Pressure closed-loop self-calibration method for ultrahigh-speed jet injection process

Abstract

The invention relates to the technical field of automatic control, and discloses a pressure closed-loop self-calibration method for an ultrahigh-speed jet injection process, and the method comprises the steps: initializing a system, and building a pressure and temperature combined compensation model; a high-precision sensor is used for collecting real-time pressure and temperature data; by calculating a pressure deviation intelligent switching control strategy, PID is combined with feed-forward compensation in a steady state, and a deep reinforcement learning algorithm is switched to realize quick response in a transient state; automatically triggering closed-loop self-calibration according to operation time or a deviation threshold value, executing zero point and full scale calibration and updating compensation parameters; and carrying out residual analysis by using a digital twinborn model to realize fault diagnosis and early warning. According to the invention, the problems of low control precision and poor long-term stability of an ultra-high-speed jet flow process under nonlinear and large-lag working conditions are solved, and high-precision stable control and system self-adaptive calibration under a wide speed range are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automatic control, in particular to a pressure closed-loop self-calibration method for super-high-speed jet injection process. BACKGROUND

[0002] At present, as a core means of high-energy conversion and material processing, super-high-speed jet technology has been widely used in aerospace propulsion systems and advanced material precision machining fields. Through pressurizing fluid medium to hundreds of megapascals, the technology uses a nozzle to form a high-energy jet flow several times the speed of sound, thereby realizing fine cutting of high-hardness materials such as titanium alloy or high-efficiency atomization injection of fuel. With the extension of industrial demand to extreme working conditions, the fourth generation of jet equipment is gradually developing towards multi-functional integration, and more stringent index requirements are put forward for pressure stability and energy transmission efficiency in the jet process.

[0003] In view of the above-mentioned super-high-pressure fluid control requirements, the existing jet injection system mostly adopts a closed-loop regulation mode based on sensor feedback. The working principle is usually to arrange a pressure acquisition unit at the outlet of the jet pipeline or the accumulator, to monitor the fluid pressure signal in real time and transmit it to the controller. The controller outputs a signal to drive the hydraulic servo valve or proportional valve according to the preset target value, and adjusts the flow and direction of the hydraulic oil circuit to control the reciprocating motion frequency and stroke of the piston of the pressurizing cylinder, so as to adjust the output pressure of the high-pressure water, and try to maintain the pressure balance and flow stability in the jet process.

[0004] Although the existing technology can realize basic pressure regulation, it still has shortcomings when facing extreme working conditions of super-high-speed jet. First, the traditional system lacks a dynamic compensation mechanism for sensor aging drift and environmental temperature influence, and the measurement error accumulates after long-term operation, resulting in distorted feedback data, and manual offline calibration has a long cycle and low efficiency. Second, the conventional PID control strategy is difficult to cope with the strong nonlinearity and large lag characteristics in the jet process, and the control response has delay and overshoot when the pressure fluctuates sharply in a wide speed range, which cannot maintain pressure stability within milliseconds. In addition, the existing system lacks fault diagnosis capability based on model comparison, and it is difficult to distinguish between sensor failure and actuator abnormality, and once a fault occurs, it can only be passively shut down, which cannot guarantee the continuous safe operation of the system.

[0005] Therefore, the present application provides a pressure closed-loop self-calibration method for super-high-speed jet injection process to solve the deficiencies in the prior art. SUMMARY

[0006] In view of the deficiencies of the prior art, the present application provides a pressure closed-loop self-calibration method for an ultra-high-speed jet injection process, which solves the problems of low control precision, slow response speed and poor long-term operation stability caused by the lack of real-time self-calibration mechanism of the existing ultra-high-speed jet pressure control method under nonlinear and large-lag working conditions.

[0007] To achieve the above object, the present application is implemented by the following technical solutions: a pressure closed-loop self-calibration method for an ultra-high-speed jet injection process, comprising the following steps: Step S1, system initialization and parameter setting: a pressure and temperature joint compensation model is established, which is used for temperature drift correction of sensor data in the subsequent process; Step S2, real-time pressure data acquisition and processing: the pressure signal of the jet is collected by a pressure sensor, and the temperature data is collected by a temperature sensor, and the processed digital pressure signal is transmitted to a controller; Step S3, intelligent parameter adjustment: the deviation between the real-time pressure measurement value and the target jet pressure value is calculated, and the control strategy is selected to drive the actuator according to the state of the deviation; Step S4, full-automatic pressure closed-loop self-calibration: when the trigger condition is met, the current control task is suspended, zero-point calibration and full-scale gain calibration are performed in turn, and the parameters of the pressure and temperature joint compensation model are updated; Step S5, fault diagnosis and early warning: the theoretical prediction pressure value is calculated by using a digital twin model, and abnormality is determined by comparing the real-time pressure measurement value with the theoretical prediction pressure value.

[0008] By adopting the above technical solutions, since the pressure and temperature joint compensation model is established and combined with the full-automatic closed-loop self-calibration mechanism, the sensor zero-point drift and sensitivity drift can be periodically corrected, and the influence of environmental temperature change and device aging on measurement accuracy is eliminated. At the same time, through the synergistic effect of intelligent parameter adjustment and fault diagnosis and early warning, the system can automatically switch control strategies and monitor the health status in real time under complex working conditions. Therefore, high-precision stable control of wide-range pressure in the ultra-high-speed jet injection process is realized, the environmental adaptability and operation safety of the system are improved, and the technical problems of low control precision and difficulty in maintaining long-term measurement accuracy of traditional control methods under nonlinear and large-lag working conditions are solved.

[0009] Preferably, in the S1 step, the pressure and temperature joint compensation model is specifically reading pre-calibrated model data, and the model stores zero-point drift coefficients and sensitivity drift coefficients of the sensor under different temperatures; in the S2 step, the collected signal is processed in multiple stages, including instrument amplification and Butterworth low-pass filtering at the hardware level, and infinite impulse response filter or finite impulse response filter secondary filtering at the software level.

[0010] By adopting the technical scheme, high-frequency fluid noise and electromagnetic interference in the jet flow process are effectively filtered out by using the multi-stage filtering mechanism, ensuring high signal-to-noise ratio of the digital signal; the pre-calibrated compensation model provides an accurate reference for subsequent data correction, ensuring the rationality of the initial control parameters.

[0011] Preferably, in the S3 step, when the deviation is in the steady-state range, a PID algorithm combined with feedforward compensation is used, the feedforward compensation outputs a compensation signal according to a set fluid dynamics model for known system disturbances; when the deviation exceeds the steady-state range or the change rate is too large, a deep reinforcement learning algorithm based on a long short-term memory network and a deep deterministic policy gradient is switched to; the long short-term memory network extracts time series features, and the policy network of the deep deterministic policy gradient outputs continuous action control.

[0012] By adopting the technical scheme, the PID combined with the feedforward compensation solves the problem of regular disturbance suppression in the steady state; in the transient or large deviation condition, the long short-term memory network effectively compensates for the large lag characteristic of the super-high-speed jet flow system by extracting time series features; in cooperation with the deep deterministic policy gradient algorithm, continuous control actions can be directly output, adapting to the nonlinear pressure and flow coupling relationship in the jet injection process, and realizing fast response and no overshoot control in a wide speed range.

[0013] Preferably, in the step S4, calibration is triggered according to the running time or the pressure deviation threshold; zero-point calibration obtains the zero-point offset by pressure relief; full-scale gain calibration calculates the gain coefficient by pressurizing to the standard pressure point; and the compensation parameters are updated according to the current temperature, and the collected data are linearly corrected by using the updated parameters.

[0014] By adopting the technical scheme, the timeliness of calibration is ensured by the dual triggering mechanism of time domain and error domain; the zero-point and full-scale two-point calibration method combined with real-time temperature compensation is adopted to realize dynamic reconstruction of the sensor transfer characteristics, effectively eliminate the cumulative error generated by long-term operation, and maintain high measurement accuracy throughout the life cycle.

[0015] Preferably, in the step S5, a digital twin model is established based on the fluid mechanics equation and the system identification parameters to calculate the theoretical predicted pressure value; the residual sequence of the actual measured value and the theoretical value is analyzed to determine the anomaly; the sensor fault and the actuator fault are distinguished according to the abnormal characteristics, and the switching of the backup channel or the safety shutdown strategy is respectively executed.

[0016] By adopting the technical scheme, a digital twin model is introduced to construct a theoretical benchmark of the system, and early capture of micro-fault features is realized through residual trend analysis; the two different types of faults, i.e., sensor failure and system control instability, can be accurately distinguished, and differentiated fault tolerance or protection measures are taken, thereby avoiding unnecessary shutdown or safety accidents caused by misjudgment.

[0017] The application provides a pressure closed-loop self-calibration method for an ultra-high-speed jet injection process. 1. The application can periodically perform zero point and full range gain calibration and update compensation parameters in real time by establishing a pressure and temperature combined compensation model and combining a full-automatic closed-loop self-calibration mechanism. This design effectively eliminates the measurement errors caused by the aging drift of sensors due to long-term operation and the change of environmental temperature, realizes wide-range high-precision control of the ultra-high-speed jet injection process, and improves the long-term measurement stability and control precision of the system.

[0018] 2. The application adopts a multi-modal intelligent parameter adaptive adjustment strategy, uses PID combined with feedforward compensation to suppress interference in a steady state, and automatically switches to a deep reinforcement learning algorithm based on long short-term memory network and deep deterministic policy gradient in a transient or large deviation working condition. This technical scheme effectively overcomes the large lag and nonlinear coupling characteristics of the ultra-high-speed jet system, shortens the error convergence time, and realizes millisecond-level fast response without overshoot in a wide speed range working condition.

[0019] 3. The application introduces a fault diagnosis and early warning mechanism based on digital twinning, compares the measurement value of the physical sensor with the theoretical prediction value of the digital twinning model in real time, and analyzes the residual sequence. This scheme can accurately identify sensor faults and actuator abnormalities during system operation, and automatically trigger backup channel switching or safety protection strategies, effectively reducing unplanned downtime and improving the overall operation reliability and maintenance efficiency of the system. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 The application provides a pressure closed-loop self-calibration system architecture for an ultra-high-speed jet injection process. Figure 2 The application provides a pressure closed-loop self-calibration method flowchart for an ultra-high-speed jet injection process. DETAILED DESCRIPTION

[0021] The technical solutions in the embodiments of the application will be described below in conjunction with the drawings of the application specification. Obviously, the described embodiments are only part of the embodiments of the application, not all embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0022] Referring to the drawings Figure 1 The present application provides a pressure closed-loop self-calibration system for an ultra-high-speed jet injection process, which mainly consists of a hardware system and a software algorithm architecture. The hardware system is responsible for the collection and processing of physical signals and the driving of the execution mechanism. The software algorithm architecture runs in the hardware controller and is responsible for implementing complex control strategies and self-calibration logic.

[0023] The hardware system includes a pressure sensor module. The pressure sensor module uses a high-precision piezoelectric pressure sensor or a strain gauge pressure sensor. The measurement range of the sensor is selected and configured between 0.1 MPa and 600 MPa according to the requirements of the actual application scenario. The measurement accuracy of the sensor is configured to reach ±0.01%FS. The pressure sensor has a temperature compensation function unit integrated inside, which enables the sensor to maintain the stability of the output performance in an environmental temperature range of -20°C to 80°C.

[0024] The hardware system also includes a signal conditioning circuit electrically connected to the pressure sensor module. The signal conditioning circuit is used to process the analog signals output by the sensor, specifically including an amplification circuit, a filter circuit, and an analog-to-digital conversion circuit. The amplification circuit uses an instrument amplifier, and the gain parameter of the instrument amplifier is designed in a adjustable mode to adapt to different amplitude input signals. The filter circuit uses a Butterworth low-pass filter, and the cutoff frequency of the filter is set to 1 kHz to filter out high-frequency noise interference. The analog-to-digital conversion circuit includes an analog-to-digital converter with a resolution of not less than 16 bits and a sampling rate of not less than 10 kHz, which converts the processed analog signals into digital signals.

[0025] The hardware system also includes a controller connected to the signal conditioning circuit to receive digital pressure signals. The controller uses a high-performance microprocessor with a main frequency of not less than 400 MHz, such as an ARM Cortex-M7 architecture or a more powerful processor chip. The controller has a floating-point operation unit integrated inside for performing complex mathematical operations. The controller runs a real-time operating system, which supports a multi-task scheduling mechanism to ensure real-time execution of control algorithms.

[0026] The hardware system also includes an actuator drive circuit connected between the controller and the execution mechanism. The actuator drive circuit is designed to match the specific type of execution mechanism, including proportional valves or servo valves. The actuator drive circuit has a pulse width modulation (PWM) output function with a resolution of not less than 12 bits and a frequency of not less than 10 kHz, which is used to accurately drive the execution mechanism to adjust the pressure output.

[0027] The hardware system further comprises a communication interface integrated on the controller. The communication interface comprises at least one of an RS485 interface, a CAN bus interface, and an Ethernet interface. The communication interface supports a MODBUS communication protocol or a PROFIBUS communication protocol, for realizing data exchange between the controller and an upper computer or other external devices.

[0028] The software algorithm architecture runs in the controller, comprising a control algorithm layer and a data processing algorithm module. The control algorithm layer adopts a hierarchical control architecture design, and is specifically divided into a bottom-layer PID control module, a middle-layer model predictive control module, and a top-layer intelligent optimization control module. The bottom-layer PID control module is configured to realize basic pressure closed-loop control logic. The middle-layer model predictive control module is configured to perform predictive control calculation based on an established fluidics dynamics model. The top-layer intelligent optimization control module is configured to perform online optimization of control parameters by using a deep learning algorithm.

[0029] The data processing algorithm module is configured to pre-process and analyze collected signals. The module comprises a digital filtering unit, a feature extraction unit, and a trend analysis unit. The digital filtering unit adopts an infinite impulse response (IIR) filter or a finite impulse response (FIR) filter, for eliminating noise in digital signals. The feature extraction unit is used to calculate and extract feature parameters from real-time pressure signals. The trend analysis unit is used to analyze pressure change trends according to historical data, and to predict future states of the system.

[0030] Referring to the accompanying drawings, Figure 2 The present application provides a pressure closed-loop self-calibration method for an ultra-high-speed fluidics injection process, which comprises steps S1 to S5, each of which is executed in sequence to realize high-precision control of pressure and automatic calibration of the system.

[0031] After the system is started, the controller first reads preset configurations in the memory for initialization setting. This step specifically comprises setting a target fluidics pressure value, which is adjusted within a specified pressure range according to requirements of a specific application scenario. At the same time, the system initializes control parameters of the PID controller, including a proportional coefficient, an integral coefficient, and a differential coefficient. In addition, the system needs to set a self-calibration period and a calibration trigger threshold, as a basis for judging triggering of the subsequent automatic calibration process. In this step, the system also establishes a pressure-temperature joint compensation model, which is used to correct temperature drift of sensor data in subsequent processes.

[0032] The system utilizes a high-precision pressure sensor to acquire the pressure signal of the jet in real time, and an integrated temperature sensor to simultaneously acquire the ambient temperature and the sensor's body temperature. The acquired analog pressure signal is first amplified and filtered by a signal conditioning circuit to remove high-frequency noise interference. Subsequently, the conditioned analog signal is converted into a digital pressure signal by a high-resolution analog-to-digital converter and transmitted to the controller for further data processing and analysis.

[0033] The controller receives real-time digital pressure signals and automatically adjusts control parameters based on real-time changes in the jet flow conditions using an adaptive control algorithm. The controller monitors pressure fluctuations and control effectiveness in real time; when it detects an increase in pressure fluctuation amplitude or a decrease in control accuracy, the system automatically switches control strategies. This switching logic involves transitioning from a basic control mode to a more complex intelligent control mode, such as employing deep reinforcement learning-based algorithms, to meet the high-precision pressure control requirements under wide velocity ranges and complex nonlinear conditions.

[0034] The system continuously monitors runtime and pressure deviation during operation. When the system runtime reaches the set self-calibration cycle, or the real-time monitored pressure deviation exceeds the set calibration trigger threshold, the system automatically triggers the self-calibration procedure. During self-calibration, the system sequentially performs zero-point calibration and full-scale gain calibration, and updates the parameters of the pressure and temperature joint compensation model based on the currently acquired temperature data. This step aims to compensate for measurement errors caused by sensor aging, temperature changes, or environmental drift in real time, ensuring the long-term measurement accuracy of the system.

[0035] The system integrates a fault diagnosis module to monitor the health status of pressure sensors, the operating status of actuators, and overall control performance indicators in real time. The system uses a digital twin model to simulate and compare the calibration and operation processes. When fault characteristics such as abnormal sensor signals, sluggish actuator response, or control system divergence are detected, the system automatically switches to a safety mode and sends an alarm signal to the operator to prevent equipment damage or safety accidents.

[0036] Step S1 involves system initialization and multi-dimensional parameter setting. This step is the starting point of the entire control process and aims to establish a reliable operating benchmark for subsequent high-precision control and self-calibration. After the system powers on, the controller first executes a hardware self-test program to confirm that all sensor interfaces, actuator drive circuits, and communication modules are in normal working order. Subsequently, the controller accesses the onboard non-volatile memory and reads the stored default configuration file. This configuration file contains the system's hardware mapping relationships, sensor range definitions, and communication protocol parameters. Based on this information, the controller completes the initialization configuration of the underlying drivers, ensuring the correct connection between the software logic and the physical hardware.

[0037] After completing the basic configuration, the system enters the target parameter setting phase. The controller receives and sets the target jet pressure value through the human-machine interface or remote communication interface. This set value can be flexibly adjusted within a wide range from 0.1 MPa to 600 MPa, depending on the specific application scenario. The controller has internal safety verification logic that checks the amplitude limit of the set target pressure value to ensure it is within the safe operating range allowed by the system hardware, preventing overpressure damage to the equipment due to incorrect settings. After confirmation, the target pressure value is stored in the controller's running memory as a reference for subsequent closed-loop control circuits.

[0038] Next, the system initializes the PID controller, which is crucial for ensuring the stability of the control system during startup. The controller retrieves preset control parameters from memory and sets the initial values ​​for the proportional, integral, and derivative coefficients. These initial parameters are pre-tuned and stored based on the system's response characteristics under typical operating conditions, ensuring the system's response speed and stability during the initial startup phase. Simultaneously, the controller clears or presets the integrator's state variables to prevent excessive overshoot due to integral saturation during system startup, thus achieving soft-start control.

[0039] Meanwhile, the system is configured with automatic calibration mechanism time and threshold parameters. The controller starts its internal high-precision hardware timer and sets the self-calibration cycle to 24 hours, which is used to trigger timed system health maintenance. Simultaneously, the system sets a calibration trigger threshold of 0.5% of full scale. This means that during system operation, if the monitored pressure deviation continuously exceeds this set range, the system will be forced to enter the calibration process, regardless of whether the timed cycle has been reached. This dual-trigger mechanism balances the efficiency of continuous system operation with the accuracy of measurement data.

[0040] Finally, the system reads the pre-calibrated and stored pressure and temperature joint compensation model. This model exists in the form of a multidimensional data table or polynomial coefficients, detailing the zero-point drift and sensitivity drift characteristics of the pressure sensor under different temperature environments. The controller loads this calibration data into the cache of the high-speed computing unit, constructing a real-time temperature compensation lookup table. This step ensures that during subsequent real-time acquisition, the system can quickly retrieve or calculate the corresponding compensation coefficients based on the current ambient temperature and sensor body temperature, preparing data for high-precision measurement across the entire temperature range.

[0041] Step S2 is real-time pressure data acquisition and processing, which is mainly responsible for converting the fluid pressure changes in the physical world into accurate digital signals recognizable by the controller, and is the premise of achieving high-precision control. First, the system activates the piezoelectric or strain gauge high-precision pressure sensor located at the key nodes of the fluidic pipeline to continuously capture the fluid pressure in the pipeline at a set high sampling frequency. At the same time, the temperature sensor integrated in the pressure sensor or installed nearby is started simultaneously, and the current external environment temperature and the sensor's own body temperature are collected respectively. This synchronous acquisition mechanism ensures that each pressure sampling point corresponds to an exact temperature value, providing the necessary data basis for subsequent elimination of thermal zero drift and sensitivity drift caused by temperature changes.

[0042] The collected original analog pressure signal is usually weak and accompanied by electromagnetic noise in the industrial field, so it needs to be preprocessed by a specially designed signal conditioning circuit. The original signal first enters the precision instrument amplifier circuit, which has high common-mode rejection ratio and low temperature drift characteristics, and can amplify the weak sensor signal of microvolts or millivolts to the standard voltage range required by the analog-to-digital converter. During amplification, the circuit automatically adjusts the gain parameter to adapt to the dynamic range of the signal, ensuring that the best signal resolution is obtained under low and high pressure conditions, preventing the increase of quantization error caused by too small signal amplitude or the clipping distortion caused by too large signal amplitude.

[0043] The amplified analog signal then enters the filtering stage, and the system uses a hardware-implemented Butterworth low-pass filter to purify the signal spectrum. The filter is designed to have flat passband characteristics, which can effectively filter out mechanical vibration noise generated by the reciprocating motion of the high-pressure pump and high-frequency electromagnetic interference introduced by devices such as frequency converters. By setting a reasonable cutoff frequency, the filter preserves the dynamic characteristics of the fluidic pressure while maximizing the suppression of high-frequency noise above half the sampling frequency, thereby preventing aliasing in the subsequent analog-to-digital conversion process and ensuring the authenticity of the signal.

[0044] The conditioned and purified analog signal is finally sent to a high-resolution analog-to-digital converter. The converter works with a quantization accuracy of no less than sixteen bits and a high sampling rate, discretizing the continuously changing analog voltage signal into a sequence of digital pressure signals. The controller receives these digital signals through direct memory access or high-speed serial interface, ensuring real-time data transmission and reducing the load on the central processing unit. After receiving the original digital signals, the controller does not use them directly, but further processes them using the internal integrated infinite impulse response filter or finite impulse response filter algorithm for secondary digital filtering. This step can further eliminate residual random noise and smooth signal glitches, ultimately obtaining real-time pressure measurement values with high signal-to-noise ratio and high stability for subsequent control algorithm calls.

[0045] Step S3 is a multi-modal intelligent parameter self-adaptive adjustment, which is a core link to realize the system to give consideration to both steady-state accuracy and dynamic response speed under complex working conditions. The controller first performs a deviation calculation logic, subtracts the real-time pressure measurement value obtained in the previous link from the target jet pressure value set in advance to obtain the current pressure deviation value. The controller continuously monitors the amplitude of the pressure deviation value and its rate of change over time, and compares the absolute value of the deviation with the preset steady-state range threshold in real time, which is used as the basis for determining the current running state of the system, so as to determine the selection of the subsequent control strategy.

[0046] When the calculated pressure deviation is within the preset steady-state range, the system determines that it is currently in a stable running or fine-tuning stage, and the controller automatically locks and runs the basic control mode. In this mode, the controller mainly enables the proportional-integral-derivative algorithm, uses the proportional element for fast response, uses the integral element to eliminate static error, and uses the derivative element to predict the trend. In order to further improve the system's ability to suppress fluid dynamics interference, the controller adds a feedforward compensation link on this basis. The feedforward compensation link calculates the compensation control amount for known disturbances based on the pre-established fluid dynamics model and the current system state parameters. Specifically, the total control output of the basic control mode is calculated by the following formula: ; wherein, , are the proportional coefficient, the integral coefficient and the differential coefficient, respectively; is the pressure deviation at time , i.e. the difference between the target pressure value and the real-time measurement value; is a feedforward compensation function, which is a nonlinear function of the target pressure value and the current jet flow , used to generate a compensation component to offset the inherent damping and back pressure interference of the system. The compensation amount is directly added to the output end of the feedback controller, which adjusts the actuator in advance before the actual impact of the interference, so as to achieve high control accuracy and stability under steady-state conditions.

[0047] Once the controller detects that the pressure deviation exceeds the preset steady-state range or the time rate of change of pressure exceeds the set threshold, it usually means that the system is in a transient process of starting, load mutation or encountering strong interference. At this time, in order to overcome the lag and overshoot problem of traditional linear control in fast response, the controller immediately switches to an advanced control strategy. This strategy is based on a deep reinforcement learning framework, specifically using a combination of long short-term memory network and deep deterministic policy gradient algorithm architecture. This switching mechanism ensures that the system can flexibly call control algorithms of different complexity according to the severity of the working conditions.

[0048] During the operation of the advanced control strategy, the long short-term memory network layer acts as the system's state perception unit. It receives a multi-dimensional state vector composed of the current pressure value, the historical pressure sequence in the past period of time, and the error integral value. Using the unique gating mechanism of the long short-term memory network, the algorithm can effectively forget irrelevant information and remember key historical trends, thereby extracting time series features reflecting the dynamic characteristics of the system from the sequence data. This process enables the controller to capture and compensate for the inherent large lag characteristics in the super-high-speed fluid injection process, enabling early perception of pressure change trends.

[0049] Based on the extracted time series features, the policy network in the deep deterministic policy gradient algorithm is responsible for the final decision output. Unlike simple discrete control, the policy network directly outputs continuous action control quantities, which are converted into analog signals through the drive circuit to directly control the opening of the proportional valve or the displacement of the servo valve core. Through pre-training and learning, the algorithm has mastered the complex nonlinear two-dimensional function relationship between water flow, metering valve displacement, and valve pressure difference in the non-pressure-difference constant fluid injection water supply system. Therefore, even in wide speed range or highly nonlinear working conditions, the controller can output optimal control instructions to achieve fast response and no overshoot of pressure, ensuring the dynamic quality of the fluid injection process.

[0050] Step S4 is a full-automatic pressure closed-loop self-calibration mechanism, which aims to solve the problem of sensor drift caused by long-term operation and ensure the long-term accuracy of measurement data. The system runs a separate monitoring thread in the background, which uses a high-precision timer to accumulate the continuous working time of the system and simultaneously calculates the integral value of the real-time pressure deviation. When the monitoring logic detects that the system running time reaches the pre-set self-calibration period or the real-time monitored pressure deviation value continuously exceeds the set calibration trigger threshold, the system determines that the current measurement accuracy cannot meet the control requirements. At this time, the controller immediately triggers an interrupt request, safely suspends the current fluid control task, locks the actuator in a safe state, and formally enters the self-calibration process.

[0051] After the self-calibration process is initiated, the zero-point drift calibration subroutine is executed first. The controller sends a stop command to the high-pressure pump drive unit to stop the continuous pressure output and simultaneously controls the unloading valve to fully open, guiding the high-pressure fluid in the pipeline back to the storage tank for depressurization. The controller continuously monitors the rate of change of the sensor feedback signal. When it detects that the pipeline pressure no longer decreases and the signal fluctuation amplitude is lower than the preset noise benchmark, it determines that the system has reached a zero-pressure steady state. At this time, the system samples the analog voltage or digital code output by the sensor multiple times and takes the average value. This average value is confirmed as the current zero-point offset and stored in the temporary calibration register.

[0052] After zero-point calibration, the system automatically transitions to full-scale gain calibration. The controller closes the unloading valve, restarts the high-pressure pump, and employs a refined pressure ramp-up control strategy to drive the system pressure steadily up to the preset standard pressure point. This standard pressure point is typically set at 90% of the sensor's full scale, for example, 360 MPa. Choosing this point effectively reflects the sensor's linearity while avoiding reaching the system's safe overflow pressure. Once the system pressure stabilizes at the standard pressure point, the controller acquires the raw signal value output by the sensor at this time and, combined with the known standard physical pressure value, calculates the current sensor gain coefficient through division. This coefficient reflects the sensor's actual sensitivity in converting physical pressure into an electrical signal.

[0053] During the simultaneous execution of the zero-point calibration and full-scale gain calibration, the system also performs parallel updates to the temperature compensation parameters. The system reads the current sensor body temperature and ambient temperature values, using these as indexes to search within a preset pressure and temperature joint compensation model. Based on the current temperature conditions, the controller extracts or interpolates the corresponding temperature drift correction factor from the model, fine-tuning the newly acquired zero-point offset and gain coefficient. This eliminates the secondary impact of temperature changes on the calibration results, ensuring the validity of the calibration parameters under the current thermal environment.

[0054] After the calibration process is complete, the controller writes the updated zero-point offset and gain coefficient to the parameter area of ​​the non-volatile memory and real-time processing unit, overwriting the old calibration data. Subsequently, the system resumes its previous control task, but in each subsequent sampling cycle, the controller uses the updated parameters to correct the original acquired data in real time. This real-time correction process is implemented using the following pressure calibration mathematical model: ; in, This is the final pressure output value after calibration, used for closed-loop control feedback; The raw digital pressure signal acquired by the sensor; The current temperature of the sensor body; For the corresponding current sensor body temperature under the zero offset value, which is obtained by the last zero calibration value combined with the temperature compensation model interpolation; For the corresponding current sensor body temperature under the gain correction coefficient, which is calculated by the last full-scale calibration value combined with the temperature compensation model. Through the operation of the formula, the system can dynamically eliminate the baseline drift and sensitivity error, and finally output the accurate pressure value after double calibration for closed-loop control.

[0055] Step S5 is fault diagnosis and early warning based on digital twinning. This step constructs a virtual reference system running in parallel with the physical fluid system, which is used to realize active monitoring and prospective fault diagnosis of the system health state. The system first instantiates a digital twin model highly corresponding to the physical entity in the controller. This model is not a simple mathematical function, but a complex simulation body integrating the basic equations of fluid mechanics, system identification parameters, and historical operation data. It can accurately describe the dynamic behavior of fluid in the high-pressure pipeline, the output characteristics of the high-pressure pump, and the response characteristics of the actuator, thereby reproducing the core dynamics of the physical system in the virtual space.

[0056] During normal operation of the system, this digital twin model is synchronized with the physical system. Each frame of control instruction output by the controller, such as the voltage signal to drive the proportional valve or the frequency signal to control the pump motor speed, is sent to both the physical actuator and the digital twin model. After receiving these same control inputs, the digital twin model immediately performs real-time simulation calculations and outputs the theoretically predicted pressure value at the current time. At the same time, the controller compares the actual pressure measurement value processed by the physical sensor with the theoretically predicted pressure value output by the digital twin model in real time point by point, and obtains a continuous time series, i.e., the residual sequence, through subtraction operation.

[0057] This residual sequence is a key representation of the system health state. In an ideal and healthy system, the theoretically predicted value and the actual measured value should be highly consistent, and the amplitude of the residual sequence should be randomly fluctuated within a small range around zero. The trend analysis module inside the controller continuously performs statistical analysis on the residual sequence, including calculating its mean, variance, peak value, and change rate, etc. When the instantaneous value of the residual sequence or its statistical value within a certain time window exceeds the pre-set safety threshold, the system determines that there is a significant deviation between the behavior of the physical system and the theoretical model, indicating that the system has potential abnormalities or faults.

[0058] After determining the system abnormality, the fault diagnosis logic further initiates pattern matching and root cause analysis on the abnormal characteristics. For example, if the residual sequence presents a sudden step change or complete loss of signal, it usually matches the fault pattern of sensor hardware damage or line disconnection. In this case, the system will immediately execute the sensor fault-tolerant strategy, seamlessly switching the feedback source of the control loop to the pre-deployed backup sensor channel, ensuring the continuity of control. If the residual sequence presents gradually increasing oscillation or continuous one-way deviation, it is more consistent with the characteristic pattern of actuator response hysteresis, valve sticking, or control algorithm instability. At this time, the system will determine that a more serious actuator fault or control failure has occurred.

[0059] For the determination result of actuator fault or control instability, the system will immediately switch to the preset safety mode to prevent the situation from worsening. In the safety mode, the controller will suspend the complex intelligent control algorithm and lock the output of the actuator to a pre-set minimum pressure value or zero pressure value that can ensure the safety of the equipment and personnel. In some applications with extremely high safety requirements, the safety mode will directly trigger the emergency shutdown program to shut off the high-pressure power source. While performing the above protection actions, the controller will generate alarm information containing fault type code, occurrence time and related state parameters, and actively send it to the upper computer or central monitoring system through the on-board communication interface, providing clear guidance for the quick troubleshooting of the operator, thereby realizing the whole-process closed-loop safety protection of the entire ultra-high-speed water jet system.

[0060] The above-mentioned pressure closed-loop self-calibration method is applied to an ultra-high pressure water jet cutting device in this embodiment. The target pressure working range of the pressure control system of the device is set to 100-400 MPa, the control accuracy requirement is within ±0.5% of the full scale, the self-calibration period of the system is set to 24 hours, and the temperature compensation coverage range is -20-80°C.

[0061] After the system is started and the initialization is completed, the controller reads the pre-set proportional-integral-derivative control parameters and historical calibration data. The controller sends instructions to the high-pressure drive unit to start the multi-stage plunger pump, gradually increasing the pressure of the water medium to the set cutting pressure value. In this process, the piezoelectric pressure sensor installed at the outlet of the accumulator collects pressure signals in real time at a sampling frequency of 20 kHz. The controller receives and processes the signals, maintains pressure stability using an improved proportional-integral-derivative algorithm combined with a feedforward compensation strategy. When the controller detects that the deviation between the real-time pressure and the set pressure exceeds 0.1% of the full scale, it immediately outputs an adjustment signal to adjust the motor speed of the high-pressure pump or the opening of the overflow valve to suppress pressure fluctuations.

[0062] The timer inside the system continuously accumulates the running time, and whenever the running time reaches 24 hours, the system automatically triggers the self-calibration program. The self-calibration program first enters the zero-point calibration stage, and the controller instructs the high-pressure pump to stop running and opens the unloading valve to release pressure. After the pressure in the pipeline is completely released and stabilized, the system collects the output voltage value of the sensor under zero pressure, calculates the current zero-point offset, and updates the zero-point parameter in the memory.

[0063] Subsequently, the system enters the full-scale calibration stage, and the controller drives the high-pressure pump to stabilize the system pressure to 360 megapascals, which corresponds to 90% of the full scale. The system collects the output value of the sensor at this time, and calculates the new gain coefficient in combination with the standard pressure value. During the entire calibration process, the temperature sensor synchronously collects the ambient temperature and medium temperature, and the controller retrieves the pressure-temperature joint compensation model according to the collected temperature data, updates the compensation coefficient at this temperature point, so that the measurement accuracy after compensation reaches plus or minus 0.05% of the full scale.

[0064] During the cutting operation, if the thickness or hardness of the material being cut changes suddenly, causing the pressure fluctuation amplitude to increase, the controller automatically switches to a prediction control mode based on deep learning. In this mode, the system uses historical data sequences to predict the pressure demand at the next moment and adjusts the control parameters in advance, thereby improving the system response speed. At the same time, the system monitors the pressure stability, control accuracy, and energy consumption indicators in real time. Once it detects a loss of pressure sensor signal or a sustained divergence of control deviation, the system automatically switches to a backup sensor channel or executes a safety shutdown strategy, and sends an alarm signal through the human-machine interface.

[0065] This embodiment is aimed at the wide-range pressure control requirements in the supersonic combustion process of an aero-engine, with a pressure ratio control range covering 2 to 1000, a control accuracy requirement of plus or minus 1%, and a system response time of less than 50 milliseconds.

[0066] In this embodiment, the pressure sensor is a high-temperature-resistant special sensor that can work stably in an environment temperature of minus 50 degrees Celsius to 200 degrees Celsius. The sensor is integrated with a temperature compensation circuit for correcting measurement errors at extreme temperatures. The system not only relies on single-point pressure measurement, but also collects pressure distribution data through a sensor array arranged on the wall surface of the combustion chamber. A data-driven wall surface pressure reconstruction model runs inside the controller to reconstruct the pressure distribution state of the entire flow field based on limited sensor data points.

[0067] After the engine starts, the system automatically adjusts the control strategy according to the current flight state parameters and combustion chamber working conditions. In the low-speed flight stage, the controller uses a conventional proportional-integral-derivative control strategy to maintain the combustion chamber pressure. When the flight state enters the supersonic stage, the controller automatically switches to a deep reinforcement learning control strategy based on long short-term memory networks and deep deterministic policy gradient.

[0068] For example, during the acceleration process when the flight Mach number changes from 0.8 to 2.5, the inlet recovery characteristics change dramatically, resulting in sharp fluctuations in the combustion chamber inlet pressure. At this time, the intelligent control algorithm automatically identifies the current working point according to the real-time acquisition of the flow field parameters, and dynamically adjusts the weight parameters of the controller to ensure the stability of the pressure control.

[0069] The self-calibration process is automatically performed during maintenance after the engine is shut down. The system uses a built-in standard pressure source to calibrate the sensor at multiple pressure points such as low pressure, medium pressure, and high pressure. The correction data generated by the calibration is transmitted to the ground maintenance station through the wireless communication interface for subsequent offline analysis and control model optimization. At the same time, the fault diagnosis module integrated in the system monitors the health status of the sensor and the working performance of the actuator in real time. When a sensor failure is detected, the system automatically excludes the sensor data and maintains the control loop operation based on the reconstructed data of the remaining sensors to ensure flight safety.

Claims

1. A pressure closed-loop self-calibration method for ultra-high-speed jet injection process, characterized in that... Includes the following steps: S1. System initialization and parameter setting: Establish a pressure and temperature joint compensation model, which is used to correct temperature drift in sensor data in subsequent processes; S2. Real-time pressure data acquisition and processing: The pressure signal of the jet is acquired using a pressure sensor and the temperature data is acquired using a temperature sensor. The processed digital pressure signal is then transmitted to the controller. S3. Intelligent parameter adjustment: Calculate the deviation between the real-time pressure measurement value and the target jet pressure value, and select a control strategy to drive the actuator according to the state of the deviation; S4. Fully automatic pressure closed-loop self-calibration: When the trigger condition is met, the current control task is suspended, zero-point calibration and full-scale gain calibration are performed in sequence, and the parameters of the pressure and temperature joint compensation model are updated. S5. Fault Diagnosis and Early Warning: Calculate the theoretically predicted pressure value using a digital twin model, and determine anomalies by comparing the real-time pressure measurement value with the theoretically predicted pressure value.

2. The pressure closed-loop self-calibration method for ultra-high-speed jet injection process according to claim 1, characterized in that, The S1 step specifically includes: The controller reads the default configuration file from memory; Set the target jet pressure value, self-calibration cycle, and calibration trigger threshold; Initialize the control parameters of the PID controller, including the proportional coefficient, integral coefficient, and derivative coefficient; The pressure and temperature joint compensation model is read from the pre-calibrated model, which stores the zero-point drift coefficient and sensitivity drift coefficient of the sensor at different temperatures.

3. The pressure closed-loop self-calibration method for ultra-high-speed jet injection process according to claim 1, characterized in that, The S2 step specifically includes: A high-precision pressure sensor is used to collect real-time pressure signals inside the jet pipe, and an integrated temperature sensor is used to collect ambient temperature and sensor body temperature. The acquired analog pressure signal is transmitted to the signal conditioning circuit, amplified by the instrumentation amplifier, and filtered out high-frequency noise interference by the Butterworth low-pass filter. The conditioned analog signal is converted into a digital pressure signal sequence using an analog-to-digital converter; After receiving the digital pressure signal sequence, the controller performs secondary digital filtering using an infinite impulse response filter or a finite impulse response filter to obtain the real-time pressure measurement value with a high signal-to-noise ratio.

4. The pressure closed-loop self-calibration method for ultra-high-speed jet injection process according to claim 1, characterized in that, The S3 step specifically includes: Calculate the deviation between the real-time pressure measurement value and the target jet pressure value; Determine whether the deviation is within a preset steady-state range; When the deviation is within the preset steady-state range, the controller adopts the basic control mode, uses the PID algorithm combined with feedforward compensation to calculate the control quantity, and drives the actuator to move. The feedforward compensation outputs a compensation signal for known system disturbances based on a set fluid dynamics model.

5. The pressure closed-loop self-calibration method for ultra-high-speed jet injection process according to claim 4, characterized in that, The S3 step also includes: When the deviation exceeds the preset steady-state range or the pressure change rate exceeds the set threshold, the controller switches to an advanced control strategy. The advanced control strategy employs a deep reinforcement learning algorithm based on a long short-term memory network and a deep deterministic policy gradient. The Long Short-Term Memory Network receives a state vector containing current pressure, historical pressure sequences, and error integrals, and extracts time-series features. The policy network of the deep deterministic policy gradient outputs continuous motion control quantities to directly drive the proportional valve or servo valve based on the extracted time series features.

6. The pressure closed-loop self-calibration method for ultra-high-speed jet injection process according to claim 1, characterized in that, The triggering logic for step S4 is as follows: The system monitors the running time and calculates the pressure deviation in real time using a timer; When the running time reaches the set self-calibration cycle, or when the pressure deviation continues to exceed the set calibration trigger threshold, the system suspends the current control task and enters the self-calibration process.

7. The pressure closed-loop self-calibration method for ultra-high-speed jet injection process according to claim 1, characterized in that, The zero-point calibration in step S4 specifically includes: The controller commands the high-pressure pump to stop working and opens the unloading valve to relieve pressure. Once the system pressure stabilizes, the sensor output value is collected as the current zero-point offset.

8. The pressure closed-loop self-calibration method for ultra-high-speed jet injection process according to claim 1, characterized in that, The full-scale gain calibration in step S4 specifically includes: The controller drives the system to pressurize to a standard pressure point, which is 90% of the full scale. The system collects the sensor output value at this time and calculates the current gain coefficient by combining it with the standard pressure value.

9. The pressure closed-loop self-calibration method for ultra-high-speed jet injection process according to claim 1, characterized in that, The S4 step also includes: While completing the zero-point calibration and the full-scale gain calibration, the system reads the current sensor temperature value and updates the compensation parameters at the current temperature point according to the pressure and temperature joint compensation model. After calibration, the controller uses the updated zero-point offset and gain coefficient to correct subsequent pressure acquisition data in real time. The correction method is to subtract the zero-point offset from the original acquisition value and then multiply it by the gain coefficient.

10. The pressure closed-loop self-calibration method for an ultra-high-speed jet injection process according to claim 1, characterized in that, The S5 step specifically includes: The controller runs the digital twin model established based on fluid dynamics equations and system identification parameters; the digital twin model receives the same control input signals as the physical system and calculates the theoretically predicted pressure value in real time. The controller compares the real-time pressure measurement values ​​collected by the physical sensors with the theoretical predicted pressure values ​​in real time, calculates the residual sequence, and performs trend analysis. When the residual value of the residual sequence exceeds a preset safety threshold, the system is determined to be abnormal. If the abnormal characteristics match the sensor failure mode, the system switches to the backup sensor channel; If the abnormal characteristics match the actuator failure or control instability mode, the system switches to a safe mode. In the safe mode, the output is locked to a safe value or an emergency stop procedure is executed, and an alarm message containing a fault code is sent through the communication interface.

Images