A method and system for monitoring and controlling processing parameters of traditional Chinese medicine decoction pieces

Abstract

This application provides a method and system for monitoring and controlling processing parameters of traditional Chinese medicine (TCM) decoction pieces, relating to the field of parameter monitoring and control technology. It constructs a digital twin of the processing process of the target TCM decoction pieces; collects process parameter streams from the physical space in real time through a multimodal sensing network, and synchronously maps these streams to the digital twin, driving the real-time synchronous evolution of the digital twin and the physical space processing process; based on the real-time evolution status of the digital twin, it identifies abnormal patterns in the process parameter streams and predicts the deviation characteristics of the final processing quality under these abnormal patterns; when the deviation characteristics exceed preset multi-level tolerance thresholds, it triggers dynamic reprogramming of the processing process. This application can achieve synchronous evolution of the virtual and real processing processes by constructing a digital twin, realizing the transformation of the TCM decoction piece processing process towards intelligent control and improving the process adaptability of processing quality.

Description

Technical Field

[0001] This application relates to the field of parameter monitoring and control technology, and more specifically, to a method and system for monitoring and controlling parameters in the processing of traditional Chinese medicine decoction pieces. Background Technology

[0002] Currently, the control methods for processing equipment of traditional Chinese medicine decoction pieces are mainly divided into three categories: The first category is the traditional manual operation method, which relies on the operator's sensory experience to judge the degree of processing and manually adjust the heating power and stirring speed; the second category is the single-parameter closed-loop control method, which installs temperature sensors on the processing equipment and realizes single-loop PID control of temperature through a programmable logic controller; the third category is the formula parameter control method, which stores the pre-set process parameter curves in the controller and the equipment automatically executes according to the preset timing sequence.

[0003] While some existing technologies employ the finite element method (FEM) for temperature field simulation analysis of processing equipment, these methods are mostly used in the offline design phase and lack real-time synchronization with the physical process. This prevents them from providing a dynamic virtual mirror for online control. Current control systems lack comprehensive perception capabilities of multi-physics coupling states, cannot identify the root causes of abnormal patterns, and cannot predict the cumulative impact of anomalies on final quality. When process deviations occur, operators often only discover quality defects after processing is completed through sampling and testing, resulting in batch scrapping. Therefore, how to achieve synchronized virtual and real evolution of the processing process by constructing a digital twin, realizing the transformation of the processing of traditional Chinese medicine decoction pieces towards intelligent control, and improving the process adaptability of processing quality is a challenge facing the industry. Summary of the Invention

[0004] This application provides a method and system for monitoring and controlling the processing parameters of traditional Chinese medicine decoction pieces. By constructing a digital twin, the virtual and real evolution of the processing process can be realized, thereby transforming the processing of traditional Chinese medicine decoction pieces into intelligent control and improving the process self-adaptive capability of processing quality.

[0005] In a first aspect, this application provides a method for monitoring and controlling processing parameters of traditional Chinese medicine decoction pieces, the method comprising the following steps: Constructing a digital twin of the processing process of target Chinese herbal medicine pieces; The process parameter flow in the physical space is collected in real time through a multimodal sensing network, and the process parameter flow is synchronously mapped to the digital twin, driving the digital twin to evolve synchronously with the physical space processing process in real time. Based on the real-time evolution state of the digital twin, abnormal patterns in the process parameter flow are identified, and deviation characteristics of the final quality of the processing under the abnormal patterns are predicted. When the deviation feature exceeds a preset multi-level tolerance threshold, dynamic reprogramming of the processing technology is triggered.

[0006] In this embodiment, the digital twin includes a physical processing equipment entity in physical space and a virtual processing image in information space, wherein the virtual processing image integrates a multi-physics coupling model.

[0007] In this embodiment, the multimodal sensing network includes: A fiber optic temperature sensor array for collecting three-dimensional temperature field data is deployed on the cylinder wall, stirring blades and inside the material of the processing equipment. Terahertz spectroscopy detection device for penetrating and collecting data on the internal moisture distribution and composition changes of materials; A high-speed binocular vision system for acquiring three-dimensional motion trajectory and morphological evolution data of materials; An acoustic emission sensor array is deployed on the cylinder wall of the processing equipment to collect acoustic signature data generated by the collision between materials and the equipment wall during the processing.

[0008] In this embodiment, synchronously mapping the process parameter stream to the digital twin and driving the digital twin to evolve synchronously with the physical space fabrication process in real time specifically includes: The process parameter stream is aligned with the time sequence and spatial coordinates to construct spatiotemporally synchronized data frames; Each data frame is input as a boundary condition into the multiphysics coupling model of the digital twin, driving the multiphysics coupling model to perform real-time iterative calculations; The calculation results of the multiphysics coupling model are compared with the residuals of the real-time acquired process parameter stream, and the internal parameters of the multiphysics coupling model are adjusted.

[0009] In this embodiment, based on the real-time evolution state of the digital twin, identifying abnormal patterns in the process parameter stream and predicting the deviation characteristics of the final processing quality under the abnormal patterns specifically includes: Obtain the multiphysics evolution state sequence of the digital twin within a continuous time window; The multiphysics field evolution state sequence is input into a spatiotemporal feature extraction network based on a graph convolutional network and a long short-term memory network to extract the spatiotemporal evolution feature vector of the processing process. The spatiotemporal evolution feature vector is input into a single classifier based on support vector data description to identify abnormal patterns; When an abnormal pattern is identified, a reverse attribution analysis is performed on the spatiotemporal feature extraction network to determine the contribution of each spatial node in each physical field to the abnormal pattern and locate the root cause of the anomaly. The spatiotemporal evolution feature vector and root cause results are input into a predictor based on a temporal convolutional network to predict the deviation characteristics of the current abnormal mode from the quality index of the final processing time within the remaining processing time.

[0010] In this embodiment, the deviation feature is a vector representing the degree of deviation of each quality indicator from the expected value of the normal process under the current abnormal state.

[0011] In this embodiment, triggering dynamic reprogramming of the processing technique when the deviation characteristic exceeds a preset multi-level tolerance threshold specifically includes: A multi-level tolerance threshold is preset, wherein the multi-level tolerance threshold includes a first tolerance threshold and a second tolerance threshold, and the first tolerance threshold is less than the second tolerance threshold; When the deviation feature is less than the first tolerance threshold, the processing procedure is determined to be under normal operating conditions. When the deviation feature is greater than the first tolerance threshold and less than the second tolerance threshold, a local parameter correction mode is triggered; When the deviation feature is greater than the second tolerance threshold, the global process reprogramming mode is triggered.

[0012] In this embodiment, the global process reprogramming mode specifically includes: Freeze the current physical space process execution state and record the current snapshot state of the digital twin; The snapshot state is input into a preset generative process reconstruction model to generate multiple candidate process compensation trajectories; Multiple candidate process compensation trajectories are input into the digital twin for multiphysics simulation and deduction to determine the corresponding quality prediction value, energy consumption prediction value and process fluctuation index, and then determine the final compensation trajectory. The final compensation trajectory is smoothly connected with the remaining process trajectory of the current processing stage to generate a complete sequence of control instructions for the remaining stage.

[0013] In this embodiment, the final compensation trajectory refers to the time sequence of process parameters for the remaining stages that can achieve the best quality index at the end of the processing, the lowest energy consumption, and the smallest process fluctuation under the current processing state.

[0014] Secondly, this application provides a monitoring and control system for processing parameters of traditional Chinese medicine decoction pieces, used to execute a method for monitoring and controlling processing parameters of traditional Chinese medicine decoction pieces, the parameter monitoring and control system comprising: The twin building block is used to construct a digital twin of the processing process of the target Chinese herbal medicine pieces. The synchronous evolution module is used to collect the process parameter flow of the physical space in real time through a multimodal sensing network, and synchronously map the process parameter flow to the digital twin, driving the digital twin to evolve synchronously with the physical space processing process in real time. The cumulative deviation module is used to identify abnormal patterns in the process parameter flow based on the real-time evolution state of the digital twin, and predict the deviation characteristics of the final quality of the processing under the abnormal patterns. The dynamic programming module is used to trigger dynamic reprogramming of the processing technology when the deviation feature exceeds a preset multi-level tolerance threshold.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: A digital twin of the processing process of the target Chinese herbal medicine slices is constructed; the process parameter flow in the physical space is collected in real time through a multimodal sensing network, and the process parameter flow is synchronously mapped to the digital twin, driving the digital twin to evolve synchronously with the processing process in the physical space in real time; based on the real-time evolution status of the digital twin, abnormal patterns in the process parameter flow are identified, and the deviation characteristics of the final quality of the processing under the abnormal patterns are predicted; when the deviation characteristics exceed the preset multi-level tolerance thresholds, the dynamic reprogramming of the processing process is triggered.

[0016] Therefore, this application firstly achieves digital characterization of the multi-field coupling evolution law of materials during the processing process by constructing a digital twin of the target Chinese herbal medicine processing process. This digital twin can accurately reproduce the dynamic evolution of the physical processing process in the information space, laying a digital foundation for the intelligent and precise control of Chinese herbal medicine processing. Secondly, by constructing a multimodal sensing network and establishing a synchronous mapping mechanism between process parameter flow and the digital twin, the perception and virtual-real synchronous evolution of the Chinese herbal medicine processing process are realized. This enables the real-time and accurate mapping of multi-source heterogeneous data collected in the physical space to the digital twin, driving the virtual image and physical entity to maintain high-fidelity synchronous evolution, improving the reliability, safety, and response speed of process control, and providing a foundation for the intelligent and precise control of Chinese herbal medicine processing. The intelligent closed-loop control provides data support and model foundation. Then, through anomaly pattern recognition and deviation feature prediction based on the real-time evolution state of the digital twin, a spatiotemporal feature extraction model integrating graph convolutional networks and long short-term memory networks is constructed. This model can automatically extract deep spatiotemporal coupling features from the multi-physics field evolution state sequence output by the digital twin, accurately trace the spatial location and parameter type of anomalies, and provide a decision basis for subsequent dynamic reprogramming of the process, thereby improving the quality assurance capability of traditional Chinese medicine decoction piece processing. Finally, by constructing a multi-level tolerance threshold and dynamic reprogramming mechanism, differentiated and precise intelligent intervention on deviations in the processing technology of traditional Chinese medicine decoction pieces is realized, improving the adaptive control level, quality assurance capability, and process stability of traditional Chinese medicine decoction piece processing.

[0017] In summary, the technical solution adopted in this application can realize the synchronous evolution of the processing process in both virtual and real worlds by constructing a digital twin, thereby transforming the processing of Chinese herbal medicine slices into intelligent control and improving the process adaptability of processing quality. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this embodiment of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is an exemplary flowchart of a method for monitoring and controlling processing parameters of traditional Chinese medicine decoction pieces provided in this application; Figure 2 This is a flowchart illustrating the internal processing of a digital twin provided in this application; Figure 3 This is a schematic diagram of a system architecture for monitoring and controlling the processing parameters of traditional Chinese medicine decoction pieces, provided in this application. Figure 4 This is a module structure diagram of a monitoring and control system for processing parameters of traditional Chinese medicine decoction pieces provided in this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0021] This application provides a method and system for monitoring and controlling the processing parameters of traditional Chinese medicine decoction pieces. The core of this method is to construct a digital twin of the processing process of the target traditional Chinese medicine decoction pieces; to collect the process parameter flow in physical space in real time through a multimodal sensing network, and to synchronously map the process parameter flow to the digital twin, driving the digital twin to evolve synchronously with the physical space processing process in real time; based on the real-time evolution state of the digital twin, to identify abnormal patterns in the process parameter flow, and to predict the deviation characteristics of the final processing quality under the abnormal patterns; when the deviation characteristics exceed preset multi-level tolerance thresholds, to trigger dynamic reprogramming of the processing process.

[0022] Example 1: To better understand the above technical solution, the following will provide a detailed description of the technical solution in conjunction with the accompanying drawings and specific implementation methods. (Refer to...) Figure 1 As shown in the figure, this is an exemplary flowchart of a method for monitoring and controlling processing parameters of traditional Chinese medicine decoction pieces according to this embodiment of the application. The parameter monitoring and control method includes the following steps: In step S1, a digital twin of the processing process of the target Chinese herbal medicine is constructed.

[0023] In its specific implementation, firstly, the digital twin includes a physical processing equipment entity in physical space and a virtual processing image in information space. The virtual processing image integrates a multi-physics coupling model. For the processing equipment of the target Chinese medicinal herb, a high-precision three-dimensional geometric model is established. This model includes not only the equipment's external dimensions, cylinder structure, and the geometry and spatial layout of the stirring blades, but also the distribution of heating elements, the deployment of temperature measuring points, and the structural features of the inlet and outlet. Kinematic modeling is performed on the key moving parts of the processing equipment, defining the stirring blade's rotation speed range, rotation direction, and the material's trajectory constraints within the cylinder. The multi-physics coupling model includes temperature... The system includes a coupled temperature field-flow field sub-model, a moisture content migration sub-model, and a material motion morphology sub-model. The coupled temperature field-flow field sub-model is constructed based on computational fluid dynamics, treating the hot air inside the processing equipment as a compressible fluid. The Navier-Stokes equations describe airflow motion, and the energy conservation equation describes heat transfer. This sub-model considers the thermal radiation effect of the heating element, the convective heat transfer effect between the cylinder wall and the material, and the heat conduction effect within the material pile. It can output the temperature variation over time at any point inside the equipment. The moisture content migration sub-model is constructed based on Fick's diffusion law, treating the material as a porous medium and considering the moisture content within the material particles. The moisture content migration sub-model introduces the concept of water activity to describe the thermodynamic equilibrium relationship between material moisture content and temperature and relative humidity, and considers the evaporation phase change process of moisture on the material surface. It can output the evolution law of moisture content distribution inside the material pile over time. The material motion morphology sub-model is constructed based on the discrete element method, treating each material particle as an independent computational unit. It uses Hertz-Mundling contact theory to describe the collision and friction behavior between particles and between particles and the equipment wall. It uses Newton's second law to calculate the displacement, velocity, and acceleration of each particle under stirring. This material motion morphology sub-model can output the three-dimensional spatial distribution of the material pile. The three sub-models, including key kinematic parameters such as mixing uniformity and contact frequency with the wall, interact in a loosely coupled manner. Within each computation time step, the material motion morphology sub-model first calculates the spatial distribution of the material, which serves as the basis for mesh generation in the temperature field-flow field coupled sub-model and the moisture content migration sub-model. Subsequently, the temperature field-flow field coupled sub-model calculates the temperature field distribution under the current distribution, and the moisture content migration sub-model calculates the moisture migration process under the current distribution. Finally, the results of the temperature field and moisture content field are fed back to the material motion morphology sub-model to update the thermal property parameters of the material, thereby obtaining a digital twin of the target Chinese herbal medicine processing process.

[0024] In step S2, the process parameter flow in the physical space is collected in real time through a multimodal sensing network, and the process parameter flow is synchronously mapped to the digital twin, driving the digital twin to evolve synchronously with the physical space fabrication process in real time.

[0025] In this embodiment, the multimodal sensing network includes: A fiber optic temperature sensor array for collecting three-dimensional temperature field data is deployed on the cylinder wall, stirring blades and inside the material of the processing equipment. Terahertz spectroscopy detection device for penetrating and collecting data on the internal moisture distribution and composition changes of materials; A high-speed binocular vision system for acquiring three-dimensional motion trajectory and morphological evolution data of materials; An acoustic emission sensor array is deployed on the cylinder wall of the processing equipment to collect acoustic signature data generated by the collision between materials and the equipment wall during the processing.

[0026] In practical implementation, a multimodal sensing network can be used to collect process parameter streams in physical space in real time. Specifically, a multi-point cascaded fiber optic grating sensor is used to collect three-dimensional temperature field distribution data, including cylinder wall temperature, blade interface temperature, and temperature changes at different depths inside the material. A terahertz spectroscopy detection device is used to collect the moisture distribution and key active ingredient content changes inside the material non-contactly by utilizing the penetrating characteristics of terahertz waves. A high-speed binocular vision system, consisting of two high-speed industrial cameras, is installed outside the observation window. A binocular stereo vision algorithm is used to reconstruct the three-dimensional point cloud of the material in real time, extracting morphological evolution features such as material volume changes, centroid motion trajectory, and surface color texture. An acoustic emission sensor array is deployed circumferentially at equal angles on the outside of the barrel wall of the processing equipment to collect acoustic signature signals generated by the collision between the material and the equipment wall during the processing. After feature extraction, time-domain, frequency-domain, and time-frequency-domain feature parameters are obtained. The data collected by the above four types of sensors are used as the process parameter stream.

[0027] In this embodiment, the process parameter stream is synchronously mapped to the digital twin, driving the digital twin to evolve synchronously with the physical space fabrication process in real time. This can be achieved through the following steps: The process parameter stream is aligned with the time sequence and spatial coordinates to construct spatiotemporally synchronized data frames; Each data frame is input as a boundary condition into the multiphysics coupling model of the digital twin, driving the multiphysics coupling model to perform real-time iterative calculations; The calculation results of the multiphysics coupling model are compared with the residuals of the real-time acquired process parameter stream, and the internal parameters of the multiphysics coupling model are adjusted.

[0028] In practical implementation, firstly, the process parameter streams can be aligned with time series and spatial coordinates to construct spatiotemporally synchronized data frames. Specifically, in the time dimension, the IEEE 1588 precise time protocol is used to connect all sensors to a unified time synchronization network, ensuring that the data output from the fiber optic temperature sensing array, terahertz spectroscopy detection device, high-speed binocular vision system, and acoustic emission sensor array all carry timestamps. Then, time registration is performed on the multi-source data using fixed time windows, and all sensor data falling within the same time window are sorted according to their timestamps to form a time-aligned data set. In the spatial dimension, all sensor data are mapped to a pre-constructed three-dimensional spatial coordinate system of the processing equipment. The coordinates of each sensing point of the fiber optic temperature sensing array have been precisely calibrated during deployment; the detection spot of the terahertz spectroscopy detection device... The location is determined by a laser positioning device, which determines its coordinates in the spatial coordinate system. The three-dimensional point cloud data reconstructed by the high-speed binocular vision system itself contains spatial coordinate information. The installation positions of each sensor in the acoustic emission sensor array also have clear coordinate labels. Spatial coordinate attributes are added to each channel of sensor data to form a triplet data unit. All triplet data units in different spatial locations within the same time window are then encapsulated to construct a spatiotemporally synchronized data frame. Each data frame contains complete multi-physics field state information inside the equipment at the current moment, including three-dimensional temperature field distribution, material moisture content distribution, material morphology point cloud data, and acoustic emission characteristic parameters.Then, each data frame can be input as a boundary condition into the multiphysics coupling model of the digital twin, driving the multiphysics coupling model to perform real-time iterative calculations. That is, each data frame is transmitted to the solver of the digital twin in real time as the boundary and initial conditions of the multiphysics coupling model, driving the model to perform iterative calculations. The temperature field-flow field coupling sub-model receives the three-dimensional temperature field distribution data in the data frame and loads it onto the corresponding nodes of the model mesh as temperature boundary conditions. For areas not covered by fiber optic grating sensing points, the system uses the radial basis function interpolation method to perform spatial interpolation based on the temperature values ​​of adjacent sensing points to generate a complete temperature boundary field. The moisture content migration sub-model receives the terahertz spectrum inversion results in the data frame and loads the moisture content distribution data inside the material into the model mesh as initial conditions. At the same time, the moisture content migration sub-model also receives the temperature distribution data output by the temperature field-flow field coupling sub-model to update parameters such as the moisture diffusion coefficient and evaporation rate. The material motion morphology sub-model receives the binocular data in the data frame. Visual 3D point cloud data is used to align the spatial distribution of materials in the model with the actual point cloud using a point cloud registration algorithm. This adjusts the initial position and packing morphology of the material particles in the model. Simultaneously, the material motion morphology sub-model receives acoustic signature parameters extracted from an acoustic emission sensor array to correct parameters such as the collision recovery coefficient and friction coefficient between particles. After receiving boundary and initial conditions, the three sub-models iteratively solve the problem based on the current computation time step. Within each computation step, the three sub-models interact in a loosely coupled manner: the material motion morphology sub-model outputs the spatial distribution of materials, serving as the basis for mesh generation in the temperature field-flow field sub-model and the moisture content migration sub-model; the temperature field-flow field sub-model calculates temperature field updates; the moisture content migration sub-model calculates moisture field updates; and the updated temperature and moisture fields are fed back to the material motion morphology sub-model to update the material's thermal properties. After completing one computation step iteration, the evolution state of the digital twin advances by one time step, achieving synchronous evolution with the physical space processing process.

[0029] In addition, in practical implementation, the calculation results of the multiphysics coupling model can be compared with the real-time acquired process parameter streams using residuals. The internal parameters of the multiphysics coupling model can then be adjusted. Specifically, at the end of each synchronization cycle, the system compares the calculation results of the digital twin at the end of the current cycle with the actual process parameter streams acquired in the physical space at the same moment using residuals. The residual calculation model for temperature field shows the root mean square error between the predicted temperature and the actual measured temperature at all sensing points; the residual calculation model for moisture content shows the deviation between the predicted moisture content and the terahertz inversion moisture content; the residual calculation model for morphology field shows the relative error between the predicted material volume and the binocular vision reconstructed volume; and the residual calculation model for acoustic signature shows the difference between the predicted collision energy and the actual acoustic emission signal energy. When any of the above residuals exceeds a preset synchronization error threshold, an adaptive adjustment mechanism for model parameters is triggered. The extended Kalman filter method is used to treat the uncertain parameters in the multiphysics coupling model, such as the interparticle heat transfer coefficient, moisture diffusion coefficient, interparticle friction coefficient, and collision recovery coefficient, as state variables, and the residuals as observation values. The uncertain parameters are then estimated and corrected online using the Kalman filter recursive formula.

[0030] like Figure 2 The flowchart illustrating the internal processing of the digital twin demonstrates the complete closed-loop process of process parameter flow from acquisition to virtual-real synchronous evolution. First, the process parameter flow acquired by the multimodal sensing network undergoes spatiotemporal alignment processing to construct a spatiotemporal synchronous data frame containing timestamps, spatial coordinates, and sensor values. Subsequently, the data frame is input as a boundary condition into the multiphysics coupling model of the digital twin, driving the temperature field-flow field coupling sub-model, moisture content migration sub-model, and material motion morphology sub-model to perform interactive iterative calculations in a loosely coupled manner, completing the virtual-real evolution of one time step and outputting the evolution state of the virtual image. Finally, the calculation results of the virtual image are compared with the process parameter flow acquired in real time in physical space for residual comparison. The entire process is executed cyclically, forming a closed-loop virtual-real synchronous mapping mechanism from spatiotemporal alignment, boundary driving, iterative calculation to residual correction and parameter adaptation.

[0031] It should be noted that by constructing a multimodal perception network and establishing a synchronous mapping mechanism between process parameter flow and digital twin, the perception and virtual-real synchronous evolution of the processing of Chinese herbal medicine pieces are realized. This enables the real-time and accurate mapping of multi-source heterogeneous data collected in physical space to the digital twin, driving the virtual image and physical entity to maintain high-fidelity synchronous evolution. This improves the reliability, safety and response speed of process control, and provides data support and model foundation for the intelligent closed-loop control of Chinese herbal medicine piece processing.

[0032] In step S3, based on the real-time evolution state of the digital twin, abnormal patterns in the process parameter flow are identified, and deviation characteristics of the final quality of the processing under the abnormal patterns are predicted.

[0033] Preferably, in this embodiment, identifying abnormal patterns in the process parameter stream based on the real-time evolution state of the digital twin and predicting the deviation characteristics of the final processing quality under the abnormal patterns can be achieved through the following steps: Obtain the multiphysics evolution state sequence of the digital twin within a continuous time window; The multiphysics field evolution state sequence is input into a spatiotemporal feature extraction network based on a graph convolutional network and a long short-term memory network to extract the spatiotemporal evolution feature vector of the processing process. The spatiotemporal evolution feature vector is input into a single classifier based on support vector data description to identify abnormal patterns; When an abnormal pattern is identified, a reverse attribution analysis is performed on the spatiotemporal feature extraction network to determine the contribution of each spatial node in each physical field to the abnormal pattern and locate the root cause of the anomaly. The spatiotemporal evolution feature vector and root cause results are input into a predictor based on a temporal convolutional network to predict the deviation characteristics of the current abnormal mode from the quality index of the final processing time within the remaining processing time.

[0034] In specific implementation, firstly, the multiphysics evolution state sequence of the digital twin within a continuous time window can be obtained. That is, the time window length can be set to T, and the window sliding step size can be Δt. At each sampling moment, the digital twin outputs a snapshot of the multiphysics state at the current moment. This snapshot contains the following four types of data: temperature field data represented as a three-dimensional tensor, moisture content field data represented as a three-dimensional tensor, material morphology data represented as a three-dimensional point cloud, and acoustic signature feature data represented as a feature vector. The above four types of data collected within the continuous time window are organized in chronological order to form a multiphysics evolution state sequence. Secondly, the multiphysics evolution state sequence can be input into a graph convolutional network and long short-term memory network-based system. In the spatiotemporal feature extraction network, the spatiotemporal evolution feature vector of the processing process is extracted. That is, the multiphysics state at each moment is modeled as a graph structure, which includes a set of spatial nodes: each grid cell corresponding to the spatial grid of the processing equipment; a set of edges connecting adjacent spatial nodes; and a node feature matrix, where the features of each node include the node's temperature value, moisture content value, point cloud density, and the projection value of the acoustic signature feature at the corresponding spatial location. The graph structure is input into the graph convolutional network module, which uses Chebyshev graph convolutional layers to reduce computational complexity through polynomial approximation, thereby outputting a spatial feature map. The spatial feature maps of each moment within a continuous time window are input into the long short-term memory network module in chronological order. The network employs a two-layer stacked structure, with each layer containing 128 hidden units. The output of the Long Short-Term Memory (LSTM) network is flattened and dimensionality reduced by a fully connected layer to obtain a spatiotemporal evolution feature vector. This spatiotemporal evolution feature vector can then be input into a single classifier based on support vector data description to identify abnormal patterns. Specifically, the single classifier can be pre-trained, with training data derived from spatiotemporal evolution feature vectors extracted during multiple batches of normal processing. At the end of each time window's sliding step, the current spatiotemporal evolution feature vector is input into the pre-trained support vector data description single classifier for anomaly detection. For example, the core of anomaly detection is the kernel distance from the current spatiotemporal evolution feature vector to the center of the pre-trained hypersphere. The calculated kernel distance is compared with the hypersphere radius obtained from pre-training. If the kernel distance is less than the hypersphere radius, it indicates that the current spatiotemporal evolution feature vector is located inside the normal working condition hypersphere, and the current processing process is determined to be in a normal state, outputting a normal label. If the kernel distance is greater than the hypersphere radius, it indicates that the current spatiotemporal evolution feature vector is located outside the normal working condition hypersphere, and the current processing process is determined to have an abnormal mode. The kernel distance is subtracted from the hypersphere radius, and the result is divided by the hypersphere radius. The result is used as the anomaly index, where the anomaly index represents the relative distance of the current state from the boundary of the normal working condition hypersphere. The larger the anomaly index, the more serious the degree of abnormal deviation, thus outputting an anomaly label and an anomaly index.

[0035] Furthermore, in the specific implementation, when an abnormal pattern is identified, a reverse attribution analysis is performed on the spatiotemporal feature extraction network to determine the contribution of each spatial node in each physical field to the abnormal pattern, and to locate the root cause of the anomaly. That is, when an abnormal pattern is identified, for the multi-physics evolution state sequence input at the current moment, a baseline input is first defined. The baseline input represents the reference state under normal working conditions. In this embodiment, the average value of each physical field parameter during the historical normal processing is taken. For the temperature field and moisture content field, the baseline is taken as the historical average value of each spatial node; for the material morphology point cloud, the baseline is taken as the average material accumulation morphology at the same stage during the normal processing; for acoustic features, the baseline... The mean of the feature vectors under normal operating conditions is taken, and the contribution of the i-th input feature to the anomaly index is determined by the integral gradient method. Since the integral is difficult to solve analytically, a numerical integration method is used for approximation in actual calculation. m sampling points are uniformly selected in the interval [0,1], and the gradient values ​​at each sampling point are calculated and accumulated. The input state at the sampling points is input into the model composed of the spatiotemporal feature extraction network and a single classifier. Through the automatic differentiation mechanism of the deep learning framework, the partial derivative vector of the model output anomaly index with respect to all input features is calculated at once. This operation is executed in parallel on the GPU through the backpropagation algorithm. The above process is repeated for all m sampling points to obtain the result. For m sets of gradient vectors, each gradient vector is multiplied by its corresponding interpolation step size and then summed to obtain the contribution of each input feature. For the temperature and moisture content fields, each spatial grid node corresponds to a contribution; for the material morphology point cloud, each point cloud point corresponds to a contribution; for the acoustic signature feature, each feature dimension corresponds to a contribution. For the temperature and moisture content fields, the contribution of each grid node in the temperature field is mapped back to the 3D grid according to spatial coordinates to generate an abnormal contribution spatial distribution map. Similarly, a contribution distribution map for the moisture content field is generated. The contribution distribution maps are then thresholded, and spatial nodes whose contribution exceeds twice the standard deviation of the global contribution mean are marked as... Candidate anomaly nodes are clustered using a density-based spatial clustering algorithm, grouping spatially adjacent candidate nodes into the same anomaly region. After clustering, the average contribution of all nodes within each anomaly region is calculated and sorted from highest to lowest average contribution. The anomaly region with the highest contribution is identified as the root cause location of the anomaly, and the three-dimensional spatial coordinate range of the root cause location is output. Then, the total contribution of the temperature field, moisture content field, material morphology field, and acoustic signature field is calculated and normalized for comparison. The physical field with the highest contribution is determined as the root cause parameter type. Finally, the three-dimensional spatial coordinate range of the root cause location and the root cause parameter type are used as the root cause result.Finally, the spatiotemporal evolution feature vector and root cause results can be input into a predictor based on a temporal convolutional network to predict the deviation features of the current abnormal mode from the final quality indicators of the processing time within the remaining processing time. Specifically, the root cause localization results and spatiotemporal evolution feature vectors are input into the predictor based on a temporal convolutional network. The predictor uses a dilated causal convolutional network as its core computational architecture. The network consists of four stacked dilated causal convolutional layers with dilation rates of 1, 2, 4, and 8, and a kernel size of 3. After each convolutional layer, batch normalization, ReLU activation, and residual connections are performed sequentially, and a dropout mechanism is introduced to prevent overfitting. After four convolutional layers and global average pooling, the features are input into a three-layer fully connected network, ultimately outputting deviation features. These deviation features are vectors representing the degree of deviation of each quality indicator from the expected value of the normal process under the current abnormal state.

[0036] It should be noted that, by recognizing abnormal patterns and predicting deviation features based on the real-time evolution state of digital twins, a spatiotemporal feature extraction model integrating graph convolutional networks and long short-term memory networks was constructed. This model can automatically extract deep spatiotemporal coupling features from the multi-physics field evolution state sequence output by the digital twin, accurately trace the spatial location and parameter type of the anomaly, provide a decision basis for subsequent dynamic reprogramming of the process, and improve the quality assurance capability of traditional Chinese medicine decoction piece processing.

[0037] In step S4, when the deviation feature exceeds the preset multi-level tolerance threshold, the dynamic reprogramming of the processing technology is triggered.

[0038] In this embodiment, when the deviation feature exceeds a preset multi-level tolerance threshold, triggering the dynamic reprogramming of the processing technology can be achieved through the following steps: A multi-level tolerance threshold is preset, wherein the multi-level tolerance threshold includes a first tolerance threshold and a second tolerance threshold, and the first tolerance threshold is less than the second tolerance threshold; When the deviation feature is less than the first tolerance threshold, the processing procedure is determined to be under normal operating conditions. When the deviation feature is greater than the first tolerance threshold and less than the second tolerance threshold, a local parameter correction mode is triggered; When the deviation feature is greater than the second tolerance threshold, the global process reprogramming mode is triggered.

[0039] In practical implementation, firstly, multiple tolerance thresholds can be preset, that is, multiple tolerance thresholds can be preset based on expert recommendations. These multiple tolerance thresholds include a first tolerance threshold and a second tolerance threshold, where the first tolerance threshold is less than the second tolerance threshold. These include: an effective ingredient content tolerance threshold, a moisture content tolerance threshold, and a color tolerance threshold. Secondly, when the deviation characteristic is less than the first tolerance threshold, the processing is determined to be under normal operating conditions; that is, when all deviation characteristics are less than the first tolerance threshold, the processing is determined to be under normal operating conditions, and no parameter correction is performed. Then, when the deviation characteristic is greater than the first tolerance threshold and less than... When the deviation characteristic reaches the second tolerance threshold, a local parameter correction mode is triggered. That is, when the deviation characteristic is greater than any tolerance threshold in the first tolerance threshold but less than the second tolerance threshold, the local parameter correction mode is triggered. Based on the root cause localization result, the type of parameter to be corrected is determined, and the target parameter is tuned online using a PID controller. Control commands are sent to the corresponding actuators to bring the process parameters back to the normal range. During the correction process, the change of deviation characteristic is continuously monitored. When the deviation characteristic falls back below the first threshold, the correction stops. Finally, when the deviation characteristic is greater than the second tolerance threshold, a global process reprogramming mode is triggered.

[0040] In this embodiment, the global process reprogramming mode specifically includes: Freeze the current physical space process execution state and record the current snapshot state of the digital twin; The snapshot state is input into a preset generative process reconstruction model to generate multiple candidate process compensation trajectories; Multiple candidate process compensation trajectories are input into the digital twin for multiphysics simulation and deduction to determine the corresponding quality prediction value, energy consumption prediction value and process fluctuation index, and then determine the final compensation trajectory. The final compensation trajectory is smoothly connected with the remaining process trajectory of the current processing stage to generate a complete sequence of control instructions for the remaining stage.

[0041] In practical implementation, firstly, the current physical space process execution state can be frozen, and a snapshot state of the current digital twin can be recorded. That is, the freeze operation includes two aspects: pausing further adjustments to the physical space process parameters to maintain the current output state of the actuators, avoiding control command conflicts and jumps during reprogramming; pausing the autonomous evolution calculation of the digital twin, using the current state as the baseline for reprogramming. The snapshot record covers complete information about the current processing state: the current processing stage identifier, remaining processing time, and the multi-physics distribution state of the materials, including: three-dimensional temperature field distribution data, moisture content field distribution data, and material three-dimensional temperature field distribution data. The system first provides point cloud morphology data and equipment operating status parameters. Secondly, the snapshot state can be input into a preset generative process reconstruction model to generate multiple candidate process compensation trajectories. This model employs a conditional variational autoencoder architecture, using a historical successful fabrication trajectory database as training samples to generate multiple candidate process compensation trajectories under the current state conditions. Then, these candidate process compensation trajectories can be input into a digital twin for multiphysics simulation to determine the corresponding quality prediction values, energy consumption prediction values, and process volatility indicators. The trajectories are input into the digital twin for multiphysics simulation to determine the corresponding predicted quality values, predicted energy consumption values, and process volatility indicators, thereby determining the final compensation trajectory. For example, the predicted quality indicators include the final predicted values ​​for effective ingredient content, moisture content, and appearance color; the predicted energy consumption value is the total predicted energy consumption value for the remaining stages; and the process volatility indicator is a weighted score of the amplitude and frequency of control command fluctuations, reflecting the stability of the process. A Pareto-optimal multi-objective optimization algorithm is used to maximize the predicted quality value, minimize the predicted energy consumption value, and optimize the process volatility indicator, performing a non-dominated ranking of each candidate trajectory. The non-dominated solution set on the Pareto front is selected, and the candidate trajectories in the non-dominated solution set are comprehensively scored using the analytic hierarchy process (AHP). The trajectory with the highest comprehensive score is selected as the final compensation trajectory. Finally, the final compensation trajectory can be smoothly connected with the remaining process trajectory of the current processing stage to generate a complete control command sequence for the remaining stage. That is, the final compensation trajectory can be smoothly connected with the executed process trajectory. The control command jump at the connection point is eliminated by cubic spline interpolation, generating a smooth control command sequence for the remaining stage. This command sequence is sent to the actuator through the distributed control system to achieve precise control of the remaining processing process.

[0042] like Figure 3The diagram shows a system architecture for monitoring and controlling processing parameters of traditional Chinese medicine decoction pieces. In the physical space, the processing equipment is equipped with a multimodal sensing network to collect multi-dimensional process parameter streams such as temperature field, moisture content field, material morphology, and acoustic characteristics in real time. After preprocessing by the data acquisition unit, the data is uploaded to the information space. In the information space, a high-performance edge computing server carries a digital twin construction module, which receives the process parameter stream and drives the digital twin to evolve synchronously with the physical space in real time through a virtual-real synchronization mapping module. An anomaly early warning module identifies and predicts deviations in the evolution state of the digital twin. When the predicted deviation exceeds the limit, a dynamic reprogramming module is triggered to generate the optimal compensation trajectory. Finally, the control command is sent to the execution mechanism in the physical space through a composite control execution module, forming a complete closed-loop intelligent control system.

[0043] Therefore, this application firstly achieves digital characterization of the multi-field coupling evolution law of materials during the processing process by constructing a digital twin of the target Chinese herbal medicine processing process. This digital twin can accurately reproduce the dynamic evolution of the physical processing process in the information space, laying a digital foundation for the intelligent and precise control of Chinese herbal medicine processing. Secondly, by constructing a multimodal sensing network and establishing a synchronous mapping mechanism between process parameter flow and the digital twin, the perception and virtual-real synchronous evolution of the Chinese herbal medicine processing process are realized. This enables the real-time and accurate mapping of multi-source heterogeneous data collected in the physical space to the digital twin, driving the virtual image and physical entity to maintain high-fidelity synchronous evolution, improving the reliability, safety, and response speed of process control, and providing a foundation for the intelligent and precise control of Chinese herbal medicine processing. The intelligent closed-loop control provides data support and model foundation. Then, through anomaly pattern recognition and deviation feature prediction based on the real-time evolution state of the digital twin, a spatiotemporal feature extraction model integrating graph convolutional networks and long short-term memory networks is constructed. This model can automatically extract deep spatiotemporal coupling features from the multi-physics field evolution state sequence output by the digital twin, accurately trace the spatial location and parameter type of anomalies, and provide a decision basis for subsequent dynamic reprogramming of the process, thereby improving the quality assurance capability of traditional Chinese medicine decoction piece processing. Finally, by constructing a multi-level tolerance threshold and dynamic reprogramming mechanism, differentiated and precise intelligent intervention on deviations in the processing technology of traditional Chinese medicine decoction pieces is realized, improving the adaptive control level, quality assurance capability, and process stability of traditional Chinese medicine decoction piece processing.

[0044] In summary, the technical solution adopted in this application can realize the synchronous evolution of the processing process in both virtual and real worlds by constructing a digital twin, thereby transforming the processing of Chinese herbal medicine slices into intelligent control and improving the process adaptability of processing quality.

[0045] Example 2: This application provides a monitoring and control system for processing parameters of traditional Chinese medicine decoction pieces, referring to... Figure 4As shown in the figure, this is a modular structure diagram of a monitoring and control system for processing parameters of traditional Chinese medicine decoction pieces according to this embodiment of the present application. The parameter monitoring and control system includes: The twin construction module 100 is used to construct a digital twin of the processing process of the target Chinese herbal medicine pieces; The synchronous evolution module 200 is used to collect the process parameter flow of the physical space in real time through a multimodal sensing network, and synchronously map the process parameter flow to the digital twin, driving the digital twin to evolve synchronously with the physical space processing process in real time. The cumulative deviation module 300 is used to identify abnormal patterns in the process parameter flow based on the real-time evolution state of the digital twin, and predict the deviation characteristics of the final quality of the processing under the abnormal patterns. The dynamic programming module 400 is used to trigger dynamic reprogramming of the processing technology when the deviation feature exceeds a preset multi-level tolerance threshold.

[0046] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0047] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compactdisc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0048] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

Claims

1. A method for monitoring and controlling processing parameters of traditional Chinese medicine decoction pieces, characterized in that, The parameter monitoring and control method includes the following steps: Constructing a digital twin of the processing process of target Chinese herbal medicine pieces; The process parameter flow in the physical space is collected in real time through a multimodal sensing network, and the process parameter flow is synchronously mapped to the digital twin, driving the digital twin to evolve synchronously with the physical space processing process in real time. Based on the real-time evolution state of the digital twin, abnormal patterns in the process parameter flow are identified, and deviation characteristics of the final quality of the processing under the abnormal patterns are predicted. When the deviation feature exceeds the preset multi-level tolerance threshold, dynamic reprogramming of the processing technology is triggered.

2. The method for monitoring and controlling processing parameters of traditional Chinese medicine decoction pieces as described in claim 1, characterized in that, The digital twin includes a physical processing equipment entity in physical space and a virtual processing image in information space, wherein the virtual processing image integrates a multi-physics coupling model.

3. The method for monitoring and controlling processing parameters of traditional Chinese medicine decoction pieces as described in claim 1, characterized in that, The multimodal sensing network includes: A fiber optic temperature sensor array for collecting three-dimensional temperature field data is deployed on the cylinder wall, stirring blades and inside the material of the processing equipment. Terahertz spectroscopy detection device for penetrating and collecting data on the internal moisture distribution and composition changes of materials; A high-speed binocular vision system for acquiring three-dimensional motion trajectory and morphological evolution data of materials; An acoustic emission sensor array is deployed on the cylinder wall of the processing equipment to collect acoustic signature data generated by the collision between materials and the equipment wall during the processing.

4. The method for monitoring and controlling processing parameters of traditional Chinese medicine decoction pieces as described in claim 1, characterized in that, The process parameter stream is synchronously mapped to the digital twin, driving the digital twin to evolve in real time synchronously with the physical space preparation process. This specifically includes: The process parameter stream is aligned with the time sequence and spatial coordinates to construct spatiotemporally synchronized data frames; Each data frame is input as a boundary condition into the multiphysics coupling model of the digital twin, driving the multiphysics coupling model to perform real-time iterative calculations; The calculation results of the multiphysics coupling model are compared with the residuals of the real-time acquired process parameter stream, and the internal parameters of the multiphysics coupling model are adjusted.

5. The method for monitoring and controlling processing parameters of traditional Chinese medicine decoction pieces as described in claim 1, characterized in that, Based on the real-time evolution state of the digital twin, identifying abnormal patterns in the process parameter flow and predicting the deviation characteristics of the final processing quality under the abnormal patterns specifically includes: Obtain the multiphysics evolution state sequence of the digital twin within a continuous time window; The multiphysics field evolution state sequence is input into a spatiotemporal feature extraction network based on a graph convolutional network and a long short-term memory network to extract the spatiotemporal evolution feature vector of the processing process. The spatiotemporal evolution feature vector is input into a single classifier based on support vector data description to identify abnormal patterns; When an abnormal pattern is identified, a reverse attribution analysis is performed on the spatiotemporal feature extraction network to determine the contribution of each spatial node in each physical field to the abnormal pattern and locate the root cause of the anomaly. The spatiotemporal evolution feature vector and root cause results are input into a predictor based on a temporal convolutional network to predict the deviation characteristics of the current abnormal mode from the quality index of the final processing time within the remaining processing time.

6. The method for monitoring and controlling processing parameters of traditional Chinese medicine decoction pieces as described in claim 1, characterized in that, The deviation feature is a vector representing the degree of deviation of each quality indicator from the expected value of the normal process under the current abnormal state.

7. The method for monitoring and controlling processing parameters of traditional Chinese medicine decoction pieces as described in claim 1, characterized in that, When the deviation characteristic exceeds a preset multi-level tolerance threshold, the dynamic reprogramming of the processing technique is triggered, specifically including: A multi-level tolerance threshold is preset, wherein the multi-level tolerance threshold includes a first tolerance threshold and a second tolerance threshold, and the first tolerance threshold is less than the second tolerance threshold; When the deviation feature is less than the first tolerance threshold, the processing procedure is determined to be under normal operating conditions. When the deviation feature is greater than the first tolerance threshold and less than the second tolerance threshold, a local parameter correction mode is triggered; When the deviation feature is greater than the second tolerance threshold, the global process reprogramming mode is triggered.

8. The method for monitoring and controlling processing parameters of traditional Chinese medicine decoction pieces as described in claim 7, characterized in that, The global process reprogramming mode specifically includes: Freeze the current physical space process execution state and record the current snapshot state of the digital twin; The snapshot state is input into a preset generative process reconstruction model to generate multiple candidate process compensation trajectories; Multiple candidate process compensation trajectories are input into the digital twin for multiphysics simulation and deduction to determine the corresponding quality prediction value, energy consumption prediction value and process fluctuation index, and then determine the final compensation trajectory. The final compensation trajectory is smoothly connected with the remaining process trajectory of the current processing stage to generate a complete sequence of control instructions for the remaining stage.

9. The method for monitoring and controlling processing parameters of traditional Chinese medicine decoction pieces as described in claim 1, characterized in that, The final compensation trajectory refers to the time sequence of process parameters for the remaining stages that, under the current processing conditions, can achieve the optimal quality index at the processing endpoint, the lowest energy consumption, and the smallest process fluctuations.

10. A monitoring and control system for processing parameters of traditional Chinese medicine decoction pieces, used to execute the monitoring and control method for processing parameters of traditional Chinese medicine decoction pieces as described in any one of claims 1 to 9, characterized in that, The parameter monitoring and control system includes: The twin building block is used to construct a digital twin of the processing process of the target Chinese herbal medicine pieces. The synchronous evolution module is used to collect the process parameter flow of the physical space in real time through a multimodal sensing network, and synchronously map the process parameter flow to the digital twin, driving the digital twin to evolve synchronously with the physical space processing process in real time. The cumulative deviation module is used to identify abnormal patterns in the process parameter flow based on the real-time evolution state of the digital twin, and predict the deviation characteristics of the final quality of the processing under the abnormal patterns. The dynamic programming module is used to trigger dynamic reprogramming of the processing technology when the deviation feature exceeds a preset multi-level tolerance threshold.

Images