Dynamic pressure compensation method for zero-working-zone aluminum profile extrusion die

Abstract

The invention relates to the technical field of metal plastic processing, in particular to a dynamic pressure compensation method for a zero-working-band aluminum profile extrusion die, which comprises the following steps of: synchronously acquiring the pressure of a main cylinder, the actual speed of an extrusion rod, the temperature of the die and process variable sensor data for representing an extrusion process in an extrusion process according to a set control period; the method comprises the following steps: constructing a multi-dimensional input feature vector containing time sequence features and context features based on currently acquired sensor data and sensor data in a past preset time window; and inputting the input feature vector into a pre-trained neural network prediction model, and calculating and outputting an extrusion rod speed control quantity for pressure compensation in real time. According to the method, pressure and flow fluctuation caused by dynamic factors such as temperature change, material difference and mold abrasion is effectively compensated, and high-stability control over the outlet speed is achieved.

Description

Technical Field

[0001] This invention relates to a dynamic pressure compensation method for a zero-working-strip aluminum profile extrusion die, belonging to the field of metal plastic processing technology. Background Technology

[0002] Aluminum profile extrusion production is a complex and dynamic process. Its product quality (such as dimensional accuracy, surface quality, and microstructure) and productivity are greatly affected by process stability. During the extrusion process, a variety of dynamic factors can cause fluctuations in the main cylinder pressure and metal flow state. These factors include, but are not limited to: temperature gradients at the head, middle, and tail of the aluminum ingot; continuous heating and thermal equilibrium of the die during operation; temperature changes in the extrusion cylinder; and wear of the die working zone that gradually occurs as the number of production bars increases.

[0003] Traditional control methods typically employ constant-speed extrusion or constant-pressure extrusion modes. The former presets a fixed extrusion bar speed, while the latter maintains constant master cylinder pressure through simple PID feedback regulation. However, these methods are all "remedial" control methods, unable to anticipate and counteract the dynamic disturbances caused by the coupling of multiple variables. This leads to fluctuations in the profile exit speed, resulting in problems such as dimensional deviations and surface defects (e.g., ripples, bright bands). Simultaneously, it exacerbates abnormal wear of the die. Therefore, there is an urgent need to improve the dynamic pressure compensation method for zero-working-strip aluminum profile extrusion dies to solve the aforementioned problems. Summary of the Invention

[0004] The purpose of this invention is to provide a dynamic pressure compensation method for zero-working-strip aluminum profile extrusion dies, in order to solve the problem that traditional control methods usually adopt constant speed extrusion or constant pressure extrusion modes. The former presets a fixed extrusion bar speed, while the latter maintains a constant master cylinder pressure through simple PID feedback adjustment. However, these methods are all "post-event remedial" control, which cannot anticipate and offset the dynamic interference of the above-mentioned multi-variable coupling, resulting in fluctuations in the profile exit speed, and thus causing problems such as dimensional deviations and surface defects (such as ripples and bright bands), while also aggravating the problem of abnormal wear of the die.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A dynamic pressure compensation method for a zero-working-strip aluminum profile extrusion die includes the following steps:

[0007] Step 1: During the extrusion process, data from various sensors installed on the extruder and die are collected synchronously at a fixed control cycle (preferably 10-100 milliseconds);

[0008] The collected data should include at least: master cylinder pressure Actual displacement / velocity of the compression rod Mold temperature and variables characterizing the squeezing process. (Extrusion time) Or already compressed length );

[0009] Step 2: Transform the original time-series data into feature vectors that can fully characterize the current dynamic state of the system. Specifically, it includes:

[0010] Sliding window capture: Define a window containing the current time t and the previous time t. A sliding time window is defined for each historical moment, with the window length w set according to the dynamic characteristics of the process (corresponding to data of 5-30 seconds). The time series data matrix within the window is then obtained. ,in Let i be the sensor data vector at time i;

[0011] Feature engineering calculation: for window data Perform online calculations to extract three types of features:

[0012] Raw features: Sensor readings at the current moment ;

[0013] Statistical characteristics: Calculate the statistics of key parameters within the calculation window, including the mean value of the master cylinder pressure. and standard deviation This is used to characterize the average level and fluctuation intensity of pressure;

[0014] Change rate characteristics: Calculate the changing trend of key parameters, mold temperature change rate. and master cylinder pressure change rate This is used to capture the dynamic direction of the process;

[0015] Contextual feature fusion: Generate a static contextual feature vector C, containing the alloy grade and mold number of the current production task;

[0016] Feature concatenation: The original features, statistical features, and rate of change features obtained above are concatenated with the context feature vector C to form the final multidimensional input feature vector. ;

[0017] Step 3: Convert the constructed feature vectors Input into a pre-trained neural network prediction model In this process, the model is trained on a large amount of historical production data and can learn the mapping relationship between complex operating conditions and optimal control actions. Depending on different control strategies, there are two preferred implementation modes for this step:

[0018] First control mode: Neural network model The complete input constructed using the aforementioned steps (typically a time-series data matrix) The calculation is performed using the format of context C, and the optimal target set speed value of the squeeze lever for the next control cycle is directly output, i.e.:

[0019] Model In this mode, control commands are given directly;

[0020] The second control mode: neural network model The output is a prediction of the change in master cylinder pressure over a short period of time (1 second later). ;

[0021] Subsequently, based on the current pressure, the predicted pressure value is obtained. ;

[0022] Finally, this predicted value is fed into a PID controller, which adjusts the value according to the target pressure. With predicted pressure deviation = - Calculate the required speed compensation amount and update the target setting speed. ;

[0023] This model combines the predictive power of neural networks with the robustness of traditional control.

[0024] Step 4: Set the target speed generated in Step 3. As a command value, it is sent to the core motion controller of the extruder;

[0025] The motion controller uses its internal closed-loop control algorithm (such as position-speed-current three-loop control) to quickly and accurately adjust the drive current of the servo proportional valve or the displacement of the variable pump, thereby changing the flow of the hydraulic system and driving the actual speed of the extrusion rod to track the command value.

[0026] The dynamic characteristics of this execution process can be approximated as a first-order inertial element. ;

[0027] Through this real-time, continuous adjustment, the system actively compensates for interference, stabilizing the outlet speed or main cylinder pressure near the set target.

[0028] The effectiveness of the methods described in steps 1-4 above is based on a rigorous mathematical model, specifically manifested in the state-space modeling of the extrusion process and the approximation using neural networks:

[0029] State-space description of the process: The extrusion process is considered as a dynamic system, and its state... Evolved over time;

[0030] and controlled input and measurable disturbance The influence of (alloy, mold type); the state transition of the system is represented as ,in It is a complex nonlinear function. To address process noise, the key lies in using neural networks. To approximate this unknown true function with high precision ;

[0031] Neural network model structure: Long Short-Term Memory (LSTM) network is preferred as the core building block;

[0032] For the input sequence LSTM units utilize their gating mechanism (forget gate) Input gate Output gate ) and cell state To learn and memorize long-term dependencies; its mathematical expression is:

[0033]

[0034] in, σ is the hidden state, sigmoid function, and ⊙ is element-wise multiplication. The output of the LSTM layer is then mapped to the final control command or predicted value through a fully connected layer. This structure gives the model a powerful ability to capture the nonlinearity and temporal dynamics of the squeezing process.

[0035] Model training and optimization: The neural network parameters θ are obtained through supervised learning, collecting a large amount of historical production data, where each data sample includes the input feature X and the corresponding "label" under a specific working condition - the optimal extrusion bar speed that can maintain a stable exit speed under that working condition;

[0036] The training objective is to minimize the difference between the model's predicted output speed and the true optimal speed. The loss function is the mean squared error.

[0037] ;

[0038] The parameters θ are iteratively updated using optimization algorithms such as Adam until the model converges.

[0039] This invention has at least the following beneficial effects:

[0040] In this invention, information from multiple sensor sources can be fused in real time, and neural network models can be used to accurately predict the trend of process state changes. The extrusion speed can be adjusted proactively and in advance, thereby effectively compensating for pressure and flow fluctuations caused by dynamic factors such as temperature changes, material differences, and die wear, and achieving high stability control of the outlet speed. Attached Figure Description

[0041] Figure 1 This is a flowchart of a dynamic pressure compensation method for a zero-working-strip aluminum profile extrusion die according to the present invention.

[0042] Figure 2 This is a logic block diagram of the first and second control modes of the present invention. Detailed Implementation

[0043] The following will describe in detail the implementation of this application with reference to the accompanying drawings and embodiments, so that the implementation process of how this application uses technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly.

[0044] See Figure 1-2 A dynamic pressure compensation method for a zero-working-strip aluminum profile extrusion die, specifically as follows:

[0045] Example 1:

[0046] This embodiment provides a complete implementation process for a dynamic pressure compensation system:

[0047] First, install the necessary sensors on the extruder: main cylinder pressure sensor, extrusion rod displacement encoder, and die thermocouples (at least 2 points). In the control system, set the data acquisition cycle to 50ms.

[0048] When compression begins, the system starts a data acquisition thread to synchronously read the main cylinder pressure. Actual speed of the extrusion rod Mold temperature , and the current squeezing time obtained from the PLC ;

[0049] Next, the feature construction module starts working, setting the sliding window length w=100 (corresponding to 5 seconds of historical data). The module maintains a real-time updated data buffer. Every 50ms, when a new set of data arrives, the oldest set in the buffer is removed and the newest set is added.

[0050] Subsequently, the calculation module immediately calculates the mean and standard deviation of the pressure within the current window, as well as the average rate of change of the mold temperature. Simultaneously, the alloy grade and mold number entered by the operator on the HMI are encoded as a feature vector C; all features are concatenated into a feature vector of approximately 15 dimensions. ;

[0051] Then, the inference module loads the pre-trained neural network model and adopts the first control mode;

[0052] The model is a two-layer LSTM (64 units per layer) followed by two fully connected layers; The input model consists of a sequence of its first 99 historical feature vectors. After forward computation, the model outputs a scalar value, which is the suggested extrusion bar speed for the next cycle, for example, from the current 2.1 mm / s to 2.05 mm / s.

[0053] Finally, the motion control module receives the command. Its embedded PID controller uses the suggested extrusion bar speed for the next cycle as the setpoint and the actual encoder feedback as the process value to calculate the current signal output to the servo valve. The servo valve responds by changing the oil flow rate, thereby bringing the extrusion bar speed towards 2.05 mm / s in the next control cycle. The entire process is repeated cyclically until the extrusion is complete.

[0054] Example 2:

[0055] This embodiment adopts the second control mode, and the data acquisition and feature construction steps are the same as in embodiment 1;

[0056] The trained neural network model is constructed as a stress predictor in this mode, which will use the temporal feature vectors As input, the output is the change in master cylinder pressure in the next second (i.e., after 20 control cycles). For example, a prediction that the pressure will drop by 0.5 MPa;

[0057] The control system maintains a target pressure value of 25 MPa; the control algorithm adds the current pressure to the predicted change to obtain the predicted pressure one second later. The prediction deviation is then calculated. ;

[0058] This deviation It is fed into an independent pressure PID controller, whose proportional gain is... Integral Time Differential time It has been pre-configured;

[0059] PID controller according to Based on its historical values, the speed adjustment required to eliminate this prediction bias is calculated, and this adjustment is added to the current speed setpoint to obtain a new speed value.

[0060] The subsequent servo execution steps are exactly the same as in Example 1. In this way, the system starts deceleration compensation in advance before the actual pressure drops significantly, achieving a smoother and more proactive control effect.

[0061] If certain terms are used in the specification and claims to refer to specific components, those skilled in the art will understand that hardware manufacturers may use different names to refer to the same component. This specification and claims do not distinguish components based on differences in name, but rather on differences in function. The term "comprising" as used throughout the specification and claims is an open-ended term and should be interpreted as "comprising but not limited to." "Approximately" means that within an acceptable margin of error, those skilled in the art can solve the technical problem and substantially achieve the technical effect within a certain margin of error.

[0062] It should be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a product or system comprising a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a product or system. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the product or system that includes that element.

[0063] The foregoing description illustrates and describes several preferred embodiments of the present invention. However, as previously stated, it should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the inventive concept described herein through the foregoing teachings or techniques or knowledge in related fields. Any modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A dynamic pressure compensation method for a zero-working-strip aluminum profile extrusion die, characterized in that, Includes the following steps: Step 1: Synchronously collect the main cylinder pressure during the extrusion process according to the set control cycle. Actual speed of the extrusion rod Mold temperature and process variables used to characterize the squeezing process Sensor data; Step 2: Based on the sensor data collected currently and within the preset time window in the past, construct a multi-dimensional input feature vector that includes temporal features and contextual features. ; Step 3: Convert the input feature vector The input is fed into a pre-trained neural network prediction model, which calculates and outputs in real time the control amount of the extrusion rod speed for pressure compensation. ; Step 4: Control the speed based on the output. The servo actuator drives the extruder in real time, dynamically compensating for pressure fluctuations caused by die temperature rise, material rheological stress changes, and die wear.

2. The dynamic pressure compensation method for a zero-working-strip aluminum profile extrusion die according to claim 1, characterized in that: In step 1, process variables are used to characterize the squeezing process. For compression time Or already compressed length .

3. The dynamic pressure compensation method for a zero-working-strip aluminum profile extrusion die according to claim 1, characterized in that: Feature vector Specifically, it includes: Define a time that includes the current time and the previous time. A sliding time window representing historical moments is used to obtain the time series data matrix within the window. ; For the time series data matrix Perform feature extraction to generate a combined feature set containing original values, statistical features, and rate of change features; The combined feature set is concatenated with a static context feature vector C to form the input feature vector. The context feature vector C is composed of the encoding of the alloy grade and the mold number.

4. The dynamic pressure compensation method for a zero-working-strip aluminum profile extrusion die according to claim 3, characterized in that: The statistical characteristics include the mean value of the master cylinder pressure within the time window. with standard deviation The rate of change characteristic includes the mold temperature change rate. and master cylinder pressure change rate .

5. The dynamic pressure compensation method for a zero-working-strip aluminum profile extrusion die according to claim 1, characterized in that: The neural network prediction model is a time-series prediction model based on the Long Short-Term Memory (LSTM) network.

6. The dynamic pressure compensation method for a zero-working-strip aluminum profile extrusion die according to claim 5, characterized in that: The neural network prediction model adopts an encoder-decoder structure, where the encoder is an LSTM layer used to extract the input feature vector. Advanced temporal features in [the context]; The decoder is a fully connected layer used to map the high-level temporal features to the speed control quantity. .

7. The dynamic pressure compensation method for a zero-working-strip aluminum profile extrusion die according to claim 1, characterized in that: In step S3, the specific implementation is the first control mode: The neural network prediction model uses time-series data within the aforementioned sliding time window. With context feature C as input, the target set speed value of the extrusion rod for the next control cycle is directly output. ,and ,in This refers to the neural network prediction model.

8. The dynamic pressure compensation method for a zero-working-strip aluminum profile extrusion die according to claim 1, characterized in that: In step S3, the second control mode is specifically implemented: The neural network prediction model uses time-series data within the aforementioned sliding time window. Taking the context feature C as input, the output is a predicted value of the change in master cylinder pressure after a set time period in the future. ; Then, the predicted value With current pressure Add them together to get the predicted pressure value. ; Finally, a proportional-integral-derivative PID controller is used to adjust the pressure according to the target pressure. With the predicted pressure value deviation Calculate the speed compensation amount And update the target set speed of the extrusion bar accordingly. .

9. The dynamic pressure compensation method for a zero-working-strip aluminum profile extrusion die according to claim 1, characterized in that: The drive servo actuator mentioned in step S4 is specifically: Set the target speed The command value is sent to the motion controller of the extruder; The motion controller adjusts the current of the servo proportional valve or the displacement of the variable pump through a closed-loop control algorithm, so that the actual speed of the extrusion rod follows the command value. change.

10. The dynamic pressure compensation method for a zero-working-strip aluminum profile extrusion die according to claim 1, characterized in that: It also includes an online model adaptation step: During system operation, the stability indicators of profile export speed are continuously monitored; When the stability index exceeds a preset threshold or a production batch is switched, the model update process is triggered, and the neural network prediction model is incrementally learned or fine-tuned using the latest collected production data.

Images