Automobile injection molding part production process control system and method

Abstract

The invention relates to the technical field of automobile part manufacturing, and discloses an automobile injection molding part production process control system and method, and the system comprises the following modules: a data collection module which is used for collecting technological parameters, molds, raw materials and equipment operation original data, attaching timestamps, and storing the data in a database; the process parameter prediction module is used for reading original data to construct a time sequence data set, inputting the time sequence data set into a TFT model to obtain a pre-training model, and predicting a short-term process parameter fluctuation range in combination with current production working condition parameters; and the quality risk index acquisition module is used for inputting the process parameter data and the mold data into a quality risk index calculation formula to obtain a quality risk index. Through the system, data-driven comprehensive production optimization is realized, the process control accuracy and adaptability are improved, the quality control scientificity and reliability are enhanced, the intelligent level of the production process is improved, the production efficiency is effectively improved, the defective rate is reduced, and the product quality is stabilized. The problem that process control lacks system intelligence is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automobile parts manufacturing, in particular to a production process control system and method for automobile injection molded parts. BACKGROUND

[0002] With the development of the automobile industry, the production scale and quality requirements of automobile injection molded parts as a key component of automobiles are constantly improving. Under the trend of modern manufacturing, consumers' expectations for the overall quality and performance of automobiles are increasing.

[0003] However, the traditional production method of automobile injection molded parts is derived from a relatively extensive industrial manufacturing mode, and has long relied on manual experience and relatively simple and fixed processes in process control. In the face of today's complex and changing market demand, raw material supply differences and increasingly precise product design requirements, traditional process control methods are inadequate, and the lack of system intelligence is increasingly prominent. SUMMARY

[0004] To overcome the shortcomings of the prior art, the present application provides a production process control system and method for automobile injection molded parts, which solves the problem of lack of system intelligence in process control.

[0005] To achieve the above purpose, the present application realizes a production process control system for automobile injection molded parts by the following technical solutions, which includes the following modules:

[0006] A data acquisition module is used to collect process parameters, molds, raw materials and equipment operation raw data, and store them in a database with a time stamp;

[0007] A process parameter prediction module is used to read the raw data to construct a time series data set, input a TFT model to obtain a pre-trained model, and predict the short-term process parameter fluctuation range in combination with the current production condition parameters;

[0008] A quality risk index acquisition module is used to input process parameter data and mold data into a quality risk index calculation formula to obtain a quality risk index;

[0009] A defect feature acquisition module is used to collect product surface images and injection sound frequencies, and input them into a multi-source data recognition model after preprocessing to obtain a defect feature index;

[0010] A reinforcement learning control module is used to design the state space, action space and reward function of the DQN algorithm according to the raw data, short-term process parameter fluctuation range, quality risk index and defect feature index, train and update the online Q network and target Q network, and minimize the loss function;

[0011] The process execution adjustment module is configured to adjust the temperature controller, the pressure regulating valve and the cooling system of the injection molding machine according to the optimal DQN control model, so as to complete the real-time optimization control of the process parameters.

[0012] By adopting the technical scheme, data-driven decision making is realized based on the cooperation of the modules, quality risks can be early warned and controlled, defects can be accurately detected, process parameters can be optimized in real time, production stability and consistency can be ensured, and the system intelligence level and enterprise competitiveness are improved.

[0013] Preferably, the data acquisition module comprises:

[0014] A temperature sensor array is arranged on the barrel, screw and mold cavity of the injection molding machine, and is configured to acquire melt temperature and mold temperature data.

[0015] A pressure sensor group is arranged on the surface of the mold cavity and the hydraulic system of the injection molding machine, and is configured to acquire melt filling pressure and injection pressure data.

[0016] A displacement sensor is configured to acquire screw displacement data to determine the injection amount.

[0017] A data interface unit is configured to access raw material batch parameters, product design drawing data and production equipment operating state data.

[0018] Preferably, the process parameter prediction module comprises: when constructing the time series data set, the original data is subjected to sliding window processing, and the window size is 24 hours of production data; the TFT model is a time frequency transformer model, the input layer of which receives normalized time series data, and the output layer predicts the fluctuation range of the process parameters in the future 1-4 hours, including the temperature fluctuation threshold, the pressure fluctuation threshold and the cooling time fluctuation threshold.

[0019] Preferably, the quality risk index acquisition module comprises: the quality risk index calculation formula is wherein RI is the quality risk index, T is the real-time temperature, T0 is the standard temperature, P is the real-time pressure, P0 is the standard pressure, D is the mold wear amount, D max is the maximum allowable mold wear amount, and w1, w2 and w3 are weight coefficients and w1+w2+w3=1.

[0020] Preferably, the defect feature acquisition module comprises: the pre-processing of the surface image removes noise through Gaussian filtering and extracts the defect area by threshold segmentation; the pre-processing of the injection sound frequency converts the time domain signal into a frequency domain feature spectrum through short-time Fourier transform; the multi-source data recognition model is a fusion model of a convolutional neural network and a recurrent neural network, the input layer of which receives an image feature matrix and a sound frequency feature spectrum, and the output layer outputs a defect type and a severity score.

[0021] Preferably, the reinforcement learning control module includes: the state space is defined as S={X,F,RI,FI,T pred ,P pred}, where X is the original data feature vector, F is the defect feature index, RI is the quality risk index, FI is the process parameter fluctuation range, T pred To predict temperature fluctuations, P pred To predict pressure fluctuations; the action space is defined as A = {ΔT, ΔP, Δt}, where ΔT is the temperature adjustment, ΔP is the pressure adjustment, and Δt is the cooling time adjustment; the reward function is designed as R = -RI·w r -FI·w f +δ·δ good Among them, w r 、w f is the weight coefficient, δ is the process stability index, δ good It is the factor for determining product qualification.

[0022] Preferably, the reinforcement learning control module further includes: a loss function is Among them, θ is the online Q network parameter, θ - is the target Q network parameter, γ is the discount factor, and the parameters are updated by stochastic gradient descent method. Where α is the learning rate.

[0023] Preferably, the process execution adjustment module includes: the adjustment accuracy of the temperature controller is ±0.5°C, the response time of the pressure regulating valve is ≤0.05 seconds, and the flow adjustment range of the cooling system is 30%-100% of the rated flow; during adjustment, according to the action parameters output by the optimal DQN control model, it is executed in the following order: first adjust the temperature controller, adjust the pressure regulating valve after an interval of 15-30 seconds, and then adjust the cooling system after an interval of 10-20 seconds.

[0024] A control method for a production process control system for automobile injection molded parts, used in the production process control system for automobile injection molded parts, comprises the following steps:

[0025] S1, the data acquisition module collects the original data and stores it in the database with a timestamp;

[0026] S2, the process parameter prediction module reads data to construct a time series data set and predicts the short-term process parameter fluctuation range through the TFT model;

[0027] S3, the quality risk index acquisition module calculates the quality risk index based on the process parameter data and mold data;

[0028] S4, the defect feature acquisition module collects surface images and audio, and after pre-processing, inputs them into the multi-source data recognition model to obtain the defect feature index;

[0029] S5, the reinforcement learning control module trains the DQN algorithm based on the original data, the fluctuation range, the quality risk index and the defect feature index to obtain an optimal control model;

[0030] S6, the process execution adjustment module adjusts the temperature, the pressure and the cooling parameter according to the optimal control model to complete real-time optimization control.

[0031] The application provides a system and method for controlling the production process of an automobile injection molding part.

[0032] 1. The application realizes comprehensive production optimization driven by data through the system, improves the accuracy and adaptability of process control, enhances the scientific nature and reliability of quality control, improves the intelligent level of the production process, effectively improves the production efficiency, reduces the rate of defective products and stabilizes the product quality. The problem of lack of system intelligence in process control is solved.

[0033] 2. The application collects various physical parameters and information parameters of injection molding production through the use of multiple types of sensors and data interface integration, constructs a full-factor data file, realizes fine control through precise monitoring and time stamp marking, reduces quality risks from the source, improves system control efficiency through multi-module data interconnection, ensures production consistency and stability, and provides assistance for process optimization and quality control. The problem that single temperature measurement cannot fully reflect the melt temperature field and mold thermal balance, which easily causes warping defects, is solved.

[0034] 3. The application eliminates dimensional differences by using relative deviation and wear ratio calculation, makes the quality risk index have cross-process and cross-product comparability, realizes accurate quantitative risk level, relies on a weighted model to cooperatively evaluate the influence of multiple parameters on quality risk, clearly defines the process adjustment priority, and improves the efficiency of quality control. By quantifying mold wear into the index, the trend of mold wear can be used to plan mold maintenance or replacement in advance, reduce the rate of defective products and production cost, and ensure production continuity and economic benefits. The problems of subjectivity, non-quantization and blind area in risk analysis under the coupling effect of multiple factors in traditional qualitative assessment of quality risk relying on manual experience are solved.

[0035] 4. The application realizes multi-factor cooperative optimization control, cooperatively adjusts process parameters such as temperature, pressure and cooling time according to the comprehensive influence of each factor, improves the accuracy and comprehensiveness of process control, and stabilizes product quality; improves the intelligent and automatic level of process adjustment, relies on reinforcement learning to automatically make decisions, outputs the optimal adjustment scheme according to real-time state feedback, improves the adjustment efficiency and reduces human errors; enhances the adaptability and generalization ability of the model to actual production, so that it can cope with various working condition changes, output appropriate suggestions, reduce adjustment cost, improve the robustness and flexibility of the injection molding production system. The problem of lack of system in multi-factor comprehensive decision-making and process adjustment in injection molding production is solved. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 A module architecture diagram of an automobile injection molding part production process control system is provided for the present application.

[0037] Figure 2 A control method flowchart of an automobile injection molding part production process control system is provided for the present application. DETAILED DESCRIPTION

[0038] The technical solutions of the present application will be described clearly and completely below in combination with the drawings of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0039] Please refer to the drawings of the present application Figure 1 The present application provides an automobile injection molding part production process control system, which comprises the following modules:

[0040] A data acquisition module is used to acquire process parameters, molds, raw materials and equipment operation raw data, and store them into a database with time stamps.

[0041] The data acquisition module comprises:

[0042] A temperature sensor array is arranged in the injection molding machine cylinder, screw and mold cavity to acquire melt temperature and mold temperature data;

[0043] A pressure sensor group is arranged on the surface of the mold cavity and the hydraulic system of the injection molding machine to acquire melt filling pressure and injection pressure data;

[0044] A displacement sensor is used to acquire screw displacement data to determine the injection amount;

[0045] A data interface unit is used to access raw material batch parameters, product design drawing data and production equipment operation state data.

[0046] Specifically, temperature sensors are arranged in the injection molding machine cylinder, screw and mold cavity. Based on the heat conduction principle, the temperature changes of the plastic melt during the cylinder heating, screw conveying and mold cavity injection molding stages are sensed to acquire melt temperature and mold temperature data. These temperature data reflect the plasticization state of the plastic melt and the heat transfer situation during the injection molding process, which are the basic monitoring data for avoiding defects such as warping and weld lines of the injection molded parts due to uneven temperature, and provide temperature basis for subsequent process adjustment.

[0047] Pressure sensors are arranged on the surface of the mold cavity and the hydraulic system of the injection molding machine. According to the pressure conversion principle, the melt filling pressure and injection pressure data are collected. The melt filling pressure data reflect the flow resistance of the melt in the cavity and the filling effect, and the injection pressure data reflect the power output of the hydraulic system.

[0048] The screw displacement data are collected based on the electromagnetic induction or optical detection principle by installing the screw driving mechanism. Since the screw propulsion distance is directly related to the injection amount, the injection amount is accurately controlled through displacement monitoring.

[0049] The industrial data communication protocol is followed to access the raw material batch parameters, product design drawing data, and production equipment operation state data, and to open up the production multi-source information channel.

[0050] After the collected raw data are attached with time stamps, they are uniformly stored in the database to provide basic data support with time sequence characteristics for subsequent modules of the system.

[0051] Through the cooperation of multiple types of sensors and the integration of data interfaces, the comprehensive collection of physical parameters such as temperature, pressure, and displacement and information parameters such as raw materials, design, and equipment in injection molding production is realized, a full-factor data file is constructed, and the fine control of injection amount, temperature field, and pressure field is achieved through precise monitoring and time stamp marking, thereby reducing quality risks from the source. At the same time, through the data interconnection of multiple modules, the overall control efficiency of the system is improved, the consistency and stability of injection molding production are ensured, and solid data support is provided for process optimization and quality control. The problem of single temperature measurement being unable to comprehensively reflect the melt temperature field and mold thermal balance and easily causing warping defects is solved.

[0052] The process parameter prediction module is used to read the raw data to construct a time series data set, input the TFT model to obtain a pre-trained model, and predict the short-term process parameter fluctuation range in combination with the current production condition parameters.

[0053] The process parameter prediction module includes: when constructing the time series data set, the raw data are subjected to sliding window processing, and the window size is 24 hours of production data; the TFT model is a time frequency transformer model, the input layer of which receives normalized time series data, and the output layer predicts the process parameter fluctuation range in the future 1-4 hours, including temperature fluctuation threshold, pressure fluctuation threshold, and cooling time fluctuation threshold.

[0054] Specifically, the time series data set is constructed by reading the raw data output by the data collection module and using the sliding window method to construct the data set. With a window size of 24 hours of production data, continuous data segments are sequentially intercepted according to time sequence to form multiple groups of time series samples. This method can retain the continuous change characteristics of process parameters over time and adapt to the time sequence correlation of the injection molding production process.

[0055] Data normalization processing is performed on the constructed time series data to convert parameters of different dimensions and value ranges into a uniform distribution, eliminate the interference of dimension differences on model training, and make the input data meet the training requirements of the TFT model.

[0056] TFT model prediction process: The TFT model serves as the core prediction unit, and the input layer receives the normalized time series data. Through the multi-layer attention mechanism and the time-frequency feature extraction module, the long and short-term time series rules of the parameters are mined. Based on the learned time series patterns, the model output layer predicts the fluctuation range of the process parameters in the next 1-4 hours, including temperature fluctuation threshold, pressure fluctuation threshold, and cooling time fluctuation threshold, providing advance and parameter adjustment boundaries for subsequent process control.

[0057] Through the combination of sliding window and TFT model, the long and short-term time series rules of process parameters are accurately extracted, and the fluctuation prediction accuracy is improved. Normalization processing realizes the collaborative adaptation of multiple parameters, and outputs fluctuation thresholds covering temperature, pressure, and cooling time, providing complete reference for comprehensive process adjustment. Predicting the parameter fluctuation range 1-4 hours in advance supports the control system to plan adjustment strategies in advance, effectively improves the stability and controllability of injection molding production, reduces the scrap rate and process debugging time. It solves the problems of insufficient utilization of time series characteristics of process parameters in traditional control, model training failure, and short-term fluctuation prediction deficiency leading to process out of control and quality defects.

[0058] The quality risk index acquisition module is used to input the process parameter data and mold data into the quality risk index calculation formula to obtain the quality risk index.

[0059] The quality risk index acquisition module includes: the quality risk index calculation formula is where RI is the quality risk index, T is the real-time temperature, T0 is the standard temperature, P is the real-time pressure, P0 is the standard pressure, D is the mold wear amount, D max is the maximum allowable mold wear amount, and w1, w2, and w3 are weight coefficients and w1+w2+w3=1.

[0060] Specifically, first, the corresponding key data is obtained from the data acquisition module, including the real-time temperature T and the real-time pressure P in the process parameter data, which are monitored and recorded in real time by the temperature sensor and pressure sensor installed in the relevant parts of the injection molding machine. As for the mold data, the focus is on the mold wear amount D, which can be obtained through regular detection of the mold or with the help of some equipment that can monitor the mold wear condition in real time. In addition, the standard temperature T0 and the standard pressure P0 are fixed values that are pre-set during the product design stage according to the design requirements of the product, and they represent the temperature and pressure parameters that guarantee the quality of the injection molded parts in the ideal state. The maximum allowable mold wear amount D maxD0 is the maximum wear of the mold, which is determined by considering the material properties of the mold and its initial design life, etc. For example, for a mold made of a certain material, its maximum wear value can be determined by professional engineering calculation or experience data of similar molds according to the wear resistance, structural strength, and expected number of production uses of the material.

[0061] The quality risk index RI is calculated using a specially designed quality risk index calculation formula In this formula, each part has a clear physical meaning and a characterization effect on quality risk. For This item reflects the deviation of real-time temperature T from standard temperature T0, and the temperature fluctuation is shown in the form of relative proportion. If the real-time temperature T is higher than the standard temperature T0 by a certain value, the relative size of the temperature deviation from the standard value can be intuitively seen by this formula calculation, and the influence of temperature factor on quality risk can be reflected.

[0062] Similarly, represents the relative deviation of real-time pressure P from standard pressure P0, that is, the proportional relationship of pressure fluctuation. In the injection molding process, the stability of pressure has a crucial influence on the molding quality of injection molded parts. For example, excessive pressure may cause flash, material shortage, and other defects in injection molded parts, while insufficient pressure may cause the injection molded parts to be underfilled, etc. Through this relative proportion calculation, the risk of pressure fluctuation to quality can be clearly measured.

[0063] represents the proportional relationship of mold wear D to the maximum allowable mold wear D max It reflects the position of the current wear degree of the mold in the whole range of allowable wear. With the continuous production, the mold will inevitably wear, and the wear of the mold will directly affect the dimensional accuracy, surface quality, etc. of the injection molded parts. Through this proportional calculation, the factor of mold wear can be included in the overall quality risk consideration.

[0064] The weight coefficients w1, w2, w3 play a role in balancing the influence degree of different factors on the quality risk, and they satisfy the relationship w1+w2+w3=1. The specific values of these weight coefficients are set in advance according to the characteristics of the product itself and the specific process requirements. For some special plastic materials that are extremely sensitive to temperature changes, small fluctuations in temperature during injection molding can cause significant quality problems. Therefore, the value of w1 is set relatively high to highlight the importance of the temperature factor in the overall quality risk assessment. Through such a weighted summation method, the synergistic effects of temperature, pressure, and mold wear on quality risk can be quantitatively integrated to form a single index that comprehensively reflects the degree of quality risk - the quality risk index RI.

[0065] By using the relative deviation and the wear ratio to calculate the dimensionless difference, the quality risk index has cross-process and cross-product comparability, and can accurately quantify the risk level; relying on the weighted model to evaluate the influence of multiple parameters on quality risk, the process adjustment priority is clear, and the quality control efficiency is improved; and by quantifying the mold wear into the index, the mold maintenance or replacement can be planned in advance according to the change trend, the scrap rate and production cost are reduced, and the production continuity and economic benefits are ensured. The problems of subjectivity, non-quantification and blind area in risk analysis under the coupling effect of multiple factors in traditional qualitative assessment of quality risk by relying on manual experience are solved.

[0066] The defect feature acquisition module is used to collect product surface images and injection sound, and after preprocessing, the multi-source data recognition model is input to obtain the defect feature index.

[0067] The defect feature acquisition module includes: the preprocessing of the surface image removes noise by Gaussian filtering and extracts the defect area by threshold segmentation; the preprocessing of the injection sound converts the time domain signal into a frequency domain feature spectrum by short-time Fourier transform; the multi-source data recognition model is a fusion model of convolutional neural network and recurrent neural network, which receives the image feature matrix and sound feature spectrum at the input layer, and outputs the defect type and severity score at the output layer.

[0068] Specifically, first, data collection, reasonable arrangement of equipment on the injection molding production line, use of high-definition industrial cameras to shoot product surface images from multiple angles to capture appearance defects such as scratches and pores; at the same time, acoustic sensors are installed at key parts of the injection molding machine to collect sound data, because the injection molding states and defect conditions have corresponding characteristics in the sound, which provides a basis for subsequent analysis.

[0069] Next comes surface image preprocessing. First, Gaussian filtering is used for denoising. This method calculates the weighted average of pixels and their neighborhood based on a Gaussian function. By setting appropriate Gaussian kernel size and standard deviation parameters, it suppresses salt-and-pepper and Gaussian noise, making the image smoother and more responsive to realistic features. Second, threshold segmentation is used to extract defect areas. Based on the grayscale values ​​of image pixels, experience or analysis is used to determine the threshold range, classify pixels, and accurately segment the defect areas for subsequent analysis of their shape, size, and other characteristics.

[0070] For injection molding audio preprocessing, short-time Fourier transform is used to divide the long audio signal into multiple short time segments. After determining the range of each segment with the help of window function and window length parameters, Fourier transform is performed separately to convert the time domain signal into a frequency domain feature map, which clearly shows the changes of different frequency components over time, making it easier to extract audio features related to defects.

[0071] Finally, a multi-source data recognition model is employed. Its input layer receives preprocessed image feature matrices and audio feature maps, providing multi-source data support for defect identification. The convolutional neural network in this model is responsible for deeply mining image features, extracting multi-layer features such as edges and textures. The recurrent neural network processes the temporal information of the audio feature maps, exploring temporal and long-term dependencies. These two layers are integrated for collaborative analysis. After internal computational learning, the output layer outputs defect type and severity scores, providing intuitive and valuable reference for quality assessment and process improvement.

[0072] The comprehensive use of multi-source data improves the accuracy of defect detection, with different types of data complementing and verifying each other, reducing misjudgments and missed detections. Preprocessing image and audio data optimizes data quality, providing high-quality input for the model and helping to more accurately identify defect characteristics. A fusion model enables efficient collaborative analysis of image and audio features, automatically learning the mapping relationship between data and defects. This improves detection efficiency while outputting valuable defect type and severity scores, providing strong support for quality control in injection molding production. This addresses the limitations of traditional single data sources in defect detection and the difficulty in extracting features from audio time-domain signals.

[0073] The reinforcement learning control module is used to design the state space, action space and reward function of the DQN algorithm based on the original data, short-term process parameter fluctuation range, quality risk index and defect characteristic index, train and update the online Q network and target Q network and minimize the loss function.

[0074] The reinforcement learning control module includes: the state space is defined as S = {X,F,RI,FI,T pred ,P pred}, where X is the original data feature vector, F is the defect feature index, RI is the quality risk index, FI is the process parameter fluctuation range, Tpred To predict temperature fluctuation, P pred To predict pressure fluctuation; action space is defined as A = { AT, AP, At} where AT is temperature adjustment, AP is pressure adjustment, At is cooling time adjustment; reward function is designed as R = -RI·w r -FI·w f +δ·δ good where w r , w f are weight coefficients, δ is process stability index, δ good is product qualification factor.

[0075] The reinforcement learning control module further comprises: the loss function is where θ is online Q network parameter, θ - is target Q network parameter, γ is discount factor, and the parameter is updated by stochastic gradient descent method where α is learning rate.

[0076] Specifically, firstly, data integration and state space definition. Data from multiple modules are collected, including feature vectors X formed by extracting original data, covering injection molding parameter and raw material, mold information; short-term process parameter fluctuation range FI given by the process parameter prediction module, which can predict parameter change trend; quality risk index RI comprehensively reflects the quality risk caused by temperature, pressure deviation and mold wear; defect feature index F reflects the product defect condition; there are also predicted temperature fluctuation T pred and predicted pressure fluctuation P pred in FI. Based on this, the state space S = {X, F, RI, FI, T pred , P pred} of DQN algorithm is defined, which provides complete state description for decision making.

[0077] Then, the action space is determined. The action space A = { AT, AP, At} is determined in combination with the adjustable link of injection molding process. Wherein, the temperature adjustment amount AT determines the adjustment range according to the characteristics of injection molding material and the characteristics of mold, to ensure process optimization and product quality; the pressure adjustment amount AP is related to the flow effect of melt in the mold, which needs to be adjusted reasonably according to factors such as cavity structure and product wall thickness; the cooling time adjustment amount At is set according to product size and material thermal properties, which is of great significance to ensure product accuracy and reduce internal stress, so as to limit the specific operation direction of process optimization.

[0078] Then, the reward function is designed. The action effect is evaluated by reward function R = -RI·w r -FI·w f +δ·δ good to guide optimization. The weight coefficients w r , wf Respectively weigh the importance of RI, FI in the reward evaluation, determine comprehensively according to product quality requirements, production stability requirements, etc. For example, precise products will increase w r Control by heavy quality risk. The process stability index δ is derived from the process parameter fluctuation, equipment operation stability, etc. The product qualified factor δ is determined good Depending on whether the product meets the standard, the system selects the optimal action strategy according to the feedback, and realizes the process and quality optimization goal.

[0079] Finally, model training and parameter updating. Based on the above DQN algorithm framework, including online Q network and target Q network. During training, the agent selects action a according to the current state s, executes, gets new state s' and reward r, stores (s, a, r, s') for training. Use the loss function Measure the difference between the two networks, where the discount factor γ balances current and future rewards, and the value is between 0 and 1, affecting the action strategy selection tendency. Through the stochastic gradient descent method, use learning rate α to update the online Q network parameters θ according to Update the online Q network parameters θ, and reasonably set α to ensure convergence. After repeated training, the model converges, and the optimal process parameter adjustment strategy is output.

[0080] Through multi-factor collaborative optimization control, according to the comprehensive influence of each factor on temperature, pressure, cooling time and other process parameters, the precision and comprehensiveness of process control are improved, and the product quality is stable; The intelligent and automated level of process adjustment is improved, relying on reinforcement learning to automatically make decisions, outputting the optimal adjustment scheme according to real-time state feedback, improving adjustment efficiency and reducing human error; The adaptability and generalization ability of the model to actual production are enhanced, so that it can cope with various working condition changes, output appropriate suggestions, reduce adjustment cost, improve the robustness and flexibility of the injection molding production system. The problem of lack of systematicness in multi-factor comprehensive decision-making and process adjustment in injection molding production is solved.

[0081] Process execution adjustment module, for adjusting the temperature controller, pressure regulating valve and cooling system of the injection molding machine according to the optimal DQN control model to complete real-time optimization control of process parameters.

[0082] The process execution adjustment module includes: the adjustment accuracy of the temperature controller is ±0.5℃, the response time of the pressure regulating valve is ≤0.05 seconds, and the flow regulation range of the cooling system is 30%-100% of the rated flow; During adjustment, according to the action parameters output by the optimal DQN control model, execute in the following order: first adjust the temperature controller, then adjust the pressure regulating valve after an interval of 15-30 seconds, and then adjust the cooling system after an interval of 10-20 seconds.

[0083] Specifically, first, the action parameters are obtained from the optimal DQN control model. This model integrates multiple information and, after training and updating, can output accurate action parameters based on real-time production status, providing decision-making basis for subsequent adjustments.

[0084] In terms of temperature controller, its precision reaches ±0.5℃, and based on high-precision temperature sensors, it monitors the temperature of key parts of the injection molding machine and feeds back. After receiving the temperature adjustment parameter of the model, it uses PID control algorithm to calculate the adjustment through proportional, integral, and differential links, changes the heating or cooling power, and accurately controls the temperature within the set precision, meeting the requirements of different materials and products for temperature.

[0085] The pressure regulating valve has a response time of ≤0.05 seconds. When receiving the pressure adjustment parameter output by the model, it relies on electromagnetic or hydraulic drive devices to quickly change the opening of the valve core and adjust the flow of hydraulic oil, thereby accurately adjusting the melt pressure to ensure good filling in the mold cavity and avoid defects such as material shortage and flash.

[0086] The flow regulating range of the cooling system is 30%-100% of the rated flow. According to the cooling-related parameters given by the model, the opening of the flow regulating valve is adjusted to change the flow of the cooling medium, achieving cooling speed control. If the cooling speed needs to be increased, the flow is adjusted larger, and vice versa, to ensure the dimensional accuracy and structural stability of the product.

[0087] In terms of adjustment sequence, the temperature controller is adjusted first, followed by the pressure regulating valve after an interval of 15-30 seconds, and then the cooling system after another interval of 10-20 seconds. This is because temperature affects the plasticization of materials and the flowability of the melt, so temperature is adjusted first to make the material reach a suitable processing state. After the temperature stabilizes, the pressure is adjusted to avoid adverse coupling caused by simultaneous changes in temperature and pressure. After the pressure adjustment is complete and the melt filling is determined, the cooling system is adjusted to allow the product to cool in a suitable environment, reducing quality problems caused by improper adjustment sequence and ensuring product quality.

[0088] Through process parameter control, the precise adjustment capability of each component and strict operation according to the model instructions, the product quality is stabilized, and the diverse production demands are met; the scientificity and rationality of process adjustment are ensured, the quality problems are avoided by sequential adjustment, the overall quality is improved, the equipment life is prolonged, and the maintenance cost is reduced; the real-time and intelligent level of process control is improved, the system adaptability and flexibility are enhanced, the production efficiency is improved, and stable high-quality products are ensured. The problems of insufficient adjustment precision, unreasonable adjustment sequence, and difficulty in realizing real-time optimal control in traditional injection molding process are solved.

[0089] First is the data acquisition module, which is responsible for collecting a variety of raw data, including process parameters, mold state, raw material information, and equipment operating conditions. These data are then stored in a database with precise timestamps, providing a foundation for subsequent modules.

[0090] The process parameter prediction module reads raw data from the database and constructs a time series dataset that reflects the changes in process parameters over time. This data is then input into a TFT model, which has strong data analysis capabilities and can uncover the changing patterns of process parameters over long and short periods. By combining this information with real-time production conditions, such as the properties of the raw materials being used and the specific specifications of the products being produced, the model can predict the range of fluctuations that process parameters may experience in the near future, such as changes in temperature, pressure, and cooling time. This allows for early preparation for process adjustments.

[0091] The quality risk index acquisition module obtains data related to process parameters and molds, taking into account real-time temperature, pressure, and mold wear conditions. Through specific quantitative methods, the module analyzes key factors that affect quality and derives a quality risk index, which visually presents the level of quality risk in the current production process and provides a clear reference for controlling product quality.

[0092] The defect feature acquisition module uses high-definition industrial cameras installed on the production line to collect product surface images and high-precision acoustic sensors to collect audio data during the injection molding process. For the collected surface images, appropriate filtering algorithms are used to remove noise, and segmentation methods are used to extract defect regions and organize image features. For audio data, specific transformation methods are used to convert it from the time domain to the frequency domain feature spectrum for better analysis of the information contained. The processed image features and frequency domain feature spectrum are then input into a specialized multi-source data recognition model, which outputs the defect type and corresponding severity score of the product, helping to accurately assess product quality.

[0093] The reinforcement learning control module collects various types of key data output by the previous modules, and defines the state space, action space and reward function related to the DQN algorithm on this basis, and builds an algorithm framework containing online Q network and target Q network. In this framework, the agent selects the corresponding action according to the current state and executes it, then observes the new state after executing the action and the reward obtained, and through a specific calculation and parameter updating mechanism, the network parameters are repeatedly trained and optimized, so that the model gradually converges, and finally the optimal process parameter adjustment strategy suitable for different production states can be output.

[0094] The process execution adjustment module adjusts the temperature controller, pressure regulating valve and cooling system of the injection molding machine in order according to the optimal strategy obtained by the reinforcement learning control module. The temperature controller realizes accurate temperature adjustment by means of high-precision temperature monitoring and corresponding control algorithm; the pressure regulating valve quickly responds and accurately adjusts the pressure by relying on the internal drive device; the cooling system adjusts the flow of cooling medium by changing the opening of the flow regulating valve, and then controls the cooling speed. And adjust each component in a specific order to ensure that the process parameters are optimized and controlled in real time, and ensure stable product quality.

[0095] Through the system, comprehensive production optimization driven by data is realized, the precision and adaptability of process control are improved, the scientificity and reliability of quality control are enhanced, the intelligent level of production process is improved, the production efficiency is effectively improved, the scrap rate is reduced, the product quality is stabilized, and the competitiveness in the field of automobile injection molding production is enhanced. The problem of lack of system intelligence in process control is solved.

[0096] Please refer to the attached Figure 2 A control method of an automobile injection molding production process control system, for an automobile injection molding production process control system, comprising the following steps:

[0097] S1, the data acquisition module acquires original data and stores them into the database with time stamp;

[0098] S2, the process parameter prediction module reads the data to construct time series data set, and predicts the short-term process parameter fluctuation range through TFT model;

[0099] S3, the quality risk index acquisition module calculates the quality risk index according to the process parameter data and the mold data;

[0100] S4, the defect feature acquisition module collects surface images and audio, and inputs them into the multi-source data recognition model after preprocessing to obtain the defect feature index;

[0101] S5, the reinforcement learning control module trains the DQN algorithm based on the original data, the fluctuation range, the quality risk index and the defect feature index to obtain the optimal control model;

[0102] S6, the process execution adjustment module adjusts the temperature, pressure and cooling parameters according to the optimal control model to complete real-time optimal control.

[0103] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely divergences of the principles and spirit of the application and that numerous modifications, changes, substitutions, and alterations can be made thereto without departing from the spirit and scope of the application as defined by the appended claims and their equivalents.

Claims

1. A control system for the production process of automobile injection molding parts, characterized in that: Includes the following modules: The data acquisition module is used to collect the original data of process parameters, molds, raw materials and equipment operation, and store them in the database with a time stamp; The process parameter prediction module is used to read the original data to construct a time series data set, input the TFT model to obtain a pre-trained model, and combine the current production condition parameters to predict the short-term process parameter fluctuation range; A quality risk index acquisition module is used to input process parameter data and mold data into a quality risk index calculation formula to obtain a quality risk index; Defect feature acquisition module, used to collect product surface images and injection molding sound frequency, and input them into the multi-source data recognition model after pre-processing to obtain the defect feature index; The reinforcement learning control module is used to design the state space, action space, and reward function of the DQN algorithm based on the raw data, short-term process parameter fluctuation range, quality risk index, and defect characteristic index, train and update the online Q network and target Q network, and minimize the loss function; The process execution adjustment module is used to adjust the temperature controller, pressure regulating valve and cooling system of the injection molding machine according to the optimal DQN control model to achieve real-time optimization control of process parameters.

2. The automobile injection molding production process control system according to claim 1, characterized in that: The data acquisition module includes: Temperature sensor arrays are placed on the barrel, screw, and mold cavity of the injection molding machine to collect melt temperature and mold temperature data; The pressure sensor group is installed on the mold cavity surface and the hydraulic system of the injection molding machine to collect melt filling pressure and injection pressure data; Displacement sensor, used to collect screw displacement data to determine injection volume; The data interface unit is used to access raw material batch parameters, product design drawing data and production equipment operating status data.

3. The automobile injection molding production process control system according to claim 1, characterized in that: The process parameter prediction module includes: when constructing a time series data set, sliding window processing is performed on the original data, and the window size is 24 hours of production data; the TFT model is a time-frequency transformer model, whose input layer receives normalized time series data, and the output layer predicts the process parameter fluctuation range in the next 1-4 hours, including temperature fluctuation thresholds, pressure fluctuation thresholds, and cooling time fluctuation thresholds.

4. The automobile injection molding production process control system according to claim 1, characterized in that: The quality risk index acquisition module includes: the quality risk index calculation formula is: Where RI is the quality risk index, T is the real-time temperature, T0 is the standard temperature, P is the real-time pressure, P0 is the standard pressure, D is the mold wear, D max is the maximum allowable wear of the mold, w1, w2, w3 are weight coefficients, and w1+w2+w3=1.

5. The automobile injection molding production process control system according to claim 1, characterized in that: The defect feature acquisition module includes: preprocessing the surface image by removing noise through Gaussian filtering and extracting the defect area using threshold segmentation; preprocessing the injection molding sound frequency by converting the time domain signal into a frequency domain feature spectrum through short-time Fourier transform; the multi-source data recognition model is a fusion model of convolutional neural network and recurrent neural network, whose input layer receives the image feature matrix and the audio feature spectrum, and the output layer outputs the defect type and severity score.

6. The automobile injection molding production process control system according to claim 1, characterized in that: The reinforcement learning control module includes: the state space is defined as S={X,F,RI,FI,T pred ,P pred }, where X is the original data feature vector, F is the defect feature index, RI is the quality risk index, FI is the process parameter fluctuation range, T pred To predict temperature fluctuations, P pred To predict pressure fluctuations; the action space is defined as A = {ΔT, ΔP, Δt}, where ΔT is the temperature adjustment, ΔP is the pressure adjustment, and Δt is the cooling time adjustment; the reward function is designed as R = -RI·w r -FI·w f +δ·δ good Among them, w r 、w f is the weight coefficient, δ is the process stability index, δ good It is the factor for determining product qualification.

7. The automobile injection molding production process control system according to claim 1, characterized in that: The reinforcement learning control module also includes: the loss function is Among them, θ is the online Q network parameter, θ - is the target Q network parameter, γ is the discount factor, and the parameters are updated by stochastic gradient descent method. Where α is the learning rate.

8. The automobile injection molding production process control system according to claim 1, characterized in that: The process execution adjustment module includes: the adjustment accuracy of the temperature controller is ±0.5°C, the response time of the pressure regulating valve is ≤0.05 seconds, and the flow adjustment range of the cooling system is 30%-100% of the rated flow. During adjustment, according to the action parameters output by the optimal DQN control model, the following sequence is executed: first adjust the temperature controller, then adjust the pressure regulating valve after an interval of 15-30 seconds, and then adjust the cooling system after an interval of 10-20 seconds.

9. A control method for a production process control system of automobile injection molding parts, characterized in that: A control system for the production process of automobile injection molded parts according to any one of claims 1 to 8, comprising the following steps: S1, the data acquisition module collects the original data and stores it in the database with a timestamp; S2, the process parameter prediction module reads data to construct a time series data set and predicts the short-term process parameter fluctuation range through the TFT model; S3, the quality risk index acquisition module calculates the quality risk index based on the process parameter data and mold data; S4, the defect feature acquisition module collects surface images and audio, and after pre-processing, inputs them into the multi-source data recognition model to obtain the defect feature index; S5, the reinforcement learning control module trains the DQN algorithm based on the original data, fluctuation range, quality risk index and defect characteristic index to obtain the optimal control model; S6. The process execution adjustment module adjusts the temperature, pressure and cooling parameters according to the optimal control model to complete real-time optimization control.

Images