CN-121607598-B - Aluminum wheel die-casting mold temperature closed-loop regulation and control system and method

CN121607598BCN 121607598 BCN121607598 BCN 121607598BCN-121607598-B

Abstract

The invention discloses a temperature closed-loop regulation and control system and a temperature closed-loop regulation and control method for an aluminum wheel die-casting mould, which belong to the technical field of aluminum alloy die-casting forming and intelligent control and comprise a data acquisition module, in particular a temperature sensor, a flow sensor and a pressure sensor which are arranged at key positions of the mould, wherein the data acquisition module is in communication connection with a prediction module, the prediction module adopts a CNN-LSTM composite neural network model, the prediction module is in communication connection with an optimization decision module and provides a predicted value, the optimization decision module adopts an integrated tree-structure pareto estimator TPE, the optimization decision module provides an optimal decision to an execution control module, the execution control module is a PLC controller, and a feedback learning module is further arranged for carrying out online update on the prediction model and an optimization strategy according to control effect data. The invention can effectively reduce the energy consumption and the manual intervention cost, and has obvious practical value and economic value.

Inventors

BI JIANG
XU QINGYAN
TAN YUNXIANG
Ma Juhuai
WANG ZHENTAO
HAN JIE
LI SHIDE
WANG JI
DONG GUOJIANG
LIU DEHUA
GUO SHIWEI
Yuan Tianxiao

Assignees

燕山大学
中信戴卡股份有限公司
清华大学

Dates

Publication Date: 20260508
Application Date: 20260202

Claims (9)

1. A temperature closed-loop regulation and control system of an aluminum wheel die-casting mould comprises a data acquisition module, in particular a temperature sensor, a flow sensor and a pressure sensor which are arranged at key positions of the mould, and is characterized in that the data acquisition module is connected with a prediction module in a communication manner, and the prediction module adopts the following components The composite neural network model is characterized in that an optimization decision module is in communication connection with a prediction module and provides a predicted value, the optimization decision module adopts an integrated tree-structured pareto estimator TPE, the optimization decision module provides an optimal decision to an execution control module, the execution control module is a PLC controller, and a feedback learning module is also arranged for updating the prediction model and the optimization strategy on line according to control effect data; The regulation and control method based on the closed-loop regulation and control system comprises the following steps: Step1, collecting historical data and training by using the historical data A composite neural network model; Step 2, collecting relevant data of the die in real time by using a collecting module, preprocessing the data, and inputting the data into a trained machine Outputting a temperature predicted value of each measuring point in the future by the composite neural network model; step 3, according to the temperature predicted value and the predicted result, when the predicted result is abnormal, according to an ideal temperature curve, the TPE module intervenes in the TPE module, gives an initial cooling parameter predicted combination, converts the initial cooling parameter predicted combination into time sequence data, inputs the time sequence data into the predicted module, predicts whether the predicted result after the initial cooling parameter combination is regulated can reach the requirement of meeting the ideal temperature curve, and if not, the process iterates until the predicted result is given to meet the optimal control parameter of the ideal temperature curve; inputting the optimal control parameters into an execution control module, and driving an electric regulating valve and a variable-frequency water pump by virtue of a PLC (programmable logic controller) according to the optimal control parameters, so as to accurately regulate the flow rate and the on-off time sequence of cooling water; Step 5, collecting data, feeding the data back to the data collection of the prediction model to complete closed loop, and synchronously and adaptively adjusting the weight coefficient of the objective function And And periodically analyzing the sampling distribution of the TPE, and identifying a parameter sensitive area for key optimization.
2. The aluminum wheel die-casting mold temperature closed-loop regulation system of claim 1, wherein the CNN-LSTM composite neural network model comprises an input layer, a Conv1D layer, a MaxPooling D layer, an LSTM layer, a Dropout layer, an LSTM layer, a flame layer, a Dense layer and an output layer in sequence from input to output.
3. The aluminum wheel die-casting mold temperature closed-loop control system according to claim 2, wherein: The input layer of the composite neural network model adopts an ultra-long time sequence window design, namely a receipt window design of 1250 time steps, and captures the thermal inertia characteristic and the complete process cycle of the die casting process.
4. The aluminum wheel die-casting mold temperature closed-loop regulation system according to claim 1, wherein: The input of the tree-structured pareto estimator TPE is manually set standard process template data, namely an ideal temperature curve and a previous-round cooling parameter, the output is an optimal control parameter, the previous-round cooling parameter and the optimal control parameter are parameters for controlling cooling equipment, a hot start mechanism is introduced into the TPE, namely, the process time is divided into two controllable and uncontrollable stages, a differential processing strategy is adopted in TPE optimization, reference process prediction data are multiplexed for the uncontrollable stages, and parameter optimization searching is carried out for the controllable stages.
5. The aluminum wheel die-casting mold temperature closed-loop control system according to claim 1, wherein the step 1 is specifically as follows: Step 1.1, collecting production data for a period of time, wherein the production data comprises temperature time sequence data of each measuring point in each casting period, sampling intervals of 1 second, corresponding cooling process parameters, product quality detection results, environment parameters, and workshop temperature and humidity, wherein the corresponding cooling process parameters comprise flow, temperature and on-off time sequences; Step 1.2, data cleaning and feature engineering, namely eliminating abnormal data during equipment failure, performing receipt window processing on temperature data, performing window width 1250 step length, performing accumulation calculation on flow data, strengthening flow features, and normalizing all features to be equal to one A section; Step 1.3 training CNN-LSTM composite neural network model, wherein the input layer receives 1250 seconds of history data, the CNN layer comprises 2 convolution blocks to extract spatial characteristics, the LSTM layer comprises 2 layers of 128 neuron capturing time sequence dependencies, the output layer predicts a temperature value of one period in the future, an Adam optimizer is adopted, the learning rate is 0.001, the batch size is 64, 30 epochs are trained, and the method is achieved on a verification set And (5) storing the model.
6. The aluminum wheel die-casting mold temperature closed-loop control system according to claim 1, wherein the step2 is specifically as follows: Step 2.1, establishing a real-time data stream processing pipeline, configuring a data acquisition module to acquire all sensor data at the frequency of 1Hz, performing edge calculation pretreatment including data verification, outlier rejection and missing value interpolation, constructing a sliding time window, maintaining a history data cache of the latest 1250 seconds, and pushing the pretreated data to a prediction module in real time; step 2.2, performing on-line temperature prediction, and inputting the current time window data into the deployed device The model outputs the temperature predicted value of each measuring point in 245 seconds, calculates the deviation vector of the predicted temperature and the set target temperature When any measuring point is deviated Triggering an optimization decision module; and 2.3, carrying out dynamic correction of a prediction result, fusing a prediction value and an actual measurement value by adopting Kalman filtering, dynamically adjusting prediction confidence according to error statistics of the last 10 predictions, and triggering an online fine adjustment mechanism of the model when the continuous 5 prediction errors exceed a threshold value.
7. The aluminum wheel die-casting mold temperature closed-loop control system according to claim 1, wherein the specific steps of the step 3 are as follows: Step 3.1, constructing Bayesian optimization search space, defining decision variable space Comprising i key cooling control parameters, namely delay time Time of opening Setting an objective function I.e. root mean square error of predicted temperature and set temperature curve, defining early-stop threshold Automatically terminating the optimization when the early-stop threshold is reached; Step 3.2, performing intelligent search by a TPE algorithm, realizing a TPE sampler by using a Optuna framework, randomly sampling 5 parameter combinations in an initialization stage, evaluating RMSE values as priori knowledge, automatically sequencing historical observation points according to target values by the TPE algorithm, selecting the observation point with the best performance to construct a good sample set L (x), constructing a poor sample set G (x) by the rest, fitting a kernel density estimation model to the L (x) and the G (x) respectively, and constructing a probability density function And By maximizing the expected improvement Selecting the next sampling point, setting the maximum evaluation times to 25 times when Or ending when the upper limit of the evaluation times is reached; Step 3.3, implementing a downsampling acceleration strategy, performing 10 times downsampling on the generated 1Hz high-frequency process data, converting the data into 0.1Hz data, obviously reducing the input data quantity of a CNN-LSTM model, reserving 2450 time steps of complete casting cycle data after downsampling, shortening the single temperature prediction time from 30 seconds to less than 3 seconds through downsampling, and enabling the TPE algorithm to complete multiple iterative evaluation in the production beat; Step 3.4, generating optimized control parameters, and extracting optimal parameter combinations after TPE optimization is completed Update to In a dictionary by The function converts the optimized parameters into a complete time sequence control sequence, which comprises flow set values of each path of cooling water and accurate switch time sequence, wherein the range of the flow set values is Adding feature engineering process to calculate accumulated flow The method is used for accurately predicting the model, and storing the optimal parameters into a JSON file; Step 3.5, verifying the optimization effect, re-executing the complete prediction process by using the optimal parameters, calculating the final RMSE value for verification, and passing The method comprises the steps of evaluating a maximum absolute error MAE and a root mean square error RMSE of a curve of a predicted temperature and a set temperature by a function, recording a convergence curve of an optimization process, analyzing performance improvement of a TPE algorithm in the first 5 times of random exploration and the subsequent 20 times of Bayesian optimization, and counting total optimization time to ensure that real-time control requirements are met.
8. The aluminum wheel die-casting mold temperature closed-loop control system according to claim 1, wherein the step 4 is specifically as follows: step 4.1, issuing and executing a control instruction, encoding optimal control parameters after TPE optimization into Modbus RTU protocol frames, sending accurate opening instructions to each path of electric regulating valve through a PLC (programmable logic controller), controlling a variable-frequency water pump in a coordinated manner, and maintaining the pressure of the system stable And recording the instruction execution state and response time in real time.
9. The aluminum wheel die-casting mold temperature closed-loop control system according to claim 1, wherein the step 5 is specifically as follows: Step 5.1, monitoring control effect and performance evaluation, collecting temperature response data after control execution, calculating root mean square error of actual temperature and target temperature, evaluating temperature convergence time and overshoot, counting convergence speed and calculation time consumption of TPE algorithm, recording cooling water consumption and energy consumption data for cost analysis; step 5.2, implementing online update of Bayesian optimization, storing parameter-effect pairs controlled each time into an experience database, updating prior distribution of TPE by using new data every time 30 casting cycles are completed, adopting a migration learning strategy to migrate optimization experience of similar working conditions to a new task, and adaptively adjusting weight coefficients of an objective function based on accumulated control effects And Periodically analyzing the sampling distribution of the TPE, and identifying a parameter sensitive area for key optimization; Step 5.3, constructing a knowledge graph auxiliary decision, extracting high-frequency excellent parameter combinations in the TPE optimization process, constructing a process knowledge base, establishing a mapping relation of temperature deviation and optimal parameters, analyzing correlation and coupling relation among the parameters, and generating an optimization report.

Description

Aluminum wheel die-casting mold temperature closed-loop regulation and control system and method Technical Field The invention relates to the technical field of aluminum alloy die-casting forming and intelligent control, in particular to a closed-loop control system and method for aluminum wheel die-casting mold temperature based on machine learning prediction and optimization. Background In the field of automobile light weight and high end manufacturing, aluminum wheels are core parts because of the characteristics of high strength and light weight, low pressure casting is a main stream production process, uniformity and stability of a mold temperature field directly determine solidification rate, tissue morphology and internal quality of castings, if the mold temperature is controlled improperly, defects such as shrinkage cavities, shrinkage porosity, deformation and surface cracks of the castings are easily caused, mechanical properties and service life of the aluminum wheels are seriously affected, and therefore, accurate regulation and control of the mold temperature is a core technical link in the low pressure casting process of the aluminum wheels. At present, the temperature control technology for the aluminum wheel die casting die in the industry is mainly divided into four major categories of empirical control method, PID control, preset program control and off-line optimization, and all the major technical limitations exist in all the methods. The empirical control method is used as a traditional regulation and control mode, relies on a skilled worker to manually adjust the opening degree and the on-off time length of the cooling water valve by observing visual characteristics such as the surface color and luster of the casting, the molding state and the like, and can maintain basic production under fixed working conditions, but the mode is strong in subjectivity, difficult to unify in operation standard, obvious response lag exists in manual judgment and adjustment, and the high-beat and high-precision production requirements of a modern assembly line cannot be adapted. PID control realizes temperature deviation correction by means of combination adjustment of proportion, integration and differentiation, and is widely applied to simple industrial control scenes, however, the aluminum wheel die casting process has complex characteristics of strong nonlinearity, large time lag and multivariable coupling, the die temperature can be influenced by various factors such as raw material component fluctuation, environmental temperature and humidity change, equipment operation loss and the like, fixed parameters of the traditional PID controller are difficult to deal with working condition dynamic change, frequent manual setting parameters are needed, and adaptability and control accuracy can not meet high-end casting production requirements. The preset program control is an open loop control mode of presetting a cooling time sequence and parameter combination according to a given technological rule, is simple and convenient to operate and high in repeatability, but lacks dynamic adjustment capability, and cannot correct a control strategy in real time when facing sudden conditions such as raw material batch difference, workshop environment disturbance and the like, and is easy to cause an optimal section of a die Wen Pianli so as to cause quality fluctuation of castings. The off-line optimization method is characterized in that static optimization of technological parameters is completed before production by adopting numerical simulation software such as ProCAST, MAGMA or experimental design means such as an orthogonal test and a response surface method, and theoretical optimal parameter combinations can be obtained, but the method is long in optimization period, high in test and simulation cost, and the optimization result is a fixed parameter set, so that dynamic working condition changes in the production process cannot be responded in real time, and optimal control of the die temperature in the whole production period is difficult to ensure. Further, the existing mold temperature control scheme has five core technical defects that firstly, the prediction capability is insufficient, various methods can only implement passive adjustment according to the current temperature deviation, the prediction capability of future change trend of a mold temperature field is lacking, the control has hysteresis, the fluctuation range of the mold temperature is large, secondly, the optimization response is slow, the traditional method cannot quickly solve the multi-parameter collaborative optimization problem within the severe time constraint of the production beat, real-time optimization control is difficult to realize, thirdly, the autonomous learning capability is lacking, the control strategy is solidified for a long time, the process rule cannot be mined from massive historical production data, the regulation an