Intelligent control system and method for sintering temperature of powder metallurgy green body

Abstract

The invention discloses an intelligent control system and method for the sintering temperature of a powder metallurgy green body, and belongs to the technical field of powder metallurgy, and the system comprises a data collection unit, a data processing center, a data storage unit, a temperature adjustment execution mechanism, a data display unit and a data communication unit. A temperature prediction model constructed by the data processing center is composed of three core modules, namely a topology-aware multi-scale feature extraction module, a chaos dynamics time sequence dependency recognition module and a multi-source information fusion dynamic attention module, complex correlation among parameters is reserved through topology analysis, temperature nonlinear dynamic characteristics are captured through the chaos theory, and the temperature prediction accuracy is improved. A multi-dimensional dynamic attention mechanism pays attention to key information, high-precision prediction and control over the sintering temperature are achieved, and the system has the self-adaptive learning capacity and can automatically adjust the control strategy according to different materials and process conditions.

Description

Technical Field

[0001] This invention relates to the field of powder metallurgy technology, and in particular to an intelligent control system and method for the sintering temperature of powder metallurgy billets, which is especially suitable for precise control of the temperature of powder metallurgy billets during the sintering process. Background Technology

[0002] Powder metallurgy is a process technology that manufactures parts by forming and sintering metal or alloy powders. In powder metallurgy production, precise control of the sintering temperature has a decisive impact on product quality. Traditional sintering temperature control mainly relies on PID controllers, adjusting the heating power based on the deviation between the set temperature and the actual temperature. However, this method struggles to handle the complex nonlinear characteristics and multivariate coupling problems during the sintering process.

[0003] With the development of Industry 4.0, intelligent manufacturing technologies are gradually being applied to traditional manufacturing fields. Currently, research has applied methods such as neural networks and fuzzy logic to sintering temperature control, but these methods still have the following shortcomings: first, they lack the ability to effectively integrate multi-source heterogeneous data during the sintering process; second, they struggle to capture the complex temporal dependencies in the temperature change process; and third, they cannot adapt to the temperature control requirements of different materials and process conditions. Therefore, developing an intelligent system capable of precisely controlling the sintering temperature of powder metallurgy billets has significant practical implications. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent control system and method for the sintering temperature of powder metallurgy billets, which aims to solve the problems of insufficient sintering temperature control accuracy, poor adaptability, and low energy utilization in the prior art.

[0005] This invention proposes an intelligent control system for the sintering temperature of powder metallurgy billets, comprising: The data acquisition unit is used to acquire multi-source heterogeneous data inside the reactor in real time. The multi-source heterogeneous data includes gas temperature value, gas flow rate value, reactor electric heating data, reactor temperature regulation mechanism action value, billet temperature value, reactor internal temperature value, and gas concentration value. The data processing center, connected to the data acquisition unit, is used to preprocess and filter the multi-source heterogeneous data, and to build a temperature prediction model based on the preprocessed data. The temperature prediction model includes a topology-aware multi-scale feature extraction module, a chaotic dynamics time-series dependency identification module, and a dynamic attention module for multi-source information fusion. A data storage unit, connected to the data processing center, is used to store the multi-source heterogeneous data and the training sample data of the temperature prediction model; A temperature regulating actuator, connected to the data processing center, is used to adjust the temperature of the reactor according to the output of the temperature prediction model. The data display unit, connected to the data processing center, is used to visually display temperature prediction results and control process data. The data communication unit is used to realize data interaction between the data acquisition unit, the data processing center, the data storage unit, the temperature regulation actuator, and the data display unit.

[0006] Preferably, the temperature regulation actuator includes a heating coil, a furnace door temperature detector, and a temperature regulation controller. The temperature regulation controller receives heating data from the heating coil and temperature data inside the reactor, and performs temperature regulation based on the received data. The temperature regulation includes the calculation and application of a first temperature correction value and a second temperature correction value. The second temperature correction value takes effect when the first temperature correction value is close to zero.

[0007] Preferably, the topology-aware multi-scale feature extraction module includes: A feature space construction unit is used to construct a high-dimensional feature space from the multi-source heterogeneous data. The topology analysis unit is used to calculate the strength of topology relationships between different parameters in the high-dimensional feature space. A multi-scale feature extraction unit is used to perform hierarchical decomposition and extraction of features based on the strength of the topological relationship; Feature dimensionality reduction and preservation units are used to reduce feature dimensionality while preserving key topological information.

[0008] Preferably, the chaotic dynamics time-series dependency identification module includes: The time series data segmentation unit is used to divide temperature time series data into three time windows: short-term, medium-term, and long-term. The phase space mapping unit is used to map the time series data of each time window to the phase space; The system stability analysis unit is used to calculate system invariants in phase space and evaluate system stability. The inflection point identification unit is used to identify key inflection points and steady-state regions of temperature changes; Temporal dependency network building blocks are used to build dependency networks that represent the temperature change patterns at different time scales.

[0009] Preferably, the dynamic attention module for multi-source information fusion includes: Information representation unified unit, used to convert data from different sources and in different formats into a unified representation format; Multi-level fusion units are used to fuse information at three levels: parameter level, parameter group level, and global level. The dynamic attention calculation unit is used to calculate attention weights in the time dimension, feature dimension, and spatial dimension based on the current system state. Uncertainty quantification unit is used to construct a probabilistic graphical network to quantify the uncertainty of prediction results and generate confidence intervals.

[0010] Preferably, the data processing center further includes: The data preprocessing module is used to perform data cleaning, data transformation, and data normalization on the multi-source heterogeneous data; The abnormal data processing and filtering module is used to detect and remove abnormal data; The training data extraction module is used to extract temperature regulation-related data from the preprocessed data as a training dataset. The training data sample construction module is used to construct training samples that include input data and output data. A data model training module is used to train the temperature prediction model based on the training samples. The data model testing and evaluation module is used to test and evaluate the temperature prediction model to obtain the optimal model.

[0011] Preferably, the data preprocessing module performs data cleaning according to the following rules: For missing values, depending on the type of missing object, the missing values ​​are filled with the mean or median of the missing data for that type. For abnormal samples, delete the detected abnormal samples; For outliers, use numbers in the range of the third-quarter to the first-quarter digits within the group containing the object to estimate the outlier.

[0012] Preferably, the anomaly processing and filtering module performs anomaly detection on the data by constructing an isolated forest model, including: Use all feature data after dimensionless processing to perform preliminary screening of outlier data; Cluster the data after initial screening; Construct an isolated forest model based on the clustering results; The isolated forest model is used to further filter out outlier data, and non-outlier data is used as the training dataset.

[0013] Preferably, the input data vector constructed by the training data sample construction module includes the furnace temperature value, billet temperature value, heating coil temperature value, input furnace gas temperature value, input furnace gas concentration value, gas flow rate value, and information on whether there is human intervention. The output data includes the furnace door temperature value, output furnace gas temperature value, and output furnace gas concentration value.

[0014] A control method using the aforementioned intelligent control system for sintering temperature of powder metallurgy billets includes the following steps: (1) The data acquisition unit acquires multi-source heterogeneous data inside the reactor in real time and transmits the multi-source heterogeneous data to the data processing center; (2) The data processing center preprocesses the multi-source heterogeneous data, including data cleaning, data transformation and data normalization; (3) The data processing center uses the abnormal data processing and filtering module to detect and filter the preprocessed data to obtain non-abnormal data as the training dataset. (4) The data processing center extracts temperature regulation-related data from the training dataset and constructs training samples containing input data vectors and output data; (5) The data processing center uses the training samples to train the temperature prediction model, which includes a topology-aware multi-scale feature extraction module, a chaotic dynamics time-series dependency recognition module, and a dynamic attention module for multi-source information fusion. (6) The topology-aware multi-scale feature extraction module extracts multi-scale features that preserve topological relationships from the multi-source heterogeneous data; (7) The chaotic dynamics time-series dependency identification module analyzes the nonlinear dynamic characteristics of the temperature time-series data and identifies key turning points in temperature changes; (8) The dynamic attention module for multi-source information fusion integrates features from different sources and focuses on key information through a multi-dimensional attention mechanism; (9) The data processing center predicts the future temperature under the current condition based on the temperature prediction model; (10) The temperature regulation actuator calculates the temperature correction amount and correction curve based on the prediction results; (11) The temperature regulating actuator adjusts the reactor temperature according to the temperature correction amount and the correction curve; (12) The data display unit displays the prediction results and control process data in real time.

[0015] This invention integrates topology, chaos theory, and deep learning techniques to construct a deep neural network temperature prediction model based on multi-dimensional tensor analysis, achieving high-precision prediction and control of sintering temperature. This system offers the following advantages: 1. Significantly improves temperature control accuracy, reducing control error from ±5-8℃ in traditional systems to within ±2℃, reducing temperature fluctuation by approximately 65%, and improving product quality consistency; 2. By accurately identifying the temperature change patterns and key inflection points during the sintering process, the system can predict and adjust the control strategy in advance, reducing temperature over-adjustment and under-adjustment, and improving the control response speed by approximately 40%. 3. The optimized temperature control strategy significantly reduced energy consumption, with energy consumption in the sintering process reduced by approximately 22% and gas consumption reduced by approximately 18%, thus improving the economics of production; 4. The system has adaptive learning capabilities, which can automatically adjust the control strategy according to different materials and process conditions, significantly improving the system's versatility and adaptability; 5. By effectively integrating and utilizing multi-source heterogeneous data during the sintering process, the system provides comprehensive process monitoring and quality traceability capabilities, which is helpful for process improvement and quality control. Attached Figure Description

[0016] Figure 1 This is a structural framework diagram of an intelligent control system for sintering temperature of powder metallurgy billets according to the present invention. Figure 2 This is a functional module structure diagram of the data processing center of the present invention; Figure 3 This is a schematic diagram of the structure of the topology-aware multi-scale feature extraction module of the present invention; Figure 4 This is a schematic diagram of the structure of the chaotic dynamics time-series dependency identification module of the present invention; Figure 5 This is a schematic diagram of the structure of the dynamic attention module for multi-source information fusion of the present invention; Figure 6 This is a schematic diagram of the temperature regulation actuator of the present invention; Figure 7 This is a flowchart of the control method of the present invention. Detailed Implementation

[0017] Please refer to Figure 1 - Figure 7 The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Those skilled in the art should understand that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention.

[0018] Reference Figure 1 The present invention provides an intelligent control system for the sintering temperature of powder metallurgy billets, including a data acquisition unit 1, a data processing center 2, a data storage unit 3, a temperature regulation actuator 4, a data display unit 5, and a data communication unit 6.

[0019] Data acquisition unit 1 is used to acquire multi-source heterogeneous data inside the reactor in real time. This multi-source heterogeneous data includes key parameters such as gas temperature, gas flow rate, reactor electric heating data, reactor temperature control mechanism operation values, billet temperature, reactor internal temperature, and gas concentration. Preferably, this data is acquired in discrete time series format, with a sampling frequency set to 5 times / second. This frequency can fully capture the temperature change characteristics during sintering without generating excessive redundant data. In practical applications, this sampling frequency can be adjusted according to different sintering processes. For the rapid heating stage, it can be increased to 10 times / second; while for the isothermal stage, it can be reduced to 2 times / second, achieving optimized resource allocation.

[0020] Data processing center 2 is connected to data acquisition unit 1 via a high-speed data bus. Data processing center 2 is mainly responsible for preprocessing and filtering the multi-source heterogeneous data, and constructing a temperature prediction model based on the preprocessed data. This temperature prediction model includes three key modules: a topology-aware multi-scale feature extraction module 21, a chaotic dynamics time-series dependency identification module 22, and a multi-source information fusion dynamic attention module 23. These three modules work together to achieve accurate prediction of sintering temperature changes.

[0021] Data storage unit 3 is connected to data processing center 2 and is used to store multi-source heterogeneous data and training sample data for temperature prediction models. In addition, this unit also stores historical sintering data, model parameters, and system configuration information, providing data support for model training and system optimization. In actual sintering production line applications, a typical powder metallurgy sintering system can generate approximately 20GB of raw data per day. After initial compression and screening, about 5GB of this data requires long-term storage. Therefore, data storage unit 3 adopts a hierarchical storage strategy: hot data (recent data) is stored in high-speed storage media, and cold data (historical data) is stored in large-capacity storage media, ensuring both system performance and cost-effectiveness.

[0022] Temperature control actuator 4 is connected to data processing center 2 via an industrial control network to adjust the furnace temperature based on the output of the temperature prediction model. In typical powder metallurgy sintering applications, the temperature control response time is no more than 2 seconds, and the control accuracy can reach within ±2℃, meeting the process requirements of most powder metallurgy products.

[0023] Data display unit 5 is connected to data processing center 2 via data communication unit 6, used to visualize temperature prediction results and control process data. This unit provides an intuitive graphical interface, facilitating operator monitoring and analysis of the sintering process. In practical applications, operators can use data display unit 5 to view temperature curves, comparisons between predicted and actual values, and trends in key parameters in real time, providing decision support for process adjustments and quality control.

[0024] Data communication unit 6 serves as the connecting link of the entire system, enabling data interaction between data acquisition unit 1, data processing center 2, data storage unit 3, temperature control actuator 4, and data display unit 5. This unit supports multiple industrial communication protocols, such as Modbus, PROFIBUS, and OPCUA, ensuring seamless connection and efficient communication between all parts of the system. Data transmission rates are typically above 10Mbps, meeting the real-time control requirements of the system.

[0025] Reference Figure 6 This embodiment details the structure and working principle of the temperature regulating actuator 4. The temperature regulating actuator 4 includes a heating coil 41, a furnace door temperature detector 42, and a temperature regulating controller 43. The temperature regulating controller 43 receives heating data from the heating coil 41 and temperature data inside the reactor furnace, and performs temperature regulation based on the received data.

[0026] The temperature control process involves two key steps: the calculation and application of a first temperature correction and a second temperature correction. In powder metallurgy sintering, temperature control requires simultaneous consideration of accuracy and stability. The system first calculates the first temperature correction based on the current temperature error and its trend, and applies it to temperature regulation. Subsequently, the system continues to monitor temperature changes and calculates the second temperature correction. When the first temperature correction approaches zero (typically defined as an absolute value less than 0.5℃), the second temperature correction takes effect, further refining the temperature adjustment.

[0027] In a typical iron-based powder metallurgy sintering process, the sintering temperature needs to be raised from room temperature to 1120℃ and held for 2 hours before slow cooling. During the heating phase, the first temperature correction primarily addresses large temperature adjustments, with a correction amount of 10-15℃. In the isothermal phase, the second temperature correction plays a major role, with a correction amount typically between 0.5-2℃, ensuring the temperature remains stable around the target value. This two-stage control strategy significantly improves the accuracy and stability of temperature control, making the sintering process more controllable and the product quality more consistent.

[0028] Reference Figure 3The topology-aware multi-scale feature extraction module 21 includes a feature space construction unit 211, a topology relationship analysis unit 212, a multi-scale feature extraction unit 213, and a feature dimensionality reduction and preservation unit 214.

[0029] The feature space construction unit 211 is used to construct a high-dimensional feature space from multi-source heterogeneous data. In its implementation, this unit first standardizes different types of data to eliminate dimensional differences, and then maps them to a unified high-dimensional feature space. For example, different normalization methods are used for temperature data, flow rate data, and gas concentration data. 1. Temperature data are normalized using maximum-minimum values: , in, This is the original temperature value. The normalized temperature value. and These represent the minimum and maximum temperatures, respectively. In the sintering of copper-based powder metallurgy billets, It is usually set to room temperature 25°C. The maximum sintering temperature is set at 850°C; however, for iron-based powder metallurgy billets... At the same room temperature of 25°C, The temperature could reach as high as 1200°C.

[0030] 2. Traffic data is standardized using Z-score: , in, The original flow rate value. The standardized flow rate value. This represents the average flow rate. This represents the standard deviation of the flow rate. In practical applications, the average gas flow rate is... and standard deviation It is usually determined based on production experience, such as the hydrogen flow rate. Typically 5L / min, Approximately 0.5 L / min; while the nitrogen flow rate It could be as high as 20L / min. Approximately 2 L / min.

[0031] The topology analysis unit 212 is used to calculate the strength of topological relationships between different parameters in a high-dimensional feature space. This unit uses a neighborhood information-preserving matrix to record the topological relationships between parameters, and the matrix elements... This represents the correlation strength between parameters i and j. The correlation strength is calculated as follows: , in, For parameters and parameters The strength of the correlation between them Indicates parameters and The Euclidean distance between them and This is a local-scale parameter used to adjust the decay rate of the correlation strength. In practical applications, for physically closely related parameters, such as furnace interior temperature and furnace door temperature, and Typically, a small value (around 0.1-0.2) is set so that only very close values ​​show a high correlation. For indirectly correlated parameters, such as temperature and gas concentration, and Setting it to a larger value (approximately 0.5-0.8) allows for more lenient association determinations.

[0032] The multi-scale feature extraction unit 213 is used for hierarchical decomposition and extraction of features based on the strength of topological relationships. In the sintering temperature control of powder metallurgy billets, this unit realizes three-level feature extraction: 1. Parameter-level Single Feature Extraction: Extracting the unique characteristics of each parameter, such as the rate of temperature change and fluctuation amplitude. For example, the rate of temperature change... The calculation is as follows: , in, Let t be the temperature value at time t. for Temperature value at time, This is the time interval, usually set to 1 second.

[0033] Parameter combination feature extraction: Extracting the interaction features of related parameter groups. For example, temperature-flow coupling features. , in, The standardized temperature value. The standardized flow rate value. , and The weighting coefficients are determined through data analysis and are typically used in the early stages of sintering. Adjusted to during the later stages of sintering This dynamic adjustment reflects the changing importance of each parameter at different stages.

[0034] 3. System-level global feature extraction: Extracting macroscopic features that reflect the state of the entire sintering system, such as the system energy state. : , in, For the first The standardized values ​​of each parameter These are the weighting coefficients. This represents the total number of parameters. The weight of each parameter is determined based on its impact on the system state. For example, the weight of the temperature parameter is usually 0.4-0.5, the weight of the flow rate parameter is 0.2-0.3, and the weight of the concentration parameter is 0.2-0.25.

[0035] Feature dimensionality reduction and preservation unit 214 is used to reduce the dimensionality of features while preserving key topological information. In large sintering furnace control systems, the original features may have tens or even hundreds of dimensions. Directly using these features would lead to excessively high computational complexity, thus requiring dimensionality reduction. This unit employs the Locally Linear Embedding (LLE) algorithm, which can preserve the local geometric relationships between data points. For a sintering control system containing more than 50 original features, dimensionality reduction can reduce the number of features to 15-20 while retaining more than 90% of the information, significantly improving computational efficiency.

[0036] Reference Figure 4 The chaotic dynamics time-series dependency identification module 22 includes a time-series data segmentation unit 221, a phase space mapping unit 222, a system stability analysis unit 223, a turning point identification unit 224, and a time-series dependency network construction unit 225.

[0037] The time-series data segmentation unit 221 is used to divide the temperature time-series data into three time windows: short-term, medium-term, and long-term. In the control of powder metallurgy sintering processes, temperature changes exhibit characteristics across multiple time scales, thus requiring multi-scale analysis. The short-term window is set to 30 seconds, primarily capturing rapid temperature fluctuations, such as those caused by the switching of heating elements; the medium-term window is set to 5 minutes, used to identify temperature change trends, such as acceleration or deceleration during the heating phase; and the long-term window is set to 30 minutes, used to analyze the overall sintering cycle characteristics, such as changes during preheating, isothermal heating, and cooling stages. This multi-scale time window design enables the system to comprehensively capture temperature change characteristics at different time scales.

[0038] Phase space mapping unit 222 is used to map the timing data of each time window to the phase space. This unit constructs the phase space using the time delay coordinate method, specifically: , in, Let be the phase space vector at time t. The temperature value at time t. For time delay parameters, For embedded dimension. In the sintering temperature control of copper-based powder compacts, It is usually set to 2 seconds (corresponding to 10 sampling points). The value is set to 4; however, for the more complex sintering of tungsten-based powder compacts, It may need to be increased to 3-4 seconds. The number of parameters was increased to 5-6 to capture more complex dynamic characteristics. These parameters were determined through optimization using mutual information functions and the pseudo-nearest neighbor method, which can best represent the dynamic characteristics of the system.

[0039] System stability analysis unit 223 is used to calculate system invariants in phase space and evaluate system stability. The key invariants calculated by this unit include: 1. Maximum Lyapunov index (MLE): , in, The maximum Lyapunov index, This represents the separation distance between two initially similar trajectories in phase space at time t. This represents the norm of the separation distance. This represents the norm of the initial separation distance (MLE). In the powder metallurgy sintering process, MLE > 0 indicates that the system exhibits chaotic characteristics, making long-term temperature changes difficult to predict, typically occurring during the rapid heating phase in the early stages of sintering; MLE < 0 indicates that the system is stable, with highly predictable temperature changes, typically occurring during the isothermal phase. In practical applications, for MLE > 0.01, the system shortens the prediction time domain and increases the control frequency; while for MLE < 0.005, the prediction time domain can be extended, the control frequency reduced, and system resources optimized.

[0040] 2. Relevant Dimensions: , in, For the relevant dimension, The correlation integral represents the distance in phase space less than 1 / 2. Point-to-point ratio, This is the distance threshold. The higher the value, the higher the system complexity. In iron-based powder sintering, the preheating stage... A value typically between 2.5 and 3.5 indicates high system complexity; while the value during the isothermal phase... A drop to 1.5-2.0 indicates that the system is approaching a stable state.

[0041] The inflection point identification unit 224 is used to identify key inflection points and steady-state regions of temperature changes. This unit identifies inflection points, intersections, and stable regions of the trajectory by analyzing the geometric characteristics of the phase space trajectory. In powder metallurgy sintering, inflection points typically correspond to changes in process stages, such as the end of preheating and the start of isothermal control. Identifying these key points helps optimize control strategies and improve temperature control accuracy. For example, when transitioning from the preheating stage to the isothermal stage, the system adjusts control parameters 3-5 minutes in advance to slow the heating rate and avoid temperature overshoot. This strategy reduces the temperature overshoot from 5-8℃ in traditional control to 1-2℃.

[0042] The temporal dependency network building unit 225 is used to establish a dependency network representing the temperature change patterns at different time scales. This network is a directed weighted graph, where nodes represent states at different time points, and edge weights represent state transition probabilities. By analyzing the structural characteristics of this network, the system can identify typical patterns and laws of temperature change, providing a basis for prediction and control. In copper-based powder sintering, the system identified four typical temperature change patterns: rapid heating, slow heating, isothermal fluctuation, and slow cooling. Different control strategies were adopted for each pattern, significantly improving the control effect.

[0043] Reference Figure 5 The dynamic attention module 23 for multi-source information fusion includes an information representation unification unit 231, a multi-level fusion unit 232, a dynamic attention calculation unit 233, and an uncertainty quantification unit 234.

[0044] The information representation unification unit 231 is used to convert data from different sources and formats into a unified representation. In the powder metallurgy sintering temperature control system, data sources are diverse, including temperature sensors, flow meters, gas concentration analyzers, etc., and the data output formats and units of these devices are different. This unit realizes the standardized representation of data, ensuring that data from different sources can be processed and analyzed within the same framework. Specifically, for each type of data... Convert to standard representation : , in, For the standardized data representation, The original data, It refers to data types. The transformation function includes operations such as standardization, encoding, and feature extraction. For example, for temperature data, This includes processing techniques such as sliding window averaging and trend extraction; for gas concentration data, It includes processes such as gradient calculation and volatility analysis. This unified representation enables the system to effectively integrate information from different sources.

[0045] The multi-level fusion unit 232 is used to fuse information at three levels: parameter level, parameter group level, and global level. This unit implements a hierarchical information fusion strategy, gradually constructing a complete system representation from a single parameter to the entire system. 1. Parameter-level fusion: Merging information on the same parameter at different time points, such as calculating the average value, rate of change, and volatility of temperature over the past 5 minutes; 2. Parameter Group Fusion: Integrates information from related parameters, such as temperature-flow rate groups and temperature-concentration groups. For temperature-flow rate groups, the fusion function... It can be represented as: , in, For the fusion result, Temperature characteristics, For traffic characteristics, It is a temperature-flow interaction feature. , and These are the corresponding weighting coefficients. In practical applications, Typically in the range of 0.4-0.6, Within the range of 0.2-0.3, These weights are dynamically adjusted during the sintering stage, within the range of 0.2-0.3.

[0046] 3. Global Fusion: Integrating information from all parameter groups to form a global representation of the system's overall state. Global Representation The calculation is as follows: , in, This represents the global state of the system. For the first The fusion result of the parameter groups For the corresponding weighting coefficients, This represents the total number of parameter groups. The importance of different parameter groups in the current state is determined through a dynamic attention mechanism.

[0047] The dynamic attention calculation unit 233 is used to calculate attention weights in the time, feature, and spatial dimensions based on the current system state. In powder metallurgy sintering temperature control, the importance of data at different time points, with different parameters, and at different locations varies, requiring dynamic adjustment of the focus. The attention weights are calculated as follows: , in, It is the attention weight vector along the time dimension. It is a weight matrix. It is currently in a hidden state. This is the current context vector, and softmax is the normalization function that ensures the sum of all weights is 1. Similarly, attention is applied along the feature dimension. Spatial Dimension Attention It is also calculated using the corresponding weight matrix.

[0048] In practical applications, attention weights reflect the importance of different pieces of information. For example, in the initial rapid heating stage of aluminum-based powder sintering, gas flow rate and concentration have higher attention weights. Within the range of 0.4-0.6, these parameters directly affect the heating rate; during the intermediate isothermal phase, the attention weight of the rate of temperature change increases ( Reaching 0.5-0.7), as a stable temperature becomes more critical; in the later, slow cooling stage, the attention weight of the billet temperature becomes dominant. (Usually higher than 0.6), because this is directly related to product quality.

[0049] Uncertainty quantification unit 234 is used to construct a probabilistic graphical network to quantify the uncertainty of the prediction results and generate confidence intervals. In the complex sintering process, temperature prediction is uncertain, and understanding this uncertainty is crucial for control decisions. Uncertainty representation employs a Bayesian neural network method, obtaining the prediction distribution through Monte Carlo sampling. , in, Given input Prediction results The probability distribution, It is the first parameter of the neural network. One sample, It refers to the number of samples. In practical applications, It is usually set to 100, but when computing resources are limited, it can be reduced to 50, which can still obtain a good uncertainty estimate.

[0050] The predicted confidence interval is obtained from the prediction distribution using the quantile method: , in, The 95% confidence interval is... and These are the 2.5% and 97.5% quantiles of the predicted distribution, respectively. In actual sintering control, the system dynamically adjusts the control strategy based on the width of the confidence interval: when the confidence interval is narrow (e.g., within ±3℃), a more aggressive control strategy is adopted; when the confidence interval is wide (e.g., exceeding ±5℃), a more conservative control strategy is adopted to avoid temperature fluctuations caused by over-adjustment.

[0051] Reference Figure 2 The data processing center 2 also includes a data preprocessing module 24, an abnormal data processing and screening module 25, a training data extraction module 26, a training data sample construction module 27, a data model training module 28, and a data model testing and evaluation module 29.

[0052] The data preprocessing module 24 is used for data cleaning, data transformation, and data normalization of multi-source heterogeneous data. In the powder metallurgy sintering temperature control system, the raw data often has problems such as noise, missing data, and inconsistencies, requiring preprocessing to improve data quality. Data cleaning handles data anomalies caused by factors such as sensor failure and electrical interference; data transformation converts raw data of different formats into a unified format to facilitate subsequent processing; and data normalization eliminates scale differences between different data through dimensionless processing, improving the effectiveness of model training.

[0053] The abnormal data processing and filtering module 25 is used to detect and remove abnormal data. In the sintering process control, factors such as sensor malfunctions and human interference may lead to data anomalies, which can affect the accuracy of model training and prediction. This module uses statistical methods and machine learning techniques to identify outliers in the data, thereby improving data quality.

[0054] The training data extraction module 26 is used to extract temperature-related data from the preprocessed data as a training dataset. In powder metallurgy sintering temperature control, not all parameters are directly related to temperature prediction; therefore, it is necessary to selectively extract relevant parameters to construct a targeted training dataset. For example, for stainless steel powder blank sintering, furnace temperature, gas flow rate, and oxygen concentration are key parameters that need to be extracted; while for aluminum-based powder blanks, additional attention needs to be paid to parameters such as hydrogen concentration and heating power.

[0055] The training data sample construction module 27 is used to construct training samples that include input and output data. In practical applications, this module organizes the input and output parameters into training samples according to their temporal relationship, ensuring that the model can learn the mapping relationship between input and output.

[0056] The data model training module 28 is used to train the temperature prediction model based on training samples. This module implements the model training process, including parameter optimization and loss function calculation. A batch training strategy is adopted during model training, with each batch containing 64 to 128 samples. The learning rate is dynamically adjusted during training to improve training efficiency and model performance.

[0057] The data model testing and evaluation module 29 is used to test and evaluate the temperature prediction model to obtain the optimal model. This module comprehensively evaluates the model performance through various evaluation metrics, such as mean absolute error (MAE), root mean square error (RMSE), and correlation coefficient (R²). In practical applications, an excellent temperature prediction model typically has an MAE of less than 3°C, an RMSE of less than 4°C, and an R² greater than 0.95 on the test set, meeting the requirements for sintering temperature control.

[0058] In powder metallurgy sintering temperature control systems, data quality is crucial to model performance. This invention employs three different cleaning strategies to address different types of data problems: For missing values, depending on the type of missing data, the missing values ​​are filled using either the mean or median of the missing data for that type. In practical applications, temperature data is usually relatively smooth, and missing values ​​can be filled using the average of the previous and next 10 time points (approximately 2 seconds of data). For gas flow data with greater fluctuations, the median of the same type of data within the most recent 30 minutes is used for filling, which avoids the influence of extreme values ​​and provides a more reasonable estimate.

[0059] For abnormal samples, the detected abnormal samples are deleted. In the powder metallurgy sintering process, abnormal samples usually manifest as sudden jumps in temperature or gas parameters. These jumps are often caused by sensor malfunctions or external interference, rather than actual process changes. The system identifies abnormal samples by setting reasonable thresholds, such as a temperature change rate exceeding 50℃ / second (far higher than the maximum change rate of 10℃ / second in the normal sintering process) or a sudden change in gas flow rate exceeding 200%. These thresholds are determined based on extensive production data analysis and expert experience, and can effectively identify abnormal situations.

[0060] For outliers, the values ​​within the 3 / 4 to 1 / 4 quartile range of the object's group are used to estimate the outliers. This method, based on the statistical principle of box plots, effectively handles outliers while maintaining the overall distribution characteristics of the data. In powder metallurgy sintering temperature control, the quartile range method typically controls the correction error for temperature data within ±3℃, meeting the system's data quality requirements.

[0061] Example 8 This embodiment details the working principle of the anomaly data processing and filtering module 25. This module employs the Isolation Forest algorithm for anomaly detection, a particularly suitable unsupervised learning method for handling outliers in high-dimensional data.

[0062] The specific process for anomaly data detection is as follows: First, all feature data after dimensionless processing are used for preliminary screening of outliers. Dimensionless processing employs the aforementioned maximum-minimum normalization and Z-score standardization methods to ensure that different features are compared on the same scale. The preliminary screening uses the 3-sigma rule, marking data points deviating from the mean by more than 3 standard deviations as potential outliers. For powder metallurgy sintering data, this step typically identifies approximately 5%–10% of potential outliers, significantly reducing the amount of data required for subsequent processing.

[0063] Secondly, cluster analysis was performed on the initially screened data. During the powder metallurgy sintering process, the data characteristics differ significantly at different stages, thus requiring stage-specific anomaly detection. The K-means algorithm was used for cluster analysis, dividing the data into 3 to 5 categories, representing different working stages or states. For example, for standard iron-based powder compact sintering, the data is typically divided into four categories: preheating stage (room temperature to 600℃), rapid heating stage (600-1000℃), isothermal stage (1000-1120℃), and cooling stage.

[0064] Then, an isolation forest model is constructed based on the clustering results. The core idea of ​​the isolation forest algorithm is that outliers are more easily isolated; therefore, anomalies can be detected by constructing random trees and measuring the ease with which data points are isolated. For each cluster, a dedicated isolation forest model is trained to improve the accuracy of anomaly detection. The key parameter settings for the isolation forest model are as follows: Number of trees: 100-200 trees are sufficient to ensure detection results without causing excessive computational burden. Subsampling size: 60% of the original data, balancing the representativeness and randomness of the sampling; Maximum tree depth: usually set to (Subsampling size) to prevent overfitting; Finally, an isolation forest model is used to further filter out outlier data, obtaining non-outlier data as the training dataset. The outlier threshold is typically set at the 95th percentile of the model's output score, meaning that approximately 5% of the data will be marked as outliers and removed from the training set. In powder metallurgy sintering processes, this ratio is a balanced point that has been validated through extensive experiments, effectively removing outlier data while retaining sufficient training samples.

[0065] This embodiment details the specific implementation of the training data sample construction module 27. In the powder metallurgy billet sintering temperature control system, the design of the training samples directly affects the model's learning effect and prediction performance.

[0066] The input data vector constructed by the training data sample construction module 27 includes the furnace temperature, billet temperature, heating coil temperature, input furnace gas temperature, input furnace gas concentration, gas flow rate, and information on whether human intervention exists. These parameters comprehensively reflect the state of the sintering process, providing rich information input for the prediction model.

[0067] The output data includes the furnace door temperature, the output furnace gas temperature, and the output furnace gas concentration. These parameters are key indicators for evaluating the sintering effect and are also target parameters that the control system needs to predict and regulate.

[0068] When constructing training samples, the system needs to consider the time lag between input and output parameters. In copper-based powder sintering, this time lag is typically 30–45 seconds; while in iron-based powder sintering, due to the larger heat capacity, the time lag can reach 60–90 seconds. To capture this time lag relationship, a sliding window method is used for sample construction: the input data at time t is used to predict the output data at time t+Δt, where Δt is the time delay determined according to the process characteristics.

[0069] In addition, to improve the model's generalization ability, data augmentation techniques were employed on the training data samples, such as adding small-amplitude random noise (typically ±2% of the original value) and adjusting the sampling window size. These techniques effectively improved the model's adaptability to different operating conditions, enabling it to maintain good predictive performance even when faced with production fluctuations.

[0070] This embodiment details a control method using an intelligent control system for the sintering temperature of powder metallurgy billets according to the present invention. The method includes the following steps: (1) Data acquisition unit 1 acquires multi-source heterogeneous data inside the reactor in real time and transmits the multi-source heterogeneous data to data processing center 2. In practical applications, the data acquisition frequency is set to 5Hz, that is, 5 data acquisitions per second. This sampling frequency is determined based on the temperature change characteristics of the powder metallurgy sintering process: in the rapid heating stage, the temperature change rate can reach 5-10℃ / minute, and the 5Hz sampling frequency can capture a temperature change of about 0.02℃, meeting the requirements of fine control; while in the isothermal stage, the temperature change rate drops to 0.5-1℃ / minute, and the same sampling frequency provides richer data, which helps to identify small fluctuations and trend changes. Data transmission adopts the industrial Ethernet protocol, with a communication bandwidth of 10Mbps and a transmission delay of less than 10ms, ensuring the real-time performance of data transmission.

[0071] (2) Data Processing Center 2 performs preprocessing on multi-source heterogeneous data, including data cleaning, data transformation, and data normalization. Data cleaning removes noise, missing values, and outliers; data transformation converts raw data in different formats into a unified format; and data normalization eliminates scale differences between different data through dimensionless processing. In the powder metallurgy billet sintering control system, the preprocessing stage can improve data quality by approximately 25% to 30%, laying the foundation for subsequent analysis and modeling.

[0072] (3) Data processing center 2 uses the abnormal data processing and filtering module 25 to perform anomaly detection and filtering on the preprocessed data to obtain non-abnormal data as the training dataset. As mentioned above, this module uses the isolated forest algorithm for anomaly detection, which can identify and exclude about 5% of abnormal data points, significantly improving the quality and representativeness of the dataset.

[0073] (4) Data processing center 2 extracts temperature regulation-related data from the training dataset and constructs training samples containing input data vectors and output data. In practical applications, the system dynamically adjusts the feature selection strategy for different sintering materials and processes. For example, for powder materials with high carbon content, the system increases the weight of gas concentration features; while for easily oxidized materials, it pays more attention to features such as gas flow rate and temperature change rate.

[0074] (5) Data processing center 2 uses training samples to train the temperature prediction model, which includes a topology-aware multi-scale feature extraction module 21, a chaotic dynamics time-series dependency identification module 22, and a multi-source information fusion dynamic attention module 23. The model training adopts the batch gradient descent method, with a batch size of 64 and an initial learning rate of 0.001. A cosine annealing learning rate scheduling strategy is used, that is, the learning rate gradually decays from the initial value to the minimum value (usually 1 / 10 of the initial value) according to the cosine function, and then returns to the initial value, and repeats. This strategy can effectively avoid local optima and improve model performance. A typical iron-based powder blank sintering temperature prediction model requires about 5,000 samples to train, and the training time is about 30 minutes (on a standard industrial server).

[0075] (6) The topology-aware multi-scale feature extraction module 21 extracts multi-scale features that preserve topological relationships from multi-source heterogeneous data. As mentioned earlier, this module preserves the inherent structure of the data and improves the expressive power of the features by constructing a topological relationship network between features. In practical applications, the system dynamically adjusts the level and depth of feature extraction for sintering processes of different complexities. For example, for simple copper-based powder sintering, two-level feature extraction (parameter level and parameter group level) is sufficient; while for complex tungsten-cobalt alloy powder sintering, full three-level feature extraction is required to fully capture the system characteristics.

[0076] (7) Chaotic Dynamics Time-Series Dependency Identification Module 22 analyzes the nonlinear dynamic characteristics of temperature time-series data and identifies key turning points in temperature changes. This module reveals the intrinsic laws of temperature changes and improves prediction accuracy through phase space reconstruction and system invariant analysis. For different sintering stages, the system adjusts the size and overlap of the analysis window. For example, in the rapid heating stage, the window size is set to 20-30 seconds and the overlap is 50%; while in the isothermal stage, the window size can be increased to 60-90 seconds and the overlap reduced to 30%. This dynamic adjustment strategy balances computational efficiency and analysis accuracy.

[0077] (8) The dynamic attention module 23, which integrates features from different sources, focuses on key information through a multi-dimensional attention mechanism. As mentioned earlier, this module can automatically adjust the attention given to different information, focusing on the most relevant parameters and time points according to the current system state. In the powder metallurgy sintering process, the dynamic attention mechanism can adapt to the control requirements of different stages and improve prediction and control performance.

[0078] (9) Data Processing Center 2 predicts the future temperature under the current condition based on the temperature prediction model. In the standard sintering process, the prediction time range is set to the next 5 to 10 minutes, and the prediction step size is 30 seconds. This setting achieves a good balance between prediction accuracy and computational complexity. The system not only predicts the temperature value, but also predicts the temperature change trend and possible fluctuation range, providing comprehensive information for control decisions.

[0079] (10) Temperature regulation actuator 4 calculates the temperature correction amount and correction curve based on the prediction results. The calculation of the correction amount comprehensively considers the current temperature error, the temperature change trend, and the predicted future temperature change, realizing a control strategy that combines feedforward and feedback. This strategy is more forward-looking than traditional pure feedback control, and can respond to temperature changes in advance, reducing control lag.

[0080] (11) Temperature regulation actuator 4 adjusts the reactor temperature according to the temperature correction amount and correction curve. In practical applications, temperature regulation adopts a smooth change strategy to avoid drastic equipment operation and large temperature fluctuations. For example, for the adjustment of heating power, the system limits the maximum adjustment range to no more than 15% of the current power to protect the heating element and reduce temperature overshoot.

[0081] (12) Data display unit 5 displays the prediction results and control process data in real time. The displayed content includes the current temperature, predicted temperature, control error, and the changing trends of key parameters, presented in various chart formats, such as line graphs, heat maps, and dashboards. This visualization information helps operators to fully understand the system status, promptly identify potential problems, and optimize control strategies.

[0082] The above steps form a complete closed-loop control process. Through deep learning technology and advanced data processing methods, accurate prediction and control of the sintering temperature of powder metallurgy billets are achieved. In actual production, this method has been successfully applied to the sintering processes of various powder metallurgy products, such as copper-based self-lubricating bearings, stainless steel filters, and tungsten-cobalt cemented carbide tools, significantly improving product quality and production efficiency. Compared with traditional PID control, this method improves temperature control accuracy by approximately 60%, reduces energy consumption by approximately 20%, and increases product yield by approximately 15%, demonstrating significant technical and economic value.

[0083] The embodiments described above are merely illustrative of specific implementations of the present invention, and while the descriptions are detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. An intelligent control system for the sintering temperature of powder metallurgy billets, characterized in that, include: The data acquisition unit is used to acquire multi-source heterogeneous data inside the reactor in real time. The multi-source heterogeneous data includes gas temperature value, gas flow rate value, reactor electric heating data, reactor temperature regulation mechanism action value, billet temperature value, reactor internal temperature value, and gas concentration value. The data processing center, connected to the data acquisition unit, is used to preprocess and filter the multi-source heterogeneous data, and to build a temperature prediction model based on the preprocessed data. The temperature prediction model includes a topology-aware multi-scale feature extraction module, a chaotic dynamics time-series dependency identification module, and a dynamic attention module for multi-source information fusion. The topology-aware multi-scale feature extraction module includes: A feature space construction unit is used to construct a high-dimensional feature space from the multi-source heterogeneous data. The topology analysis unit is used to calculate the strength of topology relationships between different parameters in the high-dimensional feature space. A multi-scale feature extraction unit is used to perform hierarchical decomposition and extraction of features based on the strength of the topological relationship; Feature dimensionality reduction and preservation unit, used to reduce feature dimensionality while preserving key topological information; A data storage unit, connected to the data processing center, is used to store the multi-source heterogeneous data and the training sample data of the temperature prediction model; A temperature regulating actuator, connected to the data processing center, is used to adjust the temperature of the reactor according to the output of the temperature prediction model. The data display unit, connected to the data processing center, is used to visually display temperature prediction results and control process data. The data communication unit is used to realize data interaction between the data acquisition unit, the data processing center, the data storage unit, the temperature regulation actuator, and the data display unit.

2. The intelligent control system for sintering temperature of powder metallurgy billets according to claim 1, characterized in that, The temperature regulation actuator includes a heating coil, a furnace door temperature detector, and a temperature regulation controller. The temperature regulation controller receives heating data from the heating coil and temperature data inside the reactor, and performs temperature regulation based on the received data. The temperature regulation includes the calculation and application of a first temperature correction value and a second temperature correction value. The second temperature correction value takes effect when the first temperature correction value is close to zero.

3. The intelligent control system for sintering temperature of powder metallurgy billets according to claim 1, characterized in that, The chaotic dynamics time-series dependency identification module includes: The time series data segmentation unit is used to divide temperature time series data into three time windows: short-term, medium-term, and long-term. The phase space mapping unit is used to map the time series data of each time window to the phase space; The system stability analysis unit is used to calculate system invariants in phase space and evaluate system stability. The inflection point identification unit is used to identify key inflection points and steady-state regions of temperature changes; Temporal dependency network building blocks are used to build dependency networks that represent the temperature change patterns at different time scales.

4. The intelligent control system for sintering temperature of powder metallurgy billets according to claim 1, characterized in that, The dynamic attention module for multi-source information fusion includes: Information representation unified unit, used to convert data from different sources and in different formats into a unified representation format; Multi-level fusion units are used to fuse information at three levels: parameter level, parameter group level, and global level. The dynamic attention calculation unit is used to calculate attention weights in the time dimension, feature dimension, and spatial dimension based on the current system state. Uncertainty quantification unit is used to construct a probabilistic graphical network to quantify the uncertainty of prediction results and generate confidence intervals.

5. The intelligent control system for sintering temperature of powder metallurgy billets according to claim 1, characterized in that, The data processing center also includes: The data preprocessing module is used to perform data cleaning, data transformation, and data normalization on the multi-source heterogeneous data; The abnormal data processing and filtering module is used to detect and remove abnormal data; The training data extraction module is used to extract temperature regulation-related data from the preprocessed data as a training dataset. The training data sample construction module is used to construct training samples that include input data and output data. A data model training module is used to train the temperature prediction model based on the training samples. The data model testing and evaluation module is used to test and evaluate the temperature prediction model to obtain the optimal model.

6. The intelligent control system for sintering temperature of powder metallurgy billets according to claim 5, characterized in that, The data preprocessing module performs data cleaning according to the following rules: For missing values, depending on the type of missing object, the missing values ​​are filled with the mean or median of the missing data for that type. For abnormal samples, delete the detected abnormal samples; For outliers, use numbers in the range of the third-quarter to the first-quarter digits within the group containing the object to estimate the outlier.

7. The intelligent control system for sintering temperature of powder metallurgy billets according to claim 5, characterized in that, The abnormal data processing and filtering module detects anomalies in the data by constructing an isolated forest model, including: Use all feature data after dimensionless processing to perform preliminary screening of outlier data; Cluster the data after initial screening; Construct an isolated forest model based on the clustering results; The isolated forest model is used to further filter out outlier data, and non-outlier data is used as the training dataset.

8. The intelligent control system for sintering temperature of powder metallurgy billets according to claim 5, characterized in that, The input data vector constructed by the training data sample construction module includes the furnace temperature value, billet temperature value, heating coil temperature value, input furnace gas temperature value, input furnace gas concentration value, gas flow rate value, and information on whether there is human intervention. The output data includes the furnace door temperature value, output furnace gas temperature value, and output furnace gas concentration value.

9. A control method for a powder metallurgy billet sintering temperature intelligent control system according to any one of claims 1-8, characterized in that, The following steps are involved: (1) The data acquisition unit acquires multi-source heterogeneous data inside the reactor in real time and transmits the multi-source heterogeneous data to the data processing center; (2) The data processing center preprocesses the multi-source heterogeneous data, including data cleaning, data transformation and data normalization; (3) The data processing center uses the abnormal data processing and filtering module to detect and filter the preprocessed data to obtain non-abnormal data as the training dataset. (4) The data processing center extracts temperature regulation-related data from the training dataset and constructs training samples containing input data vectors and output data; (5) The data processing center uses the training samples to train the temperature prediction model, which includes a topology-aware multi-scale feature extraction module, a chaotic dynamics time-series dependency recognition module, and a dynamic attention module for multi-source information fusion. (6) The topology-aware multi-scale feature extraction module extracts multi-scale features that preserve topological relationships from the multi-source heterogeneous data; (7) The chaotic dynamics time-series dependency identification module analyzes the nonlinear dynamic characteristics of the temperature time-series data and identifies key turning points in temperature changes; (8) The dynamic attention module for multi-source information fusion integrates features from different sources and focuses on key information through a multi-dimensional attention mechanism; (9) The data processing center predicts the future temperature under the current condition based on the temperature prediction model; (10) The temperature regulation actuator calculates the temperature correction amount and correction curve based on the prediction results; (11) The temperature regulating actuator adjusts the reactor temperature according to the temperature correction amount and the correction curve; (12) The data display unit displays the prediction results and control process data in real time.

Images