Hot compress patch process parameter dynamic optimization control method and system

Abstract

The invention provides a hot compress patch process parameter dynamic optimization control method and system, and relates to the field of industrial control and digital processing, and the method comprises the steps: collecting a point cloud data set and an infrared thermogram time sequence, and constructing a thermodynamic state data set; generating a thermodynamic state sequence through a joint transformation function, extracting high-dimensional data features by using a physical prior processing unit containing a static space bias matrix, and outputting a comprehensive performance score through a linear data decoding function and a performance score calculation model; when the score is lower than a threshold value, fusing the dynamic thermal performance characteristics and the initial static characteristics on a process diagram data structure, and outputting fault probability distribution through a multi-round state iterative calculation method; and fault nodes are identified based on fault probability distribution, an early warning instruction is generated in a rule base through matching, a hybrid optimization objective function is constructed in combination with prediction and target thermal performance feature vectors, and an optimal control instruction is output. According to the method, the thermal performance of the product is evaluated by fusing multi-modal perception, root cause tracing is supported, and dynamic optimization control of process parameters is realized.

Description

Technical Field

[0001] This invention relates to the fields of industrial control and digital processing, and in particular to a method and system for dynamic optimization control of process parameters of heat therapy patches. Background Technology

[0002] For precision thermal control products such as heat therapy patches, accurate quality control and dynamic process optimization during production have become key technologies to ensure the consistency of their core performance and safety in use. How to scientifically optimize the uniformity and stability of product heating in complex and variable production environments, and avoid product failures caused by defects such as localized overheating or insufficient heating, has become an important issue that urgently needs to be addressed in the intelligent manufacturing and quality control upgrade of this field.

[0003] Chinese patent application CN120370717A discloses a high-precision adaptive thermal compensation system and method for a pick-and-place machine. The method includes: real-time acquisition of temperature and deformation data of multiple components of the placement head through a temperature-deformation sensing unit; reconstruction of a three-dimensional temperature field based on a finite element algorithm by a thermo-mechanical coupling analysis unit, dynamic calculation of thermal expansion distribution to solve the problem of coarse model in traditional single-point temperature measurement; prediction of thermal drift within the next 5ms by an online parameter identification and long short-term memory network model by a dynamic compensation control unit, adaptive adjustment of compensation strategy based on motion conditions; and decomposition of the compensation amount into displacement and torsional correction commands by a closed-loop execution unit to achieve precise cancellation of thermal deformation and construct a closed-loop thermal compensation system for the entire process.

[0004] However, current technology still faces many challenges. In the large-scale continuous production of heat therapy patches, when key process parameters experience slight drifts due to equipment aging or fluctuations in operating conditions, traditional quality control methods rely on single-point temperature measurement and offline sampling. Limited by low spatial coverage and temporal discontinuity, these methods struggle to timely and comprehensively monitor the early deterioration of the product's surface temperature field uniformity caused by this drift. If this early deterioration is not detected in time, process parameter deviations will accumulate and amplify, potentially leading to a large number of critically defective products or even the scrapping of entire batches, resulting in material waste and economic losses. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for dynamic optimization control of heat therapy patch process parameters, the specific technical solution of which is as follows:

[0006] The method for dynamic optimization control of heat therapy patch process parameters includes:

[0007] A point cloud dataset and an infrared thermal image time series are collected. Based on the spatial transformation relationship between sensors, the infrared thermal image time series is mapped onto a three-dimensional mesh model constructed from the point cloud dataset to construct a thermodynamic state dataset.

[0008] Based on the thermodynamic state dataset, a joint transformation function is used to generate a thermodynamic state sequence. A physical prior processing unit with an introduced static spatial bias matrix is ​​used to perform matrix calculations on the thermodynamic state sequence and output high-dimensional data features. The high-dimensional data features are mapped to the corresponding performance indicators through a parallel linear data decoding function. The performance indicators are input into the performance scoring calculation model and output a comprehensive performance score.

[0009] Based on the comprehensive performance score, it is determined whether the quality is below a preset quality threshold to activate the root cause diagnosis process. If the root cause diagnosis process is activated, the dynamic thermal performance feature vector is fused with the initial static feature vector of each node in the pre-constructed production process diagram data structure to generate node input features. The node input features are iteratively updated using a multi-round state iteration calculation method to output the failure probability distribution of all nodes.

[0010] Based on the fault probability distribution, the fault node is identified by the maximum a posteriori probability decision method, and data is matched in the preset diagnosis-disposal rule base to generate production early warning instructions. By comprehensively utilizing the fault probability distribution, the predicted thermal performance feature vector generated based on online sensor data, and the preset target thermal performance feature vector, a feedforward-feedback coupled hybrid optimization objective function is constructed and solved to output the optimal control instruction.

[0011] Furthermore, the method for constructing the thermodynamic state dataset includes:

[0012] The initial three-dimensional geometric shape of the heat patch before activation is collected, and a point cloud dataset representing the surface geometric features of the heat patch is generated.

[0013] The time-series infrared radiation intensity signal of the heat therapy patch is periodically collected during its full thermal life cycle, and the infrared radiation intensity signal is mapped in real time in combination with a calibration function to generate an infrared thermogram time series characterizing the surface temperature field of the heat therapy patch.

[0014] A three-dimensional mesh model is constructed by reconstructing a Poisson surface based on a point cloud dataset. The spatial transformation relationship between heterogeneous coordinate systems is solved by jointly calibrating the sensors. Based on the spatial transformation relationship and the thermal imager intrinsic parameter matrix, the infrared thermal image time series is mapped to the three-dimensional mesh model to construct a thermodynamic state dataset that integrates dynamic thermodynamic behavior.

[0015] Furthermore, the output step of the comprehensive performance score includes:

[0016] The static spatial position and dynamic instantaneous temperature of each three-dimensional grid vertex in the thermodynamic state dataset are concatenated into a state vector. The state vector is then subjected to a nonlinear data transformation using a joint transformation function. The transformed state vectors are arranged in chronological order to form a thermodynamic state sequence.

[0017] The static spatial bias matrix is ​​calculated based on the geometric features between the vertices of each three-dimensional mesh in the three-dimensional mesh model. The static spatial bias matrix is ​​used as an additive bias term to perform matrix calculation on the thermodynamic state sequence, thereby realizing dynamic thermodynamic dependency modeling guided by physical priors and outputting high-dimensional data features.

[0018] Global average pooling is performed on high-dimensional data features along the spatiotemporal dimension to aggregate them into a context vector. The context vector is then input into a set of parallel preset linear data decoding functions to map to the corresponding performance indicators. Multiple performance indicators are concatenated along the feature dimension to construct a dynamic thermal performance feature vector.

[0019] The dynamic thermal performance feature vector is input into a performance scoring calculation model trained under supervision based on a training dataset labeled with authoritative quality scores, and the model outputs a comprehensive performance score.

[0020] Furthermore, the static spatial bias matrix is ​​calculated by a learnable spatial bias encoding function, which uses the geometric features of the Euclidean distance between the vertices of the 3D mesh and the height difference along the Z-axis as input features to generate the corresponding bias matrix elements.

[0021] Furthermore, the performance indicators include a thermal dispersion gradient vector, an entropy increase uniformity index, and an abnormal hotspot spatiotemporal trajectory matrix. Specifically, the thermal dispersion gradient vector quantifies the macroscopic gradient distribution of the temperature field on the surface of the heat patch in space and its evolution trend over time; the entropy increase uniformity index is a scalar quantity used to globally quantify the uniformity of the heating process; and the abnormal hotspot spatiotemporal trajectory matrix is ​​used to parametrically describe the motion trajectory and temperature changes of key hotspots in two-dimensional space.

[0022] Furthermore, the output step of the fault probability distribution includes:

[0023] Using a structured data modeling method, the key factors affecting the thermal performance of heat therapy patches in the manufacturing process are abstracted into entity nodes, and the dependencies between these entity nodes are defined as directed edges to construct a production process graph data structure.

[0024] The dynamic thermal performance feature vector and the initial static feature vector of each node in the production process diagram data structure are concatenated to generate node input features. The multi-round state iteration calculation method is used to iteratively update the production process diagram data structure to obtain the final feature state vector of each node. The final feature state vector of each node is input into the fault probability function to construct the fault probability distribution of all nodes.

[0025] Furthermore, the entity nodes are divided into material attribute nodes, process parameter nodes, equipment status nodes, and intermediate quality nodes; wherein, the material attribute nodes are used to characterize the key physicochemical properties of the input materials, the process parameter nodes are used to characterize the operating parameters set and controlled in the production process, the equipment status nodes are used to characterize the working status parameters of the production equipment during operation, and the intermediate quality nodes are used to characterize the intermediate product attributes that affect subsequent processes.

[0026] Furthermore, the directed edges are divided into causal relationship edges, temporal relationship edges, and subordinate relationship edges; wherein, the causal relationship edge indicates that a change in the state of one node will cause a change in the state of another node, the temporal relationship edge indicates the sequence of the production process, and the subordinate relationship edge indicates that the process parameter node is an inherent attribute of the material attribute node or the equipment state node.

[0027] Furthermore, the generation of the production early warning instruction includes: identifying fault nodes based on the fault probability distribution using the maximum a posteriori probability decision method, using the fault nodes as query indexes to perform data matching in a preset diagnosis-disposal rule base, generating a production early warning instruction, and pushing the production early warning instruction to the monitoring terminal to trigger the corresponding disposal process.

[0028] Furthermore, the output step of the optimal control command includes:

[0029] Real-time acquisition of online sensor data, input of the online sensor data into a pre-trained performance predictor model, and generation of predicted thermal performance feature vectors;

[0030] Based on the predicted thermal performance feature vector, the fault probability distribution, and the preset target thermal performance feature vector, a hybrid optimization objective function with feedforward-feedback coupling is constructed and solved to determine the optimal control command.

[0031] Furthermore, the performance predictor model is preferably a multilayer perceptron (MLP) or a simplified graph neural network, which learns a complex nonlinear mapping relationship between online sensor data that is easy to collect in real time and the final product performance characteristics that are difficult to measure directly in real time by training on a large amount of historical data.

[0032] Furthermore, the hybrid optimization objective function is coupled with a performance error term and a diagnostic regularization term; wherein, the performance error term, based on a pre-trained forward process model, performs forward inference on the process parameters after applying the parameter adjustment vector with the assumptions, predicts the predicted thermal performance feature vector of the future product, and minimizes the parameter adjustment vector by calculating the square of the Euclidean distance between the predicted thermal performance feature vector and the target thermal performance feature vector; the diagnostic regularization term uses the failure probability distribution as an adaptive penalty weight to constrain the parameter adjustment vector.

[0033] The dynamic optimization control system for heat therapy patch process parameters includes a multimodal information acquisition module, a data scoring module, a root cause reasoning module, and a control decision module.

[0034] The multimodal information acquisition module is used to acquire point cloud datasets and infrared thermal image time series. Based on the spatial transformation relationship between sensors, the infrared thermal image time series is mapped onto a three-dimensional mesh model constructed from the point cloud dataset to construct a thermodynamic state dataset.

[0035] The data scoring module: Based on the thermodynamic state dataset, a joint transformation function is used to generate a thermodynamic state sequence. Using a physical prior processing unit that introduces a static spatial bias matrix, matrix calculations are performed on the thermodynamic state sequence to output high-dimensional data features. The high-dimensional data features are mapped to corresponding performance indicators through a parallel linear data decoding function. The performance indicators are input into the performance scoring calculation model to output a comprehensive performance score.

[0036] The root cause reasoning module determines whether the comprehensive performance score is below a preset quality threshold to activate the root cause diagnosis process. If the root cause diagnosis process is activated, the dynamic thermal performance feature vector is fused with the initial static feature vector of each node in the pre-constructed production process diagram data structure to generate node input features. The node input features are iteratively updated using a multi-round state iteration calculation method to output the failure probability distribution of all nodes.

[0037] The control decision module identifies fault nodes based on the fault probability distribution using the maximum a posteriori probability decision method, performs data matching in a preset diagnosis-disposal rule base, generates production early warning instructions, and comprehensively utilizes the fault probability distribution, the predicted thermal performance feature vector generated based on online sensor data, and the preset target thermal performance feature vector to construct and solve a feedforward-feedback coupled hybrid optimization objective function, and outputs the optimal control instructions.

[0038] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0039] This invention integrates three-dimensional point cloud data representing static geometric topology and infrared thermogram time series representing dynamic thermodynamic behavior as two independent and complementary information sources to construct a unified thermodynamic state dataset. This avoids the problems of inaccurate assessment of heating uniformity and inability to trace the root cause of defects caused by traditional detection methods ignoring the physical influence of the micro-morphology of the product surface on heat transfer.

[0040] This invention uses a spatial bias matrix representing static geometric topology as a physical prior and inputs it into the dynamic correlation calculation of the physical prior processing unit to perform matrix calculations on the thermodynamic state sequence, outputting high-dimensional data features. This allows the learning and representation of heat conduction laws that are more consistent with real physical processes, thus improving the problem that traditional black-box deep learning models do not learn complex spatiotemporal evolution laws sufficiently due to the lack of physical constraints.

[0041] This invention combines a graph data structure containing causal relationships in the production process with a multi-round state iteration calculation method. It iteratively updates the feature vectors representing product performance anomalies on the topology of the graph data structure for attribution reasoning. This improves the problem of poor interpretability and susceptibility to spurious associations in traditional fault diagnosis based on statistical correlation, which is caused by a lack of structured understanding of the overall causal chain of the production process.

[0042] This invention constructs a hybrid optimization objective function that couples feedforward performance prediction and feedback fault diagnosis. When calculating the adjustment amount of process parameters, it can both pre-compensate for future performance deviations caused by process disturbances and constrain the adjustment behavior to the most likely root cause of fault diagnosed by historical data. This improves the problem of lag in response of traditional single feedback control or lack of robustness of single feedforward control, and achieves a balance between speed and accuracy in control. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a flowchart illustrating the principle of the dynamic optimization control method for the process parameters of the hot compress patch of the present invention.

[0045] Figure 2 This is a functional block diagram of the dynamic optimization control system for the process parameters of the heat therapy patch of the present invention. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Example 1:

[0048] Please see Figure 1 As shown, this embodiment provides a method for dynamic optimization control of heat therapy patch process parameters, including:

[0049] S1000, collecting point cloud datasets and infrared thermal image time series Based on the spatial transformation relationship between sensors, the infrared thermal image time series is... Mapped to the point cloud dataset Constructed 3D mesh model Above, construct a thermodynamic state dataset. .

[0050] Specifically, this step aims to systematically capture and deeply integrate the static three-dimensional geometric features and the full-cycle dynamic surface temperature field evolution behavior of the physical entity of the heat patch through the collaborative application of multimodal sensing technologies, thereby constructing a thermodynamic state dataset. .

[0051] Further, step S1000 includes:

[0052] Step S1100: Collect the initial three-dimensional geometric shape of the heat patch before activation, and generate a point cloud dataset characterizing the geometric features of the heat patch surface. .

[0053] Specifically, this step aims to obtain the macroscopic three-dimensional geometric morphology of the heat patch in its static state before activation, i.e., a point cloud dataset. This morphological data serves as a unified spatial benchmark for subsequent multimodal data fusion. It also contains information on surface micro-undulations caused by key process parameters such as differences in the density of the internal heating material and the compaction process. This information has a priori value for analyzing heating uniformity and process stability.

[0054] In the specific implementation process, a non-contact, high-precision structured light 3D scanner is used to perform a full-field 3D scan of the upper surface of the heat therapy patch to obtain the original point cloud data of the heat therapy patch. During the scanning process, a pre-coded grating pattern is projected onto the surface of the heat patch sample, and the image sensor captures the coded pattern that is deformed due to the undulations of the heat patch sample surface. Then, based on the principle of triangulation, the captured pattern is analyzed to obtain the set of three-dimensional coordinate points of the heat patch covering the entire field of view.

[0055] To improve the signal-to-noise ratio and reduce the computational load of subsequent processing, the original point cloud data was processed. Perform comprehensive filtering preprocessing, which involves constructing comprehensive filtering operators. This involves transforming raw point cloud data containing noise and redundant information. Mapped to an optimized point cloud dataset .

[0056] Among them, the comprehensive filtering operator It is a collection of one or more point cloud preprocessing algorithms, such as the Statistical Outlier Removal (SOR) algorithm for removing isolated outliers caused by environmental noise or reflection, and the Voxel Grid Downsampling (VGD) algorithm for reducing data density and achieving data homogenization while maintaining key geometric contours. A point cloud dataset is a digital object used to reconstruct a high-fidelity 3D surface model in a computer as a set of discrete points. Representing point cloud datasets The i-th three-dimensional spatial coordinate vector contains all the spatial information of a specific sampling point on the surface of the heat patch sample, that is, the location of the sampling point in the three-dimensional Cartesian coordinate system. Three coordinate components; Represents a three-dimensional spatial coordinate vector The Cartesian coordinate components of are high-precision floating-point numbers, where ... represents the projection values ​​of the sampling point onto the X, Y, and Z axes of the measuring device coordinate system, respectively. These three values ​​together determine the precise location of the sampling point in three-dimensional space; i represents the point cloud dataset. The index is used to uniquely identify and traverse the point cloud dataset. All three-dimensional spatial coordinate vectors Its value ranges from 1 to N, which are consecutive positive integers; N represents the point cloud dataset. Three-dimensional spatial coordinate vector The total number of point clouds.

[0057] Step S1200: Periodically collect time-series infrared radiation intensity signals throughout the entire thermal lifecycle of the heat therapy patch, and combine them with a calibration function to map the infrared radiation intensity signals in real time, generating an infrared thermogram time series characterizing the surface temperature field of the heat therapy patch. .

[0058] Specifically, this step aims to capture the dynamic thermodynamic behavior of the heat therapy patch throughout its entire lifespan in real time using a non-contact, full-field imaging method, and obtain the spatiotemporal evolution data of its surface temperature field, i.e., infrared thermogram time series. The infrared thermal image time series It serves not only as a basis for subsequent objective quantitative evaluation of heating performance and intelligent diagnosis of thermal anomaly patterns, but also as the fundamental input data for establishing a dynamic correlation between process parameters and heating performance.

[0059] In the specific implementation process, the activated heat therapy patch is placed in a test chamber where environmental parameters can be precisely controlled to simulate the real-world usage scenario of the heat therapy patch. For example, when simulating the human body surface environment, the ambient temperature is kept constant at a certain level. The relative humidity is maintained at A high-precision infrared thermal imager calibrated with a standard blackbody radiation source was used, with a preset time sampling frequency. and sampling time interval Continuous and uninterrupted infrared thermal imaging is performed on the surface of the heat-treated patch sample until its average surface temperature falls below a preset effective working threshold. Recording ends.

[0060] At each preset sampling time t, the raw data collected by the detector array of the infrared thermal imager is the unprocessed infrared radiation intensity signal. To obtain this original infrared radiation intensity signal Converted into pixel temperature values ​​with specific and clear physical meaning A calibration function is required. Perform a mapping transformation. The calibration function... It is a multivariable physical model based on the physical laws of thermal radiation, comprehensively incorporating the core factors affecting thermal radiation measurement. The specific process formula is as follows:

[0061] ;

[0062] in, The pixel temperature value is a precise temperature scalar obtained after radiometric calibration and environmental correction. Its physical dimension is Celsius or Kelvin, and it is used to characterize the surface temperature of the heat patch. The instantaneous thermodynamic state at position and time t; Representing the discrete sampling time, it is a timestamp or index used to identify a specific data point in a time series; The original infrared radiation intensity signal represents the raw physical quantity collected at time t by the sensing unit located in the m-th row and n-th column of the detector array of the infrared thermal imager. This represents the surface emissivity of the heat-treated patch sample, with a value ranging from 0 to 1. It is used to correct for deviations in infrared radiation intensity caused by differences in the material's own properties. This represents the ambient temperature compensation parameter, which is a precise temperature value used to subtract interference from stray environmental radiation on the measurement results. This represents the atmospheric path compensation parameter, used to correct the energy attenuation of infrared signals during propagation caused by atmospheric absorption and scattering.

[0063] At sampling time t, for the entire detector array Each sensing unit performs the above temperature calculation process in parallel and outputs the calculated pixel temperature values. Constructed as a two-dimensional temperature field matrix This refers to the single-frame heatmap matrix. This matrix is ​​a discretized spatial representation of the instantaneous thermodynamic state of the heat patch surface at sampling time t. Here, m and n represent the values ​​in the single-frame heatmap matrix, respectively. The row and column indices, with values ​​ranging from 1 to H and 1 to W respectively, together form a coordinate pair. Used in In a two-dimensional array, W and H uniquely identify the position of a pixel; W and H respectively represent the position of a single frame in the heatmap matrix. The number of pixels in the horizontal and vertical directions together define the spatial sampling density and data dimension of the temperature field in a single frame.

[0064] Over time, at fixed sampling intervals The above single-frame heatmap matrix is ​​continuously generated and recorded. This constitutes a complete time series recording the thermodynamic behavior of the heat therapy patch sample throughout its entire life cycle, namely, an infrared thermogram time series. The specific formula is as follows:

[0065] ;

[0066] in, Representing an infrared thermal image time series, it is a matrix of single-frame thermal images. The resulting ordered set provides a data foundation for subsequent dynamic performance evaluation and fault diagnosis; The sampling time interval is the time step between two consecutive heatmap matrices, and its reciprocal is the dynamic detection frequency. The end time of the test is the point at which the heating process of the heat patch stops and data sampling ceases.

[0067] Step S1300, based on point cloud dataset A 3D mesh model is constructed by reconstructing the Poisson surface. Furthermore, by jointly calibrating the sensors to solve the spatial transformation relationship between heterogeneous coordinate systems, and based on the spatial transformation relationship and the thermal imager's intrinsic parameter matrix, the infrared thermal image time series is generated. Mapped to the three-dimensional mesh model Construct a thermodynamic state dataset that integrates dynamic thermodynamic behavior .

[0068] Specifically, this step aims to address the inconsistencies in coordinate system, resolution, and sampling dimension between three-dimensional geometric data from heterogeneous sensors and two-dimensional time-series temperature data. A thermodynamic state dataset is generated by registering and mapping dynamic thermodynamic information onto a static three-dimensional geometric model. .

[0069] In practice, this step is implemented through the following three consecutive technical phases:

[0070] Phase 1: Constructing the digital geometric basis. This involves using the point cloud dataset output from step S1100. A standardized three-dimensional mesh model representing the macroscopic and microscopic geometry of the thermal patch was constructed using the Poisson Surface Reconstruction (PSR) algorithm. The three-dimensional mesh model It is the static spatial basis that carries subsequent temperature field data, and its specific expression formula is as follows:

[0071] ;

[0072] in, Representing a 3D mesh model The set of vertices constitutes the three-dimensional mesh model. An ordered list of all vertices, used for discretizing the 3D mesh model. The geometric surface; Representing a 3D mesh model A set of triangular facets used to describe a 3D mesh model. Surface continuity.

[0073] The second stage involves establishing the spatial mapping relationship for multimodal data. To ensure the accuracy of the fusion between temperature data and the geometric model, systematic errors introduced by differences in sensor layout must be eliminated. This is achieved by jointly calibrating the 3D scanner and the infrared thermal imager, solving for the coordinates from the infrared thermal imager. To the coordinate system of the 3D scanner The spatial transformation relationship is defined by a 4×4 homogeneous transformation matrix. This homogeneous transformation matrix is ​​uniquely described. The rotation matrix R and the translation vector t are coupled, and the specific formula is as follows:

[0074] ;

[0075] in, This represents a homogeneous transformation matrix, used to unify the coordinate space of data from different sensors. The coordinate system of the 3D scanner is a three-dimensional Cartesian coordinate system established based on the 3D scanner. The coordinate system of the infrared thermal imager is a three-dimensional Cartesian coordinate system with the optical center of the infrared thermal imager as the origin. This represents the rotation matrix, used to describe the coordinate system of the infrared thermal imager. Relative to the coordinate system of the 3D scanner The rotational posture; This represents the translation vector, used to describe the coordinate system of the infrared thermal imager. Origin relative to the coordinate system of the 3D scanner Spatial displacement of the origin.

[0076] The third stage: Projecting the dynamic temperature field and constructing a spatiotemporal dataset. This involves the infrared thermal image time series constructed in step S1200. As a dynamic data source, each frame contains a single-frame heatmap matrix. Projected onto the aforementioned constructed 3D mesh model Above. Specifically, for three-dimensional mesh models. Any three-dimensional mesh vertex on Using the known thermal imager intrinsic parameter matrix and the spatial transformation relationships already obtained Calculate vertices using perspective projection model Projected pixel coordinates on a two-dimensional heatmap The specific process formula is as follows:

[0077] ;

[0078] in, Representing a 3D mesh model vertex set The j-th vertex of the 3D mesh; These represent the j-th 3D mesh vertex. The coordinates of the three-dimensional mesh on the X, Y, and Z axes; j represents the vertex of the three-dimensional mesh. The index is used to traverse the vertex set. ; This represents the thermal imager intrinsic parameter matrix, the values ​​of which are obtained in advance through camera calibration and are used to project three-dimensional points in the camera coordinate system onto the two-dimensional image plane. This represents the scale factor, used to convert homogeneous coordinates to non-homogeneous coordinates. This represents the coordinates of the projected pixels, used to locate the temperature interpolation region.

[0079] Considering the projected pixel coordinates Due to the non-integer nature of the graph, a bilinear interpolation algorithm is used, based on the projected pixel coordinates. Calculate the coordinates of the projected pixel by taking the temperature values ​​and weights of its four neighboring pixels. The precise interpolated temperature is used as the vertex of the 3D mesh. The instantaneous temperature at time t By traversing all vertices of the 3D mesh And all time frames, generate a four-dimensional state-space representation that fully describes the four-dimensional thermodynamic behavior of the heat patch, i.e., a thermodynamic state dataset. The specific formula is as follows:

[0080] ;

[0081] in, Represents the vertices of a 3D mesh The instantaneous temperature at time t is a temperature scalar calculated through interpolation, used to characterize the vertices of the 3D mesh. The thermodynamic state at time t.

[0082] S2000, based on thermodynamic state dataset Joint transformation functions are used to generate thermodynamic state sequences. By introducing a static space bias matrix The physical prior processing unit processes the thermodynamic state sequence. Perform matrix calculations and output high-dimensional data features. The high-dimensional data features are decoded using a parallel linear data decoding function. Mapping these metrics to corresponding performance indicators, inputting these performance indicators into the performance scoring calculation model, and outputting a comprehensive performance score. .

[0083] Specifically, this step aims to utilize an end-to-end deep learning model to process the high-dimensional and complex thermodynamic state dataset generated in step S1300. Deep spatiotemporal feature extraction and information compression are performed to transform them into a low-dimensional structured feature vector that can comprehensively characterize the dynamic thermal performance of the heat therapy patch sample, namely, the dynamic thermal performance feature vector. Furthermore, the dynamic thermal performance feature vector is... Transformed into a comprehensive performance score that can be directly used for quality control and process optimization. This is a key process that connects intelligent analysis and intelligent control.

[0084] Further, step S2000 includes:

[0085] Step S2100, the thermodynamic state dataset is... Each 3D mesh vertex Static spatial location and dynamic instantaneous temperature are concatenated to form a state vector. A joint transformation function is used to perform a nonlinear data transformation on the state vector. The transformed state vectors are arranged in chronological order to form a thermodynamic state sequence. .

[0086] Specifically, this step aims to process the thermodynamic state dataset. This is transformed into a structured, sequential representation suitable for processing by temporal deep learning models, namely, a sequence of thermal states in a high-dimensional feature space. .

[0087] In the specific implementation process, for the thermodynamic state dataset Any three-dimensional mesh vertex at any time t Its complete instantaneous state consists of two parts: one is a static spatial position vector that does not change with time. Used to characterize the vertices of a three-dimensional mesh The two are: a fixed position in the three-dimensional mesh model and a dynamic instantaneous temperature that changes over time. , used to characterize the vertices of the three-dimensional mesh The thermodynamic state at time t.

[0088] To achieve a unified characteristic representation of the aforementioned heterogeneous information, this step employs a joint transformation function, which consists of a series of multi-level nonlinear data transformations, receiving data from the thermodynamic state dataset. Static spatial position vector and dynamic instantaneous temperature The concatenated state vectors are then used to learn the complex coupling relationship between spatial location and temperature state through multi-layer nonlinear data transformation. Finally, a high-dimensional and unified input token is output to characterize the comprehensive state features of the corresponding spatiotemporal point.

[0089] At any time t, all 3D mesh vertices The input tokens together constitute the token set of the entire field state at time t. Furthermore, the token sets at all times are arranged sequentially in chronological order to form a thermodynamic state sequence embedded in a high-dimensional feature space. The thermodynamic state sequence It not only fully preserves the spatiotemporal evolution of the surface temperature field of the heat patch, but also can be used as input for a temporal attention network for subsequent dynamic thermal performance feature extraction and process parameter optimization control.

[0090] Step S2200, based on the three-dimensional mesh model Calculate the static spatial bias matrix for geometric features between vertices of each 3D mesh. The static space bias matrix As an additive bias term, for the thermodynamic state sequence Perform matrix calculations to achieve dynamic thermodynamic dependency modeling guided by physical priors and output high-dimensional data features. .

[0091] Specifically, this step aims to propose an improved self-attention encoder layer (Physics-Informed Self-Attention, PISA) that incorporates prior physical knowledge to achieve high-precision modeling of the dynamic evolution of the temperature field on the surface of a heat-treated patch. This encoder layer serves as the core computational unit, and multiple layers can be stacked to construct a complete multi-layer temporal attention network. Its core lies in the deep coupling of two heterogeneous information elements: one is the topological information representing the static geometric structure, derived from the point cloud dataset output in step S1100. The three-dimensional mesh model constructed in step S1300 Secondly, there is time-series data characterizing dynamic thermodynamic behavior, namely the thermodynamic state sequence output in step S2100. By fusing cross-modal features, a dynamic system identification model that conforms to real physical processes is constructed, outputting high-dimensional data features. .

[0092] In its implementation, while the standard self-attention mechanism possesses strong long-distance dependency modeling capabilities in sequence data processing, its inherent permutation invariance prevents it from directly perceiving stable physical spatial topological relationships between input heat tokens. To overcome this deficiency, this step proposes a matrix computation method that incorporates prior physical information into the core computation layer. The implementation of this step includes the following three stages:

[0093] Phase 1: Quantization and encoding of static geometric associations. Based on the point cloud dataset output from step S1100. Constructed 3D mesh model A static spatial bias matrix is ​​pre-calculated. The static space bias matrix bias matrix elements Used to quantify the Three-dimensional mesh vertices and the Three-dimensional mesh vertices The static geometric correlation between them. This static geometric correlation can be encoded by a learnable spatial bias coding function. The calculated encoding function is... Based on the Euclidean distance between the vertices of the 3D mesh Z-axis height difference Using geometric features as input features, the corresponding bias matrix elements are generated. The specific process formula is as follows:

[0094] ;

[0095] in, Represents the static space bias matrix The bias matrix element in the equation is located in the static space bias matrix. No. line, number Scalar values ​​in the column; These represent the vertex indices of the 3D mesh vertices, used in the vertex set. Each vertex uniquely identifies a specific vertex and also corresponds to a static spatial bias matrix. Row indexes and column indexes; This represents a spatial bias encoding function used to convert geometric relationships into bias values; Represents the vertices of a 3D mesh and In three-dimensional space, the non-negative scalar of straight-line distance is called Euclidean distance. This represents the L2 norm operator, used to quantify the physical proximity of two 3D mesh vertices; Represents the vertices of a 3D mesh and The non-negative scalar value of the degree of separation in the vertical direction, namely the Z-axis height difference, is used to reflect the local unevenness or thickness difference on the surface of the heat patch.

[0096] Phase Two: Spatially Bias-Guided Dynamic Attention Calculation. This involves processing the thermodynamic state sequence. The mapping is performed using a linear transformation to the query matrix Q, key matrix K, and value matrix V. When calculating the attention weights, the pre-computed static spatial bias matrix is ​​used. As an additive bias, it is directly injected into the calculation of the dot product similarity between the query matrix Q and the key matrix K. This operation enables the evaluation of the dynamic thermodynamic correlation between any two 3D mesh vertices. In this process, the fixed physical spatial relationships between them are considered, thereby achieving dynamic thermodynamic dependency modeling guided by physical priors. Finally, the attention feature matrix modulated with spatial information is output. .

[0097] The third stage: Layer stacking and generation of the final hidden state. A single PISA encoder layer takes the input attention feature matrix... After processing through a feedforward neural network, residual connections, and layer normalization, the final output of the encoder layer is generated. All PISA encoder layers are stacked sequentially to construct a multi-layer temporal attention network, where the output of the previous layer serves as the input to the next. After the data stream has passed through all encoder layers, the high-dimensional feature representation of the output of the last encoder layer becomes the high-dimensional data feature of the multi-layer temporal attention network. .

[0098] Step S2300, high-dimensional data features Global average pooling is performed along the spatiotemporal dimension to aggregate the data into a context vector. This context vector is then input into a set of parallel, pre-defined linear data decoding functions, which map it to corresponding performance metrics. Multiple performance metrics are then concatenated along the feature dimension to construct a dynamic thermal performance feature vector. .

[0099] Specifically, this step aims to extract the high-dimensional abstract features, i.e., high-dimensional data features, from the output of the multi-layer temporal attention network in step S2200. This is decoded and mapped into a set of interpretable low-dimensional performance indicators with clear physical or statistical significance, namely, dynamic thermal performance feature vectors. .

[0100] In the specific implementation process, high-dimensional data features It is a high-dimensional tensor that contains a comprehensive and highly abstract representation of the thermal patch's evolution throughout its entire spatiotemporal process. To extract specific performance indicators applicable to engineering applications, it is necessary to analyze the features of this high-dimensional data. Information aggregation and decoding are performed. By applying Global Average Pooling (GAP) along the spatial and temporal dimensions, high-dimensional data features are... They are aggregated into a fixed-dimensional context vector.

[0101] Next, the context vector is fed in parallel into multiple independent and pre-defined linear data decoding functions. Each linear data decoding function is an independent linear transformation operation, and its internal parameters, including a weight matrix and a bias vector, are obtained through supervised learning on a large-scale labeled dataset, rather than being preset with fixed values. During training, these parameters are iteratively optimized using the backpropagation algorithm, so that any context vector is mapped to a corresponding specific performance metric. Since the physical meanings of each performance metric are different, each linear data decoding function has independent and specific weight and bias parameters.

[0102] The performance metrics include the thermal dispersion gradient vector, the entropy uniformity index, and the spatiotemporal trajectory matrix of abnormal hot spots. The thermal dispersion gradient vector quantifies the macroscopic gradient distribution of the temperature field on the surface of the heat patch in space and its evolution trend over time, i.e., the trend of heat diffusion. The entropy uniformity index is a scalar used to globally quantify the uniformity of the heating process, i.e., whether the overall heating is uniform. The spatiotemporal trajectory matrix of abnormal hot spots is used to parametrically describe the movement trajectory and temperature changes of key hot spots in two-dimensional space, thereby monitoring the occurrence of abnormal hot spots and their movement trajectories.

[0103] Finally, the performance metrics output by all linear data decoding functions are concatenated along the feature dimension to form a dynamic thermal performance feature vector. .

[0104] Step S2400, the dynamic thermal performance feature vector Input to dynamic thermal performance feature vector The performance scoring calculation model is trained under supervision using a training dataset labeled with authoritative quality scores, and outputs a comprehensive performance score. .

[0105] Specifically, this step aims to extract the dynamic thermal performance feature vector from step S2300. The input performance score calculation model is mapped to a single and standardized comprehensive performance score. This comprehensive performance score This enables a nonlinear mapping from a multidimensional feature space to a one-dimensional quality evaluation space, providing intuitive and quantifiable feedback evaluation indicators for automated quality grading, real-time process monitoring, and dynamic optimization of process parameters in heat therapy patches.

[0106] In the specific implementation process, this step adopts a pre-trained lightweight performance scoring calculation model to realize the calculation of high-dimensional dynamic thermal performance feature vectors. Process and output a comprehensive performance score. The performance scoring calculation model is based on a multilayer perceptron, and its internal parameters are predetermined through supervised learning training on a large number of labeled samples.

[0107] The training dataset consists of a series of feature vector-expert rating pairs Composition. Among them, It is the dynamic thermal performance feature vector of the kth training sample; It is an authoritative quality score for the k-th training sample, determined by domain experts or standardized experiments. During training, the performance scoring calculation model receives the dynamic thermal performance feature vectors of the training samples. And output the predicted score Quantitatively predict scores using a loss function With authoritative quality rating The error between them is used to iteratively optimize the internal parameters of the network using the backpropagation algorithm until a mapping relationship that can reproduce the expert evaluation criteria is learned, thus realizing a stable mapping relationship from dynamic thermal performance feature vector to performance score.

[0108] In practical applications, the dynamic thermal performance eigenvectors The input is fed into a pre-trained performance scoring model, and after one forward propagation calculation, a standardized comprehensive performance score is output. The overall performance score This is a normalized value between 0 and 1; a higher value indicates better overall heating performance of the heat therapy patch. This overall performance score can be used as a basis for quality grading, enabling automated graded management of products and rapid identification and data rejection of defective products.

[0109] S3000, based on comprehensive performance rating Determine if the quality is below the preset threshold. To activate the root cause diagnosis process, if the root cause diagnosis process is activated, then in the pre-built production process diagram data structure Above, the dynamic thermal performance feature vector With the aforementioned production process diagram data structure The initial static feature vectors of each node are fused to generate node input features. A multi-round state iteration calculation method is used to iteratively update these node input features, outputting the fault probability distribution of all nodes. .

[0110] Specifically, this step aims to define a condition-triggered intelligent diagnostic mechanism, which uses the comprehensive performance score output in step S2400 as the basis for the diagnostic process. Based on this, only when the aforementioned comprehensive performance score Below the preset quality threshold At this point, if the heat patch sample exhibits potential thermal performance defects, the intelligent diagnostic mechanism of this step is activated. This intelligent diagnostic mechanism is based on a multi-round state iteration calculation method, and uses the dynamic thermal performance feature vector output from step S2300... With pre-built production process diagram data structure Perform fusion inference and output the failure probability distribution pointing to specific process parameter deviations. This enables the attribution of fault probabilities for deviations in various process parameters that cause current thermal performance anomalies.

[0111] Further, step S3000 includes:

[0112] Step S3100: Using a structured data modeling method, key factors affecting the thermal performance of the heat-converting patch in the manufacturing process are abstracted into entity nodes, and the dependencies between these entity nodes are defined as directed edges, thus constructing a production process graph data structure. .

[0113] Specifically, this step aims to formalize and structure the unstructured prior process knowledge contained in production procedures, expert experience, and bills of materials through data extraction and structured modeling. The goal is to construct a production process diagram data structure that can be understood by machines and processed automatically. .

[0114] In practice, this step is a one-time offline construction process that uses structured data modeling to deconstruct and formalize the complete manufacturing process of the heat therapy patch. The core of this deconstruction and formalization process lies in the data structure of the production process diagram. The basic elements are defined as nodes and edges.

[0115] During the node definition phase, all key factors in the manufacturing process that may affect the thermal performance of the final heat-application patch product are abstracted into a production process diagram data structure. The entity nodes in the data. To ensure the data structure of the production process diagram... To ensure completeness, entity nodes are divided into four categories: material attribute nodes, process parameter nodes, equipment status nodes, and intermediate quality nodes.

[0116] The material attribute nodes are used to characterize the key physicochemical properties of the input materials, such as "iron powder mesh size", "activated carbon moisture content", and "binder viscosity". The process parameter nodes are used to characterize the settable and controllable operating parameters in the production process, such as "stirring time", "roller pressure", and "drying temperature curve". The equipment status nodes are used to characterize the working status parameters of the production equipment during operation, such as "mixer cylinder temperature" and "roller gap". The intermediate quality nodes are used to characterize intermediate product attributes that have an important impact on subsequent processes, such as "mixture uniformity" and "packet compaction density".

[0117] During the edge definition phase, the physical, chemical, logical, or temporal relationships between the aforementioned nodes are abstracted into a production process diagram data structure. Directed edges in the algorithm are used to characterize the dependencies between knowledge units. To accurately express causal and constraint relationships, each edge is assigned a different semantic type, including causal relationship edges, temporal relationship edges, and subordinate relationship edges.

[0118] Among them, the causal relationship edge indicates that a change in the state of one node will cause a change in the state of another node, for example, the influence of the "mixing time" process parameter node on the intermediate quality node of "mixed material uniformity"; the temporal relationship edge indicates the sequence of the production process, for example, the "mixing process" node precedes the "rolling process" node; the subordinate relationship edge indicates that the process parameter node is an inherent attribute of a certain material attribute node or equipment state node, for example, the "rolling pressure" process parameter node is subordinate to the "rolling equipment" equipment state node.

[0119] By systematically defining the nodes and edges mentioned above, a comprehensive and in-depth data structure for a production process diagram that reflects the actual manufacturing process is constructed. .

[0120] Step S3200: The dynamic thermal performance feature vector... and production process diagram data structure The initial static feature vectors of each node are concatenated to generate node input features. A multi-round state iteration calculation method is then used in the production process diagram data structure. The process involves iterative updates to obtain the final feature state vector of each node. This final feature state vector is then input into the fault probability function to construct the fault probability distribution for all nodes. .

[0121] Specifically, this step aims to build upon the production process diagram data structure constructed in step S3100. The dynamic thermal performance feature vector of the heat therapy patch was calculated using a multi-round state iteration method. The analysis identifies the process parameters most likely to cause abnormal thermal performance and outputs the corresponding failure probability distribution. This enables automated root cause diagnosis and quantitative assessment.

[0122] In the specific implementation process, this step includes the following stages:

[0123] Phase 1: Node feature initialization and diagnostic information input. This is for the production process diagram data structure. Each node in the algorithm initializes its initial static feature vector. This initial static feature vector is used to encode the prior process information of each node. For example, process parameter nodes can encode design nominal values ​​and tolerance ranges, material attribute nodes can encode key technical indicators, and equipment status nodes can encode operating parameter ranges, etc.

[0124] Subsequently, the dynamic thermal performance feature vector characterizing the abnormal thermal performance of the current heat therapy patch sample will be used. The initial symptom information used in this diagnosis is input into the multi-round state iteration calculation method. This is achieved by using the dynamic thermal performance feature vector... Data structure of production process diagram The initial static feature vectors of each node are concatenated to generate a node input feature vector that integrates static attributes and dynamic states, providing richer input data for the initial calculation of the multi-round state iteration calculation method.

[0125] Phase Two: Multi-round Iterative Information Propagation and Data Aggregation. This multi-round state iteration calculation method uses an iterative process consisting of L calculation loops within the production process diagram data structure. Information propagation and aggregation are performed to update the feature state vector of each node. In the (l+1)th iteration, the feature state vector of each node is calculated through an aggregation function. This function collects the feature state information of all its neighboring nodes in the lth iteration and combines it with its own feature state information in the lth iteration. Then, it is transformed by a learnable weight matrix and a nonlinear activation function to form the updated node feature state vector. , This represents the total number of iterations.

[0126] After L rounds of iterative calculations, the final feature state vector of each node not only contains its own initial static information, but also deeply integrates its contextual information in the entire process chain and the current product performance abnormality symptom information, forming a high-level feature representation containing rich diagnostic clues, namely the final feature state vector.

[0127] Phase Three: Fault Probability Decoding and Output. This involves structurally modifying the production process diagram data. The final feature state vectors of all nodes are input into the fault probability function to calculate the probability that each node is the root cause of the current fault. The probabilities of all nodes together constitute the final fault probability distribution. The failure probability distribution Quantifying and prioritizing potential process problems provides direct input to the upper-level dynamic optimization control system for decision-making. This enables the system to adjust the most suspicious process parameters based on the probability distribution of failures, thereby achieving precise control of the production process and real-time quality improvement.

[0128] S4000, based on fault probability distribution Fault nodes are identified using the maximum a posteriori probability decision method, and data is matched against a pre-defined diagnosis-handling rule base to generate real-time production early warning instructions. Comprehensively utilize the aforementioned fault probability distribution Based on online sensor data The generated predictive thermal performance feature vector and the preset target thermal performance feature vector Construct and solve the feedforward-feedback coupled hybrid optimization objective function, and output the optimal control command. .

[0129] Specifically, this step aims to construct a hybrid intelligent control strategy that couples feedforward and feedback, transforming the diagnostic and predictive information generated in previous steps into adaptive regulation of the production process, forming a complete technical closed loop. This strategy comprehensively utilizes three types of information: first, feedback diagnostic information, namely the fault probability distribution output in step S3200. Second, there is feedforward disturbance information, which is online sensor data collected in real time by sensors on the production line. Thirdly, the control target, namely the target thermal performance characteristic vector pre-defined by experts to characterize the ideal performance. Using the aforementioned multi-source information, a constrained optimization problem is solved, and the optimal control command is output to guide production adjustments. .

[0130] Further, step S4000 includes:

[0131] Step S4100, based on the fault probability distribution The maximum a posteriori probability decision method is used to identify faulty nodes. These faulty nodes are then used as query indexes to perform data matching in a preset diagnosis-treatment rule base, generating production early warning instructions. and the production early warning instruction The message is pushed to the monitoring terminal to trigger the corresponding handling process.

[0132] Specifically, this step aims to quantify the fault probability distribution output by the previous steps. This is transformed into human-machine readable early warning or intervention instructions with clear guiding significance, i.e., production early warning instructions. It constitutes the feedback control and decision support link in the control system, providing production managers or automated systems with immediate and actionable fault location and handling suggestions, enabling rapid response and timely adjustment to quality deviations.

[0133] In the specific implementation process, the fault probability distribution of the input is... The system analyzes the data and uses the Maximum A posteriori Probability (MAP) decision-making method to find and determine the process node with the highest probability value. The determined process node is considered the most likely root cause of the failure, i.e., the failure node. The system uses the identified failure node as a query index to perform data matching in the pre-defined Diagnosis-Action Rule Base within the process knowledge base. This rule base predefines the mapping relationship between each potential failure node and a set of recommended operations, maintenance suggestions, or parameter calibration ranges. Based on the matched rules, the system generates structured production early warning instructions. This production early warning instruction It not only includes fault location information, but also provides specific handling guidance.

[0134] Finally, production early warning instructions The data is pushed to multiple preset monitoring terminals, such as the human-machine interface of the operator control panel, the factory's manufacturing execution system (MES), or the mobile devices of relevant engineers, to trigger corresponding manual or automated processing procedures.

[0135] Step S4200: Real-time acquisition of online sensor data The online sensor data The input is fed into a pre-trained performance predictor model to generate a feature vector for predicted thermal performance. .

[0136] Specifically, this step aims to utilize readily available and rapid online sensor data from the production line. Construct an advanced prediction model for the key performance of the final product, and output the predicted thermal performance feature vector of the final product. This enables proactive sensing and compensation for process disturbances such as raw material fluctuations and environmental changes during product manufacturing, thereby improving the response speed and robustness of the control system.

[0137] The online sensor data Data is collected through various non-destructive real-time sensors deployed at key process nodes on the production line, including but not limited to the feeding port, mixing section, and coating station. These sensors encompass technical equipment from diverse fields, including but not limited to: near-infrared spectrometers for analyzing the chemical composition of materials, machine vision systems for monitoring the mixing state of materials, and process sensors for reading equipment operating status. The system periodically collects the output signals from these heterogeneous sensors, preprocesses and fuses the data, and ultimately generates a unified process state vector, i.e., online sensor data. .

[0138] In practice, traditional feedback control typically relies on the testing results of the final product, inevitably resulting in an inherent time lag. To overcome this deficiency, this step introduces a feedforward prediction mechanism.

[0139] The feedforward prediction mechanism employs a pre-trained and lightweight performance predictor model. This performance predictor model is preferably a multilayer perceptron (MLP) or a simplified graph neural network, trained on a large amount of historical data to learn from readily available real-time online sensor data. The complex nonlinear mapping relationship between the final product performance characteristics that are difficult to measure directly in real time and the final product performance characteristics.

[0140] In practical applications, the system collects online sensor data of the current production batch in real time. This data is then input into the performance predictor model. Through a single forward propagation calculation, a predicted thermal performance feature vector is obtained for the final product before it is fully formed. .

[0141] Crucially, this predicted thermal performance eigenvector The data structure and dimensions, and the dynamic thermal performance feature vector generated in step S2300. and the target thermal performance feature vector used in subsequent steps Completely consistent. Ensuring mathematical comparability between the vectors, the system can directly quantify and compare the deviation between the predicted performance and the target performance in step S4300, thereby calculating the control command used for advance compensation.

[0142] Step S4300, based on the predicted thermal performance feature vector Failure probability distribution and the preset target thermal performance feature vector The optimal control command is determined by constructing and solving a hybrid optimization objective function that combines feedforward and feedback. .

[0143] Specifically, this step aims to utilize the fault probability distribution output in step S3200. The predicted thermal performance feature vector output by step S4200 and the preset target thermal performance feature vector A feedforward-feedback coupled hybrid optimization objective function is constructed and solved to calculate the optimal process parameter adjustment, i.e., the optimal control command. This enables the intelligent balancing of minimizing future performance deviations during the production of heat therapy patches, i.e., the feedforward control objective, and the adjustment robustness based on historical diagnostics, i.e., the feedback constraint.

[0144] In practice, this step is the core computational step for achieving dynamic optimization. Let the parameter vector of the current production process be... The parameter adjustment vector to be solved is ,in These are the optimization variables that are continuously iterated and updated during the optimization process. The received predicted thermal performance feature vector... Revealing the current process parameters And the expected performance endpoint of the product under the influence of disturbances. When the predicted thermal performance eigenvector With the target thermal performance eigenvector When deviations exist, this step employs a hybrid optimization objective function that combines feedforward and feedback to calculate the final optimal control command. To compensate for this deviation. The optimal control command This includes specific adjustments to various process parameters, which are directly sent to the underlying process control system (PCS) or programmable logic controller (PLC) to execute the corresponding adjustments. The specific process formula is as follows:

[0145] ;

[0146] in, This represents the minimum parameter solving operator, used to find the parameter adjustment vector that minimizes the subsequent objective function expression. And return it as the result.

[0147] The hybrid optimization objective function is composed of the following two key coupled parts, as detailed below:

[0148] The first item is the performance error term. This is based on the core idea of ​​Model Predictive Control (MPC). It utilizes a pre-trained forward process model. Adjust the parameter vector for the assumptions applied. Subsequent process parameters Forward reasoning is performed to predict the performance vector of the future product, namely the second predicted thermal performance feature vector. This step involves calculating the second predicted thermal performance feature vector. With the desired target thermal performance eigenvector Minimize the square of the Euclidean distance between them, and adjust the parameter vector. .in, The square of the L2 norm is a mathematical metric used to measure the distance between two vectors, and is used to calculate the second predicted thermal performance eigenvector. With the target thermal performance eigenvector The square of the Euclidean distance between them is used as the performance error of the scalarization.

[0149] The second item is the diagnostic regularization term. This item will determine the failure probability distribution. That is, the feedback diagnostic results are used as an adaptive penalty weight to adjust the parameter vector. Constraints are imposed to improve the robustness of the control process. Among these, This represents the regularization coefficient, a non-negative hyperparameter set by the user to balance the weight between performance error and adjustment penalty. This represents the diagnostic regularization function.

[0150] Example 2:

[0151] This embodiment, based on Embodiment 1, provides a dynamic optimization control system for the process parameters of the heat therapy patch, such as... Figure 2 As shown, it includes a multimodal information acquisition module, a data scoring module, a root cause reasoning module, and a control decision module;

[0152] The multimodal information acquisition module is used to acquire point cloud datasets that characterize the static geometric features of the heat therapy patch. Infrared thermogram time series of dynamic thermodynamic behavior Based on the spatial transformation relationship between sensors, the infrared thermal image time series is... Mapped to the point cloud dataset Constructed 3D mesh model Above, construct a thermodynamic state dataset. ;

[0153] The data scoring module is based on a thermodynamic state dataset. Joint transformation functions are used to generate thermodynamic state sequences. By introducing a static space bias matrix The physical prior processing unit processes the thermodynamic state sequence. Perform matrix calculations and output high-dimensional data features. The high-dimensional data features are decoded using a parallel linear data decoding function. Mapping these metrics to corresponding performance indicators, inputting these performance indicators into the performance scoring calculation model, and outputting a comprehensive performance score. ;

[0154] The root cause reasoning module is based on a comprehensive performance score. Determine if the quality is below the preset threshold. To activate the root cause diagnosis process, if the root cause diagnosis process is activated, then in the pre-built production process diagram data structure Above, the dynamic thermal performance feature vector With the aforementioned production process diagram data structure The initial static feature vectors of each node are fused to generate node input features. A multi-round state iteration calculation method is used to iteratively update these node input features, outputting the fault probability distribution of all nodes. ;

[0155] The control decision module is based on the fault probability distribution. Fault nodes are identified using the maximum a posteriori probability decision method, and data is matched against a pre-defined diagnosis-handling rule base to generate real-time production early warning instructions. Comprehensively utilize the aforementioned fault probability distribution Based on online sensor data The generated predictive thermal performance feature vector and the preset target thermal performance feature vector Construct and solve the feedforward-feedback coupled hybrid optimization objective function, and output the optimal control command. .

[0156] The parts of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of corresponding technical solutions in the prior art have not been described in detail to avoid excessive elaboration.

[0157] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for dynamic optimization control of process parameters for heat therapy patches, characterized in that, include: A point cloud dataset and an infrared thermal image time series are collected. Based on the spatial transformation relationship between sensors, the infrared thermal image time series is mapped onto a three-dimensional mesh model constructed from the point cloud dataset to construct a thermodynamic state dataset. Based on the thermodynamic state dataset, a joint transformation function is used to generate a thermodynamic state sequence. A physical prior processing unit with an introduced static spatial bias matrix is ​​used to perform matrix calculations on the thermodynamic state sequence and output high-dimensional data features. The high-dimensional data features are mapped to the corresponding performance indicators through a parallel linear data decoding function. The performance indicators are input into the performance scoring calculation model and output a comprehensive performance score. Based on the comprehensive performance score, it is determined whether the quality is below a preset quality threshold to activate the root cause diagnosis process. If the root cause diagnosis process is activated, the dynamic thermal performance feature vector is fused with the initial static feature vector of each node in the pre-constructed production process diagram data structure to generate node input features. The node input features are iteratively updated using a multi-round state iteration calculation method to output the failure probability distribution of all nodes. Based on the fault probability distribution, the fault node is identified by the maximum a posteriori probability decision method, and data is matched in the preset diagnosis-disposal rule base to generate production early warning instructions. By comprehensively utilizing the fault probability distribution, the predicted thermal performance feature vector generated based on online sensor data, and the preset target thermal performance feature vector, a feedforward-feedback coupled hybrid optimization objective function is constructed and solved to output the optimal control instruction.

2. The method for dynamic optimization control of process parameters of a heat therapy patch according to claim 1, characterized in that, The method for constructing the thermodynamic state dataset includes: The initial three-dimensional geometric shape of the heat patch before activation is collected, and a point cloud dataset representing the surface geometric features of the heat patch is generated. The time-series infrared radiation intensity signal of the heat therapy patch is periodically collected during its full thermal life cycle, and the infrared radiation intensity signal is mapped in real time in combination with a calibration function to generate an infrared thermogram time series characterizing the surface temperature field of the heat therapy patch. A three-dimensional mesh model is constructed by reconstructing a Poisson surface based on a point cloud dataset. The spatial transformation relationship between heterogeneous coordinate systems is solved by jointly calibrating the sensors. Based on the spatial transformation relationship and the thermal imager intrinsic parameter matrix, the infrared thermal image time series is mapped to the three-dimensional mesh model to construct a thermodynamic state dataset that integrates dynamic thermodynamic behavior.

3. The method for dynamic optimization control of process parameters of a heat therapy patch according to claim 1, characterized in that, The steps for outputting the comprehensive performance score include: The static spatial position and dynamic instantaneous temperature of each three-dimensional grid vertex in the thermodynamic state dataset are concatenated into a state vector. The state vector is then subjected to a nonlinear data transformation using a joint transformation function. The transformed state vectors are arranged in chronological order to form a thermodynamic state sequence. The static spatial bias matrix is ​​calculated based on the geometric features between the vertices of each three-dimensional mesh in the three-dimensional mesh model. The static spatial bias matrix is ​​used as an additive bias term to perform matrix calculation on the thermodynamic state sequence, thereby realizing dynamic thermodynamic dependency modeling guided by physical priors and outputting high-dimensional data features. Global average pooling is performed on high-dimensional data features along the spatiotemporal dimension to aggregate them into a context vector. The context vector is then input into a set of parallel preset linear data decoding functions to map to the corresponding performance indicators. Multiple performance indicators are concatenated along the feature dimension to construct a dynamic thermal performance feature vector. The dynamic thermal performance feature vector is input into a performance scoring calculation model trained under supervision based on a training dataset labeled with authoritative quality scores, and the model outputs a comprehensive performance score.

4. The method for dynamic optimization control of process parameters of a heat therapy patch according to claim 1, characterized in that, The steps for outputting the fault probability distribution include: Using a structured data modeling method, the key factors affecting the thermal performance of heat therapy patches in the manufacturing process are abstracted into entity nodes, and the dependencies between these entity nodes are defined as directed edges to construct a production process graph data structure. The dynamic thermal performance feature vector and the initial static feature vector of each node in the production process diagram data structure are concatenated to generate node input features. The multi-round state iteration calculation method is used to iteratively update the production process diagram data structure to obtain the final feature state vector of each node. The final feature state vector of each node is input into the fault probability function to construct the fault probability distribution of all nodes.

5. The method for dynamic optimization control of process parameters of a heat therapy patch according to claim 4, characterized in that, The entity nodes are divided into material attribute nodes, process parameter nodes, equipment status nodes, and intermediate quality nodes. The material attribute nodes are used to characterize the key physicochemical properties of the input materials, the process parameter nodes are used to characterize the operating parameters set and controlled in the production process, the equipment status nodes are used to characterize the working status parameters of the production equipment during operation, and the intermediate quality nodes are used to characterize the intermediate product attributes that affect subsequent processes.

6. The method for dynamic optimization control of process parameters of a heat therapy patch according to claim 4, characterized in that, The directed edges are divided into causal relationship edges, temporal relationship edges, and subordinate relationship edges; wherein, the causal relationship edge indicates that a change in the state of one node will cause a change in the state of another node, the temporal relationship edge indicates the sequence of the production process, and the subordinate relationship edge indicates that the process parameter node is an inherent attribute of the material attribute node or the equipment state node.

7. The method for dynamic optimization control of process parameters of a heat therapy patch according to claim 1, characterized in that, The generation of the production early warning instruction includes: identifying fault nodes based on the fault probability distribution using the maximum a posteriori probability decision method, using the fault nodes as query indexes to perform data matching in a preset diagnosis-disposal rule base, generating a production early warning instruction, and pushing the production early warning instruction to the monitoring terminal to trigger the corresponding disposal process.

8. The method for dynamic optimization control of process parameters of a heat therapy patch according to claim 1, characterized in that, The steps for outputting the optimal control command include: Real-time acquisition of online sensor data, input of the online sensor data into a pre-trained performance predictor model, and generation of predicted thermal performance feature vectors; Based on the predicted thermal performance feature vector, the fault probability distribution, and the preset target thermal performance feature vector, a hybrid optimization objective function with feedforward-feedback coupling is constructed and solved to determine the optimal control command.

9. The method for dynamic optimization control of process parameters of a heat therapy patch according to claim 8, characterized in that, The hybrid optimization objective function is coupled with a performance error term and a diagnostic regularization term. The performance error term, based on a pre-trained forward process model, performs forward inference on the process parameters after applying the parameter adjustment vector with assumptions, predicts the predicted thermal performance feature vector of the future product, and minimizes the parameter adjustment vector by calculating the square of the Euclidean distance between the predicted thermal performance feature vector and the target thermal performance feature vector. The diagnostic regularization term uses the failure probability distribution as an adaptive penalty weight to constrain the parameter adjustment vector.

10. A dynamic optimization control system for heat therapy patch process parameters, used to implement the dynamic optimization control method for heat therapy patch process parameters as described in any one of claims 1-9, characterized in that, The system includes a multimodal information acquisition module, a data scoring module, a root cause reasoning module, and a control decision module; The multimodal information acquisition module is used to acquire point cloud datasets and infrared thermal image time series. Based on the spatial transformation relationship between sensors, the infrared thermal image time series is mapped onto a three-dimensional mesh model constructed from the point cloud dataset to construct a thermodynamic state dataset. The data scoring module: Based on the thermodynamic state dataset, a joint transformation function is used to generate a thermodynamic state sequence. Using a physical prior processing unit that introduces a static spatial bias matrix, matrix calculations are performed on the thermodynamic state sequence to output high-dimensional data features. The high-dimensional data features are mapped to corresponding performance indicators through a parallel linear data decoding function. The performance indicators are input into the performance scoring calculation model to output a comprehensive performance score. The root cause reasoning module determines whether the comprehensive performance score is below a preset quality threshold to activate the root cause diagnosis process. If the root cause diagnosis process is activated, the dynamic thermal performance feature vector is fused with the initial static feature vector of each node in the pre-constructed production process diagram data structure to generate node input features. The node input features are iteratively updated using a multi-round state iteration calculation method to output the failure probability distribution of all nodes. The control decision module identifies fault nodes based on the fault probability distribution using the maximum a posteriori probability decision method, performs data matching in a preset diagnosis-disposal rule base, generates production early warning instructions, and comprehensively utilizes the fault probability distribution, the predicted thermal performance feature vector generated based on online sensor data, and the preset target thermal performance feature vector to construct and solve a feedforward-feedback coupled hybrid optimization objective function, and outputs the optimal control instructions.

Images