Circulation method, system, equipment and storage medium for cooling water path of mould of ultra-vacuum die-casting machine

Abstract

The invention discloses an ultra-vacuum die-casting machine die cooling water channel circulation method, system, equipment and storage medium, and relates to the field of vacuum die-casting, and the method comprises the steps: building a dynamic correlation neural network through the correlation modeling of technological parameters and quality performance and the collection and processing of parameters, optimizing the weight, and generating an original feature vector; fusing features to predict defect positions, and analyzing a formation mechanism; optimizing process parameters and a scheme according to a result; the temperature of the mold is accurately controlled through the Internet of Things and a fuzzy-feedback control algorithm; and finally, collaborative optimization and loop iteration are carried out by using an incremental learning algorithm and taking energy consumption and temperature deviation as targets. Compared with an existing technology depending on experience and small-batch trial and error, the method has the advantages that defects can be accurately predicted, the process can be efficiently optimized, the mold temperature can be accurately controlled, continuous optimization iteration capacity is achieved, product quality and production efficiency can be improved, energy consumption can be reduced, and intelligent and green development of the die-casting industry can be promoted.

Description

Technical Field

[0001] The present invention relates to the field of vacuum die-casting, and in particular to a cooling water circulation method, system, equipment and storage medium for a mold of an ultra-vacuum die-casting machine. Background Art

[0002] In modern manufacturing, die-casting, with its advantages of high efficiency and precision, has become a key technology for producing various parts. This is particularly true in the manufacturing of new energy vehicle parts, 5G communications, and laptop computer components, where it plays a crucial role in achieving the goal of "lightweighting" products. Demand for large, precision, thin-walled magnesium alloy die-castings continues to grow. However, these die-castings are prone to defects such as air entrapment and shrinkage during production, seriously impacting product quality and performance.

[0003] Currently, addressing these defects primarily relies on process optimization based on empirical knowledge and reducing filling defects through small-batch trial and error. However, this traditional approach has significant drawbacks. On the one hand, the small-batch trial and error production process is lengthy, involving multiple mold adjustments, parameter changes, and product trials, consuming significant time, manpower, and material resources. On the other hand, defect prediction lacks a high degree of identifiability, making it difficult to accurately pinpoint the cause and location of defects. This results in a lack of targeted process improvements, an inability to fundamentally address the problem, and a significant limitation on production efficiency and product quality.

[0004] During the die-casting process, the mold cooling water circulation system plays a crucial role in product quality. Improper cooling water circulation can lead to uneven mold temperature distribution, which in turn can cause defects such as deformation and cracking in die-cast parts. Traditional cooling water circulation methods are mostly fixed and cannot be dynamically adjusted based on actual production process parameters and product quality performance, making them difficult to meet the complex and ever-changing needs of die-casting production.

[0005] With the rapid development of science and technology, the new energy vehicle and key component industries are moving towards large-scale, high-end, intensive, and green development. This trend places higher demands on die-casting processes, requiring not only improved production efficiency and reduced costs, but also the stability and consistency of product quality. Against this backdrop, the development of an advanced cooling water circulation method for ultra-vacuum die-casting machines is urgent. Summary of the Invention

[0006] The purpose of the present invention is to provide a method, system, equipment, and storage medium for the cooling water circulation of an ultra-vacuum die-casting machine mold. By collecting and processing process and quality performance parameters to construct a neural network to obtain the original feature vector, the features are then integrated to predict defects and analyze the mechanism. The process parameters and solutions are then optimized accordingly. The Internet of Things and fuzzy-feedback control algorithms are then used to accurately control the mold temperature. Finally, with the help of an incremental learning algorithm, the solution is collaboratively optimized with energy consumption and temperature deviation as the target and iterated cyclically to solve the above-mentioned problems.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A cooling water circulation method for an ultra-vacuum die-casting machine mold comprises the following steps:

[0009] S1: Modeling the correlation between process parameters and quality performance, collecting and preprocessing process and quality performance parameters, building a dynamic correlation neural network, introducing a dynamic weight adjustment mechanism to optimize the weights, characterizing the complex nonlinear relationship between the two, and generating highly correlated original feature vectors;

[0010] S2: Defect prediction and formation mechanism analysis: Using feature fusion extraction methods, weighted fusion of original feature vectors is performed, and new feature vectors are obtained after optimizing weights. Defect locations are predicted based on probabilistic graphical models, the formation mechanism is analyzed, and the direction of process improvement is identified.

[0011] S3: Process parameter optimization and solution formulation: Determine the optimized process parameters based on the defect prediction results, use the adaptive particle swarm optimization algorithm to find the optimal combination, combine virtual simulation with error correction, compare simulation and actual data to adjust process parameters, and optimize the process solution;

[0012] S4: Precise and coordinated control of mold temperature. An IoT control system is built based on the optimized process plan. A fuzzy-feedback control algorithm is used to determine the control variable using temperature deviation and deviation change rate as inputs. The mold temperature is then adjusted in combination with feedback control.

[0013] S5: Iterative optimization of the scheme, using incremental learning algorithm to update model parameters based on new samples, collaboratively optimize the process scheme, cooling water flow and temperature, with energy consumption and temperature deviation as the objective function, and feed the optimization scheme back to the first step of the loop iteration.

[0014] Step S1: Deep correlation modeling of process parameters and quality performance. Specifically, process parameters and quality performance indicators are collected and preprocessed, dynamic correlation neural network is constructed, dynamic weight adjustment mechanism is introduced, and the weight matrix from input layer to hidden layer is W ij (t), its update formula is:

[0015]

[0016] Among them, α is the learning rate, β is the momentum factor, and E is the loss function;

[0017] By continuously optimizing the weights through the formula, the nonlinear relationship between process parameters and quality performance indicators is modeled, and the model output is Arrange the results with high correlation first to form the original feature vector, which is recorded as F1, F2, ..., F k .

[0018] Step S2 defect prediction and formation mechanism analysis is specifically to use the original feature vectors F1, F2, ..., F obtained in the previous step k ,Using the feature fusion extraction method to process the data,construct a model based on probability graph to predict the defect location and analyze the formation mechanism;

[0019] In the feature fusion stage, a weighted fusion strategy is adopted to combine the original feature vectors F1, F2, ..., F k Perform feature fusion and assign specific weight ω i , to strengthen the key features and weaken the redundant features, weight ω i Optimization is performed through the scientific method of cross-validation, which divides the data set into multiple subsets and performs model training and validation on different subsets to ensure the rationality and effectiveness of the weights; the fused feature vector F fusion The calculation formula is:

[0020]

[0021] Among them, ω i is the weight of each eigenvector, obtained through cross-validation optimization, based on the probabilistic graphical model;

[0022] The probability distribution function of the defect position D is defined as:

[0023]

[0024] Where E is the edge set of the probability graph, ψ uv It is the potential function between nodes u and v, which is used to realize the probability prediction of defect location and formation mechanism analysis. By calculating and analyzing the probability distribution function, the location where defects may occur in the product is predicted, providing a clear direction for subsequent process improvements.

[0025] Step S3: Process parameter optimization and solution formulation. Specifically, the defect prediction results indicate the location and type of defects that may occur in the product under the current process parameters, which sets the direction for process parameter optimization.

[0026] With the goal of reducing or eliminating defects, the process parameters that need to be optimized are determined. These parameters are used as the particle dimensions in the particle swarm optimization algorithm. The adaptive particle swarm optimization algorithm is used to optimize the process parameters. The process plan is optimized by combining virtual simulation and error correction. In the adaptive particle swarm optimization algorithm, the particle velocity update improvement formula is:

[0027]

[0028] Among them, (ω(t)) is the inertia weight that changes dynamically with the number of iterations, (c1) and (c2) are learning factors, (r 1id )、(r2id ) is a random number, (p id ) is the optimal position of the individual particle, (g d ) is the global optimal position, (γ) is the gradient adjustment coefficient, is the gradient of the objective function at the current position;

[0029] This formula is used to guide particles to search for the optimal process parameter combination more efficiently, and the error correction model E of virtual simulation is combined correction =θ·(Y simulated -Y actual ), optimize the process plan, and compare the virtual simulation results Y simulated and actual production data Y actual , calculate the error and make corrections; according to the error feedback, adjust the parameters in the process plan.

[0030] Step S4: precise coordinated control of mold temperature is specifically to build an Internet of Things-based control system based on the optimized process plan and use a fuzzy-feedback control algorithm to achieve precise control of mold temperature;

[0031] The fuzzy control rule adopts a multi-input single-output structure, assuming that the temperature deviation (e) and the deviation change rate is the input, the control quantity (u) is the output, and the fuzzy control rule is obtained through fuzzy reasoning. Its core formula is:

[0032]

[0033] Among them, μ i is the membership degree of the i-th fuzzy rule, u i is the output value of the corresponding rule. Combined with feedback control, the temperature control model is:

[0034]

[0035] Among them, T next and T current are the mold temperatures at the next moment and the current moment, respectively, k p 、k i 、k d The proportional, integral and differential coefficients are used to achieve precise coordinated control of the mold temperature.

[0036] Step S5, iterative optimization of the scheme, specifically involves iteratively optimizing the process scheme and the cooling water circulation method after the optimization in the above steps using an incremental learning algorithm; collaboratively optimizing the process scheme and the flow rate and temperature of the cooling water circulation; then feeding the optimized scheme back to the first step to further adjust the model, and repeating the iterative cycle;

[0037] By continuously updating the model parameters according to new samples, optimizing the process plan, and co-optimizing the flow rate Q and temperature T of the cooling water circulation, the optimization objective function is:

[0038] J=λ1·EnergyConsumption(Q)+λ2·TemperatureDeviation(T)

[0039] Among them, λ1 and λ2 are weight coefficients, EnergyConsumption(Q) is the energy consumption corresponding to the flow rate Q, and TemperatureDeviation(T) is the deviation between the temperature T and the target temperature, which realizes the iterative optimization of the cooling water circulation method;

[0040] By adjusting the flow rate Q and temperature T of the cooling water circulation, the objective function value is minimized, and the coordinated optimization of the process plan and the cooling water circulation system is achieved. The optimized plan is fed back to the first step, and the model parameters are readjusted to start a new round of iterative optimization.

[0041] The incremental learning adopts the online gradient descent algorithm, and the update formula of the loss function L is:

[0042]

[0043] Among them, L old is the old loss function value, N is the number of new samples, y i is the true value, is the predicted value.

[0044] A method for circulating cooling water circuits for ultra-vacuum die-casting machine molds is characterized by comprising a process-related modeling module, a defect prediction and analysis module, a parameter optimization and decision module, a temperature control execution module, and a scheme iteration and optimization module. The process-related modeling module collects and processes process and quality performance parameters, constructs a dynamic association neural network to generate original feature vectors; the defect prediction and analysis module fuses the original feature vectors, uses a probabilistic graphical model to predict the defect location and analyze the mechanism; the parameter optimization and decision module optimizes process parameters and schemes based on defect prediction results using an adaptive particle swarm optimization algorithm combined with virtual simulation and error correction; the temperature control execution module builds an Internet of Things control system based on the optimization scheme, and uses a fuzzy-feedback control algorithm to accurately control the mold temperature; the scheme iteration and optimization module utilizes an incremental learning algorithm to collaboratively optimize the process scheme, cooling water circuit flow, and temperature with energy consumption and temperature deviation as objective functions, and feeds back the optimization scheme for cyclic iteration.

[0045] A terminal device, characterized in that it includes a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the method according to any one of claims 1 to 7 is implemented.

[0046] A computer-readable storage medium, characterized in that a computer program is stored in the storage medium, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

[0047] First, the process-related modeling module collects and preprocesses process and quality performance parameters. A dynamic correlation neural network is constructed, incorporating a dynamic weight adjustment mechanism to generate highly correlated raw feature vectors, characterizing the complex nonlinear relationship between the two. Next, the defect prediction and analysis module applies a feature fusion extraction method to weightedly fuse the raw feature vectors. Using a probabilistic graphical model, it predicts defect locations and analyzes formation mechanisms, identifying areas for process improvement. Based on the defect prediction results, the parameter optimization and decision-making module employs an adaptive particle swarm optimization algorithm combined with virtual simulation and error correction to determine and optimize process parameters and formulate a process plan. The temperature control execution module then builds an IoT control system based on the optimized process plan. Using a fuzzy-feedback control algorithm, it uses temperature deviation and deviation change rate as inputs to determine the control variable, combining feedback control to precisely control the mold temperature. Finally, the iterative solution optimization module utilizes an incremental learning algorithm, using energy consumption and temperature deviation as objective functions, to collaboratively optimize the process plan, cooling water flow rate, and temperature. The optimized solution is fed back to the process-related modeling module for iterative iteration, continuously improving the cooling water circulation efficiency of the die-casting machine mold.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] Accurate prediction and efficient process optimization: Existing technologies often rely on experience and small-batch trial and error to optimize processes and reduce defects, resulting in lengthy processes and limited identifiable defect prediction. This technology, however, models the correlation between process parameters and quality performance, employing a dynamic correlation neural network and weight adjustment mechanism to accurately characterize the complex nonlinear relationship between the two and generate raw feature vectors. It then uses a probabilistic graphical model to predict defect locations and analyze formation mechanisms, providing clear guidance for process improvement. Furthermore, it employs an adaptive particle swarm optimization algorithm, combined with virtual simulation and error correction to optimize process parameters and solutions, enabling more efficient and precise defect reduction or elimination, thereby improving product quality.

[0050] Achieve precise mold temperature control: Traditional cooling water circulation methods are mostly fixed modes, making it difficult to dynamically adjust according to actual conditions. This technology builds an IoT control system based on an optimized process plan. Utilizing a fuzzy-feedback control algorithm, it uses temperature deviation and deviation change rate as inputs to determine the control variable. Combined with feedback control, this achieves precise and coordinated control of mold temperature, effectively preventing product defects such as deformation and cracking caused by uneven mold temperature, and ensuring stable and consistent product quality.

[0051] Continuous optimization and iteration capabilities: This technology utilizes an incremental learning algorithm to update model parameters based on new samples. Using energy consumption and temperature deviation as objective functions, it collaboratively optimizes the process plan, cooling water flow, and temperature, and iterates the optimized solution through feedback loops. This enables the technology to continuously adapt to new production needs and changes, continuously improving die-casting production efficiency and reducing energy consumption. Compared to existing technologies, it is more flexible and sustainable, effectively driving the die-casting industry towards intelligent and green development and enhancing the industry's overall competitiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a flow chart of a cooling water circulation method for an ultra-vacuum die-casting machine mold according to the present invention. Specific implementation methods

[0053] The technical solutions in the embodiments of the present invention will be fully described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0054] like Figure 1 As shown, the following steps S1 are included: modeling the correlation between process parameters and quality performance, collecting and preprocessing process and quality performance parameters, building a dynamic correlation neural network, introducing a dynamic weight adjustment mechanism to optimize the weight, characterizing the complex nonlinear relationship between the two, and generating a highly correlated original feature vector;

[0055] S2: Defect prediction and formation mechanism analysis: Using feature fusion extraction methods, weighted fusion of original feature vectors is performed, and new feature vectors are obtained after optimizing weights. Defect locations are predicted based on probabilistic graphical models, the formation mechanism is analyzed, and the direction of process improvement is identified.

[0056] S3: Process parameter optimization and solution formulation: Determine the optimized process parameters based on the defect prediction results, use the adaptive particle swarm optimization algorithm to find the optimal combination, combine virtual simulation with error correction, compare simulation and actual data to adjust process parameters, and optimize the process solution;

[0057] S4: Precise and coordinated control of mold temperature. An IoT control system is built based on the optimized process plan. A fuzzy-feedback control algorithm is used to determine the control variable using temperature deviation and deviation change rate as inputs. The mold temperature is then adjusted in combination with feedback control.

[0058] S5: Iterative optimization of the scheme, using incremental learning algorithm to update model parameters based on new samples, collaboratively optimize the process scheme, cooling water flow and temperature, with energy consumption and temperature deviation as the objective function, and feed the optimization scheme back to the first step of the loop iteration.

[0059] Step S1: Deep correlation modeling of process parameters and quality performance. Specifically, process parameters and quality performance indicators are collected and preprocessed, dynamic correlation neural network is constructed, dynamic weight adjustment mechanism is introduced, and the weight matrix from input layer to hidden layer is W ij (t), its update formula is:

[0060]

[0061] Among them, α is the learning rate, β is the momentum factor, and E is the loss function;

[0062] By continuously optimizing the weights through the formula, the nonlinear relationship between process parameters and quality performance indicators is modeled, and the model output is Arrange the results with high correlation first to form the original feature vector, which is recorded as F1, F2, ..., F k .

[0063] Step S2 defect prediction and formation mechanism analysis is specifically to use the original feature vectors F1, F2, ..., F obtained in the previous step k ,Using the feature fusion extraction method to process the data,construct a model based on probability graph to predict the defect location and analyze the formation mechanism;

[0064] In the feature fusion stage, a weighted fusion strategy is adopted to combine the original feature vectors F1, F2, ..., F k Perform feature fusion and assign specific weight ω i , to strengthen the key features and weaken the redundant features, weight ω i Optimization is performed through the scientific method of cross-validation, which divides the data set into multiple subsets and performs model training and validation on different subsets to ensure the rationality and effectiveness of the weights; the fused feature vector F fusion The calculation formula is:

[0065]

[0066] Among them, ω i is the weight of each eigenvector, obtained through cross-validation optimization, based on the probabilistic graphical model;

[0067] The probability distribution function of the defect position D is defined as:

[0068]

[0069] Where E is the edge set of the probability graph, ψ uv It is the potential function between nodes u and v, which is used to realize the probability prediction of defect location and formation mechanism analysis. By calculating and analyzing the probability distribution function, the location where defects may occur in the product is predicted, providing a clear direction for subsequent process improvements.

[0070] Step S3: Process parameter optimization and solution formulation. Specifically, the defect prediction results indicate the location and type of defects that may occur in the product under the current process parameters, which sets the direction for process parameter optimization.

[0071] With the goal of reducing or eliminating defects, the process parameters that need to be optimized are determined. These parameters are used as the particle dimensions in the particle swarm optimization algorithm. The adaptive particle swarm optimization algorithm is used to optimize the process parameters. The process plan is optimized by combining virtual simulation and error correction. In the adaptive particle swarm optimization algorithm, the particle velocity update improvement formula is:

[0072]

[0073] Among them, (ω(t)) is the inertia weight that changes dynamically with the number of iterations, (c1) and (c2) are learning factors, (r 1id )、(r 2id ) is a random number, (p id ) is the optimal position of the individual particle, (g d ) is the global optimal position, (γ) is the gradient adjustment coefficient, is the gradient of the objective function at the current position;

[0074] This formula is used to guide particles to search for the optimal process parameter combination more efficiently, and the error correction model E of virtual simulation is combined correction =θ·(Y simulated -Y actual ), optimize the process plan, and compare the virtual simulation results Y simulated and actual production data Y actual , calculate the error and make corrections; according to the error feedback, adjust the parameters in the process plan.

[0075] Step S4: precise coordinated control of mold temperature is specifically to build an Internet of Things-based control system based on the optimized process plan and use a fuzzy-feedback control algorithm to achieve precise control of mold temperature;

[0076] The fuzzy control rule adopts a multi-input single-output structure, assuming that the temperature deviation (e) and the deviation change rate is the input, the control quantity (u) is the output, and the fuzzy control rule is obtained through fuzzy reasoning. Its core formula is:

[0077]

[0078] Among them, μ i is the membership degree of the i-th fuzzy rule, u i is the output value of the corresponding rule. Combined with feedback control, the temperature control model is:

[0079]

[0080] Among them, T next and T current are the mold temperatures at the next moment and the current moment, respectively, k p 、k i 、k d The proportional, integral and differential coefficients are used to achieve precise coordinated control of the mold temperature.

[0081] Step S5, iterative optimization of the scheme, specifically involves iteratively optimizing the process scheme and the cooling water circulation method after the optimization in the above steps using an incremental learning algorithm; collaboratively optimizing the process scheme and the flow rate and temperature of the cooling water circulation; then feeding the optimized scheme back to the first step to further adjust the model, and repeating the iterative cycle;

[0082] By continuously updating the model parameters according to new samples, optimizing the process plan, and co-optimizing the flow rate Q and temperature T of the cooling water circulation, the optimization objective function is:

[0083] J=λ1·EnergyConsumption(Q)+λ2·TemperatureDeviation(T)

[0084] Among them, λ1 and λ2 are weight coefficients, EnergyConsumption(Q) is the energy consumption corresponding to the flow rate Q, and TemperatureDeviation(T) is the deviation between the temperature T and the target temperature, which realizes the iterative optimization of the cooling water circulation method;

[0085] By adjusting the flow rate Q and temperature T of the cooling water circulation, the objective function value is minimized, and the coordinated optimization of the process plan and the cooling water circulation system is achieved. The optimized plan is fed back to the first step, and the model parameters are readjusted to start a new round of iterative optimization.

[0086] The incremental learning adopts the online gradient descent algorithm, and the update formula of the loss function L is:

[0087]

[0088] Among them, L old is the old loss function value, N is the number of new samples, y i is the true value, is the predicted value.

[0089] A method for circulating cooling water circuits for ultra-vacuum die-casting machine molds is characterized by comprising a process-related modeling module, a defect prediction and analysis module, a parameter optimization and decision module, a temperature control execution module, and a scheme iteration and optimization module. The process-related modeling module collects and processes process and quality performance parameters, constructs a dynamic association neural network to generate original feature vectors; the defect prediction and analysis module fuses the original feature vectors, uses a probabilistic graphical model to predict the defect location and analyze the mechanism; the parameter optimization and decision module optimizes process parameters and schemes based on defect prediction results using an adaptive particle swarm optimization algorithm combined with virtual simulation and error correction; the temperature control execution module builds an Internet of Things control system based on the optimization scheme, and uses a fuzzy-feedback control algorithm to accurately control the mold temperature; the scheme iteration and optimization module utilizes an incremental learning algorithm to collaboratively optimize the process scheme, cooling water circuit flow, and temperature with energy consumption and temperature deviation as objective functions, and feeds back the optimization scheme for cyclic iteration.

[0090] A terminal device, characterized in that it includes a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the method according to any one of claims 1 to 7 is implemented.

[0091] A computer-readable storage medium, characterized in that a computer program is stored in the storage medium, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

[0092] The specific implementation method is S1: modeling the association between process parameters and quality performance, collecting and preprocessing process and quality performance parameters, building a dynamic association neural network, introducing a dynamic weight adjustment mechanism to optimize the weight, characterizing the complex nonlinear relationship between the two, and generating a highly correlated original feature vector; specifically, collecting process parameters and quality performance indicators and preprocessing them, building a dynamic association neural network, introducing a dynamic weight adjustment mechanism, and the weight matrix from the input layer to the hidden layer is W ij (t), its update formula is:

[0093]

[0094] Among them, α is the learning rate, β is the momentum factor, and E is the loss function;

[0095] By continuously optimizing the weights through the formula, the nonlinear relationship between process parameters and quality performance indicators is modeled, and the model output is Arrange the results with high correlation first to form the original feature vector, which is recorded as F1, F2, ..., F kS2: Defect prediction and formation mechanism analysis, using feature fusion extraction method, weighted fusion of the original feature vector, optimize the weight to obtain a new feature vector, predict the defect location based on the probabilistic graph model, analyze the formation mechanism, and clarify the direction of process improvement; specifically, using the original feature vectors F1, F2, ..., F obtained in the previous step k ,Using the feature fusion extraction method to process the data,construct a model based on probability graph to predict the defect location and analyze the formation mechanism;

[0096] In the feature fusion stage, a weighted fusion strategy is adopted to combine the original feature vectors F1, F2, ..., F k Perform feature fusion and assign specific weight ω i , to strengthen the key features and weaken the redundant features, weight ω i Optimization is performed through the scientific method of cross-validation, which divides the data set into multiple subsets and performs model training and validation on different subsets to ensure the rationality and effectiveness of the weights; the fused feature vector F fusion The calculation formula is:

[0097]

[0098] Among them, ω i is the weight of each eigenvector, obtained through cross-validation optimization, based on the probabilistic graphical model;

[0099] The probability distribution function of the defect position D is defined as:

[0100]

[0101] Where E is the edge set of the probability graph, ψ uv It is the potential function between nodes u and v, which is used to realize the probability prediction of defect location and formation mechanism analysis. By calculating and analyzing the probability distribution function, the location where defects may occur in the product is predicted, providing a clear direction for subsequent process improvements.

[0102] S3: Process parameter optimization and solution formulation: Determine the optimized process parameters based on the defect prediction results. Use the adaptive particle swarm optimization algorithm to find the optimal combination. Combine virtual simulation with error correction, compare simulation and actual data to adjust process parameters and optimize the process solution. The defect prediction results indicate the location and type of defects that may occur in the product under the current process parameters, which sets the direction for process parameter optimization.

[0103] With the goal of reducing or eliminating defects, the process parameters that need to be optimized are determined. These parameters are used as the particle dimensions in the particle swarm optimization algorithm. The adaptive particle swarm optimization algorithm is used to optimize the process parameters. The process plan is optimized by combining virtual simulation and error correction. In the adaptive particle swarm optimization algorithm, the particle velocity update improvement formula is:

[0104]

[0105] Among them, (ω(t)) is the inertia weight that changes dynamically with the number of iterations, (c1) and (c2) are learning factors, (r 1id )、(r 2id ) is a random number, (p id ) is the optimal position of the individual particle, (g d ) is the global optimal position, (γ) is the gradient adjustment coefficient, is the gradient of the objective function at the current position;

[0106] This formula is used to guide particles to search for the optimal process parameter combination more efficiently, and the error correction model E of virtual simulation is combined correction =θ·(Y simulated -Y actual ), optimize the process plan, and compare the virtual simulation results Y simulated and actual production data Y actual , calculate the error and make corrections; according to the error feedback, adjust the parameters in the process plan.

[0107] S4: Precise and coordinated control of mold temperature. Based on the optimized process plan, an IoT control system is built. A fuzzy-feedback control algorithm is used to determine the control variable using temperature deviation and deviation change rate as inputs. The mold temperature is then adjusted using feedback control. Specifically, an IoT-based control system is built based on the optimized process plan, and a fuzzy-feedback control algorithm is used to achieve precise control of mold temperature.

[0108] The fuzzy control rule adopts a multi-input single-output structure, assuming that the temperature deviation (e) and the deviation change rate is the input, the control quantity (u) is the output, and the fuzzy control rule is obtained through fuzzy reasoning. Its core formula is:

[0109]

[0110] Among them, μ i is the membership degree of the i-th fuzzy rule, u i is the output value of the corresponding rule. Combined with feedback control, the temperature control model is:

[0111]

[0112] Among them, T next and T current are the mold temperatures at the next moment and the current moment, respectively, k p 、k i 、k dThe proportional, integral and differential coefficients are used to achieve precise coordinated control of the mold temperature.

[0113] S5: Iterative optimization of the scheme, using the incremental learning algorithm to update the model parameters based on the new samples, collaboratively optimize the process scheme, cooling water flow and temperature, with energy consumption and temperature deviation as the objective function, and feed the optimization scheme back to the first step of the loop iteration. Specifically, the incremental learning algorithm is used to iteratively optimize the process scheme and cooling water circulation method after the above steps; the process scheme and the flow and temperature of the cooling water circulation are collaboratively optimized, and then the optimized scheme is fed back to the first step to further adjust the model, and the cycle is repeated;

[0114] By continuously updating the model parameters according to new samples, optimizing the process plan, and co-optimizing the flow rate Q and temperature T of the cooling water circulation, the optimization objective function is:

[0115] J=λ1·EnergyConsumption(Q)+λ2·TemperatureDeviation(T)

[0116] Among them, λ1 and λ2 are weight coefficients, EnergyConsumption(Q) is the energy consumption corresponding to the flow rate Q, and TemperatureDeviation(T) is the deviation between the temperature T and the target temperature, which realizes the iterative optimization of the cooling water circulation method;

[0117] By adjusting the flow rate Q and temperature T of the cooling water circulation, the objective function value is minimized, and the coordinated optimization of the process plan and the cooling water circulation system is achieved. The optimized plan is fed back to the first step, and the model parameters are readjusted to start a new round of iterative optimization.

[0118] The incremental learning adopts the online gradient descent algorithm, and the update formula of the loss function L is:

[0119]

[0120] Among them, L old is the old loss function value, N is the number of new samples, y i is the true value, is the predicted value.

Claims

1. A cooling water circulation method for an ultra-vacuum die-casting machine mold, characterized in that: The following steps are involved: S1: Modeling the correlation between process parameters and quality performance, collecting and preprocessing process and quality performance parameters, building a dynamic correlation neural network, introducing a dynamic weight adjustment mechanism to optimize the weights, characterizing the complex nonlinear relationship between the two, and generating highly correlated original feature vectors; S2: Defect prediction and formation mechanism analysis: Using feature fusion extraction methods, weighted fusion of original feature vectors is performed, and new feature vectors are obtained after optimizing weights. Defect locations are predicted based on probabilistic graphical models, the formation mechanism is analyzed, and the direction of process improvement is identified. S3: Process parameter optimization and solution formulation: Determine the optimized process parameters based on the defect prediction results, use the adaptive particle swarm optimization algorithm to find the optimal combination, combine virtual simulation with error correction, compare simulation and actual data to adjust process parameters, and optimize the process solution; S4: Precise and coordinated control of mold temperature. An IoT control system is built based on the optimized process plan. A fuzzy-feedback control algorithm is used to determine the control variable using temperature deviation and deviation change rate as inputs. The mold temperature is then adjusted in combination with feedback control. S5: Iterative optimization of the scheme, using incremental learning algorithm to update model parameters based on new samples, collaboratively optimize the process scheme, cooling water flow and temperature, with energy consumption and temperature deviation as the objective function, and feed the optimization scheme back to the first step of the loop iteration.

2. The method for cooling water circuit of ultra-vacuum die-casting machine mold according to claim 1, characterized in that: Step S1: Deep correlation modeling of process parameters and quality performance. Specifically, process parameters and quality performance indicators are collected and preprocessed, dynamic correlation neural network is constructed, dynamic weight adjustment mechanism is introduced, and the weight matrix from input layer to hidden layer is W ij (t), its update formula is: Among them, α is the learning rate, β is the momentum factor, and E is the loss function; By continuously optimizing the weights through the formula, the nonlinear relationship between process parameters and quality performance indicators is modeled, and the model output is Arrange the results with high correlation first to form the original feature vector, which is recorded as F1, F2, ..., F k .

3. The method for cooling water circuit of ultra-vacuum die-casting machine mold according to claim 1, characterized in that: Step S2 defect prediction and formation mechanism analysis is specifically to use the original feature vectors F1, F2, ..., F obtained in the previous step k ,Using the feature fusion extraction method to process the data,construct a model based on probability graph to predict the defect location and analyze the formation mechanism; In the feature fusion stage, a weighted fusion strategy is adopted to combine the original feature vectors F1, F2, ..., F k Perform feature fusion and assign specific weight ω i , to strengthen the key features and weaken the redundant features, weight ω i Optimization is performed through the scientific method of cross-validation, which divides the data set into multiple subsets and performs model training and validation on different subsets to ensure the rationality and effectiveness of the weights; the fused feature vector F fusion The calculation formula is: Among them, ω i is the weight of each eigenvector, obtained through cross-validation optimization, based on the probabilistic graphical model; The probability distribution function of the defect position D is defined as: Where E is the edge set of the probability graph, ψ uv It is the potential function between nodes u and v, which is used to realize the probability prediction of defect location and formation mechanism analysis. By calculating and analyzing the probability distribution function, the location where defects may occur in the product is predicted, providing a clear direction for subsequent process improvements.

4. The method for cooling water circuit of ultra-vacuum die-casting machine mold according to claim 1, characterized in that: Step S3: Process parameter optimization and solution formulation. Specifically, the defect prediction results indicate the location and type of defects that may occur in the product under the current process parameters, which sets the direction for process parameter optimization. With the goal of reducing or eliminating defects, the process parameters that need to be optimized are determined. These parameters are used as the particle dimensions in the particle swarm optimization algorithm. The adaptive particle swarm optimization algorithm is used to optimize the process parameters. The process plan is optimized by combining virtual simulation and error correction. In the adaptive particle swarm optimization algorithm, the particle velocity update improvement formula is: Among them, (ω(t)) is the inertia weight that changes dynamically with the number of iterations, (c1) and (c2) are learning factors, (r 1id )、(r 2id ) is a random number, (p id ) is the optimal position of the individual particle, (g d ) is the global optimal position, (γ) is the gradient adjustment coefficient, is the gradient of the objective function at the current position; This formula is used to guide particles to search for the optimal process parameter combination more efficiently, and the error correction model E of virtual simulation is combined correction =θ·(Y simulated -Y actual ), optimize the process plan, and compare the virtual simulation results Y simulated and actual production data Y actual , calculate the error and make corrections; according to the error feedback, adjust the parameters in the process plan.

5. The method for cooling water circuit of ultra-vacuum die-casting machine mold according to claim 1, characterized in that: Step S4: precise coordinated control of mold temperature is specifically to build an Internet of Things-based control system based on the optimized process plan and use a fuzzy-feedback control algorithm to achieve precise control of mold temperature; The fuzzy control rule adopts a multi-input single-output structure, assuming that the temperature deviation (e) and the deviation change rate is the input, the control quantity (u) is the output, and the fuzzy control rule is obtained through fuzzy reasoning. Its core formula is: Among them, μ i is the membership degree of the i-th fuzzy rule, u i is the output value of the corresponding rule. Combined with feedback control, the temperature control model is: Among them, T next and T current are the mold temperatures at the next moment and the current moment, respectively, k p 、k i 、k d The proportional, integral and differential coefficients are used to achieve precise coordinated control of the mold temperature.

6. The method for cooling water circuit of ultra-vacuum die-casting machine mold according to claim 1, characterized in that: Step S5, iterative optimization of the scheme, specifically involves iteratively optimizing the process scheme and the cooling water circulation method after the optimization in the above steps using an incremental learning algorithm; Co-optimize the process plan and the flow and temperature of the cooling water circuit. Feed the optimized plan back to the first step to further adjust the model, and repeat the cycle. By continuously updating the model parameters according to new samples, optimizing the process plan, and co-optimizing the flow rate Q and temperature T of the cooling water circulation, the optimization objective function is: J=λ1·EnergyConsumption(Q)+λ2·TemperatureDeviation(T) Among them, λ1 and λ2 are weight coefficients, EnergyConsumption(Q) is the energy consumption corresponding to the flow rate Q, and TemperatureDeviation(T) is the deviation between the temperature T and the target temperature, which realizes the iterative optimization of the cooling water circulation method; By adjusting the flow rate Q and temperature T of the cooling water circulation, the objective function value is minimized, and the coordinated optimization of the process plan and the cooling water circulation system is achieved. The optimized plan is fed back to the first step, and the model parameters are readjusted to start a new round of iterative optimization.

7. The method for cooling water circuit of ultra-vacuum die-casting machine mold according to claim 6, characterized in that: The incremental learning adopts the online gradient descent algorithm, and the update formula of the loss function L is: Among them, L old is the old loss function value, N is the number of new samples, y i is the true value, is the predicted value.

8. A cooling water circulation method for ultra-vacuum die-casting machine mold, characterized in that: It includes a process-related modeling module, a defect prediction and analysis module, a parameter optimization and decision-making module, a temperature control execution module, and a solution iterative optimization module. The process-related modeling module collects and processes process and quality performance parameters, constructs a dynamic association neural network to generate original feature vectors; the defect prediction and analysis module integrates original feature vectors, uses a probabilistic graphical model to predict defect locations and analyze the mechanism; the parameter optimization and decision-making module optimizes process parameters and solutions based on defect prediction results using an adaptive particle swarm optimization algorithm combined with virtual simulation and error correction; the temperature control execution module builds an Internet of Things control system based on the optimization solution, and uses a fuzzy-feedback control algorithm to accurately control the mold temperature; The scheme iterative optimization module uses an incremental learning algorithm to collaboratively optimize the process scheme, cooling water flow and temperature with energy consumption and temperature deviation as objective functions, and feeds back the optimized scheme for cyclic iteration.

9. A terminal device, characterized in that: The method comprises a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images