A side mold stroke control method and system for a casting apparatus

Abstract

This invention discloses a method and system for controlling the side mold stroke of a casting equipment. The method includes: acquiring real-time position data of the side mold, pressure data of the hydraulic system, and temperature data of the mold contact surface through a multi-sensor fusion module; constructing a dynamic friction and thermal deformation coupling model during the side mold movement based on the real-time position data, pressure data, and temperature data, and predicting the position deviation trend in the current stroke stage; generating a dynamic compensation strategy for the side mold stroke based on fuzzy adaptive PID according to the position deviation trend and the preset mold closing accuracy requirements; executing the dynamic compensation strategy by adjusting the hydraulic cylinder flow and pressure in real time through a high-frequency servo valve group, and completing the position closed-loop correction within one stroke cycle. Using this invention, real-time prediction and accurate compensation of complex disturbances in the side mold stroke can be achieved, improving mold closing accuracy and stability, while ensuring the reliability of the equipment and the consistency of casting quality under high-speed continuous production.

Description

Technical Field

[0001] This invention belongs to the field of casting equipment control technology, and in particular to a method and system for controlling the side mold stroke of casting equipment. Background Technology

[0002] For casting equipment, especially the side molds in anti-gravity casting equipment, the stroke control accuracy of the side mold directly affects the casting quality and mold life. In existing technologies, the side mold is typically driven by a hydraulic system using conventional PID control. However, in actual operation, the movement of the side mold is subject to multiple disturbances, including complex dynamic friction, nonlinearity of the hydraulic system, and thermal deformation caused by uneven heating of the mold. These factors make it difficult to maintain stable positional accuracy of the side mold during its stroke, especially during long-cycle continuous production, where accumulated errors can easily lead to problems such as incomplete mold closing, flash on products, or dimensional deviations. Traditional control strategies struggle to perceive and compensate for this dynamic deviation coupled with multiple physical fields in real time, often relying on manual experience for post-production adjustments or slowing down production to ensure quality, severely limiting production efficiency and automation levels. Therefore, there is an urgent need for an intelligent control method for the side mold stroke that can perceive operating conditions in real time, accurately predict deviations, and adaptively compensate for them. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for controlling the side mold stroke of a casting equipment, so as to overcome the shortcomings of the prior art. It can realize real-time prediction and accurate compensation of complex interferences in the side mold stroke, improve the mold closing accuracy and stability, and at the same time ensure the reliability of the equipment and the consistency of casting quality under high-speed continuous production.

[0004] One embodiment of this application provides a method for controlling the side mold stroke of a casting equipment, the method comprising: The real-time position data of the side mold, the pressure data of the hydraulic system, and the temperature data of the mold contact surface are collected by the multi-sensor fusion module. Based on the real-time position data, the pressure data, and the temperature data, a dynamic friction and thermal deformation coupling model is constructed during the side mold movement process, and the position deviation trend at the current stroke stage is predicted. Based on the position deviation trend and the preset mold closing accuracy requirements, a dynamic compensation strategy for the side mold stroke based on fuzzy adaptive PID is generated. The dynamic compensation strategy is implemented by adjusting the flow and pressure of the hydraulic cylinder in real time through the high-frequency servo valve group, and completing the closed-loop position correction within one stroke cycle.

[0005] Another embodiment of this application provides a side mold stroke control system for a casting equipment, the system comprising: The data acquisition module is used to acquire real-time position data of the side mold, pressure data of the hydraulic system, and temperature data of the mold contact surface through the multi-sensor fusion module; The construction module is used to construct a dynamic friction and thermal deformation coupling model during the movement of the side mold based on the real-time position data, the pressure data, and the temperature data, and to predict the position deviation trend during the current stroke stage. The generation module is used to generate a dynamic compensation strategy for the side mold stroke based on fuzzy adaptive PID according to the position deviation trend and the preset mold closing accuracy requirements. The execution module is used to execute the dynamic compensation strategy, which adjusts the flow and pressure of the hydraulic cylinder in real time through the high-frequency servo valve group, and completes the position closed-loop correction within one stroke cycle.

[0006] Another embodiment of this application provides a storage medium storing a computer program, wherein the computer program is configured to execute the method described in any of the preceding claims when running.

[0007] Another embodiment of this application provides an electronic device including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the method described in any of the preceding claims.

[0008] Compared with the prior art, the side mold stroke control method of the casting equipment provided by the present invention can realize real-time prediction and accurate compensation of complex interference in the side mold stroke, improve the mold closing accuracy and stability, and at the same time ensure the reliability of the equipment and the consistency of casting quality under high-speed continuous production. Attached Figure Description

[0009] Figure 1 Hardware structure block diagram of a computer terminal for a side mold stroke control method for casting equipment provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating a method for controlling the side mold stroke of a casting equipment according to an embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of a side mold stroke control system for a casting equipment provided in an embodiment of the present invention. Detailed Implementation

[0010] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0011] The present invention first provides a method for controlling the side mold stroke of a casting equipment. This method can be applied to electronic devices, such as computer terminals, specifically ordinary computers.

[0012] The following detailed explanation uses a computer terminal as an example. Figure 1 This is a hardware structure block diagram of a computer terminal for a side mold stroke control method for casting equipment provided in an embodiment of the present invention. Figure 1 As shown, the computer device includes a processor, memory, and network interface connected via a system bus, wherein the memory may include non-volatile storage media and internal memory.

[0013] See Figure 2 The present invention provides a method for controlling the side mold stroke of a casting equipment, which may include the following steps: S201 collects real-time position data of the side mold, pressure data of the hydraulic system, and temperature data of the mold contact surface through a multi-sensor fusion module; Specifically, a high-precision magnetic scale sensor can be installed on the side mold guide rail to collect the displacement signal of the side mold in real time and generate the original position pulse sequence; The core of this step is to capture the real-time position information of the side mold during its movement using high-precision displacement sensing equipment, converting the mechanical displacement into pulse signals that can be processed subsequently, thus providing basic position data support for side mold stroke control. The specific implementation method is as follows: When selecting the installation location for the side mold guide rail, priority should be given to areas where the guide rail is not under stress and is parallel to the side mold's movement trajectory. This ensures that the sensor's acquisition direction is completely consistent with the actual movement direction of the side mold, avoiding displacement measurement errors caused by installation angle deviations. During installation, the installation gap between the sensor reading head and the magnetic scale should be controlled at 0.1-0.2 mm. This gap range has been experimentally verified to balance signal stability and anti-interference capability. Too small a gap can easily lead to mechanical friction damage to components, while too large a gap will weaken the electromagnetic induction signal strength and affect measurement accuracy. Simultaneously, using an elastic mounting base can absorb slight vibrations during the side mold's movement, reducing the impact of vibration on the sensor.

[0014] The core technical parameters of the selected high-precision magnetic scale sensor are: measurement resolution of 0.001 mm, sampling frequency of 1000 Hz, and measurement stroke matching the maximum stroke of the side mold (e.g., if the maximum stroke of the side mold is 500 mm, the sensor measurement stroke is set to 600 mm, with a safety redundancy). The measurement resolution of 0.001 mm determines the minimum accuracy unit of position measurement, meeting the basic requirement of ±0.01 mm for mold closing accuracy in casting equipment. The sampling frequency of 1000 Hz means that 1000 position samples can be completed per second, accurately capturing instantaneous position changes during high-speed movement of the side mold (maximum movement speed ≤10 mm / s), avoiding data loss due to untimely sampling.

[0015] The displacement signal acquisition and pulse sequence generation process is as follows: When the side mold moves along the guide rail, the magnetic grid of the magnetic grating ruler and the sensor reading head move relative to each other. The electromagnetic induction element inside the reading head senses the change in the magnetic poles of the magnetic grating and converts it into a periodic electrical signal. After being shaped and amplified by the internal signal processing circuit of the sensor, a standard square wave pulse sequence is output. The frequency of the pulse sequence is linearly positively correlated with the movement speed of the side mold, and the cumulative number of pulses corresponds one-to-one with the displacement of the side mold. The conversion relationship is: displacement = cumulative number of pulses × sensor resolution. For example, when the side mold moves at a constant speed of 5 mm / s, the number of pulses generated per second is 5 mm / s ÷ 0.001 mm / pulse = 5000, and the corresponding pulse sequence frequency is 5000 Hz; when the side mold moves 100 mm, the cumulative number of pulses is 100 mm ÷ 0.001 mm / pulse = 100000. The generated raw position pulse sequence must include the acquisition timestamp of each pulse frame (with an accuracy of 1 microsecond), in the format of "timestamp-pulse count", such as "2025-10-01 14:30:00.000001-1000", to provide a time reference for subsequent multi-sensor signal synchronization.

[0016] Piezoelectric pressure sensors are installed in the inlet and outlet oil circuits of the hydraulic cylinder to collect pressure fluctuation signals of the hydraulic system in real time and generate raw pressure simulation signals. The core of this step is to accurately capture the pressure changes when the hydraulic system drives the side mold movement, especially the details of pressure fluctuations, to provide hydraulic system status data for subsequent analysis of side mold movement load changes. The specific implementation method is as follows: The piezoelectric pressure sensor is installed at the dedicated pressure measurement interface of the hydraulic cylinder's inlet and outlet, using a vertical installation method with the sensor's pressure measurement port facing downwards. This effectively prevents air bubbles from accumulating at the sensor's sensitive element, thus preventing pressure measurement distortion caused by air bubbles. During installation, a sealing connector matching the hydraulic circuit pressure rating must be used, with a sealing pressure rating of no less than 31.5 MPa, ensuring no leakage in the high-pressure hydraulic circuit. Simultaneously, a small filter is installed between the sensor and the hydraulic circuit to filter impurities in the hydraulic circuit, preventing impurities from damaging the sensor's sensitive element.

[0017] The core technical parameters of the sensor are set as follows: measurement range 0 - 31.5 MPa, measurement accuracy class 0.1, response frequency 1000 Hz, and the output signal type is a charge signal. Among them, the measurement range of 0 - 31.5 MPa covers the normal working pressure range (5 - 25 MPa) of the hydraulic system of the casting equipment, leaving a certain pressure redundancy to cope with the instantaneous pressure peak of the system; the measurement accuracy class 0.1 means that the maximum measurement error within the full measurement range does not exceed ±0.0315 MPa, enabling precise capture of minute pressure fluctuations; the response frequency of 1000 Hz is consistent with the sampling frequency of the displacement sensor, ensuring the time synchronization of the pressure signal and the position signal; the charge signal output needs to be used in conjunction with a subsequent charge amplifier to enhance the signal anti-interference ability.

[0018] The process of pressure fluctuation signal acquisition and analog signal generation is as follows: When the hydraulic system is operating, the pressure in the oil circuit acts on the piezoelectric crystal sensitive element of the sensor. The piezoelectric crystal generates a charge signal proportional to the pressure magnitude under the action of the pressure. This charge signal is converted into a voltage-type analog signal by the charge amplifier supporting the sensor, and the conversion gain is set to 1000 pC / V (1000 picocoulombs of charge corresponding to each volt of voltage). For example, when the pressure at the inlet of the hydraulic cylinder is 15 MPa, the charge generated by the piezoelectric crystal is 15000 picocoulombs, and after conversion by the charge amplifier, a voltage of 15 V is output (15000 pC ÷ 1000 pC / V = 15 V); when there is a pressure fluctuation in the system, such as the pressure instantaneously rising from 15 MPa to 16 MPa, the output voltage synchronously rises from 15 V to 16 V, and the fluctuation response time ≤ 1 ms, enabling precise capture of the dynamic pressure changes of the hydraulic system. The generated original pressure analog signal is a continuously varying DC voltage signal with a voltage range of 0 - 20 V, corresponding to a pressure range of 0 - 31.5 MPa. The signal contains key information such as the steady-state pressure, instantaneous pressure peak, and pressure fluctuation frequency of the hydraulic system. At the same time, each sensor channel needs to record the time stamp of signal acquisition, and the time stamp accuracy is consistent with that of the displacement signal (1 microsecond) to ensure the time alignment of multi-channel signals.

[0019] Install an infrared thermocouple array on the contact surface of the mold to collect the temperature distribution signal of the mold contact surface in real time and generate an original temperature analog signal; The core of this step is to comprehensively capture the temperature distribution and change trend in the contact area between the side mold and the mold, providing temperature data for subsequent analysis of the influence of thermal deformation on the position of the side mold. The specific implementation method is as follows: The infrared thermocouple array is installed using an embedded mounting method. Mounting holes are pre-drilled on the end face where the side mold contacts the mold, with a depth of 5 mm. This ensures the sensor's sensitive element is flush with the mold contact surface, preventing poor mold fit due to sensor protrusion. The array density is set according to the size of the mold contact surface, typically one infrared thermocouple measuring point per square centimeter. For example, if the mold contact surface is 200 mm × 300 mm, a total of 600 measuring points are arranged to form a uniform temperature measuring array, fully covering the contact surface and avoiding temperature measurement blind spots. After installation, the sensor surface is sealed with a high-temperature resistant sealant. The sealant has a temperature resistance rating of no less than 300 degrees Celsius, suitable for the working temperature environment of the casting mold.

[0020] The core technical parameters of the infrared thermocouple array are set as follows: temperature measurement range 0-300 degrees Celsius, measurement accuracy ±1 degree Celsius, response time 50 milliseconds, and output signal type millivolt-level analog signal. The temperature measurement range of 0-300 degrees Celsius covers the conventional working temperature of casting molds (80-250 degrees Celsius), meeting the temperature measurement requirements of different casting conditions; the measurement accuracy of ±1 degree Celsius ensures the accuracy of temperature data, providing a reliable basis for thermal deformation calculation; the 50 millisecond response time can promptly capture dynamic changes in mold temperature, avoiding thermal deformation prediction errors caused by temperature response lag; the millivolt-level analog signal output voltage range is 0-60 millivolts, corresponding to the temperature measurement range of 0-300 degrees Celsius, with a linear conversion relationship between voltage and temperature, and a conversion factor of 0.2 millivolts / degree Celsius (0.2 millivolts per degree Celsius of temperature).

[0021] The temperature distribution signal acquisition and analog signal generation process is as follows: Infrared thermocouples absorb infrared radiation energy from the mold contact surface through their sensitive elements. Simultaneously, utilizing the thermoelectric effect of the thermocouples, temperature changes are converted into corresponding millivolt-level voltage signals. Each measuring point's infrared thermocouple independently acquires the temperature signal at its corresponding location. After being aggregated by an array signal collector, multiple parallel raw temperature analog signals are output, each signal corresponding to a single temperature measuring point. For example, when the temperature of the mold contact surface at a measuring point is 180 degrees Celsius, the output voltage signal at that measuring point is 36 millivolts (180 degrees Celsius × 0.2 millivolts / degree Celsius = 36mV). When a localized high-temperature zone appears on the contact surface, such as when the temperature at a measuring point rises to 220 degrees Celsius, the corresponding output voltage rises synchronously to 44 millivolts. If the temperature distribution on the contact surface is uneven, such as an edge area temperature of 150 degrees Celsius and a central area temperature of 200 degrees Celsius, the corresponding measuring points in these areas will output voltages of 30 millivolts and 40 millivolts respectively, clearly reflecting the differences in temperature distribution. The generated raw temperature simulation signal is a multi-channel parallel millivolt-level DC simulation signal. Each signal is accompanied by an independent timestamp and measurement point number. The timestamp accuracy is 1 microsecond, which is consistent with the displacement and pressure signals, facilitating the subsequent synchronous processing of multi-sensor signals. The signal contains key information such as the overall temperature level of the mold contact surface, local temperature peaks, temperature distribution gradient, and temperature change rate.

[0022] The original position pulse sequence, original pressure analog signal and original temperature analog signal are acquired synchronously by a multi-channel data acquisition card, and the signals are conditioned and converted from analog to digital to generate a multi-sensor digital signal set with timestamp alignment.

[0023] The core of this step is to achieve synchronous acquisition, signal optimization processing, and digital conversion of signals from multiple sensors, generating a unified format, time-aligned digital signal set to provide standardized data for subsequent model building. The specific implementation method is as follows: The core parameters of the multi-channel data acquisition card are set as follows: 16 analog input channels, 8 digital input channels, 1000 Hz sampling frequency, 16-bit analog-to-digital conversion resolution, and 1 microsecond system clock accuracy. The analog input channels are used to acquire raw pressure and temperature analog signals (requiring a total of 2 pressure signal channels + 600 temperature signal channels, which can be achieved by cascading multiple acquisition cards). The digital input channels are used to acquire raw position pulse sequences (1 pulse signal channel). The 1000 Hz sampling frequency matches the output signal frequency of each sensor, ensuring distortion-free signal acquisition. The 16-bit analog-to-digital conversion resolution converts the analog signal into 65,536 quantization levels, guaranteeing the accuracy of the digitized signal. The 1 microsecond system clock accuracy ensures precise synchronization of the acquisition timestamps for all channels, with a timestamp error ≤ 1 microsecond.

[0024] During signal acquisition and conditioning, the original position pulse sequence is a digital signal, which is directly acquired through the digital input channel of the acquisition card. Before acquisition, the signal needs to be shaped by the signal buffer to remove high-frequency noise and glitches. The shaping threshold is set to 2.5 volts (above 2.5 volts is considered high level, and below 2.5 volts is considered low level) to ensure the integrity of the pulse signal. The original pressure and temperature analog signals are analog signals and need to be preprocessed by the signal conditioning module of the acquisition card. The conditioning module includes three functions: amplification, filtering, and isolation. The amplification stage is designed for the weak millivolt-level temperature signal, setting the amplification gain to 100 times, amplifying the 0-60 millivolt temperature signal to 0-6 volts, which is compatible with the 0-20 volt range of the pressure signal. The filtering stage uses a second-order low-pass filter with a cutoff frequency set to 100 Hz, which can effectively filter out high-frequency interference in the signal (such as high-frequency noise caused by hydraulic system vibration and electromagnetic radiation interference). The filtered signal waveform is smoother and retains the effective signal components. The isolation stage uses opto-isolation technology with an isolation voltage ≥2500 volts, which can avoid signal interference between different sensor channels and protect the acquisition card from damage by high-voltage signals.

[0025] The analog-to-digital conversion process is completed by the core processing unit of the acquisition card, converting the conditioned analog signal into a digital quantity. The conversion adopts the successive approximation conversion principle, with a conversion rate of 1 microsecond per channel. For example, the conditioned pressure signal voltage is 15 volts, corresponding to a digital quantity of 49152 (at 16-bit resolution, 0-20 volts correspond to a digital quantity of 0-65535, the conversion formula is digital quantity = (input voltage / full-scale voltage) × 65535, i.e., 15V / 20V × 65535 = 49152); the conditioned temperature signal voltage is 3.6 volts (original 36 millivolts × 100 times amplification), corresponding to a digital quantity of 11796 (3.6V / 20V × 65535 = 11796). The digitized signal corresponds one-to-one with the original signal, and each digital signal is associated with a unique timestamp, channel number, and signal type (location / pressure / temperature) information.

[0026] The final generated multi-sensor digital signal set is stored in a structured format, with each data record containing "timestamp-channel number-signal type-digital quantity-converted value of the original physical quantity", for example... "2025-10-01 14:30:00.000001-1-Position-100000-100.000 mm", "2025-10-01 14:30:00.000001-2-Pressure-49152-15.00 MPa", "2025-10-01 14:30:00.000001-3-Temperature-11796-180.0 degrees Celsius", all data are arranged in ascending order by timestamp to ensure strict time alignment of signals from different types of sensors, providing high-quality, synchronized, multi-dimensional data support for the subsequent construction of a dynamic friction and thermal deformation coupling model.

[0027] S202, Based on the real-time position data, the pressure data and the temperature data, construct a dynamic friction and thermal deformation coupling model during the side mold movement process, and predict the position deviation trend in the current stroke stage; Specifically, feature extraction can be performed on the multi-sensor digital signal set aligned with timestamps, the instantaneous velocity and acceleration of the side mold can be calculated from the position data, the load change rate of the hydraulic system can be calculated from the pressure data, and the temperature gradient of the mold contact surface can be calculated from the temperature data to generate a motion state feature vector; The core of this step is to extract key features that characterize the motion state of the side model and the influence of the environment from synchronized multi-sensor data, transforming the raw data into physically meaningful feature parameters, and providing structured data support for subsequent model input. The specific implementation method is as follows: Before feature extraction, the multi-sensor digital signal set needs to be preprocessed to remove outlier data points (using the 3σ criterion, i.e., removing data that deviates from the mean by three times the standard deviation; for example, if the mean of location data is 100.000 mm and the standard deviation is 0.002 mm, data smaller than 99.994 mm or larger than 100.006 mm will be removed). This avoids outliers affecting the accuracy of feature calculation. The preprocessed data is arranged in timestamp order to ensure temporal continuity, and the sampling period T is fixed at 0.001 seconds (corresponding to a sensor sampling frequency of 1000 Hz, T = 1 / sampling frequency), providing a unified time reference for subsequent difference calculations.

[0028] The instantaneous velocity of the side mold is calculated using the first-order central difference method. This method calculates the instantaneous velocity at the intermediate moment using position data from three adjacent moments, effectively reducing the impact of single-point data errors. The calculation formula is v_k=(x_{k+1}-x_{k-1}) / (2T), where v_k is the instantaneous velocity at moment k, x_{k+1}, x_k, and x_{k-1} are the side mold position data at moments k+1, k, and k-1, respectively, and T is the sampling period. For example, at moment k-1, the position x_{k-1}=100.000mm, at moment k, x_k=100.005mm, and at moment k+1, x_{k+1}=100.012mm. Substituting these values ​​into the formula, we get v_k=(100.012-100.000) / (2×0.001)=6mm / s. This result accurately reflects the movement velocity of the side mold at moment k.

[0029] The instantaneous acceleration is calculated based on the instantaneous velocity results, also using the first-order central difference method. The formula is a_k=(v_{k+1}-v_{k-1}) / (2T), where a_k is the instantaneous acceleration at time k, and v_{k+1} and v_{k-1} are the instantaneous velocities at times k+1 and k-1, respectively. For example, if the velocity at time k-1 is v_{k-1}=5mm / s and the velocity at time k+1 is v_{k+1}=7mm / s, substituting these values, we get a_k=(7-5) / (2×0.001)=1000mm / s², or 1m / s². This value characterizes the rate of change of the side mold velocity and reflects the smoothness of the motion.

[0030] The calculation of the load change rate of a hydraulic system requires first converting pressure data into load data, and then calculating the load change rate. The conversion relationship between load and pressure is F = p × A, where F is the hydraulic system load, p is the hydraulic cylinder inlet pressure, and A is the effective area of ​​the hydraulic cylinder piston (preset parameter, e.g., A = 0.01 m²). The load change rate is calculated using the first-order forward difference method, with the formula ΔF / Δt = (F_k - F_{k-1}) / T, where ΔF / Δt is the load change rate, and F_k and F_{k-1} are the loads at times k and k-1, respectively. For example, at time k-1, the pressure p_{k-1} = 15MPa (15×10^6Pa), and the load F_{k-1} = 15×10^6×0.01 = 150000N; at time k, the pressure p_k = 15.2MPa, and the load F_k = 15.2×10^6×0.01 = 152000N. Substituting these values, we get the load change rate = (152000-150000) / 0.001 = 2×10^7N / s. This value reflects the dynamic change of the load driven by the hydraulic system and is directly related to the resistance of the side mold movement.

[0031] The temperature gradient of the mold contact surface is calculated based on multi-point temperature data from an infrared thermocouple array. The temperature gradient characterizes the rate of temperature change in space, and the gradients in the x-direction (side mold movement direction) and y-direction (perpendicular to the movement direction) need to be calculated separately. The calculation formulas are G_x=(T_i,j-T_i-1,j) / d and G_y=(T_i,j-T_i,j-1) / d, where G_x and G_y are the temperature gradients in the x and y directions, respectively, T_i,j is the temperature of the measuring point in the i-th row and j-th column, and d is the spacing between adjacent measuring points (preset to 10mm, as the array arrangement density is one measuring point per square centimeter). For example, if the temperature at measuring point (2,3) is T_2,3=180℃ and the temperature at the adjacent measuring point (1,3) is T_1,3=178℃, then the gradient in the x-direction is G_x=(180-178) / 10=0.2℃ / mm; and the temperature at the adjacent measuring point (2,2) is T_2,2=179℃, then the gradient in the y-direction is G_y=(180-179) / 10=0.1℃ / mm. The temperature gradient reflects the uniformity of temperature distribution on the mold contact surface and is a key parameter for calculating thermal deformation.

[0032] The generation of motion state feature vectors involves combining the extracted feature parameters in a fixed order to form a multi-dimensional vector. The vector dimension is set according to the number of features. For example, it may contain five features: instantaneous velocity, instantaneous acceleration, load change rate, x-direction temperature gradient, and y-direction temperature gradient, forming a 5-dimensional feature vector. The example feature vector is [6mm / s, 1000mm / s², 2×10^7N / s, 0.2℃ / mm, 0.1℃ / mm]. To facilitate processing by deep learning networks, the feature vector needs to be normalized, mapping each feature value to the interval [0,1]. The normalization formula is f_norm=(f-f_min) / (f_max-f_min), where f_norm is the normalized feature value, f is the original feature value, and f_min and f_max are the historical minimum and maximum values ​​of the feature (for example, if the velocity f_min=0mm / s and f_max=20mm / s, then 6mm / s is normalized to 0.3). Finally, the normalized motion state feature vector is generated.

[0033] The motion state feature vector is input into a pre-trained deep learning network, which learns the nonlinear friction characteristics and thermal deformation laws in the side mode motion through historical data, and generates dynamic friction coefficient estimates and thermal deformation compensation amounts. The core of this step is to utilize a pre-trained deep learning network to mine the complex mapping relationship between motion state features and nonlinear friction and thermal deformation, thereby achieving accurate estimation of key parameters. The specific implementation method is as follows: The deep learning network adopts a hybrid architecture of "Convolutional Neural Network (CNN) + Long Short-Term Memory Network (LSTM)". CNN is used to extract spatial features (such as the spatial distribution of temperature gradient) from the motion state feature vector, and LSTM is used to capture temporal features (such as the changing trends of speed and pressure over time). The combination of the two can fully explore multi-dimensional and temporal feature information. The network structure from the input layer to the output layer is as follows: input layer (receives the normalized motion state feature vector, dimension 5), CNN layer (2 layers, 3×1 kernel size, 32 neurons, ReLU activation function, used to extract spatial correlation features), LSTM layer (2 layers, 64 hidden neurons, used to capture temporal dependencies), fully connected layer (2 layers, 32 and 16 neurons respectively, ReLU activation function), output layer (2 neurons, outputting the dynamic friction coefficient estimate and thermal deformation compensation amount respectively, sigmoid activation function, the output value is mapped to the actual physical quantity range).

[0034] The network pre-training process uses historical inspection data as the training set, which contains over 100,000 samples. Each sample consists of a motion state feature vector, the actual dynamic friction coefficient, and the actual thermal deformation compensation amount. The historical data covers different casting conditions (such as different mold temperatures, different side mold movement speeds, and different hydraulic loads) to ensure that the network can learn comprehensive nonlinear laws. During training, the mean squared error (MSE) is used as the loss function, calculated as Loss = 1 / N × Σ[(y_pred - y_true)²], where N is the number of samples, y_pred is the network prediction value, and y_true is the actual value. The optimizer uses adaptive momentum estimation (Adam), with a learning rate set to 0.001 and 5000 iterations. Training stops when the loss function value converges to below 0.0001, ensuring high estimation accuracy of the network.

[0035] The learning of nonlinear friction characteristics focuses on the variation of the friction coefficient under different motion states. For example, when the side mold moves at low speed (≤2mm / s), friction is mainly static friction, with a large coefficient that changes drastically with speed; when moving at high speed (≥10mm / s), friction is mainly kinetic friction, with the coefficient tending to be stable but significantly affected by temperature. The network establishes a nonlinear mapping by learning the correlation between "speed, temperature, load" and the friction coefficient in historical data. For example, when the input feature vector has a speed of 2mm / s and a temperature gradient of 0.5℃ / mm, it can accurately estimate the corresponding static friction coefficient of 0.12.

[0036] The study of thermal deformation laws focuses on the relationship between temperature distribution and thermal deformation. The thermal deformation of the side mold follows the law of linear expansion, meaning the deformation is proportional to the temperature change, the material's coefficient of linear expansion, and the length. However, the actual deformation is nonlinear due to factors such as temperature gradient and material inhomogeneity. The network learns the relationship between "temperature gradient, average temperature, and motion time" and the actual thermal deformation from historical data. For example, when the average temperature of the mold contact surface is 200℃ and the temperature gradient in the x-direction is 0.3℃ / mm, it can estimate a thermal deformation compensation of 0.003mm along the x-direction for the side mold. This compensation provides a basis for subsequent model corrections.

[0037] The inference process after the motion state feature vector is input into the pre-trained network is as follows: the normalized feature vector is processed by a CNN layer to extract spatial features, then processed by an LSTM layer to handle temporal correlation, and finally mapped to the output layer through a fully connected layer. Example input normalized feature vector is [0.3, 0.05, 0.4, 0.2, 0.1] (corresponding to the original features [6mm / s, 1000mm / s², 2×10^7 N / s, 0.2℃ / mm, 0.1℃ / mm]). The network outputs a dynamic friction coefficient estimate of 0.05 and a thermal deformation compensation of 0.002mm. The output needs to be denormalized to restore it to the actual physical quantity range (friction coefficient range 0.01-0.2, thermal deformation compensation range 0-0.01mm).

[0038] Based on the estimated value of dynamic friction coefficient and thermal deformation compensation, a dynamic friction and thermal deformation coupling model is established during the movement of the side mold. The model takes the current control input as the independent variable and outputs the predicted value of the side mold position. The core of this step is to integrate the effects of dynamic friction and thermal deformation to construct a coupled model that can accurately describe the motion law of the side mold, thereby achieving accurate prediction of the side mold position. The specific implementation method is as follows: The dynamic friction and thermal deformation coupled model is constructed using a hybrid modeling approach of "mechanism model + data-driven correction". The mechanism model, based on classical mechanics and thermal expansion theory, describes the basic laws of side mold motion. The data-driven correction term, based on the dynamic friction coefficient and thermal deformation compensation amount output by a deep learning network, compensates for the mechanism model's shortcomings in describing nonlinear factors. The core expression of the model is x_pred = x_ideal - Δx_friction - Δx_thermal, where x_pred is the predicted side mold position, x_ideal is the ideal position based on the control input, Δx_friction is the position deviation caused by friction, and Δx_thermal is the position deviation caused by thermal deformation.

[0039] The ideal position x_ideal is calculated based on the control input: the control input is the target flow rate Q of the hydraulic cylinder, and the relationship between the hydraulic cylinder piston speed v_ideal and the flow rate is Q=A×v_ideal, therefore v_ideal=Q / A. The ideal position x_ideal is the integral of velocity over time, i.e., x_ideal=x_0+∫v_idealdt (the integration interval is from 0 to the current time t), where x_0 is the initial position of the side mold. For example, if the current control input target flow rate Q=0.00001m³ / s, and the effective area of ​​the hydraulic cylinder piston A=0.01m², then v_ideal=0.00001 / 0.01=0.001m / s=1mm / s. If the initial position x_0=100.000mm and the movement time t=0.1s, then x_ideal=100.000+1×0.1=100.100mm.

[0040] The positional deviation Δx_friction caused by friction is calculated based on a dynamic friction model, employing a composite model of Coulomb friction and viscous friction. The formula is Δx_friction = (F_friction / m) × t² / 2, where F_friction is the frictional force, m is the mass of the side mold, and t is the motion time. The frictional force F_friction = μ × F_N, where μ is the dynamic friction coefficient output by the deep learning network, and F_N is the normal force of the side mold (equal to the hydraulic system load F, i.e., F_N = F = p × A). In the example, the dynamic friction coefficient μ=0.05, the hydraulic load F=150000N, the side mold mass m=500kg, and the movement time t=0.1s. Then F_friction=0.05×150000=7500N, Δx_friction=(7500 / 500)×(0.1)² / 2=0.075×0.01 / 2=0.000375mm, that is, friction causes the actual position of the side mold to lag behind the ideal position by 0.000375mm.

[0041] The positional deviation Δx_thermal caused by thermal deformation is directly calculated using the thermal deformation compensation amount output by the deep learning network. This compensation amount already considers nonlinear factors such as temperature gradient and material properties, requiring no additional correction. In the example, the network outputs a thermal deformation compensation amount of 0.002mm, i.e., Δx_thermal = 0.002mm. Due to thermal expansion, the actual position of the side mold is offset by 0.002mm from the ideal position (if the temperature increases, the side mold expands, and the actual position is greater than the ideal position, with the deviation taking a positive value; the opposite is true when the temperature decreases).

[0042] Integration and Validation of the Coupled Model: The ideal position, friction deviation, and thermal deformation deviation are substituted into the core expression to calculate the predicted position value. In the example, x_ideal = 100.100 mm, Δx_friction = 0.000375 mm, Δx_thermal = 0.002 mm, then x_pred = 100.100 - 0.000375 - 0.002 = 100.097625 mm. Model validation involves comparing real-time position data with the predicted values. A prediction error threshold of 0.005 mm is set. If the error exceeds the threshold, model parameters need to be adjusted (e.g., correcting the side mold mass m, piston area A, etc.) to ensure the model's prediction accuracy meets the requirements.

[0043] The model's independent and output variables are defined as follows: The independent variable is the current control input, i.e., the target flow rate Q (or target pressure p, which can be converted between each other through the characteristics of the hydraulic system) of the hydraulic cylinder; the output is the predicted value x_pred of the side mold position. The time resolution of the predicted value is consistent with the sensor sampling frequency (1000Hz), i.e., 1000 predicted positions are output per second to ensure real-time tracking of the side mold's motion state. In addition, the model also needs to output the confidence level of the predicted value. The confidence level is calculated based on the historical prediction error, and the formula is confidence=1-|x_pred-x_actual| / x_actual, where x_actual is the historical actual position, the confidence level range is [0,1], and the prediction is considered valid when the confidence level is ≥0.95.

[0044] The current control input and real-time position data are input into the dynamic friction and thermal deformation coupling model. The residual between the predicted value and the actual measured value of the side mold position is calculated. The position deviation trend of the next few control cycles is predicted through time series analysis.

[0045] The core of this step is to predict the future trend of positional deviation by analyzing the residual between the model's predicted values ​​and the actual values, combined with time series analysis methods. This provides a basis for generating subsequent dynamic compensation strategies. The specific implementation method is as follows: The core logic of residual calculation is to quantify the deviation between the model's predicted value and the actual measured value. The formula is e_k = x_pred,k - x_actual,k, where e_k is the residual at time k, x_pred,k is the model's predicted position at time k, and x_actual,k is the real-time measured position at time k (position data from a multi-sensor digital signal set). The sign of the residual indicates the direction of the deviation; a positive value indicates that the predicted position is greater than the actual position, and a negative value indicates that the predicted position is less than the actual position. The absolute value of the residual indicates the magnitude of the deviation, reflecting the model's prediction accuracy. In the example, at time k, x_pred,k = 100.097625mm, x_actual,k = 100.096mm, then e_k = 100.097625 - 100.096 = 0.001625mm, indicating that the predicted position is slightly greater than the actual position, and the deviation is within an acceptable range (preset residual threshold 0.005mm).

[0046] Residual preprocessing: To eliminate the influence of random noise on trend analysis, the residual sequence needs to be smoothed. A moving average filtering algorithm is used, with a window size set to 5 (i.e., the average of the residuals at the current time and the previous 4 time steps is taken as the smoothed residual). The formula is as follows: e_smooth,k=(e_k+e_{k-1}+e_{k-2}+e_{k-3}+e_{k-4}) / 5. Example: The original residual sequence is [0.001625,0.0018,0.002,0.0021,0.0023], then the smoothed... e_smooth,k=(0.001625+0.0018+0.002+0.0021+0.0023) / 5=0.001965mm. The smoothed residual sequence better reflects the true trend of deviation change.

[0047] The time series analysis employed an Autoregressive Integrated Moving Average (ARIMA) model, suitable for analyzing stationary or non-stationary time series data, effectively capturing the trends and periodicity of the data. The parameters of the ARIMA model are represented as ARIMA(p,d,q), where p is the autoregressive order (representing the correlation between the current value and the values ​​of the previous p time points), d is the differencing order (used to transform a non-stationary series into a stationary one), and q is the moving average order (representing the correlation between the current value and the error term of the previous q time points). By performing a stationarity test on the smoothed residual series (using the ADF test), d=1 was determined (the series is stationary after one differencing). Through analysis of the autocorrelation function (ACF) and partial autocorrelation function (PACF), p=2 and q=1 were determined, resulting in the final model ARIMA(2,1,1).

[0048] The model parameters were estimated using the least squares method. The autoregressive coefficients φ_1, φ_2, and moving average coefficient θ_1 were estimated using historical residual data (smoothed residuals from the first 100 time points). Example estimation results were φ_1=0.6, φ_2=0.3, and θ_1=0.2. The model expression is as follows: Δe_smooth,k=φ_1×Δe_smooth,k-1+φ_2×Δe_smooth,k-2+ε_k+θ_1×ε_k-1, where Δe_smooth,k is the first-order difference of the smoothed residual, and ε_k is the white noise error term.

[0049] Prediction of future position deviation trends: Based on a trained ARIMA(2,1,1) model, the smoothed residuals from the most recent p time steps and the error terms from the last q time steps are input to predict the residuals for the next few control cycles. The prediction step is set to 3 control cycles (each control cycle T=0.001s), that is, predicting the residuals at 0.001s, 0.002s, and 0.003s in the future. Example input: the difference residuals of the two most recent time steps Δe_smooth,k-1=0.0001 and Δe_smooth,k-2=0.00008, and the error term ε_k-1=0.00005 of the most recent time step. Substituting these into the model, we get Δe_smooth,k+1=0.6×0.0001+0.3×0.00008+ε_k+1+0.2×0.00005. Assuming ε_k+1=0 (the white noise mean is 0), then Δe =_smooth,k+1=0.00006+0.000024+0.00001=0.000094, corresponding to e_smooth,k+1=e_smooth,k+Δe_smooth,k+1=0.001965+0.000094=0.002059mm; similarly, we predict e_smooth,k+2=0.00215mm and e_smooth,k+3=0.00224mm.

[0050] Characterization of position deviation trend: The predicted residuals are transformed into a position deviation trend, where the position deviation at future times equals the predicted residuals. The deviation trend for the next three control cycles is 0.002059mm, 0.00215mm, and 0.00224mm, respectively, showing a slow increasing trend. This indicates that the side mold position deviation will gradually increase, requiring timely compensation strategies for correction. Simultaneously, the confidence interval for the deviation trend is output (based on three times the standard deviation of the model prediction error). An example confidence interval is ±0.0001mm, ensuring the reliability of the trend prediction.

[0051] S203, Based on the position deviation trend and the preset mold closing accuracy requirements, generate a dynamic compensation strategy for the side mold stroke based on fuzzy adaptive PID. Specifically, the position deviation trend can be compared with the preset mold closing accuracy requirements, the deviation change rate and the cumulative deviation amount can be calculated, and the position control deviation index can be generated. The core of this step is to extract key parameters that reflect the dynamic characteristics of the deviation by quantitatively comparing the deviation trend with the accuracy threshold, forming standardized control deviation indicators to provide accurate input for subsequent fuzzy inference. The specific implementation method is as follows: The preset mold closing accuracy requirements need to be set in conjunction with the process requirements of the casting equipment. The core parameter is the allowable deviation threshold for the mold closing position, which is set to ±0.01 mm. That is, the deviation between the actual position and the target position of the side mold needs to be controlled within the range of -0.01 mm to 0.01 mm. If it exceeds this range, it is judged as not meeting the accuracy requirements, and compensation control needs to be activated. This threshold is determined based on the casting forming quality requirements. Excessive deviation will lead to defects such as misalignment of the casting parting surface and excessive flash. Therefore, it is used as the core benchmark for deviation trend evaluation.

[0052] The comparison process between the position deviation trend and the accuracy requirements requires first clarifying the quantitative representation of the deviation trend. Assuming that, based on the time series analysis described earlier, the predicted position deviations for the next three control cycles (each control cycle T = 0.001 seconds, corresponding to a sensor sampling frequency of 1000Hz) are e_1 = 0.002059 mm, e_2 = 0.00215 mm, and e_3 = 0.00224 mm, respectively, all within the mold closing accuracy threshold range, but showing a continuously increasing trend, compensation is needed to suppress the expansion of the deviation. The comparison outputs basic information such as the deviation exceeding the limit status (currently not exceeding the limit but with an increasing trend), the current maximum deviation (0.00224 mm), and the difference between the deviation and the threshold (0.01 - 0.00224 = 0.00776 mm), providing a basis for subsequent parameter calculations.

[0053] The rate of change of deviation is used to characterize the rate of increase or decrease of deviation. The first-order forward difference method is employed, with the formula ec_k = (e_k - e_{k-1}) / T, where ec_k is the rate of change of deviation in the k-th control period, e_k is the prediction deviation in the k-th period, e_{k-1} is the prediction deviation in the (k-1)-th period, and T is the control period. For example, the rate of change in the second period relative to the first period is calculated as follows: ec_2 = (0.00215 - 0.002059) / 0.001 = 0.091 mm / s; the rate of change in the third period relative to the second period is calculated as follows: ec_3 = (0.00224 - 0.00215) / 0.001 = 0.09 mm / s. A positive deviation change rate indicates that the deviation is increasing positively (approaching the upper limit of the accuracy threshold), while a negative value indicates that the deviation is decreasing. The larger the absolute value, the more drastic the deviation change. Here, a change rate of 0.09 mm / s indicates that the deviation is increasing at a steady rate, and the compensation response speed needs to be appropriately enhanced.

[0054] The calculation of cumulative deviation is used to characterize the continuous cumulative effect of deviation, avoiding the over-accuracy caused by the long-term accumulation of small deviations. It employs discrete-time integration (summation), with the formula ei_k = Σ(e_i) × T (i from 1 to k), where ei_k is the cumulative deviation in the k-th period, and the summation range covers the deviations of the current and all previous prediction periods. For example, the cumulative deviation for the first 3 periods... ei_3 = (0.002059 + 0.00215 + 0.00224) × 0.001 = 6.449 × 10^-6 mm·s. The larger the accumulated deviation, the longer the deviation duration, requiring enhanced compensation through an integral stage to eliminate steady-state error.

[0055] The position control deviation index is generated by combining the three core parameters mentioned above along fixed dimensions to form a three-dimensional feature vector, namely the index vector [e_k, ec_k, ei_k]. Each dimension is normalized (mapped to the interval [-1, 1]) for unified processing by the fuzzy inference system. The normalization formula is f_norm=(f-f_mid) / (f_max-f_mid), where f_mid is the intermediate value of the parameter, and f_max is the maximum threshold of the parameter (e.g., the maximum deviation threshold is 0.01 mm, the maximum deviation change rate threshold is 0.5 mm / s, and the maximum deviation accumulation threshold is 1×10^-4 mm·s). In the example, the normalized index vector for the third cycle is [0.224, 0.18, 0.064], corresponding to the normalized values ​​of deviation, change rate, and accumulation, respectively, clearly quantifying the current position control deviation status.

[0056] The position control deviation index is input into the fuzzy inference system, and the adjustment factors of the proportional, integral, and derivative parameters of the PID controller are inferred through the preset fuzzy rule base to generate a set of PID parameter adjustment factors. The core of this step is to utilize the fuzzy inference system to handle the nonlinear characteristics of the deviation index, and dynamically output the PID parameter adjustment factor through empirical fuzzy rules to achieve adaptive adaptation of the PID parameters. The specific implementation method is as follows: The fuzzy inference system adopts a Mamdani-type architecture, with core components including four modules: input variable fuzzification, fuzzy rule base, fuzzy inference engine, and output variable defuzzification. The input variables are the three-dimensional position control deviation indices generated in step one, namely deviation e, deviation change rate ec, and deviation accumulation ei; the output variables are the three core parameter adjustment factors of the PID controller: proportional coefficient adjustment factor α, integral time adjustment factor β, and derivative time adjustment factor γ. The adjustment factors all range from -0.5 to 0.5, where positive values ​​indicate an increase in the corresponding parameter, and negative values ​​indicate a decrease.

[0057] The fuzzification process for input variables requires defining fuzzy subsets and membership functions. The fuzzy subsets for the three input variables are divided into five levels: negative large (NB), negative small (NS), zero (ZO), positive small (PS), and positive large (PB). The membership function uses a triangular membership function (balancing computational efficiency and fuzzy partitioning accuracy). For example, the normalized range of the deviation e is [-1, 1]. The universe of discourse for each fuzzy subset is as follows: NB corresponds to [-1, -0.6], NS to [-0.8, -0.2], ZO to [-0.4, 0.4], PS to [0.2, 0.8], and PB to [0.6, 1]. When the normalized value of e is 0.224, its membership degree to ZO is 0.88, and its membership degree to PS is 0.12, thus achieving the conversion from precise values ​​to fuzzy quantities.

[0058] The preset fuzzy rule base is built based on engineering experience in the side mold control of casting equipment. The rule form is "If e is A and ec is B and ei is C, then α is D, β is E, and γ is F", where A, B, and C are fuzzy subsets of input variables, and D, E, and F are fuzzy subsets of output variables. The rule base contains 5×5×5=125 rules, covering all input variable combinations. Core rule examples are as follows: 1. If e is PB, ec is PS, and ei is PS, then α is PB, β is NS, and γ is PS (large and increasing deviation requires increasing the proportional gain to enhance response, decreasing the integration time to accelerate error reduction, and increasing the derivative coefficient to suppress overshoot); 2. If e is ZO, ec is ZO, and ei is ZO, then α is ZO, β is ZO, and γ is ZO (very small deviation, no parameter adjustment required); 3. If e is NS, ec is NB, and ei is NS, then α is NS, β is PB, and γ is NS (small and decreasing deviation requires decreasing the proportional gain to avoid overshoot, increasing the integration time to prevent integration saturation, and decreasing the derivative coefficient to reduce response sensitivity).

[0059] The fuzzy inference engine employs a max-min synthesis method. It first determines the trigger strength of each rule based on the membership degree of the input variables, and then synthesizes the fuzzy set of output variables using these trigger strengths. For example, given inputs e=0.224 (ZO:0.88, PS:0.12), ec=0.18 (ZO:0.9, PS:0.1), and ei=0.064 (ZO:0.936, PS:0.064), the rule with the highest trigger strength is "If e is ZO, ec is ZO, and ei is ZO, then α=ZO, β=ZO, γ=ZO," with a trigger strength of 0.88×0.9×0.936≈0.73. Simultaneously, it triggers "If e is ZO, ec is ZO, and ei is PS, then α=PS, β=ZO, γ=ZO," with a trigger strength of 0.88×0.9×0.064≈0.05.

[0060] Output variable defuzzification employs the centroid method, converting the synthesized fuzzy set into precise adjustment factor values. The centroid method formula is u = ∫μ(u) × udu / ∫μ(u)du, where μ(u) is the membership function of the output variable, and u is the adjustment factor value. For example, the fuzzy set of α is synthesized from ZO (membership 0.73) and PS (membership 0.05), yielding α = 0.02; similarly, β = 0.01 and γ = 0.03 are calculated. The final generated PID parameter adjustment factor set is [α = 0.02, β = 0.01, γ = 0.03], clearly defining the adjustment direction and magnitude of each PID parameter.

[0061] Based on the PID parameter adjustment factor set, the proportional coefficient, integral time and derivative time of the PID controller are dynamically adjusted to generate adaptive PID controller parameters. The core of this step is based on the dynamic coupling relationship between the adjustment factor and the initial PID parameters. By quantifying the adjustment formula, PID parameters adapted to the current deviation state are generated, enabling the controller to accurately respond to deviation trends. The specific implementation method is as follows: First, the initial parameters of the PID controller are set. These initial parameters are calibrated based on the static characteristics of the hydraulic control system for the side mold of the casting equipment. Experimental verification confirmed the following initial values: proportional coefficient Kp = 5.0 (dimensionless, characterizing the response strength of proportional control), integral time Ti = 0.1 seconds (characterizing the strength of the integral action; a larger value indicates a weaker integral action), and derivative time Td = 0.01 seconds (characterizing the strength of the derivative action; a larger value indicates a stronger derivative action). These initial parameters are suitable for operating conditions with small deviations and stable motion. When the deviation trend changes, dynamic correction is required through adjustment factors.

[0062] The dynamic adjustment formula needs to be designed based on the control characteristics of each PID parameter to ensure that the adjustment factor can effectively regulate the parameter response. The adjustment formula for the proportional coefficient Kp is Kp'=Kp×(1+α), where Kp' is the adjusted proportional coefficient and α is the proportional adjustment factor. A positive value increases Kp to enhance the system response speed, while a negative value decreases Kp to avoid overshoot. For example, if α=0.02, substituting, we get Kp'=5.0×(1+0.02)=5.1. Appropriately increasing the proportional coefficient can cope with the increasing trend of deviation and improve the system's sensitivity to deviation.

[0063] The formula for adjusting the integration time Ti is Ti' = Ti × (1 - β), where Ti' is the adjusted integration time and β is the integration adjustment factor. A positive value decreases Ti to enhance the integration effect and accelerate the elimination of steady-state error, while a negative value increases Ti to weaken the integration effect and prevent integration saturation. For example, if β = 0.01, substituting this into the formula gives Ti' = 0.1 × (1 - 0.01) = 0.099 seconds, which slightly reduces the integration time, enhancing the response of the integration element to the accumulated deviation and preventing the long-term accumulation of small deviations.

[0064] The adjustment formula for the derivative time Td is Td' = Td × (1 + γ), where Td' is the adjusted derivative time and γ is the derivative adjustment factor. A positive value increases Td, enhancing the derivative effect and suppressing the trend of deviation change in advance; a negative value decreases Td, weakening the derivative effect and reducing the system's sensitivity to noise. For example, γ = 0.03, substituting into the formula, we get Td' = 0.01 × (1 + 0.03) = 0.0103 seconds. This slightly increases the derivative time, strengthening the prediction of the rate of change of deviation and suppressing further increase in deviation.

[0065] Verification after parameter adjustment must ensure that the adjusted parameters meet the system stability requirements. The critical proportional gain method is used to verify stability, calculating the system's phase margin and gain margin. The phase margin must be ≥45° and the gain margin ≥6dB. In the example, the adjusted parameters Kp'=5.1, Ti'=0.099s, and Td'=0.0103s have a calculated phase margin of 52° and a gain margin of 8dB, meeting the stability requirements and suitable for subsequent compensation control.

[0066] The final adaptive PID controller parameter set is [Kp'=5.1,Ti'=0.099s,Td'=0.0103s]. This parameter set is optimized for the current trend of increasing deviation, which not only improves the system's response speed to deviation, but also balances the error elimination effect and stability through fine-tuning of integral and derivative, providing core parameter support for accurate calculation of compensation control quantity.

[0067] Based on the parameters of the adaptive PID controller and the position deviation trend, the target flow rate and pressure setpoint of the hydraulic cylinder in the next control cycle are calculated to generate a dynamic compensation strategy.

[0068] The core of this step is to convert the output of the PID controller into control commands (target flow rate and pressure) that can be executed by the hydraulic system, and to plan the compensation action for the next cycle based on the deviation trend, thus forming a complete dynamic compensation strategy. The specific implementation method is as follows: First, clarify the mapping relationship between the output of the PID controller and the control quantity of the hydraulic system. The output of the PID controller is the speed compensation increment Δv of the hydraulic cylinder, that is, the position deviation is corrected by adjusting the movement speed of the side mold. The speed compensation increment is calculated using the discrete PID control formula: Δv_k=Kp'×(e_k-e_{k-1})+(Kp'×T / Ti')×e_k+(Kp'×Td' / T)×(e_k-2e_{k-1}+e_{k-2}), where Δv_k is the speed compensation increment in the k-th control cycle, and e_k, e_{k-1}, and e_{k-2} are the deviation values ​​of the current cycle and the two previous cycles, respectively.

[0069] Substituting the adaptive PID parameters and deviation data, calculate Δv_k. In the example, e_k = 0.00224mm, e_{k-1} = 0.00215mm, e_{k-2} = 0.002059mm, T = 0.001s, Kp' = 5.1, Ti' = 0.099s, Td' = 0.0103s. The calculation process is as follows: Proportional term =5.1×(0.00224-0.00215)=5.1×0.00009=4.59×10^-4mm / s; Integral term =(5.1×0.001 / 0.099)×0.00224≈0.0515×0.00224≈1.15×10^-4mm / s; Differential term =(5.1×0.0103 / 0.001)×(0.00224-2×0.00215+0.002059)=52.53×(0.00224-0.0043+0.002059)=52.53×(-0.000001)≈-5.25×10^-5mm / s; Total speed compensation increment Δv_k=4.59×10^-4+1.15×10^-4-5.25×10^-5≈5.215×10^-4mm / s, that is, it is necessary to increase the current ideal speed by 5.215×10^-4mm / s to offset the increasing trend of deviation.

[0070] The target speed for the next control cycle is v_target = v_ideal + Δv_k, where v_ideal is the ideal speed based on the initial control input (v_ideal = 1 mm / s in the example). Substituting these values, we get v_target = 1 + 5.215 × 10^-4 ≈ 1.0005215 mm / s. The target flow rate Q_target is calculated based on the relationship between the hydraulic cylinder flow rate and speed: Q_target = A × v_target, where A is the effective area of ​​the hydraulic cylinder piston (preset A = 0.01 m² = 1 × 10^4 mm²). Substituting these values, we get... Q_target = 1 × 10^4 mm² × 1.0005215 mm / s = 1.0005215 × 10^4 mm³ / s = 1.0005215 × 10^-5 m³ / s, meaning the target flow rate is 1.0005215 × 10^-5 m³ / s. Speed ​​compensation is achieved by adjusting the flow rate.

[0071] The calculation of the target pressure P_target needs to be combined with the load change rate of the hydraulic system. The relationship between pressure and load is P_target = (F + ΔF) / A, where F is the current hydraulic load (F = 150000N in the example), ΔF is the load compensation amount, ΔF = m × a_comp, m is the mass of the side mold (m = 500kg), and a_comp is the acceleration compensation amount. (a_comp = Δv_k / T = 5.215 × 10^-4 mm / s ÷ 0.001s = 0.5215 mm / s² = 5.215 × 10^-4 m / s²). Calculation yields ΔF = 500 kg × 5.215 × 10^-4 m / s² ≈ 0.26075 N. P_target=(150000+0.26075)N÷0.01m²≈150000.26075Pa≈15.000026MPa, that is, the target pressure is 15.000026MPa, and the pressure is slightly increased to match the speed to compensate for the required driving force.

[0072] The generation of the dynamic compensation strategy requires integrating the above calculation results and clarifying the core control parameters and execution logic for the next control cycle. The strategy includes: a control cycle of 0.001 seconds, a target flow rate of 1.0005215×10^-5 m³ / s, a target pressure of 15.000026 MPa, and a compensation direction of increasing the hydraulic cylinder's inlet flow rate and pressure to enhance the side mold's movement speed and offset the current increasing position deviation. Simultaneously, a compensation effect judgment condition is set: after the next cycle, the side mold position deviation must be ≤0.0022 mm. If this is not met, the PID parameters and compensation amount are readjusted based on the new deviation trend. The strategy is presented in structured text format to ensure that the high-frequency servo valve group can accurately parse and execute control commands.

[0073] S204, execute the dynamic compensation strategy, adjust the hydraulic cylinder flow and pressure in real time through the high-frequency servo valve group, and complete the position closed-loop correction within one stroke cycle.

[0074] Specifically, the target flow and pressure setpoints in the dynamic compensation strategy can be converted into control signals for the high-frequency servo valve group, generating pulse width modulation (PWM) control waveforms. The core of this step is to establish a quantitative mapping relationship between the target parameters of the hydraulic system and the control signal of the servo valve. Precise and controllable electrical signals are generated through pulse width modulation technology, providing the foundation for the precise actuation of the servo valve. The specific implementation method is as follows: First, the core target parameters in the dynamic compensation strategy are defined, including the target flow rate Q_target and the target pressure P_target. In this example, they are 1.0005215×10^-5 m³ / s and 15.000026 MPa, respectively. Since the control signal of the high-frequency servo valve assembly is an electrical signal (usually a voltage or current signal), a mapping relationship between the target parameters and the control signal amplitude must be established first, and the conversion coefficient is determined using a linear calibration method. For flow control, the maximum flow rate Q_max corresponding to the maximum opening of the servo valve is set to 2×10^-5 m³ / s, and the corresponding maximum control voltage U_max is 10V. Therefore, the flow-voltage conversion coefficient... K_q=U_max / Q_max=10V / (2×10^-5m³ / s)=5×10^5V・s / m³; For pressure control, the maximum regulating pressure of the servo valve is P_max=31.5MPa, and the corresponding maximum control voltage is also 10V. The pressure-voltage conversion coefficient K_p=U_max / P_max=10V / 31.5MPa≈0.3175V / MPa.

[0075] The target control voltage is calculated based on the conversion factor. The target control voltage for flow rate is U_q = K_q × Q_target = 5 × 10^5 V・s / m³ × 1.0005215 × 10^-5 m³ / s ≈ 5.0026 V; the target control voltage for pressure is U_p = K_p × P_target ≈ 0.3175 V / MPa × 15.000026 MPa ≈ 4.7625 V. Since the high-frequency servo valve group is driven by pulse width modulation (PWM) signal, the target control voltage needs to be converted into the duty cycle of the PWM control waveform. The core parameters of the PWM signal include carrier frequency, amplitude and duty cycle. The carrier frequency is set to 10kHz (this frequency matches the response characteristics of the high-frequency servo valve, which can ensure control accuracy and avoid high-frequency signal attenuation. The servo valve response frequency is ≥1kHz, and a 10kHz carrier can make the valve core following error ≤0.5%). The signal amplitude is fixed at 12V (to match the power supply voltage of the servo valve drive circuit). The duty cycle D is the ratio of the high level duration to the carrier period, and the value range is 0-100%.

[0076] The mapping relationship between duty cycle and control voltage is based on the linear principle, and the formula is D = (U_target / U_amp) × 100%, where U_target is the target control voltage and U_amp is the amplitude of the PWM signal. For flow control, the duty cycle D_q = (5.0026V / 12V) × 100% ≈ 41.688%; for pressure control, the duty cycle D_p = (4.7625V / 12V) × 100% ≈ 39.688%. Considering the forward and reverse adjustment requirements of the servo valve (such as when flow or pressure needs to be reduced), a bipolar PWM waveform is adopted, with a 50% duty cycle as the midpoint (corresponding to zero adjustment). A duty cycle > 50% indicates forward adjustment (increasing flow / pressure), and a duty cycle < 50% indicates reverse adjustment (reducing flow / pressure). Therefore, the above duty cycle needs to be corrected for midpoint offset. The correction formula is D_bipolar = 50% + D_unipolar - 50% = D_unipolar (when the target adjustment is forward, in the example, both flow and pressure are forward adjustments, after correction D_q = 41.688% + 50% - 50% = 41.688%. Note that if the target control voltage is less than 5V, the corresponding duty cycle is less than 50%, which is reverse adjustment).

[0077] The final PWM control waveform is generated. The PWM waveform parameters corresponding to flow control are: carrier frequency 10kHz (period T=0.1ms), duty cycle 41.688%, which is the duration of the high level in each cycle. t_high = D_q × T = 41.688% × 0.1ms ≈ 0.041688ms, low-level duration t_low = T - t_high ≈ 0.058312ms; the PWM waveform parameters corresponding to pressure control are: carrier frequency 10kHz, duty cycle 39.688%, high-level duration ≈ 0.039688ms, low-level duration ≈ 0.060312ms. To ensure signal synchronization, the PWM waveforms corresponding to flow rate and pressure are generated using the same carrier clock, with a timestamp accuracy of 1μs, to avoid adjustment conflicts caused by signal phase differences.

[0078] The PWM control waveform is output to the drive circuit of the high-frequency servo valve group through a digital signal processor to control the opening size and direction of the servo valve core, thereby realizing the real-time adjustment of the hydraulic cylinder flow and pressure. The core of this step is to achieve precise output and drive amplification of the PWM signal through a digital signal processor, converting the electrical signal into the mechanical action of the servo valve, thereby regulating the flow and pressure of the hydraulic cylinder. The specific implementation method is as follows: Core parameter settings for the digital signal processor: A processor with high-speed PWM output is selected, with a clock frequency set to 100MHz (ensuring PWM signal generation accuracy and duty cycle adjustment resolution ≤0.1%). Two independent PWM output channels are configured, corresponding to flow control and pressure control respectively. Internally, the processor generates a 10kHz carrier signal via a timer. Based on the duty cycle parameters generated in step one, the high-level duration is set through a comparison register to achieve accurate PWM waveform generation. For example, when generating the flow control PWM waveform, the timer period register value is set to 10000 (corresponding to a clock frequency of 100MHz, period = 10000 / 100MHz = 0.1ms), and the comparison register value is set to 4169 (corresponding to a duty cycle of 4169 / 10000 = 41.69%, with an error of ≤0.002% compared to the target duty cycle of 41.688%).

[0079] The core design of the drive circuit is to amplify and level-convert the PWM signal (3.3V amplitude, which cannot directly drive the servo valve) output by the digital signal processor. The drive circuit mainly consists of an optocoupler isolation module, a power amplifier module, and an overcurrent protection module. The optocoupler isolation module uses a high-speed optocoupler with an isolation voltage ≥2500V to isolate the processor control circuit from the servo valve power circuit, preventing electromagnetic interference from the power circuit from affecting the processor's operation. The optocoupler response time is ≤10ns, ensuring that the PWM signal is transmitted without delay. The power amplifier module uses a MOSFET full-bridge circuit with a peak output current ≥5A, meeting the drive current requirements of the high-frequency servo valve (servo valve rated drive current 1-3A). The amplification factor is set to 3.6 times (amplifying the 3.3V PWM signal to 12V to match the servo valve's rated control voltage).

[0080] The overcurrent protection module ensures the safety of the drive circuit and servo valve. It collects the output current in real time through a sampling resistor, setting the overcurrent threshold to 5A (1.6 times the servo valve's maximum allowable current). When the collected current value exceeds the threshold, the protection module immediately triggers, cutting off the power output and sending an interrupt signal to the digital signal processor. Upon receiving the interrupt signal, the processor stops the PWM output to prevent component damage. For example, if the servo valve spool jams, causing the drive current to rise to 6A, the voltage signal collected by the sampling resistor, determined by a comparator, exceeds the threshold, immediately triggering protection with a response time ≤1μs.

[0081] The control logic of the servo valve core: The PWM control signal is amplified by the drive circuit and then input to the electromagnetic coil of the high-frequency servo valve. The electromagnetic force generated by the coil is proportional to the average voltage of the PWM signal (average voltage = duty cycle × signal amplitude). The electromagnetic force drives the valve core to overcome the spring force and generate displacement. The valve core displacement is linearly related to the average voltage, thereby controlling the opening size of the servo valve. For example, the average voltage of the flow control PWM signal = 41.688% × 12V ≈ 5.0026V, corresponding to a valve core displacement = 5.0026V × 0.02mm / V = 0.10005mm (servo valve voltage-displacement conversion coefficient 0.02mm / V). The valve core opening area is proportional to the displacement, opening area = 0.10005mm × 10mm (effective valve core width) = 1.0005mm², corresponding to a flow rate = opening area × flow velocity. The flow velocity is determined by the pressure difference of the hydraulic system, ultimately achieving precise control of the target flow rate of 1.0005215 × 10^-5 m³ / s.

[0082] The valve spool direction is controlled via a bipolar PWM signal. When the duty cycle is >50%, the average voltage is positive, and the electromagnetic force drives the valve spool to move in the forward direction, increasing the inlet opening. When the duty cycle is <50%, the average voltage is negative, and the electromagnetic force drives the valve spool to move in the reverse direction, increasing the return port opening. When the duty cycle is 50%, the average voltage is zero, and the valve spool returns to the neutral position under spring force, closing the port. For example, to reduce the hydraulic cylinder flow, the flow control PWM duty cycle is adjusted to 35%, with an average voltage of 35% × 12V = 4.2V. The valve spool moves in the reverse direction, reducing the inlet opening and thus decreasing the flow rate, achieving bidirectional flow regulation. Through this control process, real-time adjustment of the hydraulic cylinder flow and pressure is achieved, with an adjustment response time ≤1ms, meeting the real-time requirements of side mold stroke control.

[0083] The actual flow and pressure feedback signals of the hydraulic cylinder are collected in real time, compared with the target set value, and fine-tuned through the proportional regulating valve to generate a flow and pressure closed-loop control signal. The core of this step is to construct a closed-loop control circuit for flow and pressure. By calculating the deviation between the feedback signal and the target value, precise fine-tuning is achieved to ensure that the hydraulic system parameters remain stable within the target range. The specific implementation method is as follows: High-precision sensors are used to acquire the actual flow and pressure feedback signals. The flow sensor is installed at the hydraulic cylinder inlet, with a measurement range of 0-5×10^-5 m³ / s and a measurement accuracy of ±0.1%, outputting a 4-20mA current signal. The pressure sensor is installed in the rodless chamber of the hydraulic cylinder, with a measurement range of 0-31.5MPa and a measurement accuracy of ±0.1%, also outputting a 4-20mA current signal. The acquisition circuit converts the current signal into a voltage signal through a sampling resistor (250Ω, corresponding to a voltage range of 1-5V). After amplification by an instrumentation amplifier (5x amplification, voltage range of 5-25V), the signal is input to a multi-channel data acquisition card. The sampling frequency of the acquisition card is set to 10kHz (consistent with the PWM control signal frequency to ensure synchronization between the feedback and control signals), with an analog-to-digital conversion resolution of 16 bits and a conversion error ≤0.0015%.

[0084] Preprocessing of feedback signals: To eliminate the influence of high-frequency noise, a first-order low-pass filter is used to filter the collected flow and pressure signals. The filter cutoff frequency is set to 100Hz (lower than 1 / 10 of the acquisition frequency, which conforms to Shannon's sampling theorem and can effectively filter out high-frequency interference above 100Hz). The filtering formula is y_k=α×x_k+(1-α)×y_{k-1}, where y_k is the current filtered signal, x_k is the current acquired signal, y_{k-1} is the filtered signal at the previous moment, α is the filtering coefficient, α=2πf_cT / (1+2πf_cT), f_c is the cutoff frequency of 100Hz, and T is the sampling period of 0.1ms. Calculation yields α≈0.0617. For example, if the collected flow signal is 1.0008×10^-5 m³ / s, and the filtered signal at the previous moment is 1.0006×10^-5 m³ / s, then after filtering, y_k = 0.0617×1.0008×10^-5 + 0.9383×1.0006×10^-5 ≈ 1.0006×10^-5 m³ / s. The signal fluctuation amplitude has decreased from 0.0002×10^-5 m³ / s to 0.00001×10^-5 m³ / s, demonstrating a significant filtering effect.

[0085] Deviation Calculation and Proportional Adjustment: The pre-processed actual flow rate Q_actual is compared with the target flow rate Q_target, and the actual pressure P_actual is compared with the target pressure P_target. The deviation values ​​are calculated using the formulas e_q = Q_target - Q_actual and e_p = P_target - P_actual. In the example, Q_actual = 1.0006 × 10^-5 m³ / s, e_q = 1.0005215 × 10^-5 - 1.0006 × 10^-5 = -7.85 × 10^-11 m³ / s; P_actual = 15.00005 MPa, e_p = 15.000026 - 15.00005 = -2.4 × 10^-5 MPa. The fine-tuning of the proportional control valve is based on the proportional control algorithm. The adjustment amount ΔU = K × e, where K is the proportional control coefficient, the flow rate control coefficient K_q_reg = 1 × 10^5 V・s / m³, the pressure control coefficient K_p_reg = 1 × 10^4 V / MPa, and the adjustment amount ΔU is used to correct the target control voltage generated in step one.

[0086] For example, the flow regulation ΔU_q = K_q_reg × e_q = 1 × 10^5 × (-7.85 × 10^-11) = -7.85 × 10^-6 V, and the corrected flow target control voltage U_q' = 5.0026 V - 7.85 × 10^-6 V ≈ 5.00259 V. The corresponding duty cycle correction is (5.00259 / 12) × 100% ≈ 41.688%. The correction is extremely small, indicating that the deviation between the actual flow rate and the target flow rate is very small, and no significant adjustment is needed. Pressure regulation ΔU_p=K_p_reg×e_p=1×10^4×(-2.4×10^-5)=-0.24V, the corrected pressure target control voltage U_p'=4.7625V-0.24V=4.5225V, the corresponding duty cycle correction is (4.5225 / 12)×100%≈37.688%, the duty cycle of the pressure control signal needs to be slightly reduced, the servo valve opening is reduced, thereby reducing the hydraulic cylinder pressure.

[0087] The final generated flow and pressure closed-loop control signals are modified PWM control signal parameters. The flow control closed-loop signal has a carrier frequency of 10kHz and a duty cycle of 41.688%; the pressure control closed-loop signal has a carrier frequency of 10kHz and a duty cycle of 37.688%. The closed-loop control signals update the PWM output parameters through a digital signal processor, enabling real-time fine-tuning of the servo valve. The fine-tuning period is consistent with the acquisition period (0.1ms), ensuring that the flow and pressure remain stable near the target values, with deviations controlled within ±0.1%.

[0088] Within one stroke cycle, the side mold position data is continuously monitored, and the control signal of the high-frequency servo valve group is iteratively adjusted through the position closed-loop control algorithm until the actual position of the side mold reaches the mold closing accuracy requirement, thus completing the position closed-loop correction.

[0089] The core of this step is to construct the final closed-loop control loop for the side mold position. With mold closing accuracy as the target, the side mold position is precisely corrected by iteratively adjusting the servo valve control signal. The specific implementation method is as follows: First, define a stroke cycle: the stroke cycle of the side mold is the complete time from the starting position (out-of-mold position) to the target mold-closing position. Based on the casting equipment's process requirements, this is set to 5 seconds. Within this stroke cycle, the number of sampling points is set to 50,000 (corresponding to a sampling frequency of 10kHz, 5 seconds × 10kHz = 50,000 sampling points), ensuring that side mold position data can be collected in each control cycle (0.1ms). The mold-closing accuracy requirement is that the deviation between the actual side mold position and the target mold-closing position is ≤ ±0.01mm. The target mold-closing position is set to 500.000mm (calibrated according to the mold size).

[0090] Continuous monitoring of the side mold position data is achieved through a high-precision magnetic scale sensor (consistent with the position acquisition sensor in step one, with a measurement accuracy of 0.001mm). The sensor outputs position data in real time, which is synchronously acquired by a multi-channel data acquisition card and then transmitted to a digital signal processor. The processor analyzes the position data in real time and calculates the deviation between the actual position and the target position: e_pos = x_target - x_actual, where x_target = 500.000mm and x_actual is the real-time acquired side mold position data. For example, if x_actual = 499.998mm is acquired at a certain sampling moment, then e_pos = 500.000 - 499.998 = 0.002mm, and the deviation is within the mold closing accuracy requirement range; if x_actual = 499.985mm at a certain moment, then e_pos = 0.015mm, which exceeds the accuracy requirement, and position closed-loop correction needs to be initiated.

[0091] The position closed-loop control algorithm employs a fuzzy adaptive PID algorithm (consistent with the algorithm in step four, ensuring the continuity of the control strategy). The algorithm inputs are the position deviation e_pos and the deviation change rate ec_pos (ec_pos=(e_pos,k-e_pos,k-1) / T, where T is the control period of 0.1ms), and the output is the flow compensation increment ΔQ. By adjusting the flow rate, the side-mode velocity is corrected, thereby correcting the position deviation. The core parameters of the algorithm are the adaptive PID parameters, which are dynamically adjusted in real time according to the position deviation state. For example, when e_pos=0.015mm (exceeding the accuracy) and ec_pos=0.0002mm / ms (deviation increasing), the PID parameter adjustment factors output by the fuzzy inference system are α=0.05, β=0.03, and γ=0.04, and the adjusted PID parameters are Kp'=5.25, Ti'=0.097s, and Td'=0.0104s.

[0092] The flow compensation increment ΔQ is calculated using the discrete PID formula: ΔQ = Kp' × (e_pos,k - e_pos,k-1) + (Kp' × T / Ti') × e_pos,k + (Kp' × Td' / T) × (e_pos,k - 2e_pos,k-1 + e_pos,k-2). Substituting the example data e_pos,k = 0.015mm, e_pos,k-1 = 0.014mm, e_pos,k-2=0.013mm, T=0.1ms, the proportional term is calculated. =5.25×(0.015-0.014)=5.25×10^-3mm / s corresponding to flow compensation; integral term =(5.25×0.1 / 0.097)×0.015≈0.541×0.015≈8.115×10^-3mm / s corresponding to the flow compensation; differential term =(5.25×0.0104 / 0.1)×(0.015-2×0.014+0.013)=0.546×(0.015-0.028+0.013)=0.546×0=0; The velocity compensation corresponding to the total flow compensation increment ΔQ is 5.25×10^-3 + 8.115×10^-3 = 1.3365×10^-2 mm / s, converted to flow compensation ΔQ=A×Δv=1×10^4mm²×1.3365×10^-2mm / s=1.3365×10^2mm³ / s=1.3365×10^-7m³ / s.

[0093] The control signal of the high-frequency servo valve group is adjusted based on the flow compensation increment, so that the original target flow rate is adjusted accordingly. Q_target=1.0005215×10^-5m³ / s is corrected to Q_target' = 1.0005215 × 10^-5 + 1.3365 × 10^-7 = 1.0138865 × 10^-5 m³ / s. Then, using the conversion method in step one, the corrected target flow rate is converted into the corresponding PWM duty cycle: Q_max = 2 × 10^-5 m³ / s, U_q' = K_q × Q_target' = 5 × 10^5 × 1.0138865 × 10^-5 ≈ 5.0694 V, and the duty cycle D_q' = (5.0694 / 12) × 100% ≈ 42.245%. The digital signal processor updates the PWM output parameters to adjust the flow rate, thereby increasing the side mold movement speed and reducing the position deviation.

[0094] The iterative adjustment process continues throughout the entire stroke cycle. Each control cycle (0.1ms) completes a closed-loop process of position acquisition, deviation calculation, PID adjustment, flow correction, and servo valve control until the actual position of the side mold reaches the mold closing accuracy requirement. For example, after 10 iterative adjustments, the actual position of the side mold is adjusted from 499.985mm to 499.992mm, with a deviation e_pos=0.008mm (meeting the accuracy requirement). At this point, the iterative adjustment stops, the current servo valve control signal is maintained, and the side mold continues to move to the target position. If the actual position of the side mold fails to reach the mold closing accuracy requirement before the end of the stroke cycle (within 5 seconds), the system will trigger an alarm signal to prompt equipment maintenance and ensure the safety of the casting process.

[0095] When the side mold moves to the target mold closing position, and the position deviation for 10 consecutive control cycles is ≤ ±0.01mm, the position closed-loop correction is considered complete, and one stroke cycle ends. In the example, the side mold reaches 500.000mm at 4.8 seconds. The position data for 10 consecutive control cycles (1ms) are 500.000mm, 500.001mm, 499.999mm, 500.000mm, 500.000mm, 499.999mm, 500.001mm, 500.000mm, 499.999mm, and 500.000mm, respectively. The deviations are all within the accuracy range, completing the position closed-loop correction and ensuring that the mold closing accuracy meets the casting process requirements.

[0096] Another embodiment of the present invention provides a side mold stroke control system for a casting equipment, see below. Figure 3 The system may include: The acquisition module 301 is used to acquire real-time position data of the side mold, pressure data of the hydraulic system, and temperature data of the mold contact surface through the multi-sensor fusion module; The construction module 302 is used to construct a dynamic friction and thermal deformation coupling model during the movement of the side mold based on the real-time position data, the pressure data and the temperature data, and to predict the position deviation trend in the current stroke stage. The generation module 303 is used to generate a dynamic compensation strategy for the side mold stroke based on fuzzy adaptive PID according to the position deviation trend and the preset mold closing accuracy requirements. The execution module 304 is used to execute the dynamic compensation strategy, which adjusts the flow and pressure of the hydraulic cylinder in real time through the high-frequency servo valve group and completes the position closed-loop correction within one stroke cycle.

[0097] This invention also provides a storage medium storing a computer program, wherein the computer program is configured to execute the steps in any of the above method embodiments when running.

[0098] This invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.

[0099] The above description, based on the embodiments shown in the figures, details the structure, features, and effects of the present invention. The above description is only a preferred embodiment of the present invention, but the present invention is not limited to the scope of implementation shown in the figures. Any changes made in accordance with the concept of the present invention, or equivalent embodiments modified to have equivalent changes, that do not exceed the spirit covered by the specification and figures, should be within the protection scope of the present invention.

Claims

1. A side mold stroke control method of a casting apparatus, characterized by, The method includes: The real-time position data of the side mold, the pressure data of the hydraulic system, and the temperature data of the mold contact surface are collected by the multi-sensor fusion module. Based on the real-time position data, the pressure data, and the temperature data, a dynamic friction and thermal deformation coupling model is constructed during the side mold movement process, and the position deviation trend at the current stroke stage is predicted. Based on the position deviation trend and the preset mold closing accuracy requirements, a dynamic compensation strategy for the side mold stroke based on fuzzy adaptive PID is generated. The dynamic compensation strategy is implemented by adjusting the flow and pressure of the hydraulic cylinder in real time through the high-frequency servo valve group, and completing the closed-loop position correction within one stroke cycle.

2. The method according to claim 1, characterized in that, The process of acquiring real-time position data of the side mold, pressure data of the hydraulic system, and temperature data of the mold contact surface through a multi-sensor fusion module includes: A high-precision magnetic scale sensor is installed on the side mold guide rail to collect the displacement signal of the side mold in real time and generate the original position pulse sequence; Piezoelectric pressure sensors are installed in the inlet and outlet oil circuits of the hydraulic cylinder to collect pressure fluctuation signals of the hydraulic system in real time and generate raw pressure simulation signals. An infrared thermocouple array is installed on the contact surface of the mold to collect the temperature distribution signal of the mold contact surface in real time and generate the original temperature simulation signal. The original position pulse sequence, original pressure analog signal and original temperature analog signal are acquired synchronously by a multi-channel data acquisition card, and the signals are conditioned and converted from analog to digital to generate a multi-sensor digital signal set with timestamp alignment.

3. The method according to claim 2, characterized in that, The process of constructing a dynamic friction and thermal deformation coupling model during the side mold movement based on the real-time position data, pressure data, and temperature data, and predicting the position deviation trend during the current stroke stage, includes: Feature extraction is performed on the multi-sensor digital signal set with timestamp alignment. The instantaneous velocity and acceleration of the side mold are calculated from the position data, the load change rate of the hydraulic system is calculated from the pressure data, and the temperature gradient of the mold contact surface is calculated from the temperature data to generate a motion state feature vector. The motion state feature vector is input into a pre-trained deep learning network, which learns the nonlinear friction characteristics and thermal deformation laws in the side mode motion through historical data, and generates dynamic friction coefficient estimates and thermal deformation compensation amounts. Based on the estimated value of dynamic friction coefficient and thermal deformation compensation, a dynamic friction and thermal deformation coupling model is established during the movement of the side mold. The model takes the current control input as the independent variable and outputs the predicted value of the side mold position. The current control input and real-time position data are input into the dynamic friction and thermal deformation coupling model. The residual between the predicted value and the actual measured value of the side mold position is calculated. The position deviation trend of the next few control cycles is predicted through time series analysis.

4. The method according to claim 3, characterized in that, The step of generating a dynamic compensation strategy for the side mold travel based on fuzzy adaptive PID according to the position deviation trend and the preset mold closing accuracy requirements includes: The position deviation trend is compared with the preset mold closing accuracy requirements, the deviation change rate and the cumulative deviation are calculated, and the position control deviation index is generated. The position control deviation index is input into the fuzzy inference system, and the adjustment factors of the proportional, integral, and derivative parameters of the PID controller are inferred through the preset fuzzy rule base to generate a set of PID parameter adjustment factors. Based on the PID parameter adjustment factor set, the proportional coefficient, integral time and derivative time of the PID controller are dynamically adjusted to generate adaptive PID controller parameters. Based on the parameters of the adaptive PID controller and the position deviation trend, the target flow rate and pressure setpoint of the hydraulic cylinder in the next control cycle are calculated to generate a dynamic compensation strategy.

5. The method according to claim 4, characterized in that, The execution of the dynamic compensation strategy, which involves adjusting the hydraulic cylinder flow and pressure in real time via a high-frequency servo valve group and completing position closed-loop correction within one stroke cycle, includes: The target flow and pressure setpoints in the dynamic compensation strategy are converted into control signals for the high-frequency servo valve group, generating pulse width modulation (PWM) control waveforms. The PWM control waveform is output to the drive circuit of the high-frequency servo valve group through a digital signal processor to control the opening size and direction of the servo valve core, thereby realizing the real-time adjustment of the hydraulic cylinder flow and pressure. The actual flow and pressure feedback signals of the hydraulic cylinder are collected in real time, compared with the target set value, and fine-tuned through the proportional regulating valve to generate a flow and pressure closed-loop control signal. Within one stroke cycle, the side mold position data is continuously monitored, and the control signal of the high-frequency servo valve group is iteratively adjusted through the position closed-loop control algorithm until the actual position of the side mold reaches the mold closing accuracy requirement, thus completing the position closed-loop correction.

6. A side mold stroke control system for casting equipment, characterized in that, The system includes: The data acquisition module is used to acquire real-time position data of the side mold, pressure data of the hydraulic system, and temperature data of the mold contact surface through the multi-sensor fusion module; The construction module is used to construct a dynamic friction and thermal deformation coupling model during the movement of the side mold based on the real-time position data, the pressure data, and the temperature data, and to predict the position deviation trend during the current stroke stage. The generation module is used to generate a dynamic compensation strategy for the side mold stroke based on fuzzy adaptive PID according to the position deviation trend and the preset mold closing accuracy requirements. The execution module is used to execute the dynamic compensation strategy, which adjusts the flow and pressure of the hydraulic cylinder in real time through the high-frequency servo valve group, and completes the position closed-loop correction within one stroke cycle.

7. The system according to claim 6, characterized in that, The acquisition module is specifically used for: A high-precision magnetic scale sensor is installed on the side mold guide rail to collect the displacement signal of the side mold in real time and generate the original position pulse sequence; Piezoelectric pressure sensors are installed in the inlet and outlet oil circuits of the hydraulic cylinder to collect pressure fluctuation signals of the hydraulic system in real time and generate raw pressure simulation signals. An infrared thermocouple array is installed on the contact surface of the mold to collect the temperature distribution signal of the mold contact surface in real time and generate the original temperature simulation signal. The original position pulse sequence, original pressure analog signal and original temperature analog signal are acquired synchronously by a multi-channel data acquisition card, and the signals are conditioned and converted from analog to digital to generate a multi-sensor digital signal set with timestamp alignment.

8. The system according to claim 7, characterized in that, The building module is specifically used for: Feature extraction is performed on the multi-sensor digital signal set with timestamp alignment. The instantaneous velocity and acceleration of the side mold are calculated from the position data, the load change rate of the hydraulic system is calculated from the pressure data, and the temperature gradient of the mold contact surface is calculated from the temperature data to generate a motion state feature vector. The motion state feature vector is input into a pre-trained deep learning network, which learns the nonlinear friction characteristics and thermal deformation laws in the side mode motion through historical data, and generates dynamic friction coefficient estimates and thermal deformation compensation amounts. Based on the estimated value of dynamic friction coefficient and thermal deformation compensation, a dynamic friction and thermal deformation coupling model is established during the movement of the side mold. The model takes the current control input as the independent variable and outputs the predicted value of the side mold position. The current control input and real-time position data are input into the dynamic friction and thermal deformation coupling model. The residual between the predicted value and the actual measured value of the side mold position is calculated. The position deviation trend of the next few control cycles is predicted through time series analysis.

9. A storage medium, characterized in that, The storage medium stores a computer program, wherein the computer program is configured to execute the method of any one of claims 1-5 when it is run.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the method of any one of claims 1-5.

Images