure compensation component into the second-level carrier generator for frequency-phase collaborative processing to generate a radial frequency modulation carrier signal matching the current pressure balance adjustment coefficient; Input the axial drive pulse sequence and the radial frequency modulation carrier signal into the signal synthesis module for time-domain superposition processing to generate a composite servo drive control signal; The composite servo drive control signal is loaded onto the actuator of the servo drive unit through a power amplifier to drive the valve core to generate axial displacement actions and radial pressure compensation actions; The valve core displacement sensor data and the post-valve pressure transmitter data of the actuator are collected in real time to generate a dynamic response data set containing the actual displacement offset and the pressure fluctuation amplitude; The deviation integral calculation is performed on the actual displacement offset in the dynamic response data set and the valve displacement trajectory data set to generate a displacement following error value; The difference between the pressure fluctuation amplitude and a preset pressure reference value is compared to generate a pressure balance deviation index; The real-time pressure balance state data is generated based on the displacement following error value and the pressure balance deviation index.

7. The method according to claim 6, wherein After the axial drive pulse sequence and the radial frequency modulation carrier signal are input into the signal synthesis module for time-domain superposition processing to generate a composite servo drive control signal, the method further includes: The coil current waveform data of the servo drive unit is captured in real time through a current sampling circuit, and the transient rising edge slope and the steady-state holding section ripple amplitude of the coil current waveform data are extracted; The transient rising edge slope is input into a first comparator for overrun judgment processing with a preset transient safety threshold to generate a transient slope overrun flag signal; When the transient slope overrun flag signal is in an active state, a current suppression circuit is triggered to perform slope limiting processing on the pulse front edge of the composite servo drive control signal to generate a slope-limited and corrected axial drive pulse sequence; The steady-state holding section ripple amplitude is input into a band-pass filter for spectrum separation processing to separate an effective ripple component matching the commutation frequency of the servo motor and an electromagnetic interference high-frequency noise component; The effective ripple component and the electromagnetic interference high-frequency noise component are input into a waveform matcher to calculate the morphological similarity with a preset ideal steady-state waveform template to generate a waveform distortion correction coefficient; The phase offset compensation amount of the second-stage carrier generator is adjusted according to the waveform distortion correction coefficient to generate a phase-compensated radial frequency modulation carrier signal; The phase-compensated radial frequency modulation carrier signal and the slope-limited and corrected axial drive pulse sequence are input into the signal synthesis module for time-sequence alignment and superposition processing to generate an optimized composite servo drive control signal; The optimized composite servo drive control signal is loaded onto the power amplifier to drive the actuator of the servo drive unit to generate corrected axial displacement actions and radial pressure compensation actions; The valve core displacement sensor data of the actuator is collected in real time to update the actual displacement offset, and the post-valve pressure transmitter data is collected synchronously to update the pressure fluctuation amplitude; The updated actual displacement offset is input into the deviation integral calculation module of the displacement following error value to generate an updated displacement following error value; The updated pressure fluctuation amplitude is compared with the preset pressure reference value to generate an updated pressure balance deviation index; The updated displacement following error value and the updated pressure balance deviation index are fed back to the strategy selector of the force control decision model to trigger the online calibration of the dynamic matching processing parameters.

8. The method according to claim 6, characterized in that The iterative updating of the dynamic matching processing parameters of the force control decision model based on the deviation between the real-time pressure balance state data and the preset pressure reference value includes: Extracting the positive and negative polarity of the pressure balance deviation index from the real-time pressure balance state data as a basis for determining the deviation direction; When a positive pressure deviation is detected, a gradient ascent algorithm is used to incrementally adjust the displacement acceleration characteristic weight coefficient of the force control decision model to generate an updated displacement acceleration characteristic weight coefficient; When a negative pressure deviation is detected, a momentum optimization algorithm is used to decrease the position deviation integral threshold of the force control decision model to generate an updated position deviation integral threshold; Inputting the updated displacement acceleration characteristic weight coefficient and the updated position deviation integral threshold into a model stability evaluator to calculate the improvement of the pressure control stability index; When the improvement is less than a preset convergence threshold, a historical parameter rollback mechanism is triggered to restore the displacement acceleration characteristic weight coefficient and the position deviation integral threshold to the state before adjustment; When the improvement reaches the preset convergence threshold, synchronizing the updated displacement acceleration characteristic weight coefficient and the updated position deviation integral threshold to all online force control decision model instances; During the synchronization process, the fluctuation amplitude of the pressure balance deviation index is monitored in real time. If the fluctuation amplitude exceeds the dynamic stability tolerance, the synchronization is suspended and the redundant model parameter backup is enabled; After completing the parameter synchronization, the dynamic matching process is re-executed, and a new servo control instruction set is generated using the updated displacement acceleration characteristic weight coefficient and position deviation integral threshold.

9. The method according to claim 1, characterized in that, After obtaining the servo drive current data set and the valve displacement trajectory data set of the target regulating valve, the method further includes: performing abnormal fluctuation detection processing on the servo drive current data set to identify sudden change peak areas in the current waveform; Performing a time-frequency joint analysis on the identified mutation peak region to determine whether the identified mutation peak region is caused by mechanical jamming or fluid impact; When it is determined that the identified sudden change peak area is caused by mechanical jamming, a servo drive current limiting instruction is generated and a valve self-cleaning program is triggered; When it is determined that the identified sudden change peak area is caused by fluid impact, generating a pressure buffer control instruction and adjusting the valve response rate; The servo drive current data set after abnormal fluctuation detection processing is performed is marked as a cleansed data set.

10. A servo force control system for a flow regulating valve, characterized in that, The method comprises a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor is enabled to perform the steps of any one of the methods of claims 1 to 9.

Images