Intelligent control methods, systems and devices for vacuum isothermal forging presses

Abstract

This invention relates to the field of intelligent control technology for forging processes, specifically an intelligent control method, system, and device for a vacuum isothermal forging press. The method includes: collecting temperature, pressure, and stress data during the forging process using multimodal sensors to generate a preliminary raw data sequence; cleaning and standardizing the raw data sequence to generate standardized real-time operating condition data; constructing a dynamic process prediction model based on the real-time operating condition data and historical data to generate a key parameter prediction sequence; inputting the key parameter prediction sequence into a reinforcement learning model to generate an optimized set of process parameters; and dynamically adjusting the press execution system based on the optimized set of process parameters, while simultaneously monitoring the equipment's operating status in real time to generate a closed-loop control command sequence. This invention can effectively improve the consistency of forging quality and production efficiency, and reduce the risk of equipment failure.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for forging processes, and in particular to intelligent control methods, systems and devices for vacuum isothermal forging presses. Background Technology

[0002] Vacuum isothermal forging is a metal forming process performed under high temperature and low pressure conditions. This process significantly improves the internal microstructure uniformity and material properties of forgings by subjecting the metal material to prolonged isothermal deformation in a vacuum environment. However, due to the dynamic coupling characteristics of process parameters such as temperature, pressure, and press speed during forging, traditional control methods struggle to meet the high requirements for process accuracy and stability.

[0003] In existing technologies, forging presses typically rely on preset, fixed process parameters for control. This approach cannot respond in real time to the dynamic changes in multiple variables during the forging process. Furthermore, the monitoring and fault warning mechanisms for equipment operation are inadequate, which can easily lead to unstable forging product quality and increased equipment downtime.

[0004] With the rapid development of artificial intelligence technology, the application of deep learning and reinforcement learning algorithms in complex industrial scenarios has gradually matured. However, applying them to the dynamic optimization control of vacuum isothermal forging processes still faces many challenges, such as how to extract key process features from multimodal sensor data, how to build accurate dynamic prediction models based on real-time and historical data, and how to use reinforcement learning to achieve intelligent optimization of process parameters.

[0005] Therefore, given the shortcomings of existing vacuum isothermal forging presses in process parameter optimization, equipment status monitoring, and fault early warning, there is an urgent need for an intelligent control method and system that can combine multimodal sensor data, deep learning, and reinforcement learning algorithms to achieve dynamic optimization control of process parameters and efficient monitoring of equipment operating status. This innovative technology can not only improve the consistency and reliability of forging quality but also significantly reduce the risk of equipment failure and increase production efficiency. Summary of the Invention

[0006] This invention provides an intelligent control method, system, and device for a vacuum isothermal forging press to address the problem of how to dynamically optimize process parameters such as temperature, pressure, and press speed during forging based on real-time operating data and historical data collected by multimodal sensors, combined with deep learning and reinforcement learning algorithms, to monitor the equipment operating status in real time, generate closed-loop control commands, improve forging quality, and reduce equipment failure risks.

[0007] To address the aforementioned technical problems, this invention provides an intelligent control method for a vacuum isothermal forging press, comprising:

[0008] Temperature, pressure, and stress data during the forging process are collected using multimodal sensors to generate a preliminary raw data sequence;

[0009] The original data sequence is cleaned and standardized to generate standardized real-time operating condition data;

[0010] Based on the real-time operating data and historical data, key process features are extracted using a deep learning model, a dynamic process prediction model is constructed, and a key parameter prediction sequence is generated.

[0011] The predicted sequence of key parameters is input into a reinforcement learning model to optimize process parameters such as temperature, pressure and press speed during forging, and to generate an optimized set of process parameters.

[0012] Based on the optimized set of process parameters, the press execution system is dynamically adjusted, and the equipment operating status is monitored in real time to generate a closed-loop control command sequence.

[0013] Furthermore, the acquisition of temperature, pressure, and stress data during the forging process includes:

[0014] Temperature data is obtained through thermocouples and infrared detectors;

[0015] Pressure data is obtained using strain gauges;

[0016] Stress data is obtained through mechanical sensors.

[0017] Furthermore, the cleaning and standardization process for the original data sequence specifically includes:

[0018] The Kalman filter algorithm is used to filter noise from the original data sequence.

[0019] Detect and remove outliers to obtain cleaned data;

[0020] The cleaned data is standardized using the z-score standardization method to generate standardized real-time operating condition data.

[0021] Furthermore, the extraction of key process features specifically includes:

[0022] Based on the standardized real-time operating data, Fourier transform is used to extract key features of temperature change rate, pressure fluctuation amplitude, and stress response.

[0023] The extracted features are filtered using principal component analysis to generate a process feature dataset.

[0024] Furthermore, the construction of the dynamic process prediction model specifically includes:

[0025] Based on the aforementioned process feature dataset and historical process data, a dynamic process prediction model is established using a recurrent neural network model.

[0026] Input the process feature dataset, predict key parameters in the forging process, and generate a key parameter prediction sequence.

[0027] Furthermore, the optimized forging process parameters specifically include:

[0028] The predicted sequence of key parameters is used as the state input of the reinforcement learning model;

[0029] The optimal combination of process parameters is generated by optimizing the process parameter adjustment strategy through reinforcement learning algorithms.

[0030] Output optimized temperature, pressure, and press speed parameters.

[0031] Furthermore, the generation of the closed-loop control command specifically includes:

[0032] Collect real-time operating status data of the equipment, including real-time temperature, pressure, and press speed;

[0033] Based on the feedback data and the optimized set of process parameters, a closed-loop control command sequence is generated.

[0034] Continuously adjust the temperature, pressure, and press speed during the forging process.

[0035] Furthermore, the intelligent control method for the vacuum isothermal forging press also includes:

[0036] Analyze equipment operating status using time-series prediction models to generate fault early warning information;

[0037] Based on the fault warning information, the process parameters are dynamically adjusted.

[0038] Furthermore, an intelligent control system for a vacuum isothermal forging press, the system comprising:

[0039] The data acquisition module is used to collect temperature, pressure, and stress data from multimodal sensors during the forging process and generate raw data sequences.

[0040] The data processing module is used to clean and standardize the original data sequence to generate standardized real-time operating condition data.

[0041] The dynamic process prediction module is used to extract key process features based on the real-time operating data and historical data, construct a dynamic process prediction model, and generate a key parameter prediction sequence.

[0042] The process parameter optimization module is used to optimize the process parameters based on the predicted sequence of the key parameters and use a reinforcement learning model to generate an optimized set of process parameters.

[0043] The closed-loop control module is used to generate a closed-loop control command sequence based on the optimized set of process parameters, and to dynamically adjust the press execution system.

[0044] Furthermore, an intelligent control device for a vacuum isothermal forging press includes a data acquisition module, a data processing and feature extraction module, a dynamic process prediction module, a process parameter optimization module, and a closed-loop control module. The device can implement the steps of the aforementioned intelligent control method for a vacuum isothermal forging press.

[0045] The key innovations of this invention include:

[0046] (1) Data acquisition and processing technology based on multimodal sensors. The innovative combination of multimodal sensors with data cleaning and standardization generates high-precision real-time operating condition data, effectively avoiding process parameter deviations caused by data noise or outliers.

[0047] (2) Construction of dynamic process prediction model. Deep learning algorithms (such as recurrent neural networks, RNN) are used to model real-time operating data. By combining historical data and prior knowledge, the changing trend of key parameters is accurately predicted, thus solving the nonlinear coupling problem under complex operating conditions.

[0048] (3) Process parameter optimization based on reinforcement learning. By dynamically adjusting the predicted parameters through reinforcement learning algorithms, the combination of process parameters is optimized, thereby improving the adaptability and optimization efficiency of equipment and processes during production.

[0049] The following are its main beneficial effects:

[0050] (1) High-precision real-time operating condition data acquisition and processing. This invention acquires multi-dimensional real-time operating condition data such as temperature, pressure, and stress based on multi-modal sensors, and generates standardized high-quality operating condition data through noise filtering and outlier removal techniques. Compared with the traditional fixed parameter setting method, the real-time acquired and processed data is more accurate and comprehensive, laying the foundation for subsequent dynamic optimization.

[0051] (2) Dynamic prediction and optimization improve process accuracy. By using deep learning models to extract key process features and combining historical data and expert knowledge to construct dynamic process prediction models, it is possible to accurately predict the future trend of process parameters. By optimizing parameters such as temperature, pressure and press speed during the forging process through reinforcement learning models, the problem that traditional methods are difficult to adapt to complex dynamic working conditions is solved, and the consistency of forging quality is greatly improved.

[0052] (3) Closed-loop control reduces the risk of equipment failure. Based on real-time monitoring of equipment operating status and generation of closed-loop control command sequences, the operating parameters of the equipment can be dynamically adjusted to respond to equipment anomalies in real time and reduce the probability of equipment failure. Compared with traditional manual intervention or fixed rule control methods, this invention significantly improves the continuity and reliability of production. Attached Figure Description

[0053] Figure 1 A flowchart illustrating the intelligent control method for a vacuum isothermal forging press provided in this application embodiment;

[0054] Figure 2 This is a structural block diagram of the intelligent control system for a vacuum isothermal forging press provided in an embodiment of this application;

[0055] Figure 3 This is a structural block diagram of the intelligent control device for a vacuum isothermal forging press provided in an embodiment of this application. Detailed Implementation

[0056] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0057] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0058] Example 1: Refer to Figure 1 This is a flowchart illustrating the intelligent control method for a vacuum isothermal forging press provided in an embodiment of the present invention. The process may include at least steps S100-S500:

[0059] S100: Based on multimodal sensors, temperature, pressure, and stress data during the forging process are collected to generate a preliminary raw data sequence;

[0060] S200. The original data sequence is cleaned and standardized to generate standardized real-time operating condition data.

[0061] S300. Based on the real-time operating data and historical data, extract key process features using a deep learning model, construct a dynamic process prediction model, and generate a key parameter prediction sequence.

[0062] S400. Input the predicted sequence of key parameters into the reinforcement learning model to optimize the process parameters such as temperature, pressure and press speed during the forging process, and generate an optimized set of process parameters.

[0063] S500: Based on the optimized set of process parameters, dynamically adjust the press execution system, monitor the equipment operating status in real time, and generate a closed-loop control command sequence.

[0064] Step S100 includes at least steps S110-S130:

[0065] S110. Acquire temperature, pressure, and stress data during the forging process, monitor process parameters in real time using multimodal sensors, and generate preliminary raw data sequences through the sensor data acquisition module.

[0066] Sensor Data Acquisition: During the operation of the vacuum isothermal forging press, the multimodal sensors are deployed at key locations in the forging system to monitor process parameters such as temperature, pressure, and stress in real time. Specifically, the temperature sensor measures the temperature inside the press cavity using a thermocouple or infrared detector; the pressure sensor monitors pressure changes inside the press using a strain gauge or piezoelectric sensor; and the stress sensor records the stress generated during forging using a mechanical sensor.

[0067] Data Acquisition Module: The raw signals acquired by the sensors are transmitted in real time through the data acquisition module. Specifically, the data acquisition module includes a signal processing unit, which performs preprocessing such as conversion, amplification, and filtering on analog signals from different sensors to convert them into digital signals. The data from each sensor is processed synchronously in a time sequence, and multi-dimensional data such as temperature, pressure, and stress are aggregated into a preliminary raw data sequence (hereinafter referred to as the raw data sequence).

[0068] Let the temperature be T, the pressure be P, and the stress be σ. The preliminary raw data sequence is as follows: The value of each parameter changes over time. This includes temperature, pressure, and stress data at every moment during the forging process.

[0069] Real-time data recording: The raw data sequence will be recorded in real time, providing basic input data for subsequent data cleaning and processing.

[0070] S120. Noise filtering and outlier removal are performed on the original data sequence, and data standardization is carried out to generate standardized operating condition data.

[0071] Noise filtering and outlier removal: the original data sequence It contains a certain amount of noise and outliers (such as abnormal data caused by sensor malfunction or environmental interference). First, Kalman filtering or wavelet denoising algorithms are used to denoise the data. Denoising is performed to eliminate random noise introduced by sampling errors in the data. Specifically, Kalman filtering minimizes the impact of noise by predicting and updating the signal, thereby obtaining more accurate temperature, pressure, and stress data.

[0072] Outlier Removal: Further, outlier detection is performed on the denoised data. This is done by calculating the statistical characteristics (such as mean, variance, etc.) of each parameter (temperature T, pressure P, and stress σ), and by setting reasonable thresholds (e.g., outlier criteria of mean ± 3 times standard deviation) to filter out values ​​outside the normal range and remove them from the data sequence.

[0073] The normal ranges for temperature T, pressure P, and stress σ are set as follows: By removing outlier data that exceeds these ranges, standardized data with noise and outliers are obtained. .

[0074] Data standardization: To standardize the data and eliminate errors caused by different units of measurement, the z-score standardization method is used. Specifically, standardization is performed by calculating the mean μ and standard deviation σ of each parameter.

[0075]

[0076] in, This is the standardized data, where μ is the mean of temperature, pressure, and stress, and σ is the standard deviation. This step will generate a standardized operating condition data sequence. Each of its parameters has a mean of 0 and a standard deviation of 1.

[0077] Output standardized data: the standardized operating condition data This data will be passed as input to the subsequent feature extraction stage (i.e., module S130).

[0078] S130. Extract key features from the standardized operating condition data sequence, perform feature selection and feature engineering, and obtain high-quality real-time operating condition data.

[0079] Feature extraction: In standardized operating condition data Based on this, key features are extracted using signal processing and data analysis methods. Specifically, Fourier transform, waveform analysis, and other techniques are used to perform frequency and time domain analysis on the time-series data of temperature T, pressure P, and stress σ to extract key feature parameters such as frequency components, amplitude, and phase. Furthermore, statistical features such as the rate of change, maximum value, minimum value, and mean value of each parameter can be extracted by combining dynamic characteristics.

[0080] Feature selection: From the extracted multidimensional features, features with strong correlation to forging process quality are selected. Methods such as principal component analysis (PCA), mutual information method, and correlation analysis are used to screen out important features that can effectively reflect the forging process. For example, in some cases, the rate of temperature change is highly correlated with the material deformation quality, while changes in pressure and stress may have a significant impact on the dimensional accuracy of the finished product.

[0081] Let the extracted features be These features are used for subsequent process modeling.

[0082] Feature engineering: Based on the selected features, further feature engineering processes are performed. In line with the needs of machine learning and deep learning, the original features are fused, transformed, or new features are derived. For example, combined features of temperature, pressure, and stress may be weighted and averaged or synthesized based on historical data and prior knowledge to provide more accurate input for subsequent modeling.

[0083] Generating high-quality real-time operating condition data: Finally, after feature extraction and selection processing, high-quality real-time operating condition data is obtained. This data will be used as input data in the subsequent dynamic process prediction model (S200 module).

[0084] During the implementation of modules S110, S120, and S130, temperature, pressure, and stress data during the forging process are acquired in real time using multimodal sensors (module S110). This data undergoes noise filtering, outlier removal, and standardization (module S120), ultimately extracting the key features that have the greatest impact on the forging process (module S130). The output data from these steps will serve as crucial input data for subsequent deep learning and reinforcement learning algorithms, used to construct dynamic process prediction models and optimize forging process parameters.

[0085] Step S200 includes at least steps S210-S230:

[0086] S210. Based on the real-time operating data and historical data, use a deep learning model to extract key process features in the forging process and construct a process feature dataset.

[0087] Data input: the real-time operating data (Standardized data of temperature T, pressure P, and stress σ) and historical data Historical forging process data, including historical temperature, pressure, and stress parameters, will serve as input data for the deep learning model. Specifically, real-time operating data will be combined with historical data to form a multi-dimensional input dataset. .

[0088]

[0089] Deep learning model training: Using deep learning models such as deep neural networks (DNN), convolutional neural networks (CNN), or long short-term memory networks (LSTM) on the dataset. Training is then conducted. Through training, the deep learning model can extract important process features from the forging process. Specifically, the model gradually adjusts the network parameters through forward and backward propagation mechanisms, enabling it to effectively learn the key factors affecting forging quality and process stability.

[0090] Process feature extraction: The deep learning model, through training, automatically extracts features from the input data. Key process features that help predict the quality and efficiency of the forging process are extracted. Specifically, the extracted features may include temperature change rate, pressure fluctuation amplitude, stress response, etc., which play an important role in optimizing the forging process.

[0091] Output process feature dataset: The process feature dataset consists of key process features extracted by the deep learning model. This dataset will be used for subsequent process modeling and prediction.

[0092]

[0093] in, These represent key process characteristics such as temperature, pressure, and stress. This dataset provides input for subsequent dynamic process prediction models.

[0094] S220. Combining expert knowledge and prior information, process characteristic data and historical process data are integrated to construct a dynamic process prediction model, perform process modeling, and obtain a process parameter prediction sequence.

[0095] Introduction of expert knowledge and prior information: Combining expert knowledge and industry experience in the forging field to further refine process characteristic data. This process involves processing prior knowledge, such as the processing range of specific temperatures, pressures, and stresses, or the patterns of process parameter variations from historical experience, to guide the model's learning process. For example, experts may already know the deformation behavior of a material under specific pressures and temperatures; therefore, this knowledge can be incorporated into the model as prior information.

[0096] Integrating historical process data: Integrating historical process data (Including process parameters of past forging processes and their corresponding quality results) and process characteristic data Data fusion is then performed to further enhance the predictive power of the model. Methods such as weighted averaging, concatenation, or additive models can be used to combine historical process data with current real-time operating data to obtain a fused dataset. .

[0097]

[0098] Constructing a dynamic process prediction model: utilizing the fused dataset A dynamic process prediction model is constructed using time-series models such as recurrent neural networks (RNNs) or long short-term memory networks (LSTMs) in deep learning. This model can not only predict the changes of key parameters in the forging process based on the current process characteristics, but also capture the temporal correlations in the process and predict the trend of process parameter changes over a certain period of time in the future.

[0099] Set the model output as This refers to the future predicted values ​​of process parameters such as temperature, pressure, and stress during the forging process.

[0100] Output process parameter prediction sequence: After model training, the dynamic process prediction model will output the predicted process parameter sequence. This sequence will describe the changing trends of parameters such as temperature, pressure, and stress during the forging process within a future time window.

[0101] in, These are the predicted temperature, pressure, and stress parameters, representing the changes in process parameters over a future time period.

[0102] S230. Based on the dynamic process prediction model and the extracted process features, generate future predicted values ​​for parameters such as temperature, pressure, and stress during the forging process, and provide an adjustable parameter prediction range.

[0103] Generating future values ​​based on a prediction model: The dynamic process prediction model generates future values ​​based on input process feature data. and the merged data Based on this, future predicted values ​​of parameters such as temperature, pressure, and stress during the forging process are generated. Specifically, the predicted values ​​are based on the temporal characteristics of the model and can reflect the trend of changes in various parameters over time during the forging process.

[0104] Prediction range adjustment and control: Based on feedback from real-time operating data and the process range set by experts, the parameter range of the model prediction is adjusted. For example, the temperature variation range can be set to 900℃ to 1200℃ using expert knowledge, and the pressure variation range can be set to 50MPa to 100MPa. According to these settings, the predicted values ​​output by the dynamic process prediction model will automatically adjust within the set range, thereby ensuring the rationality of the prediction and its match with actual production conditions.

[0105] The adjusted temperature range is set as follows: Pressure range is The stress range is .

[0106] Output Predicted Values ​​and Adjustment Range: Ultimately, the dynamic process prediction model will generate future predicted values ​​during the forging process. In conjunction with the adjustment range set by experts, an adjustable parameter prediction range is provided. This output value will be used for subsequent process optimization and closed-loop control to ensure that the forging process is carried out within a reasonable parameter range.

[0107]

[0108] in, The adjusted predicted values ​​ensure that all process parameters meet the control requirements of the forging process.

[0109] In modules S210, S220, and S230, based on real-time operating condition data... By extracting key process features through deep learning models, a dynamic process prediction model is constructed and optimized using expert knowledge. This ultimately generates predicted future values ​​for parameters such as temperature, pressure, and stress during the forging process. These predicted values ​​and their adjustable range will be used for subsequent process optimization and real-time adjustments to ensure the stability and quality of the forging process.

[0110] Step S300 includes at least steps S310-S330:

[0111] S310: Input the predicted sequence of key process parameters into the reinforcement learning model, and dynamically adjust it through the reinforcement learning algorithm to obtain the optimal combination of process parameters.

[0112] Reinforcement learning model construction: The reinforcement learning model consists of a state space, an action space, and a reward function. The state space is composed of the input process parameter sequence, including the predicted values ​​from modules S220 and S230 (i.e., ... In this step, the reinforcement learning model observes the current state S(t) (i.e., the current temperature, pressure, stress, etc.) and determines how to adjust the process parameters to optimize the forging process.

[0113] Definition of state space and action space: Specifically, the state space S(t) can be represented as:

[0114]

[0115] Where T(t), P(t), and σ(t) are the actual values ​​of temperature, pressure, and stress at the current time t. The action space A(t) is a set of adjustable process parameters, typically including temperature. .

[0116] Dynamic Adjustment and Strategy Optimization: The reinforcement learning algorithm calculates the optimal action a(t) based on the current state S(t) and the strategy π(S(t)), which involves adjusting the specific values ​​of temperature, pressure, and press speed. The specific control strategy is as follows:

[0117]

[0118] Reinforcement learning algorithms optimize π(S(t)) through a continuous training process to find an optimal combination of parameters that can be continuously improved through multiple trials. The optimization objective is to maximize a certain reward function (e.g., improve process stability, save energy, or product quality).

[0119] Optimal process parameter combination output: The reinforcement learning algorithm outputs the optimal process parameter combination based on the policy. This refers to the specific values ​​of the adjusted temperature, pressure, and press speed, representing the most suitable process parameters during forging.

[0120]

[0121] in, These represent the temperature, pressure, and press speed after reinforcement learning optimization at time t, respectively.

[0122] S320: Based on the optimization scheme output by the reinforcement learning model, adjust process parameters such as temperature, pressure, and press speed to obtain the optimal set of process parameters.

[0123] Application of the optimization scheme: Based on the optimal combination of process parameters output by the S310 module. The control system parameters during the actual forging process are adjusted. Specifically, the control system will optimize the temperature... ,pressure and press speed Input into the forging press system.

[0124] Temperature, pressure, and press speed regulation: The control system dynamically adjusts the equipment's operating parameters based on the output optimized by a reinforcement learning algorithm.

[0125] Real-time monitoring and feedback: During the adjustment process, the control system monitors process parameters such as temperature, pressure, and press speed in real time to ensure that the actual values ​​match the optimized values. If the deviation exceeds the predetermined range, readjustment is performed. The real-time feedback system feeds this information back to the reinforcement learning algorithm so that it can further optimize the control strategy.

[0126] S330: Perform accuracy verification and parameter debugging on the optimized process parameter set to generate the final process parameter set.

[0127] Accuracy Verification: After the optimized process parameters are adjusted in the S320 module, the system will verify the accuracy of key process parameters such as temperature, pressure, and press speed. By comparing the results with actual production, the system will evaluate whether the optimized process parameters can achieve the expected results.

[0128] For example, verifying the quality of the product during the forging process and checking for problems such as uneven temperature, unstable pressure, or deviation in press speed.

[0129] At this point, the predicted and actual values ​​of temperature, pressure, and press speed should be consistent. If the deviation is large, the optimization parameters need to be adjusted.

[0130] Parameter adjustment: During the accuracy verification process, if deviations or non-compliance with production requirements are found in certain process parameters, corresponding parameter adjustments will be performed. For example, adjustments to the temperature range may be necessary. Make fine adjustments, or adjust the pressure fluctuation range. This ensures that it meets product requirements.

[0131] Generate the final set of process parameters: Through accuracy verification and parameter tuning, a final set of precisely optimized process parameters is generated. This set represents the final control parameters required in the actual forging process.

[0132]

[0133] in, These are the temperature, pressure, and press speed parameters after accuracy verification and debugging.

[0134] In modules S310, S320, and S330, process parameters are continuously optimized through reinforcement learning models, ultimately generating the optimal set of process parameters. In S310, the reinforcement learning algorithm predicts sequences based on key process parameters. The parameters are dynamically adjusted. In S320, the optimized scheme is applied to the forging control system for parameter adjustment, while in S330, the optimized process parameters are verified and debugged to generate the final set of process parameters. This process ensures the optimization and stability of the forging process.

[0135] Step S400 includes at least steps S410-S430:

[0136] S410: Based on the optimized set of process parameters, input process control commands to the press execution system to adjust the working status of the press in real time.

[0137] Input control commands: The control command sequence C(t) generated in the S400 module will be input to the compressor execution system in real time. Specifically, the command sequence includes the target temperature at time t. Target pressure and target press speed The execution system will adjust the compressor's state based on these inputs. Control commands can then be expressed as follows:

[0138] These parameters dynamically adjust the equipment to ensure the press system operates as planned.

[0139] Press execution system response: Based on the input control commands, the press execution system adjusts its operating status in real time, including the synchronous adjustment of the heating system, pressure regulation system, and press speed regulation system. Temperature, pressure, and speed will be rapidly adjusted under the instructions of the control system to achieve optimal process conditions.

[0140] Feedback status monitoring: At the same time, the press execution system monitors the current temperature T(t), pressure P(t) and press speed v(t), and feeds the monitoring data back to the control system to ensure that the equipment is adjusted according to the target process parameters.

[0141] S420: Monitors the operating status of the press in real time and feeds the monitoring data back to the closed-loop control system to dynamically adjust the control parameters.

[0142] Operational Status Monitoring: After the control commands input into the S410 module drive the press to execute the system operation, the system begins to monitor the press's operating status in real time. Specifically, the monitoring data includes the press's real-time temperature T(t), real-time pressure P(t), and real-time press speed v(t), which are collected in real time by sensors.

[0143] Condition monitoring feedback: The temperature, pressure, and compressor speed obtained from real-time monitoring will be input as feedback data F(t) into the closed-loop control system. This feedback data can be expressed as:

[0144]

[0145] in, , , These are the temperature, pressure, and press speed measured by the sensor at time t, respectively.

[0146] Dynamic adjustment of control parameters: Based on the feedback data F(t), the control system dynamically adjusts the current temperature T(t), pressure P(t), and compressor speed v(t). If the feedback data deviates from the target value, the system will correct the control parameters based on the set control algorithm (such as proportional-integral-derivative control).

[0147] Specifically, if the deviations in temperature, pressure, or speed are too large, the control system will adjust the corresponding control commands in real time to reduce the deviations and ensure that the equipment remains in an optimized operating state. These adjustments can be expressed by the control parameter adjustment formulas as follows:

[0148]

[0149] Where ΔC(t) represents the control command after dynamic adjustment based on the deviation.

[0150] S430: Based on the real-time feedback data, generate a closed-loop control command sequence to continuously adjust key process parameters such as temperature, pressure, and speed.

[0151] Real-time feedback data processing: Based on the real-time feedback data F(t) obtained from the S420 module, the control system analyzes the compressor's operating status in real time and assesses deviations. According to the control strategy, the control system calculates new control commands. This is to achieve continuous optimization of process parameters.

[0152] Generate closed-loop control command sequence: In the S430 module, based on feedback data, the control system will regenerate the closed-loop control command sequence Cloop(t), which represents the new target values ​​for temperature, pressure, and compressor speed. The specific control commands can be expressed as:

[0153]

[0154] in, The temperature, pressure, and press speed were adjusted based on feedback data.

[0155] Continuous adjustment of process parameters: generating closed-loop control command sequence Subsequently, the system continuously adjusts key process parameters such as temperature, pressure, and press speed according to this sequence to maintain the equipment operating in optimal condition. This process is a dynamic adjustment process, ensuring that the equipment can always respond to environmental changes and correct operating parameters in real time.

[0156] Step S500 includes at least steps S510-S530:

[0157] S510: Based on the real-time operating data and operating status sequence, the operating status of the equipment is analyzed using a time-series prediction model to identify potential fault risks.

[0158] Fault risk identification: In the S500 module, time series prediction model It provides predictions of the future state of the equipment. By comparing actual operating data The difference between the predicted and actual values ​​can identify potential equipment failure risks. Specifically, a failure identification algorithm is used to determine potential failure risks based on the equipment's operating deviations ΔT(t), ΔP(t), and Δv(t). For example:

[0159]

[0160] Where R(t) is the fault risk value of the equipment at time t, and f(⋅) represents the fault identification function. This function can be a simple weighted sum, or it can be dynamically adjusted according to the risk level of different process parameters.

[0161] Analysis of prediction results: The analysis results of R(t) are further processed to output the probability of failure. If R(t) reaches a certain threshold... If this is the case, it indicates that the equipment has a potential risk of failure and requires further handling.

[0162] S520: Perform anomaly detection and fault identification on the prediction results, generate fault warning information, and provide a reference for equipment fault repair.

[0163] Anomaly Detection and Fault Identification: The fault risk value R(t) generated in S510 is used to detect anomalies in the equipment's operating status. By detecting the deviation between the real-time operating status F(t) and the predicted status, it is determined whether an anomaly has occurred. The anomaly detection method can determine the existence of anomalies based on a set threshold.

[0164]

[0165] Where ϵ is a set threshold. When ΔR(t) exceeds this threshold, it indicates that the equipment has malfunctioned or is in an abnormal state.

[0166] Generating fault warning information: When the fault risk of the equipment exceeds the set threshold R_{th}, the system will generate fault warning information W(t). Specifically, the fault warning information can be represented as:

[0167]

[0168] in, These are fault warning messages for temperature, pressure, and press speed, respectively.

[0169] Fault Repair Reference: The generated fault warning information provides a reference for equipment maintenance personnel, helping to identify the possible sources of equipment failure and supporting subsequent fault repair work. Appropriate repair plans are developed based on different fault types.

[0170] S530: Based on the fault warning information, adjust the process parameters using an adaptive strategy or start the automatic repair mode to dynamically correct the process parameters.

[0171] Adaptive strategy for adjusting process parameters: Based on the fault warning information W(t) generated in S520, the system can automatically adjust process parameters or activate the repair mode. Specifically, if a fault warning signal indicates that a certain process parameter (such as temperature, pressure, or press speed) is abnormal, the system will adjust the corresponding process parameter through an adaptive strategy. For example, if... If the temperature is abnormal, the system will automatically adjust the target temperature. To correct for temperature deviations, the specific adjustment formula is as follows:

[0172]

[0173] in, This is the temperature correction amount that is dynamically adjusted based on fault warning information.

[0174] Automatic Repair Mode Activation: In certain situations, a fault warning message W(t) indicates a serious equipment failure, and the system will automatically activate the repair mode. Automatic repair mode may include stopping equipment operation, initiating emergency cooling, or other repair measures. The specific repair mode can be determined based on the fault type, and the corresponding repair operation will be initiated.

[0175] Dynamic process parameter correction: Based on fault warning information and adaptive strategy adjustments, the system dynamically corrects process parameters to ensure that the equipment can return to normal operation. The dynamic correction process includes adjusting temperature T(t), pressure P(t), and press speed v(t) to restore the equipment to its optimal operating state.

[0176] In modules S510, S520, and S530, the operating status of the equipment is first analyzed using a time-series prediction model to identify potential failure risks, and anomaly detection and fault identification are performed based on the prediction results. Next, based on fault warning information, the system adjusts process parameters or initiates an automatic repair mode through adaptive strategies to ensure that the equipment can quickly recover to optimal operating conditions in the event of a fault. All steps are closely linked and rely on real-time data and feedback information acquired from previous modules (such as S420).

[0177] The key innovations of this invention include:

[0178] (1) Data acquisition and processing technology based on multimodal sensors. The innovative combination of multimodal sensors with data cleaning and standardization generates high-precision real-time operating condition data, effectively avoiding process parameter deviations caused by data noise or outliers.

[0179] (2) Construction of dynamic process prediction model. Deep learning algorithms (such as recurrent neural networks, RNN) are used to model real-time operating data. By combining historical data and prior knowledge, the changing trend of key parameters is accurately predicted, thus solving the nonlinear coupling problem under complex operating conditions.

[0180] (3) Process parameter optimization based on reinforcement learning. By dynamically adjusting the predicted parameters through reinforcement learning algorithms, the combination of process parameters is optimized, thereby improving the adaptability and optimization efficiency of equipment and processes during production.

[0181] The following are its main beneficial effects:

[0182] (1) High-precision real-time operating condition data acquisition and processing. This invention acquires multi-dimensional real-time operating condition data such as temperature, pressure, and stress based on multi-modal sensors, and generates standardized high-quality operating condition data through noise filtering and outlier removal techniques. Compared with the traditional fixed parameter setting method, the real-time acquired and processed data is more accurate and comprehensive, laying the foundation for subsequent dynamic optimization.

[0183] (2) Dynamic prediction and optimization improve process accuracy. By using deep learning models to extract key process features and combining historical data and expert knowledge to construct dynamic process prediction models, it is possible to accurately predict the future trend of process parameters. By optimizing parameters such as temperature, pressure and press speed during the forging process through reinforcement learning models, the problem that traditional methods are difficult to adapt to complex dynamic working conditions is solved, and the consistency of forging quality is greatly improved.

[0184] (3) Closed-loop control reduces the risk of equipment failure. Based on real-time monitoring of equipment operating status and generation of closed-loop control command sequences, the operating parameters of the equipment can be dynamically adjusted to respond to equipment anomalies in real time and reduce the probability of equipment failure. Compared with traditional manual intervention or fixed rule control methods, this invention significantly improves the continuity and reliability of production.

[0185] Example 2: Figure 2 A structural block diagram of the intelligent control system for a vacuum isothermal forging press according to an embodiment of the present invention is shown. Figure 2 As shown, the system may include:

[0186] The data acquisition module 10 collects temperature, pressure, and stress data from multimodal sensors during the forging process.

[0187] The sensors include thermocouples, infrared detectors, strain gauges, and piezoelectric sensors, which acquire real-time temperature, pressure, and stress data of the press cavity.

[0188] The data undergoes preprocessing such as conversion, amplification, and filtering by the signal processing unit, and is then stored as a preliminary raw data sequence. ={T,P,σ}.

[0189] The data processing module 20 cleans and standardizes the collected data and generates high-quality operating condition data.

[0190] Using Kalman filtering or wavelet denoising algorithms to denoise data sequences Noise filtering is performed to eliminate outliers caused by sensor errors or environmental interference.

[0191] The mean μ and standard deviation σ of the data are calculated using the z-score standardization method to obtain the standardized data sequence: .

[0192] The feature extraction and dynamic process prediction module 30 extracts key process features from standardized data and uses a deep learning model to build a dynamic process prediction model.

[0193] Feature extraction: Fourier transform and principal component analysis were used to extract key features such as temperature change rate, pressure fluctuation amplitude and stress response.

[0194] Process prediction model: based on real-time operating data and historical data Predicting future process parameters using a recurrent neural network (RNN) model: .

[0195] The process parameter optimization module 40 uses a reinforcement learning model to optimize the process parameters of the forging process.

[0196] A reinforcement learning model defines a state space S, an action space A, and a reward function R. The state space S contains the prediction parameters. The action space A is the temperature, pressure, and speed adjustment amounts ΔT, ΔP, and Δv.

[0197] The optimal set of parameters is output by dynamically adjusting the process parameters through strategy π(S).

[0198] The closed-loop control and equipment management module 50 enables real-time monitoring of equipment operating status and generates closed-loop control commands.

[0199] Real-time monitoring: Collect the actual operating status of the equipment through sensors F(t)={T(t),P(t),v(t)}.

[0200] Instruction generation: based on feedback data F(t) and optimization parameters Adjust the closed-loop control commands:

[0201] .

[0202] Beneficial effects of the embodiments:

[0203] Data processing and feature extraction: Real-time acquisition and processing of high-precision operating data by multimodal sensors significantly improves data quality and reliability.

[0204] Dynamic prediction and optimization: By combining deep learning and reinforcement learning algorithms, the forging process parameters are dynamically adjusted, which effectively improves the quality consistency and production efficiency of forgings.

[0205] Closed-loop control and fault management: The system has real-time monitoring and closed-loop control functions, which can detect and repair faults in a timely manner, reduce equipment downtime and extend equipment life.

[0206] Intelligent management throughout the entire process: From data acquisition to process parameter optimization and equipment management, intelligent control of the entire forging process is achieved, which meets the requirements of intelligent manufacturing in Industry 4.0.

[0207] Example 3: Figure 3 This is a structural block diagram of the intelligent control device for a vacuum isothermal forging press provided in an embodiment of the present invention. Figure 3 As shown, the device includes:

[0208] The data acquisition module (module 101) is used to acquire multimodal data during the forging process in real time, including key process parameters such as temperature, pressure and stress.

[0209] Temperature (T), pressure (P), and stress (σ) data in the forging environment are acquired using multimodal sensors such as thermocouples, strain gauges, and infrared detectors.

[0210] The data is transformed, amplified, and filtered by the signal processing unit to generate a preliminary raw data sequence. .

[0211] The data processing and feature extraction module (module 102) performs noise filtering, outlier removal and standardization on the collected data, and extracts key process features.

[0212] Kalman filtering is used to remove noise, and statistical methods are combined to eliminate outliers.

[0213] The cleaned data is then standardized to generate standardized data. .

[0214] Extract key features (such as temperature change rate, pressure fluctuation amplitude, and stress response) and generate a feature dataset. .

[0215] The dynamic process prediction module (module 103) uses deep learning models and expert knowledge to build a dynamic process prediction model to predict key process parameters in the forging process.

[0216] Input standardized data and historical data .

[0217] A dynamic prediction model is constructed using a recurrent neural network (RNN) to output a sequence of future process parameters. .

[0218] The process parameter optimization module (module 104) optimizes key process parameters in the forging process based on reinforcement learning algorithms.

[0219] Input prediction parameters As the state space, the action space is defined as the adjustment amounts of temperature, pressure, and press speed.

[0220] The reinforcement learning strategy π(S) is used to dynamically optimize process parameters and output the optimal set of process parameters.

[0221] The closed-loop control module (module 105) performs real-time control of the press based on the optimized set of process parameters and generates closed-loop control commands.

[0222] Monitor the equipment's operating status F(t)={T(t),P(t),v(t)} and provide feedback data.

[0223] Generate a closed-loop control command sequence based on feedback data: .

[0224] Beneficial effects of the embodiments:

[0225] Precise data acquisition and processing. This invention uses multimodal sensors to acquire and process forging data in real time, ensuring the accuracy and completeness of the working condition data and providing high-quality data support for subsequent prediction and optimization.

[0226] Dynamic prediction and optimization. By combining deep learning models and reinforcement learning algorithms, the system can dynamically predict the changing trends of process parameters during forging and optimize these parameters in real time, thereby improving the consistency and accuracy of forged products.

[0227] Intelligent closed-loop control. The system uses a closed-loop control module to achieve real-time monitoring and dynamic adjustment, ensuring that the equipment is always in optimal operating condition and effectively reducing human intervention.

[0228] Full-process automation and intelligence. This device integrates data acquisition, feature extraction, parameter prediction, optimization adjustment, and closed-loop control, realizing full-process automation and intelligence in the forging process, significantly improving production efficiency and equipment utilization.

[0229] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A method for intelligent control of a vacuum isothermal forging press, characterized in that, The method includes: Based on the acquisition of temperature, pressure, and stress data during the forging process using multimodal sensors, a preliminary raw data sequence is generated; specifically, this includes: acquiring temperature data through thermocouples and infrared detectors; acquiring pressure data through strain gauges; and acquiring stress data through mechanical sensors. The original data sequence is cleaned and standardized to generate standardized real-time operating condition data. Specifically, this includes: using a Kalman filter algorithm to filter noise from the original data sequence; detecting and removing outliers to obtain cleaned data; and standardizing the cleaned data based on the z-score standardization method to generate standardized real-time operating condition data. Based on real-time operating data and historical data, a deep learning model is used to extract key process features, construct a dynamic process prediction model, and generate a key parameter prediction sequence. The extraction of key process features specifically includes: using Fourier transform to extract key features such as temperature change rate, pressure fluctuation amplitude, and stress response based on standardized real-time operating data; and using principal component analysis to filter the extracted features and generate a process feature dataset. The construction of the dynamic process prediction model specifically includes: using a recurrent neural network model to establish a dynamic process prediction model; inputting the process feature dataset to predict key parameters in the forging process and generate a key parameter prediction sequence. The predicted sequence of key parameters is input into a reinforcement learning model to optimize the process parameters of temperature, pressure, and press speed during forging, generating an optimized set of process parameters. Specifically, this includes: using the predicted sequence of key parameters as the state input of the reinforcement learning model; optimizing the process parameter adjustment strategy through a reinforcement learning algorithm to generate the optimal combination of process parameters; and outputting the optimized temperature, pressure, and press speed parameters. Based on the optimized set of process parameters, the press execution system is dynamically adjusted, and the equipment operating status is monitored in real time to generate a closed-loop control command sequence. Specifically, this includes: collecting real-time operating status data of the equipment, including real-time temperature, pressure, and press speed; generating a closed-loop control command sequence based on feedback data and the optimized set of process parameters; and continuously adjusting the temperature, pressure, and press speed during the forging process.

2. The intelligent control method for a vacuum isothermal forging press according to claim 1, characterized in that, Also includes: Analyze equipment operating status using time-series prediction models to generate fault early warning information; Based on the fault warning information, the process parameters are dynamically adjusted.

3. An intelligent control system for a vacuum isothermal forging press, applied to the method described in any one of claims 1-2, characterized in that, include: The data acquisition module is used to collect temperature, pressure, and stress data from multimodal sensors during the forging process and generate raw data sequences. The data processing module is used to clean and standardize the original data sequence to generate standardized real-time operating condition data. The dynamic process prediction module is used to extract key process features based on the real-time operating data and historical data, construct a dynamic process prediction model, and generate a key parameter prediction sequence. The process parameter optimization module is used to optimize the process parameters based on the predicted sequence of the key parameters and use a reinforcement learning model to generate an optimized set of process parameters. The closed-loop control module is used to generate a closed-loop control command sequence based on the optimized set of process parameters, and to dynamically adjust the press execution system.

4. An intelligent control device for a vacuum isothermal forging press, characterized in that, The device includes a data acquisition module, a data processing and feature extraction module, a dynamic process prediction module, a process parameter optimization module, and a closed-loop control module. The device can implement the steps of the intelligent control method for a vacuum isothermal forging press as described in any one of claims 1 to 2.

Images