A heavy-duty die steel efficient automatic production process optimization method and system

Abstract

The application discloses a kind of heavy die steel high-efficiency automation production process optimization method and system, this method is by distributed sensor network real-time acquisition production data, using wavelet transform is preprocessed;Establish multi-objective optimization model, use improved particle swarm algorithm collaborative optimization production efficiency, product quality, energy consumption and equipment wear;Construction process 3D collaborative model realizes dynamic scheduling;Based on LSTM neural network prediction energy consumption and implement time-sharing scheduling;Using support vector machine is carried out quality real-time prediction and defect early warning;By deep reinforcement learning realizes continuous optimization.The system includes data acquisition, preprocessing, multi-objective optimization, dynamic scheduling, energy management, quality prediction, data traceability and central control and so on module.The application realizes whole-process intelligent collaborative control, significantly improves production efficiency, product quality and resource utilization, has important industrial application value.

Description

Technical Field

[0001] This invention relates to the field of intelligent production technology for mold steel, specifically a method and system for optimizing the efficient and automated production process of heavy mold steel. Background Technology

[0002] Heavy-duty mold steel is a core material for high-end equipment manufacturing, possessing characteristics such as high strength, high toughness, and high wear resistance. It is widely used in key areas such as automotive body panel molds, aerospace precision components, and large-scale machinery forming molds. The production process of heavy-duty mold steel is complex, encompassing multiple core processes such as steelmaking, refining, casting, forging, and heat treatment. The process parameters of each process are highly coupled, and the production status fluctuates dynamically, requiring extremely high production precision, stability, and energy consumption control.

[0003] Currently, most domestic heavy-duty mold steel production adopts a traditional semi-automated production mode. The overall production control system has many inherent defects, which seriously restricts the improvement of product quality and production efficiency. The specific problems are as follows:

[0004] 1. Experience-based control of process parameters: The adjustment of key process parameters such as core temperature, pressure, holding time, and cooling rate in each process relies entirely on the manual experience of operators, lacking quantitative and standardized scientific control basis. This results in large fluctuations in product quality between batches, poor consistency of core indicators such as hardness, toughness, and flatness, and a persistently high defect rate.

[0005] 2. Poor coordination across the entire process: The existing production model is based on independent control of each process, with disconnected material flow and information exchange between processes. The production rhythm of the preceding and following processes is not matched, resulting in excessive waiting time for materials between processes, excessive idle time of equipment, chaotic overall production rhythm, and less than 60% comprehensive utilization rate of core production equipment, leading to low efficiency in large-scale production.

[0006] 3. Inefficient energy consumption management: High-energy-consuming processes such as steelmaking heating, forging heating, and heat treatment insulation lack refined control and are not dynamically scheduled in conjunction with production load, peak and off-peak electricity prices, and equipment status. As a result, the proportion of ineffective energy consumption is high, and production energy costs and carbon emissions remain high.

[0007] 4. Delayed and passive quality inspection: The traditional quality inspection model involves sampling inspection after production is completed. It is impossible to predict quality defects in real time during the production process. By the time problems such as cracks, insufficient hardness, and composition deviation are detected, the entire batch of products has already been produced, resulting in a large waste of raw materials, energy and human resources.

[0008] 5. Lack of data traceability and iteration capabilities: The existing production system lacks a full-process data collection and storage system. Production data for each process is fragmented and isolated. When product quality problems or equipment failures occur, it is impossible to quickly trace and locate the root cause, and it is impossible to rely on historical production data to achieve continuous optimization and iteration of production processes.

[0009] Existing technical solutions mostly focus on the automation transformation and parameter optimization of single processes, which can only improve local production efficiency and cannot cover the entire production process of heavy mold steel. Although some solutions have built a data acquisition system, they only realize data storage and display, without building a linkage mechanism between data and process optimization, scheduling control, and quality prediction. The data utilization rate is extremely low, and it is impossible to achieve dynamic and intelligent optimization of the entire process.

[0010] In summary, the industry urgently needs an intelligent production optimization method and system that can achieve data-driven, multi-objective collaborative optimization, proactive quality early warning, and refined energy consumption control throughout the entire production process of heavy mold steel, in order to address many shortcomings of existing technologies. Summary of the Invention

[0011] The purpose of this invention is to overcome the shortcomings of existing technologies in the production of heavy mold steel, such as experience-based control, poor process coordination, high energy consumption, large quality fluctuations, and lack of continuous optimization mechanisms. This invention provides a method and system for optimizing the efficient and automated production process of heavy mold steel. Through real-time data acquisition throughout the entire process, multi-objective algorithm collaborative optimization, deep learning for energy consumption and quality prediction, dynamic process scheduling, and reinforcement learning for continuous iteration, this invention achieves intelligent, refined, and collaborative control of the entire production process, effectively improving production efficiency, product qualification rate, and resource utilization, while reducing production energy consumption and costs.

[0012] To achieve the above objectives, the present invention provides the following technical solution:

[0013] A method for optimizing an efficient and automated production process for heavy-duty mold steel includes the following steps:

[0014] S1. Real-time acquisition and preprocessing of production data: The key parameters of temperature, pressure, composition, energy consumption and time of each process of steelmaking, refining, casting, forging and heat treatment are collected in real time through a distributed sensor network to build a production database. The original acquired data is preprocessed by noise reduction and outlier removal using wavelet transform method.

[0015] S2. Multi-objective collaborative optimization of process parameters: Establish a four-dimensional multi-objective optimization model that includes production efficiency, product quality, energy consumption, and equipment wear, and use an improved multi-objective particle swarm optimization algorithm to iteratively solve the problem and obtain the optimal combination of process parameters for each process.

[0016] S3. Dynamic scheduling and collaborative control between processes: Construct a three-dimensional collaborative model of process material flow, information flow and energy flow, dynamically calculate the optimal start time of each process, adjust the production cycle and process parameters in real time, and minimize the waiting time between processes;

[0017] S4. Refined Energy Consumption Prediction and Control: Build an energy consumption prediction model based on LSTM neural network to predict the energy consumption demand of the production line in real time, and implement time-of-use energy consumption scheduling strategy in combination with the peak and valley electricity price of the power grid to achieve optimal energy consumption management.

[0018] S5. Real-time product quality prediction and defect warning: Build a quality prediction model based on support vector machine to predict product quality indicators in real time. When the predicted value deviates from the preset threshold, an early warning will be automatically triggered and a parameter adjustment plan will be output.

[0019] S6. Full-process data traceability and continuous optimization: Assign a unique identification code to each batch of products, record the entire production process data, and continuously iterate and optimize the production strategy based on historical data through deep reinforcement learning algorithms.

[0020] As a further aspect of the present invention: the calculation formula for wavelet transform filtering in step S1 is as follows:

[0021]

[0022] in, For the filtered and effective production data, For wavelet basis functions, These are wavelet coefficients. This represents the wavelet decomposition level.

[0023] As a further aspect of the present invention: the objective function of the multi-objective optimization model in step S2 is:

[0024]

[0025] in, As a production efficiency indicator, Let be the quality loss function. For production energy consumption, For equipment wear, These are the weighting coefficients, and the sum of the weighting coefficients is 1;

[0026] The particle update formula for the improved multi-objective particle swarm optimization algorithm is as follows:

[0027]

[0028]

[0029] in, Let be the velocity of the i-th particle in the t-th iteration. Let i be the position of the i-th particle in the t-th iteration. For inertial weights, As a learning factor, A random number between 0 and 1. This is the optimal solution for the individual. This is the globally optimal solution.

[0030] As a further aspect of the present invention: the formula for calculating the optimal start-up time of the process in step S3 is:

[0031]

[0032] in, To determine the optimal start time for the i-th process, This is the end time of the preceding process. This refers to the time for material transportation between processes. Optimize the reduction amount for buffer time.

[0033] As a further aspect of the present invention: the formula for the LSTM neural network energy consumption prediction model in step S4 is:

[0034]

[0035] in, For the future Predicted energy consumption for different time periods Historical energy consumption sequence This is a sequence of historical process parameters. This is a historical temperature operating condition sequence;

[0036] The formula for calculating the energy efficiency optimization rate is:

[0037]

[0038] in, For energy efficiency optimization, To optimize the baseline energy consumption, This is the optimized actual energy consumption.

[0039] As a further aspect of the present invention: the support vector machine quality prediction model formula in step S5 is:

[0040]

[0041] in, To predict product quality values, For Lagrange multipliers, For kernel function, For bias terms, This represents the number of support vectors.

[0042] As a further aspect of the present invention: the deep reinforcement learning reward function in step S6 is:

[0043]

[0044] in, Let be the reward value at time t. To improve the quantity of quality, This represents the change in energy consumption. This represents the change in output. These are the weighting coefficients.

[0045] A high-efficiency automated production process optimization system for heavy-duty mold steel, characterized in that it includes:

[0046] The data acquisition module is used to collect key parameters such as temperature, pressure, composition, energy consumption, and time in real time for each process of steelmaking, refining, casting, forging, and heat treatment.

[0047] The data preprocessing module is used to clean, perform wavelet noise reduction, anomaly removal and feature extraction on the collected raw data, and output standardized production data.

[0048] The multi-objective optimization module has a built-in improved multi-objective particle swarm optimization algorithm to solve for the optimal combination of process parameters, thereby achieving coordinated optimization of production efficiency, quality, energy consumption, and equipment wear.

[0049] The dynamic scheduling module is used to build a three-dimensional collaborative model of the process and dynamically adjust the process start time and production cycle time;

[0050] The energy management module has a built-in LSTM energy prediction model for real-time energy prediction and time-sharing energy scheduling control.

[0051] The quality prediction module has a built-in SVM quality prediction model, which is used for real-time product quality prediction, defect warning and parameter optimization suggestions.

[0052] The data traceability module is used to generate unique batch identifiers, store and manage production data throughout the entire process, and realize traceability of the entire product lifecycle.

[0053] The central control module communicates with each module to coordinate the operation of each module and achieve intelligent management and optimization of the system as a whole.

[0054] As a further embodiment of the present invention: the data acquisition module includes a distributed array of temperature sensors, pressure sensors, a molten steel composition analyzer, a power detector, and a time recorder, and each acquisition device establishes real-time communication with the data preprocessing module via an industrial Ethernet.

[0055] As a further aspect of the present invention: the central control module adopts a distributed architecture of edge computing units + cloud computing center. The edge computing units are deployed on the production line site and are responsible for real-time production decision-making and parameter control; the cloud computing center is responsible for model iterative training, big data storage, and long-term production strategy optimization.

[0056] Compared with the prior art, the beneficial effects of the present invention are:

[0057] 1. Achieve intelligent collaborative optimization across the entire process: Break through the technical limitations of traditional single-process local optimization, break down data barriers across the entire process of heavy mold steel production, achieve collaborative control of the entire process from steelmaking to heat treatment, reduce ineffective waiting time between processes by more than 30%, increase the overall production efficiency of the production line by 25%, and increase the comprehensive utilization rate of equipment to more than 90%.

[0058] 2. Construct a multi-objective dynamic balance optimization mechanism: Innovatively establish a multi-objective optimization model integrating efficiency, quality, energy consumption, and equipment loss, and achieve dynamic balance of multiple indicators through adaptive weight adjustment, thoroughly solving the problem of deterioration of other production indicators caused by optimization of a single indicator, making production control more scientific and comprehensive.

[0059] 3. Achieve refined and intelligent energy consumption management: By relying on LSTM neural networks to accurately predict production energy consumption and combining peak-valley electricity pricing to implement differentiated energy consumption scheduling, production energy consumption can be reduced by 15%-20%, effectively reducing enterprise production costs, reducing carbon emissions, and achieving significant green production benefits.

[0060] 4. Quality control shifts from post-event detection to pre-event prediction: By using the SVM real-time quality prediction model, quality control is brought forward, and the defect identification accuracy rate can reach over 95%, effectively avoiding the production of batch defective products, reducing the product scrap rate by 40%, and significantly improving the utilization rate of raw materials.

[0061] 5. Possesses autonomous and continuous iterative optimization capabilities: Based on deep reinforcement learning algorithms and relying on production big data, the system continuously optimizes production strategies. It has the ability to learn and upgrade itself. As production data accumulates, the optimization accuracy continues to improve, forming a long-term optimization mechanism.

[0062] 6. Full-process traceability and control: The unique batch identification code enables data traceability throughout the entire production lifecycle, reducing the problem tracing time from the traditional 2 days to 2 hours. It can quickly locate process, equipment, and operational problems, providing data support for continuous improvement of production processes. Detailed Implementation

[0063] The technical solution of the present invention will be further described in detail below with reference to specific embodiments.

[0064] A method for optimizing an efficient and automated production process for heavy-duty mold steel includes the following steps:

[0065] S1. Real-time acquisition and preprocessing of production data

[0066] A distributed sensor network covering the entire process of steelmaking, refining, casting, forging, and heat treatment is established to collect key production parameters such as smelting temperature, refining pressure, molten steel composition, casting rate, forging temperature, heat treatment holding time, cooling rate, and equipment operating power in real time, thereby constructing a real-time production database.

[0067] To address the issues of noise and outlier interference in the collected raw data, a wavelet transform algorithm is used to perform data cleaning, noise reduction, and preprocessing. This removes outlier data caused by equipment malfunctions and signal interference, improving data accuracy. The filtering calculation formula is as follows:

[0068]

[0069] in, The effective production data after noise reduction and filtering. For the selected wavelet basis functions, These are the wavelet transform coefficients. This represents the wavelet decomposition level.

[0070] S2, Multi-objective collaborative optimization of process parameters

[0071] Abandoning traditional single-efficiency or single-quality optimization models, we construct a four-dimensional multi-objective optimization model that takes into account production efficiency, product quality, energy consumption, and equipment wear, achieving dynamic balance optimization of multi-dimensional production indicators. The optimization objective function is as follows:

[0072]

[0073] in, It is a production efficiency indicator, characterized by output per unit time; The product quality loss function is comprehensively characterized by hardness deviation, composition deviation, and defect rate. Energy consumption per batch of production; Real-time wear and tear of the equipment; For the corresponding weight coefficients, and satisfy the sum of weights It can be adaptively adjusted according to the needs of production orders.

[0074] An improved multi-objective particle swarm optimization algorithm (MOPSO) is used to iteratively solve the objective function, avoiding the problems of local optima and slow convergence speed of traditional algorithms, and obtaining the optimal combination of process parameters for each process step. The particle velocity and position update formulas are as follows:

[0075]

[0076]

[0077] in, Let $\frac{i}{t}$ be the running speed of the $i$-th particle in the $t$-th iteration. Let i be the position of the i-th particle in the t-th iteration; Inertial weights are used to balance global and local search capabilities; For learning factors; A random number in the range of 0-1; This represents the optimal solution for an individual particle. This is the globally optimal solution for the population.

[0078] S3. Dynamic scheduling and collaborative control between processes

[0079] A three-dimensional collaborative model of the process, encompassing material flow, information flow, and energy flow, is constructed to break down the barriers between independent operations of each process. Based on the completion status of the preceding process, material transportation time, and equipment idle status, the optimal start time of the current process is dynamically calculated to achieve cycle time matching across all processes. The formula for calculating the optimal start time is as follows:

[0080]

[0081] in, To determine the optimal start time for the i-th process, This is the end time of the preceding process. The time spent transporting materials between processes, Optimize the reduction amount for buffer time.

[0082] Real-time monitoring of equipment operating status, production progress, and material inventory in each process allows for dynamic fine-tuning of process parameters and start-up cycles in subsequent processes, minimizing waiting time between processes and improving overall equipment utilization and production line efficiency.

[0083] S4. Refined Energy Consumption Prediction and Control

[0084] An energy consumption prediction model based on an LSTM (Long Short-Term Memory) neural network is constructed. Relying on historical energy consumption sequences, process parameter sequences, and temperature condition sequences, it accurately predicts the overall energy consumption demand of the production line in future periods. The prediction formula is as follows:

[0085]

[0086] in, For the future Predicted energy consumption for a given period of time This is a sequence of historical energy consumption data for the previous n time periods. This is a sequence of process parameters for the first n time periods. This is the sequence of process temperature conditions for the first n time periods.

[0087] Based on energy consumption forecasts and the grid's peak-valley electricity pricing policy, a time-of-use energy dispatch strategy is formulated. High-energy-consuming processes such as steelmaking and heat treatment are prioritized for scheduling during off-peak hours, while low-energy-consuming auxiliary processes are scheduled during peak hours, achieving optimal energy costs. The energy optimization effect is calculated using the following formula:

[0088]

[0089] in, For energy efficiency optimization, To optimize the baseline energy consumption under the same production capacity, To optimize actual production energy consumption.

[0090] S5. Real-time Product Quality Prediction and Defect Early Warning

[0091] A real-time quality prediction model based on Support Vector Machine (SVM) is constructed. Taking real-time process parameters of each step as input and quality indicators such as mold steel hardness, toughness, flatness, and internal defects as output, the model predicts product quality status in real time, enabling proactive prediction of quality problems. The prediction formula is as follows:

[0092]

[0093] in, To predict product quality index values, For Lagrange multipliers, This is a kernel function used to implement high-dimensional data mapping. For bias terms, This represents the number of support vectors.

[0094] The system presets standard threshold ranges for each quality indicator. When the real-time predicted quality value deviates from the threshold range, the system automatically triggers audible and visual warnings and background pop-up warnings. At the same time, it intelligently outputs corresponding process parameter adjustment schemes to achieve proactive intervention in quality defects and avoid the generation of batch defective products.

[0095] S6, End-to-End Data Traceability and Continuous Optimization Iteration

[0096] Each batch of heavy mold steel is assigned a unique identification code, and the entire process of the batch of products, from raw material entry, production of each process, quality inspection, to finished product exit, is bound and recorded, including process parameters, equipment status, energy consumption data, and quality data, to build a full-dimensional data traceability system.

[0097] By introducing a deep reinforcement learning algorithm with quality improvement, energy consumption reduction, and capacity increase as the core optimization objectives, a dedicated reward function is designed. Based on massive amounts of historical production data, the production strategy is continuously trained and iterated to achieve autonomous system optimization and upgrading. The reward function is as follows:

[0098]

[0099] in, Let t be the algorithm reward value. To improve real-time quality, This represents the real-time change in energy consumption. This represents the real-time change in production output. This is the weighting adjustment coefficient. The longer the system runs, the richer the accumulated data, the better the optimization accuracy and effect, forming a data-driven positive iterative cycle.

[0100] A second aspect of the present invention provides a system for implementing the above-described method for optimizing the efficient automated production process of heavy mold steel, comprising:

[0101] Data acquisition module: Deployed in various processes of steelmaking, refining, casting, forging and heat treatment, including temperature sensors, pressure sensors, composition analyzers, power detectors and time recorders, to collect core production parameters of each process in real time via industrial Ethernet;

[0102] Data preprocessing module: Communicates with the data acquisition module and is used to perform outlier removal, wavelet noise reduction, data normalization and feature extraction on the raw acquired data, and output standardized and effective data;

[0103] Multi-objective optimization module: Built-in improved multi-objective particle swarm optimization algorithm, based on preprocessed production data, to solve the optimal combination of process parameters and achieve a coordinated balance between efficiency, quality, energy consumption and equipment wear;

[0104] Dynamic scheduling module: Constructs a three-dimensional collaborative model of the process, and dynamically adjusts the process start time, production cycle and process parameters according to the real-time production status to achieve collaborative operation of the entire process;

[0105] Energy management module: Built-in LSTM energy consumption prediction model to realize real-time energy consumption prediction, peak and off-peak energy consumption scheduling and energy consumption data statistical analysis;

[0106] Quality Prediction Module: Built-in SVM quality prediction model, which can predict the product quality status in real time and has the functions of defect warning and parameter optimization suggestion output;

[0107] Data traceability module: Establishes a unique batch identification system, stores and manages production data throughout the entire process, and supports data query and problem tracing throughout the product lifecycle;

[0108] Central control module: It is electrically connected to all the above modules and adopts an edge computing + cloud distributed architecture. The edge computing unit is responsible for real-time production control and rapid decision-making, while the cloud computing center is responsible for model training, data storage and long-term production strategy optimization, and overall system operation.

[0109] Example 1

[0110] This embodiment provides a method for optimizing the efficient and automated production process of heavy-duty mold steel, which is applied to a large-scale production line for H13 heavy-duty mold steel. The specific implementation steps are as follows:

[0111] Step 1: Real-time data acquisition and preprocessing throughout the entire process: High-precision sensors are deployed in each process of H13 mold steelmaking, refining, casting, forging, and heat treatment to collect 28 core parameters, including melting temperature, carbon, chromium, and molybdenum content in molten steel, forging pressure, heat treatment holding temperature, and equipment power, once per second; the db4 wavelet basis function is selected, and a 5-level decomposition layer is set. Wavelet transform is used to complete data noise reduction, eliminating sensor signal interference and invalid data generated by instantaneous equipment anomalies, ensuring the accuracy of the data input to the model.

[0112] Step 2, Multi-objective process parameter optimization: Based on the production characteristics of H13 mold steel, initialize the weights of the multi-objective optimization model, setting them under normal production scenarios. Prioritize product quality; set the particle swarm optimization algorithm to iterate 100 times, with inertia weights... Learning factor The optimal process parameters for each process are obtained through iterative solution, including the optimal combination of parameters such as melting temperature 1580-1600℃, refining pressure 0.08MPa, forging start temperature 1120℃, and heat treatment quenching temperature 1030℃.

[0113] Step 3, Dynamic Coordinated Scheduling of Processes: The system calculates the optimal start time for each process in real time. Based on the completion time of the casting process and the material conveying time, it automatically adjusts the start-up rhythm of the forging process, compresses the buffer waiting time between processes, and reduces the average waiting time of each process on the original production line from 25 minutes to 17 minutes, significantly improving the production line's efficiency.

[0114] Step 4, Intelligent Energy Consumption Prediction and Scheduling: Using an LSTM model with nearly 30 days of historical production data as the training set, the energy consumption of the production line is predicted in real time for each period of the day. High-energy-consuming heat treatment and insulation processes are scheduled to the early morning off-peak electricity price period, while low-energy-consuming material trimming and testing processes are scheduled to the daytime peak electricity price period. In this embodiment, the energy consumption of a single batch of products is reduced by 18.2%, and the energy cost is significantly reduced.

[0115] Step 5: Real-time quality early warning and control: The SVM model is used to predict the core indicators of H13 mold steel in real time, such as hardness, tempering stability, and internal density. The preset hardness standard threshold is 48-52 HRC. When the predicted hardness deviates from the range during production, the system immediately issues an early warning and automatically fine-tunes the heat treatment holding time and cooling rate. In this embodiment, the product defect identification accuracy rate reaches 96.3%, and the scrap rate is reduced from the original 3.2% to 1.9%.

[0116] Step 6, Data Traceability and Continuous Optimization: Generate a unique traceability QR code for each batch of H13 mold steel and bind it to the entire production process data; continuously iterate the production strategy through deep reinforcement learning algorithms. After 3 months of data accumulation, the system autonomously optimizes the weighting of process parameters, further improving the consistency of product batch quality and continuously optimizing production efficiency.

[0117] The production system in this embodiment adopts an edge + cloud distributed architecture. The edge computing unit is deployed on the production line site and is responsible for millisecond-level real-time parameter adjustment and scheduling decisions. The cloud platform is responsible for daily model iteration, data aggregation and analysis, and production strategy optimization to ensure the system's real-time performance and long-term optimization.

[0118] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A method for optimizing the efficient and automated production process of heavy-duty mold steel, characterized in that, Includes the following steps: S1. Real-time acquisition and preprocessing of production data: The key parameters of temperature, pressure, composition, energy consumption and time of each process of steelmaking, refining, casting, forging and heat treatment are collected in real time through a distributed sensor network to build a production database. The original acquired data is preprocessed by noise reduction and outlier removal using wavelet transform method. S2. Multi-objective collaborative optimization of process parameters: Establish a four-dimensional multi-objective optimization model that includes production efficiency, product quality, energy consumption, and equipment wear, and use an improved multi-objective particle swarm optimization algorithm to iteratively solve the problem and obtain the optimal combination of process parameters for each process. S3. Dynamic scheduling and collaborative control between processes: Construct a three-dimensional collaborative model of process material flow, information flow and energy flow, dynamically calculate the optimal start time of each process, adjust the production cycle and process parameters in real time, and minimize the waiting time between processes; S4. Refined Energy Consumption Prediction and Control: Build an energy consumption prediction model based on LSTM neural network to predict the energy consumption demand of the production line in real time, and implement time-of-use energy consumption scheduling strategy in combination with the peak and valley electricity price of the power grid to achieve optimal energy consumption management. S5. Real-time product quality prediction and defect warning: Build a quality prediction model based on support vector machine to predict product quality indicators in real time. When the predicted value deviates from the preset threshold, an early warning will be automatically triggered and a parameter adjustment plan will be output. S6. Full-process data traceability and continuous optimization: Assign a unique identification code to each batch of products, record the entire production process data, and continuously iterate and optimize the production strategy based on historical data through deep reinforcement learning algorithms.

2. The method for optimizing the efficient automated production process of heavy-duty mold steel according to claim 1, characterized in that, The calculation formula for wavelet transform filtering in step S1 is as follows: in, For the filtered and effective production data, For wavelet basis functions, These are wavelet coefficients. This represents the wavelet decomposition level.

3. The method for optimizing the efficient automated production process of heavy-duty mold steel according to claim 1, characterized in that, The objective function of the multi-objective optimization model in step S2 is: ,in, As a production efficiency indicator, Let be the quality loss function. For production energy consumption, For equipment wear, Here, represents the weighting coefficients, and the sum of these weighting coefficients is 1; the particle update formula for the improved multi-objective particle swarm optimization algorithm is: , in, Let be the velocity of the i-th particle in the t-th iteration. Let i be the position of the i-th particle in the t-th iteration. For inertial weights, As a learning factor, A random number between 0 and 1. This is the optimal solution for the individual. This is the globally optimal solution.

4. The method for optimizing the efficient automated production process of heavy-duty mold steel according to claim 1, characterized in that, The formula for calculating the optimal start-up time of the process in step S3 is as follows: in, To determine the optimal start time for the i-th process, This is the end time of the preceding process. This refers to the time for material transportation between processes. Optimize the reduction amount for buffer time.

5. The method for optimizing the efficient automated production process of heavy-duty mold steel according to claim 1, characterized in that, The formula for the LSTM neural network energy consumption prediction model in step S4 is as follows: in, For the future Predicted energy consumption for different time periods Historical energy consumption sequence This is a sequence of historical process parameters. The historical temperature operating condition series is used; the energy consumption optimization rate is calculated using the following formula: in, For energy efficiency optimization, To optimize the baseline energy consumption, This is the optimized actual energy consumption.

6. The method for optimizing the efficient automated production process of heavy-duty mold steel according to claim 1, characterized in that, The formula for the support vector machine quality prediction model in step S5 is: in, To predict product quality values, For Lagrange multipliers, For kernel function, For bias terms, This represents the number of support vectors.

7. The method for optimizing the efficient automated production process of heavy-duty mold steel according to claim 1, characterized in that, The deep reinforcement learning reward function in step S6 is: in, Let be the reward value at time t. To improve the quantity of quality, This represents the change in energy consumption. This represents the change in output. These are the weighting coefficients.

8. A high-efficiency automated production process optimization system for heavy-duty mold steel implementing the method of any one of claims 1-7, characterized in that, include: The data acquisition module is used to collect key parameters such as temperature, pressure, composition, energy consumption, and time in real time for each process of steelmaking, refining, casting, forging, and heat treatment. The data preprocessing module is used to clean, perform wavelet noise reduction, anomaly removal and feature extraction on the collected raw data, and output standardized production data. The multi-objective optimization module has a built-in improved multi-objective particle swarm optimization algorithm to solve for the optimal combination of process parameters, thereby achieving coordinated optimization of production efficiency, quality, energy consumption, and equipment wear. The dynamic scheduling module is used to build a three-dimensional collaborative model of the process and dynamically adjust the process start time and production cycle time; The energy management module has a built-in LSTM energy prediction model for real-time energy prediction and time-sharing energy scheduling control. The quality prediction module has a built-in SVM quality prediction model, which is used for real-time product quality prediction, defect warning and parameter optimization suggestions. The data traceability module is used to generate unique batch identifiers, store and manage production data throughout the entire process, and realize traceability of the entire product lifecycle. The central control module communicates with each module to coordinate the operation of each module and achieve intelligent management and optimization of the system as a whole.

9. The high-efficiency automated production process optimization system for heavy-duty mold steel according to claim 8, characterized in that, The data acquisition module includes distributed temperature sensors, pressure sensors, a molten steel composition analyzer, a power detector, and a time recorder. Each acquisition device establishes real-time communication with the data preprocessing module via an industrial Ethernet network.

10. The high-efficiency automated production process optimization system for heavy-duty mold steel according to claim 8, characterized in that, The central control module adopts a distributed architecture of edge computing units and cloud computing center. The edge computing units are deployed on the production line site and are responsible for real-time production decision-making and parameter control. The cloud computing center is responsible for model iterative training, big data storage and long-term production strategy optimization.

Images