Intelligent temperature adjusting method and system for injection molding mold

Abstract

The invention provides an intelligent temperature adjusting method and system for an injection molding mold. The method comprises the steps that real-time multi-source sensor data flow is generated by collecting temperature data and pressure data of all areas of the injection molding mold; performing feature extraction on the real-time multi-source sensor data flow to obtain temperature field distribution features; a temperature field distribution prediction model is obtained through training based on the temperature field distribution characteristics and a differential private Bayesian optimization algorithm, and then a mold temperature prediction result is output through the model; performing feature fusion on the mold temperature prediction result and the pressure data, constructing input features of a neural network controller, training based on the input features and a positive definite curvature learning method to obtain the neural network controller, and generating a power control signal of a heater and a flow control instruction of a water cooling system through the neural network controller, and based on the control signal and the instruction, the temperature field of the mold is dynamically controlled through the adjustable clearance compensation mechanism. According to the invention, the precision and response speed of temperature adjustment are improved.

Description

Technical Field

[0001] The present application relates to the field of injection molding mold control technology, and in particular to an injection molding mold intelligent temperature regulation method and system. Background Art

[0002] Injection molding is an important industrial production process, and the accuracy and stability of mold temperature control directly affect product quality and production efficiency. As industrial production continues to increase product quality requirements, precise control of mold temperature field has become a key technical difficulty.

[0003] At present, mold temperature control mainly adopts traditional control methods such as PID control, which adjusts the heater power and cooling system flow by setting fixed control parameters. Although this method is simple to implement, it is difficult to adapt to the dynamic temperature changes under complex working conditions due to the lack of accurate modeling and prediction capabilities of temperature field distribution characteristics.

[0004] In recent years, with the development of artificial intelligence technology, temperature control methods based on machine learning have gradually been applied to mold temperature control. This type of method achieves a certain degree of intelligent control by establishing temperature prediction models and control models. However, the existing technology has problems such as low temperature control accuracy, difficulty in effectively responding to dynamic changes in the molding process, and slow response speed that is difficult to meet real-time control requirements. Summary of the invention

[0005] In view of this, the present application provides an intelligent temperature control method and system for injection molding molds, which solves the problems in the prior art of low mold temperature dynamic control accuracy, difficulty in effectively responding to dynamic changes in the molding process, and slow response speed that is difficult to meet real-time control requirements.

[0006] The present application embodiment provides an intelligent temperature adjustment method for an injection molding mold, comprising: Collect temperature and pressure data from all areas of the injection molding mold to generate real-time multi-source sensor data streams; By using the Transformer model and the EM algorithm, feature extraction is performed on the real-time multi-source sensor data stream to obtain temperature field distribution features; Based on the temperature field distribution characteristics and the differential private Bayesian optimization algorithm, a temperature field distribution prediction model is trained and obtained, and a mold temperature prediction result is output through the temperature field distribution prediction model; Performing feature fusion on the mold temperature prediction result and the pressure data to construct input features of a neural network controller, and training the neural network controller based on the input features and a positive curvature learning method to obtain the neural network controller, and generating power control signals for heaters in each area of ​​the injection molding mold and flow control instructions for a water cooling system through the neural network controller; Based on the power control signal of the heater and the flow control instruction of the water cooling system, the temperature field of the injection molding mold is dynamically controlled through an adjustable gap compensation mechanism.

[0007] Optionally, the collecting of temperature data and pressure data of each area of ​​the injection molding mold to generate a real-time multi-source sensor data stream includes: The temperature data collected by the temperature sensor and the pressure data collected by the pressure sensor are converted into digital form through a high-speed data acquisition card to obtain digital signals; Performing signal filtering, data calibration, and outlier elimination on the digital signal to generate preprocessed data; A distributed data caching mechanism is established to synchronize and align the pre-processed data according to timestamps to generate a real-time multi-source sensor data stream.

[0008] Optionally, the extracting features of the real-time multi-source sensor data stream by using a Transformer model and an EM algorithm to obtain temperature field distribution features includes: By using a position encoder of a Transformer model, the real-time multi-source sensor data stream is added with time series position information to generate an initial feature representation; Based on the initial feature representation, a multi-head attention mechanism is used to calculate the similarity between the query matrix, the key matrix and the value matrix, and generate an attention weight matrix for reflecting the spatiotemporal correlation of the data; Using a residual connection structure, the attention weight matrix is ​​weightedly fused with the initial feature representation to obtain an enhanced feature representation; The enhanced feature representation is clustered and analyzed by an EM algorithm to obtain the temperature field distribution feature.

[0009] Optionally, performing cluster analysis on the enhanced feature representation by using an EM algorithm to obtain the temperature field distribution feature includes: Based on the enhanced feature representation, K cluster centers are randomly initialized to obtain initial clustering parameters, wherein the K value is adaptively determined by the silhouette coefficient; According to the initial clustering parameters, alternately performing the steps of calculating the probability of belonging of data points and updating the cluster centers, and dynamically adjusting the parameter updating step size through an adaptive learning rate mechanism until a target clustering result is obtained; The temperature field distribution characteristics are obtained through the target clustering results, wherein the temperature field distribution characteristics include a temperature gradient vector field, a hot spot distribution characteristic map, and a temperature fluctuation spectrum.

[0010] Optionally, the training to obtain a temperature field distribution prediction model based on the temperature field distribution characteristics and a differential private Bayesian optimization algorithm includes: Based on the temperature field distribution characteristics, constructing a priori distribution of a Bayesian probability model; The model parameters are randomly perturbed through the differential privacy mechanism, and the target model parameters are searched through the Bayesian optimization method; Based on the target model parameters, a temperature field distribution prediction model is trained.

[0011] Optionally, constructing a prior distribution of a Bayesian probability model based on the temperature field distribution characteristics includes: By multi-dimensional Gaussian distribution modeling and historical data analysis, an initial prior distribution is obtained based on the temperature gradient vector field and the hot spot distribution characteristic map in the temperature field distribution characteristics; Based on the initial prior distribution, the parameter distribution of different working conditions is automatically learned through a hierarchical Bayesian structure to generate a hierarchical prior model; Constructing a Bayesian probability model based on the hierarchical prior model and outputting prior distribution parameters; Furthermore, the method of randomly perturbing the model parameters by using the differential privacy mechanism and searching the target model parameters by using the Bayesian optimization method includes: Add random noise to the prior distribution parameters through the Laplace mechanism, and dynamically adjust the privacy budget based on the parameter sensitivity to obtain the disturbed parameters; A proxy model is constructed by Gaussian process regression, parameter performance evaluation is performed on the perturbed parameters, and a candidate parameter set is obtained by exploring and utilizing a balance strategy, wherein the evaluation indicators of the proxy model include prediction accuracy and computational efficiency; For the candidate parameter set, model parameter optimization is performed using a variational inference method and a method to minimize the lower bound of evidence, and the target model parameters are output.

[0012] Optionally, the step of fusing the mold temperature prediction result and the pressure data to construct input features of a neural network controller includes: The mold temperature prediction result and the real-time pressure data are constructed into a time series feature sequence through a sliding time window mechanism; By using an attention mechanism, the temporal feature sequence is weighted to obtain a weighted feature sequence; The weighted feature sequence is converted into a feature vector to generate input features of a neural network controller.

[0013] Optionally, the method of obtaining the neural network controller based on the input feature and the positive curvature learning method training includes: According to the input features, a loss function based on the Riemann metric is constructed, wherein the loss function includes a temperature control error term, a pressure balance term, and an energy consumption constraint term, and a curvature regularization term is obtained to obtain an optimization objective function; According to the optimization objective function, the Riemann gradient is calculated in the tangent space of the parameters, and the updated parameters are projected back to the manifold space satisfying the positive definite constraint through exponential mapping to obtain the optimized network parameters.

[0014] Optionally, the dynamically controlling the temperature field of the injection molding mold through an adjustable gap compensation mechanism based on the power control signal of the heater and the flow control instruction of the water cooling system includes: Based on the power control signal of the heater, the power of the heater in each area of ​​the injection molding mold is modulated through a thyristor module, and based on the flow control instruction of the water cooling system, the cooling water flow of the water cooling system in each area of ​​the injection molding mold is controlled through a high-precision proportional valve; Real-time monitoring of the displacement change of the adjustable gap compensation mechanism, and adjusting the gap size of the cavity of the injection molding mold based on the displacement change; Based on the temperature data and pressure data of the injection molding mold collected in real time, PID parameters are adjusted through a fuzzy adaptive algorithm, and the prediction period and control time domain of the model predictive controller are optimized based on the reinforcement learning method to obtain the optimized control parameters. When it is detected that the process parameters meet the preset change conditions, the parameter re-optimization process is triggered.

[0015] The present application also provides an intelligent temperature control system for injection molding molds, including: Data acquisition module, used to collect temperature and pressure data of various areas of the injection molding mold and generate real-time multi-source sensor data stream; A feature recognition module is used to extract features from the real-time multi-source sensor data stream through a Transformer model and an EM algorithm to obtain temperature field distribution features; A temperature prediction module, used for training a temperature field distribution prediction model based on the temperature field distribution characteristics and a differential private Bayesian optimization algorithm, and outputting a mold temperature prediction result through the temperature field distribution prediction model; A control strategy module, used to perform feature fusion on the mold temperature prediction result and the pressure data, construct input features of a neural network controller, and obtain the neural network controller through training based on the input features and a positive curvature learning method, and generate power control signals for heaters in each area of ​​the injection molding mold and flow control instructions for a water cooling system through the neural network controller; The execution control module is used to dynamically control the temperature field of the injection molding mold through an adjustable gap compensation mechanism based on the power control signal of the heater and the flow control instruction of the water cooling system.

[0016] This application has the following technical effects: The present application collects temperature data and pressure data of each area of ​​the injection molding mold to generate a real-time multi-source sensor data stream; uses the Transformer model and the EM algorithm to extract features of the real-time multi-source sensor data stream to obtain temperature field distribution features; based on the temperature field distribution features and the differential private Bayesian optimization algorithm, a temperature field distribution prediction model is trained to obtain the mold temperature prediction result, and the mold temperature prediction result is output through the temperature field distribution prediction model; the mold temperature prediction result and the pressure data are feature fused to construct the input features of the neural network controller, and the neural network controller is trained based on the input features and the positive curvature learning method, and the power control signal of the heater in each area of ​​the injection molding mold and the flow control instruction of the water cooling system are generated through the neural network controller; based on the power control signal of the heater and the flow control instruction of the water cooling system, the temperature field of the injection molding mold is dynamically controlled through an adjustable gap compensation mechanism. Among them, the Transformer model and EM algorithm are used to extract features of multi-source sensor data, and high-precision identification of temperature field distribution characteristics is achieved. The differential private Bayesian optimization algorithm is used to train the temperature field distribution prediction model, which improves the model prediction accuracy while protecting data privacy. The neural network controller is trained based on the positive curvature learning method, which improves the stability and convergence of the controller. The temperature field dynamic control is achieved through the adjustable gap compensation mechanism, which significantly improves the accuracy and response speed of temperature regulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following is a brief introduction to the drawings required for use in the embodiments. The drawings herein are incorporated into the specification and constitute a part of the specification. These drawings illustrate embodiments consistent with the present disclosure and are used together with the specification to illustrate the technical solutions of the present disclosure. It should be understood that the following drawings only illustrate certain embodiments of the present disclosure and should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can also be obtained based on these drawings without creative work.

[0018] Figure 1 A schematic diagram of a process for an intelligent temperature adjustment method for an injection molding mold provided in an embodiment of the present application; Figure 2 A schematic diagram of a flow chart of a method for generating real-time multi-source sensor data streams in an embodiment of the present application; Figure 3This is a schematic diagram of the structure of the feature extraction method based on the Transformer model and the EM algorithm in the embodiment of the present application; Figure 4 A schematic diagram of the training process of the temperature field distribution prediction model in the embodiment of the present application; Figure 5 This is a schematic diagram of the structure of a neural network controller in an embodiment of the present application; Figure 6 It is a structural schematic diagram of the intelligent temperature adjustment system for the injection molding mold in the application embodiment. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical scheme and advantages of the embodiments of the present disclosure clearer, the technical scheme in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, rather than all of the embodiments. The components of the embodiments of the present disclosure generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present disclosure provided in the drawings is not intended to limit the scope of the present disclosure for protection, but merely represents the selected embodiments of the present disclosure. Based on the embodiments of the present disclosure, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present disclosure.

[0020] Figure 1 The figure is a flow chart of the intelligent temperature adjustment method for injection molding mold provided in the embodiment of the present application. Figure 1 As shown, the embodiment of the present application provides an intelligent temperature adjustment method for an injection molding mold, comprising the following steps S1 to S5: S1: Collect temperature and pressure data of each area of ​​the injection molding mold to generate real-time multi-source sensor data stream; like Figure 2 As shown, in some embodiments, step S1 may specifically include steps S1.1 to S1.3: S1.1: The temperature data collected by the temperature sensor and the pressure data collected by the pressure sensor are converted into digital form through a high-speed data acquisition card to obtain digital signals; S1.2: performing signal filtering, data calibration, and outlier elimination on the digital signal to generate preprocessed data; S1.3: Establish a distributed data caching mechanism to synchronize and align the pre-processed data according to timestamps to generate a real-time multi-source sensor data stream.

[0021] In this embodiment, the temperature data and pressure data on the surface and inside of the mold are collected in real time by arranging a temperature sensor array (including temperature measuring thermocouples, infrared sensors, etc.) and a pressure sensor at key positions of the mold. Among them, the sampling frequency of the temperature sensor is set to 100Hz, and the sampling frequency of the pressure sensor is set to 200Hz. Then, a high-speed data acquisition card is used to perform A / D conversion and preprocessing on the collected sensor raw signal. Specifically, it includes signal filtering, data calibration and outlier removal to ensure the accuracy and reliability of the data. Then, a distributed data caching mechanism is established to synchronize and align the preprocessed temperature and pressure data according to the timestamp to form a unified data format. Finally, a real-time data stream processing pipeline is constructed to package the synchronized multi-source sensor data into a unified data stream format to achieve real-time transmission and storage of data.

[0022] Specifically, in the process of intelligent temperature adjustment of injection molding molds, due to differences in sampling frequency, response time, data transmission delay and other factors of temperature sensors and pressure sensors, the data collected by different sensors are often out of sync in time. To solve this problem, this application uses a distributed data caching mechanism to synchronize and align multi-source sensor data.

[0023] Specifically, a multi-level data cache is first set up in the system, including local cache and distributed cache. The local cache is used to temporarily store the raw data of each sensor, and the distributed cache uses the Redis cluster to store the pre-processed data. The system adds a unified timestamp to each sensor data packet. The timestamp is generated based on a high-precision clock source and is accurate to the microsecond level to ensure the accuracy of the time mark.

[0024] During the data synchronization process, the system sets a fixed-size sliding time window. The window size is dynamically adjusted according to the real-time requirements of the system and the data characteristics, and is usually set to 10-100ms. For different sensor data falling into the same time window, the system first caches it to the corresponding local cache area. When the amount of data in the local cache area reaches the preset threshold, the data synchronization mechanism is triggered.

[0025] The data alignment process is time-driven. The system uses the timestamp of the temperature data as a reference to find the closest pressure data in the corresponding time window. If it is found that the corresponding sensor data is missing in a certain time window, the data is compensated through the linear interpolation algorithm. In order to ensure the timeliness of the data, the system will set a data waiting timeout threshold, and the data exceeding this threshold will be discarded.

[0026] In addition, in order to handle abnormal situations such as network delays and data loss, the system implements a data retransmission and recovery mechanism. When a data packet loss is detected, the system will request retransmission from the data source and use data cache for temporary compensation to ensure the continuity of the data flow. The system also monitors the working status of each sensor in real time through a heartbeat detection mechanism, and handles the fault in a timely manner when a sensor abnormality is found.

[0027] Finally, the synchronized and aligned data are organized in a unified time series to form a real-time multi-source sensor data stream in a standard format. Each data point in the data stream contains the temperature and pressure information at the corresponding moment, providing high-quality input data for subsequent feature extraction and model training.

[0028] S2: extracting features from the real-time multi-source sensor data stream through the Transformer model and the EM algorithm to obtain temperature field distribution features; like Figure 3 As shown, in some embodiments, step S2 may specifically include steps S2.1 to S2.4: S2.1: The real-time multi-source sensor data stream is added with temporal position information through the position encoder of the Transformer model to generate an initial feature representation.

[0029] In this embodiment, when performing time series feature extraction, the system receives multi-source sensor data streams as input. Specifically, the improved Transformer encoder adopts a multi-layer stacking structure, each layer of which contains a self-attention module and a feedforward neural network. For the input temperature data sequence {T1, T2, ..., Tn} and pressure data sequence {P1, P2, ..., Pn}, the time series position information is added through the position encoder to generate an initial feature representation. This improved encoder structure can not only capture the long-term dependencies of the data, but also maintain the integrity of the time series information.

[0030] Specifically, the system first uses the position encoder of the Transformer model to process the real-time multi-source sensor data stream. The purpose of position encoding is to give relative position information to each time series data point so that the model can perceive the time series relationship of the data. Specifically, the system uses a combination of sine and cosine functions to generate a position encoding vector, and the dimension of the position encoding vector is the same as the data feature dimension.

[0031] For time step t and position i, the position encoding PE(t,i) is calculated as:

[0032] Where d is the feature dimension. By adding the position encoding vector to the original data feature vector, the initial feature representation containing the temporal position information is obtained.

[0033] S2.2: Based on the initial feature representation, a multi-head attention mechanism is used to calculate the similarity between the query matrix, the key matrix and the value matrix, and generate an attention weight matrix for reflecting the spatiotemporal correlation of the data.

[0034] In this embodiment, based on feature extraction, the system designs an innovative multi-head attention mechanism. Specifically, the number of attention heads is set to 8, and each attention head is responsible for capturing feature patterns of different scales. By calculating the similarity between the query matrix (Query), the key matrix (Key), and the value matrix (Value), the system can automatically learn the association between different sensor data, and specifically splice and linearly transform the outputs of multiple attention heads to obtain a complete attention weight matrix. For example, when the temperature in a certain area rises abnormally, the system will automatically pay attention to the temperature changes and pressure changes in adjacent areas, thereby generating an attention weight matrix that reflects the spatiotemporal correlation of the data. A multi-head attention mechanism is implemented to capture the spatiotemporal correlation between data. Each of the 8 attention heads focuses on a different feature subspace, so that the multi-scale correlation of the data can be fully captured. S2.3: Using a residual connection structure, the attention weight matrix is ​​weightedly fused with the initial feature representation to obtain an enhanced feature representation.

[0035] In this embodiment, a residual connection structure is used to perform weighted fusion of the attention weight matrix and the initial feature representation. In the specific implementation, the attention weight is first normalized to ensure that the sum of the weights is 1, and then weighted summed with the initial feature representation. In addition, a layer normalization mechanism is introduced to effectively alleviate the gradient vanishing problem in deep network training, retain the original feature information, and improve the stability of feature extraction. In the specific implementation, the system performs element-by-element addition operation on the attention weight matrix and the initial feature representation, and then undergoes layer normalization to obtain an enhanced feature representation.

[0036] Layer normalization helps stabilize the training process and accelerates convergence. Its calculation formula is:

[0037] in and are the characteristic mean and standard deviation, and are learnable scaling and bias parameters.

[0038] S2.4: Perform cluster analysis on the enhanced feature representation using an EM algorithm to obtain the temperature field distribution characteristics.

[0039] Wherein, step S2.4 may specifically include: S2.4.1: Based on the enhanced feature representation, randomly initialize K cluster centers to obtain initial clustering parameters, wherein the K value is adaptively determined by the silhouette coefficient; S2.4.2: According to the initial clustering parameters, alternately perform the steps of calculating the probability of data point belonging and updating the cluster center, and dynamically adjust the parameter update step size through an adaptive learning rate mechanism until the target clustering result is obtained; S2.4.3: Obtain the temperature field distribution characteristics through the target clustering results, wherein the temperature field distribution characteristics include a temperature gradient vector field, a hot spot distribution characteristic map, and a temperature fluctuation spectrum.

[0040] In this embodiment, an improved EM algorithm is applied to feature clustering analysis. The algorithm first determines the optimal number of clusters K through the silhouette coefficient adaptively and randomly initializes K cluster centers. The silhouette coefficient is an indicator to measure the clustering effect. Its value is between -1 and 1. The closer it is to 1, the better the clustering effect. The system will automatically test different K values ​​(generally between 3 and 10), calculate the silhouette coefficient corresponding to each K value, and select the K value with the highest score. For example, assuming that the silhouette coefficient is 0.65 when K=3, 0.78 when K=4, and 0.72 when K=5, K=4 is finally selected as the number of clusters. After determining the K value, the cluster center is selected using an improved initialization method. This method will give priority to selecting points far away from the existing cluster center as the new cluster center to avoid the initial cluster center being too concentrated.

[0041] After determining the K value, the EM algorithm alternately executes the expectation step (E-step) and the maximization step (M-step). In the E-step, the posterior probability of each data point belonging to each cluster is calculated. In the M-step, the cluster center and covariance matrix are updated based on the posterior probability. In order to improve the robustness of the algorithm, an adaptive learning rate mechanism is introduced to dynamically adjust the parameter update step until the cluster center converges to a stable state. For example, if the initial learning rate is set to 0.01, the learning rate will gradually decrease as the number of iterations increases. When the algorithm runs to the 100th iteration, the learning rate may drop to 0.001, and to 0.0001 by the 500th iteration. The system monitors parameter changes. If the parameter changes for 10 consecutive iterations are less than 0.000001, the algorithm is considered to have converged and the iteration can be stopped.

[0042] Through iterative optimization of the EM algorithm, the system finally obtains the temperature field distribution characteristics, that is, a series of key features that characterize the temperature field distribution, including: The temperature gradient vector field is constructed by calculating the temperature difference between adjacent measuring points and is used to describe the direction and intensity of heat transfer. For example, if the temperature of a point is 150°C, the temperature of the point above it is 148°C, and the temperature of the point on the right is 152°C, then the heat transfer trend of this point is to diffuse upward and receive heat from the right; The hotspot distribution feature map reflects the location and range of abnormal temperature areas. First, the average value and fluctuation range of the overall temperature are calculated, and then the areas that exceed the normal range are marked. For example, if the temperature of a certain area continues to exceed the average temperature by more than 20°C, it will be marked as a hotspot area; Temperature fluctuation spectrum, characterizing the periodic characteristics of temperature changes, etc. Specifically, the temperature data is analyzed on different time scales to identify periodic temperature fluctuation patterns. For example, the system may find that some areas have temperature fluctuations every 30 seconds, which may be related to the injection molding cycle; or find that some areas have high-frequency temperature jitters, which may indicate unstable heater control. The system focuses on temperature changes in the time scale of 0.1 seconds to 10 seconds, because this range usually contains the most important process-related fluctuations.

[0043] These multi-dimensional feature information provides an important basis for subsequent temperature field prediction and control strategy optimization. It should be noted that the system adopts the feature importance analysis method to retain only the features that have significant contributions to the temperature field characterization, which not only improves the calculation efficiency but also reduces the complexity of subsequent modeling.

[0044] In addition, in this embodiment, the Transformer model mainly processes the time series data of the mold temperature field. Taking the actual scenario as an example, assume that 12 temperature sensors and 8 pressure sensors are arranged on the mold, the sampling frequency is 10Hz, and the system processes a 30-second data window each time. At this time, the input of each time step is a 20-dimensional vector (containing 12 temperature values ​​and 8 pressure values), and the entire sequence contains 300 time steps. The input shape of the model is [batch size, 300, 20], where the batch size is usually set to 32 or 64. These data will be standardized and missing values ​​will be filled before entering the model.

[0045] Position encoding is an important part of the Transformer model. The system generates a position encoding vector for each time step with the same dimension as the feature, in this case 20. These encoding vectors help the model understand the relative position relationship of data points in the time series, allowing the model to distinguish data at different time steps. The position encoding vector is generated by trigonometric functions and can express the absolute position and relative distance information of data points.

[0046] The multi-head attention mechanism is the core of the model. This system uses 8 attention heads, each with a dimension of 64. The original 20-dimensional input features are first mapped to a 512-dimensional space and then divided into 8 64-dimensional subspaces. Each attention head independently calculates the attention weight in its corresponding subspace. This design allows the model to focus on different types of feature patterns at the same time. For example, some attention heads may focus on the long-term trend of temperature changes, while other heads focus on instantaneous fluctuations.

[0047] The features extracted by the model can be divided into three categories: The first category is time-dependent features, which describe the change pattern of temperature and pressure over time. For example, the reading of a temperature sensor has a peak every 5 seconds, which is usually related to the injection molding cycle. The second category is spatial correlation features, which describe the relationship between different sensors. For example, after the reading of pressure sensor P1 increases, the readings of temperature sensors T3 and T4 will rise after 2 seconds. The third category is multi-scale features, which are features on various time scales captured by different attention heads, including rapid temperature fluctuations at the millisecond level and temperature gradient trends at the second level.

[0048] The output of the model is a high-dimensional feature tensor with dimensions [batch size, 300, feature dimension], and the feature dimension is usually set to 256 or 512. These features are then fed into the EM algorithm for cluster analysis to identify the distribution pattern of the temperature field. The output features not only retain the time series information of the original data, but also extract the deep correlation between the data, providing a basis for the system to understand and predict the law of temperature field changes.

[0049] In practical applications, these features show good predictive capabilities. For example, based on these features, the system can predict the temperature anomaly that may occur in a certain area of ​​the mold in the next few seconds, or identify the lag relationship between certain sensor data. This information is of great value for optimizing mold temperature control strategies. In addition, by analyzing the weight distribution of different attention heads, the system can also identify the key factors that have the greatest impact on temperature field changes, which provides an important basis for process parameter optimization.

[0050] It should be noted that the combination of the improved Transformer model and the EM algorithm not only improves the accuracy of feature extraction, but also significantly improves the real-time performance of the algorithm, meeting the strict requirements of the injection molding process for real-time control.

[0051] S3: Based on the temperature field distribution characteristics and the differential private Bayesian optimization algorithm, a temperature field distribution prediction model is trained and obtained, and a mold temperature prediction result is output through the temperature field distribution prediction model; like Figure 4 As shown, step S3 may specifically include S3.1 to S3.3: S3.1: Based on the temperature field distribution characteristics, construct a prior distribution of the Bayesian probability model; Step S3.1 may specifically include: S3.1.1: Obtaining an initial prior distribution based on the temperature gradient vector field and the hot spot distribution characteristic map in the temperature field distribution characteristics through multi-dimensional Gaussian distribution modeling and historical data analysis; S3.1.2: Based on the initial prior distribution, the parameter distribution of different working conditions is automatically learned through a hierarchical Bayesian structure to generate a hierarchical prior model; S3.1.3: Construct a Bayesian probability model based on the hierarchical prior model and output prior distribution parameters.

[0052] In this embodiment, the prior distribution of the Bayesian probability model is constructed based on the temperature field distribution characteristics obtained in the previous step. For features such as temperature gradient and hot spot distribution, a multidimensional Gaussian distribution is used as a prior model. By analyzing the changing law of the temperature field in the historical data, the mean vector and covariance matrix of the prior distribution are determined, where the mean vector of the distribution is determined by the average temperature gradient in the historical data, and the covariance matrix reflects the correlation of the temperature gradients in different regions. For the hot spot distribution characteristics, the system uses the Bernoulli distribution to describe the probability of each area becoming a hot spot, where the probability parameters are estimated based on the frequency of hot spots in the historical data. The temperature fluctuation spectrum characteristics are modeled by a mixed Gaussian distribution, and each Gaussian component corresponds to a typical temperature fluctuation pattern. The selection of these prior distributions fully considers the physical characteristics of the temperature field and can effectively capture the uncertainty of the temperature distribution. In addition, in order to enhance the adaptability of the model, a hierarchical Bayesian structure is introduced so that the model can automatically learn the parameter distribution characteristics under different working conditions.

[0053] Specifically, assume that an injection mold has 16 temperature monitoring points, forming a 4×4 grid layout. In addition to the temperature value, each monitoring point also contains temperature gradient information (X direction and Y direction). Historical data shows that the mold has three typical working conditions during normal production: standard working conditions (mold temperature around 160°C), high temperature working conditions (around 180°C) and low temperature working conditions (around 140°C).

[0054] In S3.1.1, the system first processes the temperature gradient vector field data. For standard working conditions, the average value of the temperature gradient is smaller in the center area of ​​the mold (about 0.2°C / cm) and larger in the edge area (about 0.5°C / cm), which reflects the characteristic of heat diffusion from the heating point to the surrounding area. The hot spot distribution characteristics show that under standard working conditions, hot spots are easily formed in the upper right and lower left corners of the mold, with probabilities of 15% and 12% respectively. The system uses a 24-dimensional Gaussian distribution (16 temperature values ​​+ 8 gradient values) to describe these features. The mean vector reflects the typical value, and the covariance matrix captures the correlation between features.

[0055] In S3.1.2, the system constructs a three-layer Bayesian structure: the first layer is the working condition level, which describes the probability distribution of different working conditions; the second layer is the region level, which describes the temperature characteristic distribution of each region under specific working conditions; the third layer is the time level, which describes the characteristics of temperature evolution over time. Taking the high temperature working condition as an example, the system found that the standard deviation of temperature fluctuation in the upper right corner of the mold under this working condition (about 1.5°C) is significantly higher than that of the standard working condition (about 0.8°C), and this information is encoded into the conditional probability distribution of this region.

[0056] In S3.1.3, the system outputs complete prior distribution parameters based on the hierarchical structure. Take the standard working condition as an example: Overall distribution of temperature field: mean 160°C, standard deviation 2°C; Regional characteristics: The average temperature gradient in the central area is 0.2℃ / cm, and in the edge area is 0.5℃ / cm; Hot spot probability: 0.15 in the upper right corner, 0.12 in the lower left corner, and less than 0.05 in other areas; Time characteristics: The temperature fluctuation cycle is about 30 seconds, corresponding to the injection molding cycle.

[0057] For the working condition conversion, the system found that it takes about 90 seconds to heat up from the standard working condition to the high temperature working condition, during which the upper right corner area usually reaches the target temperature first. Such timing features are encoded into the conversion probability matrix. The system also identified that when the mold is cooled from the high temperature working condition, the cooling rate of the edge area (about 0.4℃ / second) is faster than that of the center area (about 0.2℃ / second).

[0058] In practical applications, this prior knowledge helps the system quickly identify the current operating conditions and predict their evolution trends. For example, when the system observes that the temperature in the upper right corner begins to fluctuate and increase, it can predict in advance that it may be transitioning to high-temperature conditions. For another example, when the temperature gradient in the central area is observed to increase abnormally (exceeding 0.4℃ / cm), the system will warn of possible heater failures in advance.

[0059] This hierarchical prior structure provides reliable initial conditions for subsequent predictive modeling. The system can quickly locate the most likely current operating conditions and development trends based on the real-time observed temperature field characteristics, thereby achieving more accurate predictions. At the same time, this structure also facilitates online updates of the model: when new data is collected, the system can automatically update the parameter distribution of the corresponding level without rebuilding the entire model.

[0060] An important advantage of this hierarchical Bayesian structure is its strong interpretability. When the system makes a prediction, it is possible to clearly trace which prior knowledge led to the prediction, which is very helpful for decision support and fault diagnosis in actual production. For example, if the system predicts that a certain area may overheat, the basis for this prediction can be understood by examining the historical hot spot probability and current temperature gradient of the area.

[0061] S3.2: Randomly perturb the model parameters through the differential privacy mechanism, and search for the target model parameters through the Bayesian optimization method; S3.2 specifically includes: S3.2.1: Add random noise to the prior distribution parameters through the Laplace mechanism, and dynamically adjust the privacy budget based on the parameter sensitivity to obtain the disturbed parameters; S3.2.2: constructing a proxy model through Gaussian process regression, performing parameter performance evaluation on the perturbed parameters, and obtaining a candidate parameter set through an exploration and utilization balance strategy, wherein the evaluation indicators of the proxy model include prediction accuracy and computational efficiency; S3.2.3: For the candidate parameter set, the model parameters are optimized by using the variational inference method and the evidence lower bound minimization method, and the target model parameters are output.

[0062] In this embodiment, a differential privacy mechanism is introduced to randomly perturb the model parameters, and the optimal parameters are searched through the Bayesian optimization method. Specifically, the Laplace mechanism is used to add random noise to sensitive parameters. The size of the noise is controlled by the privacy budget ε. A smaller ε value provides stronger privacy protection, but may affect model performance. The Bayesian optimization process uses the Gaussian process as a proxy model to select the next set of parameters to be evaluated through the expected improvement criterion. The optimization goal is to maximize the prediction accuracy of the model while considering computational efficiency and privacy protection level. At the same time, an adaptive privacy allocation strategy is designed to dynamically adjust the privacy budget according to the sensitivity of different parameters, while protecting data privacy while maintaining model performance to the greatest extent. For example, for parameters in areas with drastic temperature changes, the system will allocate a larger privacy budget to ensure prediction accuracy.

[0063] In addition, the system also adopts an adaptive sampling strategy based on Thompson sampling. This strategy dynamically balances the relationship between exploration and utilization by constructing the posterior distribution of parameters. In each iteration, a set of candidate parameters is sampled from the posterior distribution, and the expected benefits of these parameters are evaluated through the surrogate model. It should be noted that the system uses Gaussian process regression as a surrogate model, which can effectively estimate the objective function value in the unknown parameter space, greatly improving the efficiency of parameter search.

[0064] In S3.2.1, the system first evaluates the sensitivity of each parameter. Assume that the model contains key parameters such as the mean, variance, and hotspot probability of the temperature field. The sensitivity of the mean temperature is low (a change of 1°C has little effect on the prediction), while the sensitivity of the hotspot probability is high (a change of 0.1 may cause a significant change in the prediction result). Based on this analysis, the system adopts a dynamic privacy budget allocation strategy: a smaller privacy budget (ε=0.1) is allocated to the mean temperature parameter, allowing the addition of larger noise; a larger privacy budget (ε=0.5) is allocated to the hotspot probability to maintain higher accuracy. For example, the original hotspot probability of a certain area is 0.15 and the mean temperature is 165°C. After adding noise, it may become a probability of 0.16 and a temperature of 163.5°C. This differentiated privacy protection strategy protects sensitive information while maintaining the basic prediction ability of the model.

[0065] In S3.2.2, the system constructs a Gaussian process proxy model to evaluate parameter performance. Taking temperature prediction as an example, the input of the proxy model includes the perturbed parameter set (such as temperature mean, variance, hot spot probability, etc.), and the output includes prediction accuracy (average prediction error) and calculation time. The system adopts a batch sampling strategy, selecting 10 sets of candidate parameters for evaluation each round. In the balance between exploration and utilization, the system is more inclined to explore new parameter combinations in the early stage (such as the first 30% of the evaluation rounds), and later searches more near the known high-performance parameter area. For example, if it is found that the parameter combination with a hot spot probability in the range of 0.14-0.17 generally performs well, the focus will be on detailed search within this range in the later stage.

[0066] In S3.2.3, the system performs a final optimization on the candidate parameters. Assume that the system identifies three sets of candidate parameter configurations: Configuration A focuses on short-term prediction accuracy, Configuration B excels in computational efficiency, and Configuration C strikes a balance between the two. The system approximates the posterior distribution through a variational inference method and optimizes by minimizing the evidence lower bound (ELBO). The optimization process may find that although Configuration C is not optimal in a single indicator, it has the best overall performance. For example, the prediction error of Configuration C is only 0.2℃ higher than that of A, but the calculation speed is 40% faster than that of A, and the privacy protection level is comparable.

[0067] A specific example of the optimization process: Initial state: Temperature mean parameter: 165℃ (163.5℃ after disturbance); Temperature variance: 2.5°C (2.8°C after disturbance); Hot spot probability: 0.15 (0.16 after perturbation); Prediction error: ±1.5℃; Computation time: 200ms / prediction; Intermediate optimization stage: After trying multiple sets of parameter configurations, I found that: Improving the temperature mean accuracy (reducing disturbances) can reduce the prediction error to ±1.2°C, but the calculation time increases to 250ms; Reducing the hotspot probability accuracy can reduce the calculation time to 150ms, but the prediction error increases to ±1.8℃; Adjusting the variance parameter has relatively little impact on performance; Final optimization results: The system selected a balanced configuration: Temperature mean: 164.2℃ (moderate disturbance); Temperature variance: 2.7°C (large disturbance); Hot spot probability: 0.157 (small disturbance); Prediction error: ±1.3℃; Computation time: 180ms / prediction; This final configuration performs well in practical applications. For example, when the mold temperature changes rapidly, the system can predict the temperature distribution in the next 5 seconds within 200ms, and the prediction error is controlled within ±1.3℃. At the same time, through differential privacy protection, even if someone obtains the model parameters, it is impossible to accurately infer the original temperature distribution characteristics.

[0068] In production practice, the advantage of this optimization strategy is its strong adaptability. When production conditions change (such as changing different plastic raw materials), the system can quickly re-optimize parameter configurations and find a new balance point. At the same time, the differential privacy mechanism ensures that sensitive process parameter information will not be leaked during the parameter optimization process.

[0069] S3.3: Based on the target model parameters, a temperature field distribution prediction model is trained.

[0070] In S3.3, the system trains the temperature field distribution prediction model based on the optimized model parameters, i.e., the target model parameters. The model adopts a deep probabilistic graph network structure, which contains multiple random variable layers and can simultaneously capture the spatial correlation and temporal dynamic characteristics of the temperature field. The specific structure can be: the first layer handles short-term predictions with a prediction range of 0-10 seconds, mainly relying on the local characteristics and immediate dynamics of the temperature field; the second layer handles medium-term predictions with a prediction range of 10-60 seconds, taking into account the spatial transfer characteristics of the temperature field and the influence of process parameters; the third layer handles long-term predictions with a prediction range of more than 60 seconds, focusing on the overall evolution trend of the temperature field. The model training adopts a sliding window strategy, that is, the data of the most recent N time steps are used to predict the temperature distribution of the next M time steps. The system evaluates the model performance through cross-validation and dynamically adjusts the weights of the prediction time domain so that the model can maintain good accuracy in different prediction ranges.

[0071] In addition, during the training process, a variational inference method is introduced to optimize the model parameters by minimizing the evidence lower bound (ELBO). In addition, a dynamic learning rate adjustment mechanism is designed to adaptively adjust the learning step size according to the convergence of the loss function.

[0072] These steps form a complete prediction model building process. The reasonable construction of the prior distribution provides the initial knowledge basis for the model, differential privacy and Bayesian optimization ensure the security and optimality of the model, and the hierarchical prediction structure enables the model to adapt to prediction needs at different time scales. For example, in practical applications, short-term predictions can be used for real-time control and adjustment, medium-term predictions can be used for process parameter optimization, and long-term predictions can be used for equipment maintenance planning. In addition, the system will regularly update the model parameters to adapt to changes in process conditions and the evolution of equipment performance. The prediction results of the model include not only the expected value of the temperature field, but also the uncertainty estimation of the prediction, which provides a more comprehensive basis for subsequent control decisions.

[0073] S4: performing feature fusion on the mold temperature prediction result and the pressure data to construct input features of a neural network controller, and training the neural network controller based on the input features and a positive curvature learning method, and generating power control signals for heaters in each area of ​​the injection molding mold and flow control instructions for a water cooling system through the neural network controller; In the embodiment of the present application, the neural network controller is first constructed. The controller adopts a dual-branch structure, one branch processes temperature prediction information, and the other branch processes pressure data. The input of the temperature branch includes: the current temperature field distribution (16 measuring points), the predicted temperature change trend in the next 5 seconds, the temperature gradient information, and the hot spot distribution probability. The input of the pressure branch includes: the real-time readings of each pressure sensor, the pressure change rate, and the pressure fluctuation characteristics. The two branches are connected through a feature fusion layer, and the fusion layer adopts an attention mechanism to automatically adjust the weights of the temperature and pressure features according to different working conditions. For example, in the early stage of injection molding, the weight of the pressure feature is higher, and in the pressure holding stage, the weight of the temperature feature will increase accordingly.

[0074] The hidden layer of the network adopts a residual structure, which contains multiple residual blocks. Each residual block contains two fully connected layers and a skip connection. This structure helps the network learn subtle changes in the temperature field. The output layer is divided into two parts: the heater power control signal (corresponding to the heaters in each area of ​​the mold) and the water cooling system flow control instructions (corresponding to different cooling circuits). The output layer uses the Sigmoid activation function to ensure that the control signal is within a reasonable range.

[0075] In terms of feature training, the system adopts a multi-stage training strategy. The first stage is feature preprocessing, which standardizes the temperature and pressure data and extracts time series features. For example, for temperature data, in addition to the original value, the change rate in the short term (1 second), medium term (5 seconds), and long term (30 seconds) is calculated. The second stage is feature selection, using Lasso regression to identify the most influential feature combination. The third stage is feature fusion, which learns the compact representation of temperature and pressure features through autoencoders.

[0076] The application of positive curvature learning methods ensures the stability of the controller. First, the system constructs a loss function, including control accuracy terms, energy consumption terms, and stability terms. The control accuracy term measures the accuracy of temperature control, the energy consumption term optimizes the energy consumption of the heating and cooling system, and the stability term ensures the smoothness of the control response through positive curvature constraints. During training, the system updates the network parameters through second-order optimization methods (such as a variant of Newton's method) while maintaining the positive curvature characteristics of the loss function. This ensures that the controller maintains stable response characteristics under various operating conditions.

[0077] The control signal generation process is divided into three steps: first, the controller calculates the ideal temperature field distribution based on the current state and the predicted results; then, the control action required to reach the target state is determined through back propagation; finally, the control action is converted into a specific execution instruction. Taking heating control as an example, the system not only outputs the power value, but also plans the heating timing mode, such as pulse width modulation parameters. For water cooling systems, the control instructions include flow rate and cooling timing.

[0078] In actual applications, the controller shows good adaptability. For example, when it detects that the temperature in a certain area rises too quickly, the system will reduce the power of the corresponding heater in advance and appropriately increase the flow of the adjacent water cooling circuit. For another example, when the mold switches to different products, the controller can automatically adjust the control strategy according to the characteristics of the new product and quickly reach a stable production state.

[0079] The controller also has adaptive learning capabilities. During the production process, the system continuously collects control effect data and regularly updates network parameters. This online learning mechanism enables the controller to adapt to slow changes in equipment performance, such as reduced heater efficiency or waterway blockage. At the same time, the system also records control strategies for special situations to form an experience library for rapid response to similar situations.

[0080] Special attention should be paid to the safety mechanism of the controller. The system sets up multiple layers of safety constraints, including temperature upper limit control, power ramp limit and emergency shutdown strategy. When an abnormal situation is detected, the controller automatically switches to safety mode to prioritize equipment safety. These safety mechanisms are implemented in a hard-coded way and are not affected by the neural network learning process.

[0081] like Figure 5 As shown, step S4 may specifically include S4.1 to S4.5: S4.1: constructing the mold temperature prediction result and the real-time pressure data into a time series feature sequence through a sliding time window mechanism; The temperature prediction results output from the previous steps and the real-time collected pressure data are feature fused to construct the input feature space of the neural network controller. Specifically, the temperature prediction results include the mean temperature and uncertainty estimation of each region at future moments, and the pressure data include the real-time pressure value and pressure change rate of each key position of the mold. The system adopts a sliding time window mechanism to combine historical data and predicted data into a time series feature sequence, and introduces an attention mechanism to weight the features to highlight the data features at important moments.

[0082] Specifically, a 60-second main time window with a sliding step of 1 second can be used, and multiple sub-windows can be set in the main window to capture features of different time scales. For the temperature prediction results, each prediction point contains the temperature change trend for the next 5 seconds, and the sampling frequency is 10Hz, thus forming a sequence of 50 prediction points. For pressure data, a sampling frequency of 10Hz is also used to construct a pressure sequence containing 600 sampling points. For example, the characteristics of a measuring point at time t may include: the current temperature is 165℃, the predicted temperature sequence for the next 5 seconds is [165.2, 165.5, 165.8, 166.0, 166.1]℃, and the corresponding pressure sequence is [45.2, 45.5, 45.8, 46.0, 46.1]MPa. The system also calculates the statistical characteristics of these sequences, such as the mean, standard deviation, and rate of change.

[0083] S4.2: weighting the temporal feature sequence through an attention mechanism to obtain a weighted feature sequence; In S4.2, the system implements a multi-level attention mechanism. The first is the attention in the time dimension. The system assigns weights to features at different time points based on the prediction accuracy. For example, if the temperature prediction at a certain time point is found to be particularly accurate, the weight of this point will be increased accordingly. The second is the attention in the spatial dimension. The system assigns weights according to the importance of different areas. For example, hot spots and areas with large temperature gradients will receive higher weights. Finally, the attention in the feature dimension. The system will dynamically adjust the weight ratio of temperature features and pressure features. In the early stage of injection molding, the pressure feature may be given a weight of 0.7 and the temperature feature a weight of 0.3; while in the holding stage, the pressure feature may be adjusted to a weight of 0.4 and a temperature feature to a weight of 0.6.

[0084] S4.3: Perform feature vector conversion on the weighted feature sequence to generate input features of the neural network controller.

[0085] In S4.3, the system converts the weighted sequence into the input features of the controller through the feature conversion network. The conversion network contains three key modules: the time series encoding module, which converts the time series into a vector of fixed dimension; the spatial encoding module, which captures the spatial relationship between different measurement points; and the feature fusion module, which combines the time series features and spatial features into the final input vector. For example, for a mold with 16 measurement points, a 256-dimensional feature vector may be generated in the end, in which the first 128 dimensions encode temperature-related information and the last 128 dimensions encode pressure-related information.

[0086] S4.4: constructing a loss function based on the Riemann metric according to the input features, wherein the loss function includes a temperature control error term, a pressure balance term, and an energy consumption constraint term, and obtaining a curvature regularization term to obtain an optimization objective function; In S4.4, the system constructs a composite loss function based on the Riemann metric. The temperature control error term uses a weighted Euclidean distance to measure the deviation between the actual temperature and the target temperature, and the weight is dynamically adjusted according to the importance of different regions. The pressure balance term uses the condition number of the covariance matrix as an indicator to ensure uniform pressure distribution in each region. The energy consumption constraint term contains the sum of the squares of the heating power and the cooling power, as well as their rate of change. For example, if the heating power of a certain area changes frequently and significantly, it will be penalized by the energy consumption constraint term. The curvature regularization term is constructed by calculating the Riemann curvature in the parameter space to ensure the smoothness of the control response.

[0087] S4.5: According to the optimization objective function, the Riemann gradient is calculated in the tangent space of the parameters, and the updated parameters are projected back to the manifold space satisfying the positive definite constraint through exponential mapping to obtain the optimized network parameters.

[0088] In this embodiment, considering the particularity of the positive curvature constraint, the traditional gradient descent method may cause the parameters to violate the positive definiteness constraint after updating. To solve this problem, a parameter update strategy based on Riemann optimization is used in this embodiment. Specifically, at each parameter update, the Riemann gradient is first calculated in the tangent space of the parameter. The gradient takes into account the geometric structure of the parameter space, and then the updated parameters are projected back to the manifold space that satisfies the positive definiteness constraint through exponential mapping. For example, if a certain weight matrix after the update does not satisfy the positive definiteness, the system will project it to the nearest positive definite matrix. This process ensures that the controller can maintain stability under various working conditions. In addition, in order to improve the training efficiency, the system also introduces a momentum term and an adaptive learning rate mechanism, which can dynamically adjust the update step size according to the gradient changes. Specifically, for a typical control cycle: first calculate the Riemann gradient based on the current state, then use the retry mechanism to ensure that the appropriate learning rate is found, and finally update the parameters through exponential mapping.

[0089] An important feature of this training method is its stability. By optimizing on a Riemann manifold, the system is able to avoid the numerical instability issues that may be encountered with traditional gradient descent. At the same time, the positive definiteness constraint ensures that the controller response always evolves in the direction of reducing the error. For example, if a control action causes an increase in temperature deviation, the system will immediately adjust the parameters to make subsequent control actions more conservative. In addition, the system will periodically verify the stability of the controller and automatically adjust the regularization strength if potential instability is found.

[0090] Let’s take a specific injection mold temperature control scenario to illustrate: Assume a mold with a 4×4 grid layout, with a total of 16 temperature monitoring points and 8 pressure monitoring points. In S4.4, the loss function based on the Riemann metric constructed by the system contains three main terms. The temperature control error term reflects the deviation between the actual temperature and the target temperature, and is in a weighted form:

[0091] Where wi is the weight of each measurement point (hotspots have higher weights), Ti is the measured temperature, and Ti_target is the target temperature. For example, when the upper right corner temperature is detected to be 175°C, the target temperature is 170°C, and the weight of this area is 1.5, the contribution of this area to the total error is 1.5. (175-170)²=37.5.

[0092] The pressure balance term and the energy consumption constraint term form a composite constraint. The pressure balance is evaluated by the standard deviation of each pressure measurement point, and the energy consumption constraint considers the sum of the squares of the heating power and the cooling power. For example, in a certain control, the system recorded: the power of the four heating zones is [800W, 750W, 720W, 780W], and the flow of the two cooling circuits is [2.5L / min, 2.3L / min]. The system will calculate the variance and rate of change of these control quantities, and large fluctuations will be subject to larger penalties.

[0093] In S4.5, the parameter optimization process is divided into three stages. The first stage is initialization, where the system randomly generates initial parameters that satisfy the positive definite condition in the 16×16 weight matrix space. For example, the initial value of the weight matrix of a layer may be the unit matrix plus a small random perturbation. The second stage is gradient calculation and parameter update, where the system calculates the Riemann gradient in the tangent space of the parameters. Suppose that in a certain iteration, the system detects that the temperature in the upper right corner area is continuously high, and the corresponding weight update will enhance the cooling control weight related to the area, while reducing the heating control weight.

[0094] The third stage is parameter projection, which ensures that the updated parameters still meet the positive definiteness constraint. For example, after a certain update, the eigenvalue of the 4×4 weight submatrix is ​​[1.2, 0.8, -0.1, 0.5], in which the negative eigenvalue -0.1 appears. The system will project it to the nearest positive definite matrix through exponential mapping to obtain the new eigenvalue distribution [1.2, 0.8, 0.1, 0.5]. This projection ensures the stability of the controller.

[0095] In practical applications, this optimization method shows good convergence and robustness. For example, during the mold heating stage, the controller can smoothly raise the temperature to the target value to avoid overshoot. Specifically, when the mold is heated from room temperature to 165°C, the heating curve is approximately exponential, and the heating power is automatically reduced when approaching the target temperature. The final temperature deviation is controlled within ±0.5°C. At the same time, the temperature difference in different areas is controlled within 2°C, and the pressure fluctuation does not exceed 5%.

[0096] This optimization method based on Riemannian geometry is particularly suitable for dealing with control problems with strong coupling characteristics such as temperature fields. For example, when an abnormal temperature in a certain area is detected, the controller will not only adjust the control parameters of the area, but also simultaneously optimize the parameters of the adjacent areas to achieve an overall coordinated control effect. During the production process, if the efficiency of a heater decreases, the system will automatically adjust the relevant weights to compensate for this performance degradation and maintain the overall control effect.

[0097] The system also has adaptive characteristics. When process conditions change (such as changing to different materials), the controller can quickly adjust parameters while maintaining stability. For example, when switching from PET material to ABS material, the system will automatically adapt to the thermal characteristics of the new material within 2-3 production cycles and adjust the control strategy to ensure stable product quality. This adaptive process always stays within the positive constraint range, avoiding the risk of controller instability.

[0098] S5: Based on the power control signal of the heater and the flow control instruction of the water cooling system, dynamically controlling the temperature field of the injection molding mold through an adjustable gap compensation mechanism; Step S5 may specifically include steps S5.1 to S5.3: S5.1: Based on the power control signal of the heater, the power of the heater in each area of ​​the injection molding mold is modulated through the thyristor module, and based on the flow control instruction of the water cooling system, the cooling water flow of the water cooling system in each area of ​​the injection molding mold is controlled through a high-precision proportional valve; In this embodiment, the temperature field of the injection molding mold is dynamically controlled based on the power control signal of the heater in each area and the flow control instruction of the water cooling system. In the specific implementation, an industrial-grade PLC controller is used as the execution unit, and communicates with each actuator through a high-speed field bus. For the heating system, an SCR (silicon controlled rectifier) ​​module is used for precise power modulation, supporting a regulation accuracy of 0.1%. For example, when the controller outputs a certain area that requires 75% power, the silicon controlled rectifier module will accurately control the conduction angle in each power frequency cycle to achieve accurate power modulation. In addition, to avoid electromagnetic interference, the system is triggered near the zero crossing point and a minimum conduction time limit (such as 1ms) is set. For the water cooling system, a high-precision proportional valve is used to control the cooling water flow, and the control resolution reaches 0.1L / min. According to the flow control instruction, the proportional valve is driven by a PWM signal to achieve precise flow regulation. For example, when the cooling water flow in a certain area needs to be reduced from 2.5L / min to 1.8L / min, the proportional valve will smoothly complete the adjustment within 100ms to avoid temperature fluctuations caused by sudden changes in flow. It should be noted that the system is also designed with a hardware redundancy mechanism to ensure that basic temperature control functions can be maintained when a single execution unit fails.

[0099] S5.2: real-time monitoring of the displacement change of the adjustable gap compensation mechanism, and adjusting the gap size of the cavity of the injection molding mold based on the displacement change; In this embodiment, the displacement change of the gap compensation mechanism is monitored and adjusted in real time by an intelligent servo driver. The gap compensation mechanism adopts a high-precision electric push rod design with micron-level positioning accuracy. In actual operation, the system first calculates the theoretical compensation amount based on the thermal deformation model, and then achieves precise positioning through closed-loop displacement feedback. Specifically, the displacement of the gap compensation mechanism is monitored in real time by a high-precision displacement sensor (resolution 0.001mm), the displacement data sampling frequency is 100Hz, and high-frequency noise is removed by low-pass filtering. When the cavity gap change is detected to exceed the preset threshold (such as 0.02mm), the system will start compensation adjustment. For example, during the injection molding process, if it is detected that the gap in a certain area has increased by 0.03mm due to thermal expansion, the compensation mechanism will perform precise compensation within 200ms and adjust the gap back to the target value. This process is gradual to avoid sudden mechanical movement affecting the molding quality. At the same time, the system will record the gap change law under different working conditions for predictive compensation.

[0100] In addition, in order to prevent damage to the mold due to overcompensation, multiple safety protection mechanisms are set up, including displacement limit, torque protection and collision detection. It is particularly worth mentioning that the system has also established a mapping relationship database between thermal deformation and gap compensation, which can quickly determine the optimal compensation strategy based on historical experience.

[0101] S5.3: Based on the temperature data and pressure data of the injection molding mold collected in real time, PID parameters are adjusted through a fuzzy adaptive algorithm, and the prediction period and control time domain of the model predictive controller are optimized based on the reinforcement learning method to obtain optimized control parameters. When it is detected that the process parameters meet the preset change conditions, the parameter re-optimization process is triggered.

[0102] In this embodiment, a complete set of dynamic parameter adjustment mechanism is established. Through the real-time acquisition of temperature and pressure data, the system can evaluate the current control effect and dynamically optimize the control parameters accordingly. Specifically, the PID parameters are adjusted online using a fuzzy adaptive algorithm, and the prediction cycle and control time domain of the model predictive controller are optimized based on the reinforcement learning method. When a significant change in the process parameters is detected, the system will automatically trigger the parameter re-optimization process to ensure that the control performance is always maintained in the best state. In addition, the system also has a process parameter self-learning function, which can continuously improve the control strategy library based on historical production data.

[0103] Specifically, in this embodiment, a two-layer adaptive control optimization is implemented. The first is the fuzzy adaptive adjustment of PID parameters. According to the temperature error and the error change rate, the system divides the working state into 9 fuzzy areas (such as "large error positive-fast change positive", "small error-slow change negative", etc.). In each area, the system adjusts the PID parameters according to the preset fuzzy rules. For example, when it is detected that the temperature error is large and decreasing rapidly, the system will appropriately reduce the proportional gain and increase the integral time to avoid overshoot. These adjustments are smooth to ensure the stability of the control process.

[0104] The second is model predictive control optimization based on reinforcement learning. The system uses the control effect as a reward signal, including temperature control accuracy, energy efficiency and response speed. Through the Q-learning algorithm, the system learns the optimal prediction cycle and control time domain under different working conditions. For example, in the rapid heating stage, the system may choose a shorter prediction cycle (such as 2 seconds) and control time domain (such as 1 second) to obtain a quick response; in the stable production stage, a longer prediction cycle (such as 10 seconds) and control time domain (such as 5 seconds) may be used to improve stability and energy saving.

[0105] The system sets multiple trigger conditions to start the parameter re-optimization process, including: process parameter changes exceeding the set threshold (such as temperature setting value changes exceeding 5°C), control performance degradation (such as temperature fluctuation increases by 20%), equipment status changes (such as mold or material replacement), etc. When re-optimization is triggered, the system will perform parameter optimization calculations in the background while maintaining the current control. For example, when it is detected that the efficiency of a heater has dropped by 15%, the system will re-optimize the control parameters of the area, including PID parameters and predictive control parameters, and appropriately adjust the parameters of adjacent areas to maintain the overall control effect.

[0106] This adaptive optimization mechanism enables the system to cope with various changes in working conditions. For example, when changing products of different thicknesses during production, the system can automatically adjust to the optimal control state within 2-3 production cycles. For another example, when the thermal characteristics of the mold change after long-term use, the system can maintain control performance through continuous parameter optimization. At the same time, the system will record the optimization experience under different working conditions and establish an experience library for rapid response to similar situations.

[0107] In practical applications, this multi-level control optimization strategy has shown excellent adaptability and stability. For example, in a certain production run, when the moisture content of the material suddenly increased, causing the molding pressure fluctuation to increase, the system was able to suppress the pressure fluctuation by adjusting the control parameters while maintaining the temperature field stable, ensuring stable product quality. The system also has a predictive maintenance function, which can detect potential equipment problems in advance by analyzing the long-term change trend of the control parameters, such as reduced heater efficiency or waterway blockage.

[0108] like Figure 6 As shown, the embodiment of the present application also provides an intelligent temperature adjustment system 500 for injection molding molds, which specifically includes: The data acquisition module 501 is used to collect temperature data and pressure data of each area of ​​the injection molding mold and generate a real-time multi-source sensor data stream; A feature recognition module 502 is used to extract features from the real-time multi-source sensor data stream through a Transformer model and an EM algorithm to obtain temperature field distribution features; The temperature prediction module 503 is used to train and obtain a temperature field distribution prediction model based on the temperature field distribution characteristics and the differential private Bayesian optimization algorithm, and output a mold temperature prediction result through the temperature field distribution prediction model; A control strategy module 504 is used to perform feature fusion on the mold temperature prediction result and the pressure data, construct input features of a neural network controller, and obtain the neural network controller through training based on the input features and a positive curvature learning method, and generate power control signals for heaters in each area of ​​the injection molding mold and flow control instructions for a water cooling system through the neural network controller; The execution control module 505 is used to dynamically control the temperature field of the injection molding mold through an adjustable gap compensation mechanism based on the power control signal of the heater and the flow control instruction of the water cooling system.

[0109] The system of the embodiments of the present disclosure can execute the method provided by the embodiments of the present disclosure, and the implementation principles are similar. The actions performed by each module in the device of each embodiment of the present disclosure correspond to the steps in the method of each embodiment of the present disclosure. For the detailed functional description of each module of the device, please refer to the description in the corresponding method shown in the previous text, which will not be repeated here.

[0110] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present disclosure, which are used to illustrate the technical solutions of the present disclosure, rather than to limit them. The protection scope of the present disclosure is not limited thereto. Although the present disclosure is described in detail with reference to the aforementioned embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the aforementioned embodiments within the technical scope disclosed in the present disclosure, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure, and should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure shall be based on the protection scope of the claims.

Claims

1. An intelligent temperature adjustment method for an injection molding mold, characterized in that: include: Collect temperature and pressure data from all areas of the injection molding mold to generate real-time multi-source sensor data streams; By using the Transformer model and the EM algorithm, feature extraction is performed on the real-time multi-source sensor data stream to obtain temperature field distribution features; Based on the temperature field distribution characteristics and the differential private Bayesian optimization algorithm, a temperature field distribution prediction model is trained and obtained, and a mold temperature prediction result is output through the temperature field distribution prediction model; Performing feature fusion on the mold temperature prediction result and the pressure data to construct input features of a neural network controller, and training the neural network controller based on the input features and a positive curvature learning method to obtain the neural network controller, and generating power control signals for heaters in each area of ​​the injection molding mold and flow control instructions for a water cooling system through the neural network controller; Based on the power control signal of the heater and the flow control instruction of the water cooling system, the temperature field of the injection molding mold is dynamically controlled through an adjustable gap compensation mechanism.

2. The method according to claim 1, characterized in that The method of collecting temperature data and pressure data of each area of ​​the injection molding mold to generate a real-time multi-source sensor data stream includes: The temperature data collected by the temperature sensor and the pressure data collected by the pressure sensor are converted into digital form through a high-speed data acquisition card to obtain digital signals; Performing signal filtering, data calibration, and outlier elimination on the digital signal to generate preprocessed data; A distributed data caching mechanism is established to synchronize and align the pre-processed data according to timestamps to generate a real-time multi-source sensor data stream.

3. The method according to claim 1, characterized in that The method of extracting features from the real-time multi-source sensor data stream by using the Transformer model and the EM algorithm to obtain temperature field distribution features includes: By using a position encoder of a Transformer model, the real-time multi-source sensor data stream is added with time series position information to generate an initial feature representation; Based on the initial feature representation, a multi-head attention mechanism is used to calculate the similarity between the query matrix, the key matrix and the value matrix, and generate an attention weight matrix for reflecting the spatiotemporal correlation of the data; Using a residual connection structure, the attention weight matrix is ​​weightedly fused with the initial feature representation to obtain an enhanced feature representation; The enhanced feature representation is clustered and analyzed by an EM algorithm to obtain the temperature field distribution feature.

4. The method according to claim 3, characterized in that The step of performing cluster analysis on the enhanced feature representation by using an EM algorithm to obtain the temperature field distribution feature includes: Based on the enhanced feature representation, K cluster centers are randomly initialized to obtain initial clustering parameters, wherein the K value is adaptively determined by the silhouette coefficient; According to the initial clustering parameters, alternately performing the steps of calculating the probability of belonging of data points and updating the cluster centers, and dynamically adjusting the parameter updating step size through an adaptive learning rate mechanism until a target clustering result is obtained; The temperature field distribution characteristics are obtained through the target clustering results, wherein the temperature field distribution characteristics include a temperature gradient vector field, a hot spot distribution characteristic map, and a temperature fluctuation spectrum.

5. The method according to claim 1, characterized in that The training to obtain a temperature field distribution prediction model based on the temperature field distribution characteristics and the differential private Bayesian optimization algorithm includes: Based on the temperature field distribution characteristics, constructing a priori distribution of a Bayesian probability model; The model parameters are randomly perturbed through the differential privacy mechanism, and the target model parameters are searched through the Bayesian optimization method; Based on the target model parameters, a temperature field distribution prediction model is trained.

6. The method according to claim 5, characterized in that The prior distribution of the Bayesian probability model is constructed based on the temperature field distribution characteristics, including: By multi-dimensional Gaussian distribution modeling and historical data analysis, an initial prior distribution is obtained based on the temperature gradient vector field and the hot spot distribution characteristic map in the temperature field distribution characteristics; Based on the initial prior distribution, the parameter distribution of different working conditions is automatically learned through a hierarchical Bayesian structure to generate a hierarchical prior model; Constructing a Bayesian probability model based on the hierarchical prior model and outputting prior distribution parameters; Furthermore, the method of randomly perturbing the model parameters by using the differential privacy mechanism and searching the target model parameters by using the Bayesian optimization method includes: Add random noise to the prior distribution parameters through the Laplace mechanism, and dynamically adjust the privacy budget based on the parameter sensitivity to obtain the disturbed parameters; A proxy model is constructed by Gaussian process regression, parameter performance evaluation is performed on the perturbed parameters, and a candidate parameter set is obtained by exploring and utilizing a balance strategy, wherein the evaluation indicators of the proxy model include prediction accuracy and computational efficiency; For the candidate parameter set, model parameter optimization is performed using a variational inference method and a method to minimize the lower bound of evidence, and the target model parameters are output.

7. The method according to claim 1, characterized in that The step of fusing the mold temperature prediction result and the pressure data to construct input features of a neural network controller includes: The mold temperature prediction result and the real-time pressure data are constructed into a time series feature sequence through a sliding time window mechanism; By using an attention mechanism, the temporal feature sequence is weighted to obtain a weighted feature sequence; The weighted feature sequence is converted into a feature vector to generate input features of a neural network controller.

8. The method according to claim 1, characterized in that The method of obtaining the neural network controller based on the input feature and the positive curvature learning method training includes: According to the input features, a loss function based on the Riemann metric is constructed, wherein the loss function includes a temperature control error term, a pressure balance term, and an energy consumption constraint term, and a curvature regularization term is obtained to obtain an optimization objective function; According to the optimization objective function, the Riemann gradient is calculated in the tangent space of the parameters, and the updated parameters are projected back to the manifold space satisfying the positive definite constraint through exponential mapping to obtain the optimized network parameters.

9. The method according to claim 1, characterized in that: The method dynamically controls the temperature field of the injection molding die through an adjustable gap compensation mechanism based on the power control signal of the heater and the flow control instruction of the water cooling system, including: Based on the power control signal of the heater, the power of the heater in each area of ​​the injection molding mold is modulated through a thyristor module, and based on the flow control instruction of the water cooling system, the cooling water flow of the water cooling system in each area of ​​the injection molding mold is controlled through a high-precision proportional valve; Real-time monitoring of the displacement change of the adjustable gap compensation mechanism, and adjusting the gap size of the cavity of the injection molding mold based on the displacement change; Based on the temperature data and pressure data of the injection molding mold collected in real time, PID parameters are adjusted through a fuzzy adaptive algorithm, and the prediction period and control time domain of the model predictive controller are optimized based on the reinforcement learning method to obtain the optimized control parameters. When it is detected that the process parameters meet the preset change conditions, the parameter re-optimization process is triggered.

10. An intelligent temperature control system for injection molding molds, characterized in that: include: Data acquisition module, used to collect temperature and pressure data of various areas of the injection molding mold and generate real-time multi-source sensor data stream; A feature recognition module is used to extract features from the real-time multi-source sensor data stream through a Transformer model and an EM algorithm to obtain temperature field distribution features; A temperature prediction module, used for training a temperature field distribution prediction model based on the temperature field distribution characteristics and a differential private Bayesian optimization algorithm, and outputting a mold temperature prediction result through the temperature field distribution prediction model; A control strategy module, used to perform feature fusion on the mold temperature prediction result and the pressure data, construct input features of a neural network controller, and obtain the neural network controller through training based on the input features and a positive curvature learning method, and generate power control signals for heaters in each area of ​​the injection molding mold and flow control instructions for a water cooling system through the neural network controller; The execution control module is used to dynamically control the temperature field of the injection molding mold through an adjustable gap compensation mechanism based on the power control signal of the heater and the flow control instruction of the water cooling system.

Images