A weakly adhesive hydrocolloid adhesion performance prediction method and system

Abstract

The present application relates to the field of intelligent prediction, and particularly relates to a weak adhesion hydrogel adhesion performance prediction method and system, comprising the following steps: S1: obtaining hydrogel material parameters and interface state information and preprocessing; S2: identifying state labels through a classification model to form an extended data set with state labels; S3: constructing a segmented physical model system driven by the interface state to obtain state-adaptive physical prediction results; S4: embedding the state-adaptive physical prediction results into a physically-constrained neural network structure to construct a prediction model and obtain model prediction results; S5: using a multi-model collaborative fusion mechanism, comprehensively considering the state-driven physical model output and the model prediction results, and obtaining the final prediction value through a weighted fusion manner; S6: taking the final prediction results and model internal parameters as inputs, performing feature contribution degree analysis, identifying key influencing factors, and outputting parameter optimization suggestions. The present application realizes high-precision and high-generalization-capability prediction.

Description

Technical Field

[0001] This invention relates to the field of intelligent prediction, and in particular to a method and system for predicting the adhesion properties of weakly viscous hydrocolloids. Background Technology

[0002] Weakly viscous hydrocolloids, as a typical class of soft materials, are widely used in medical dressings, wearable sensors, flexible electronic devices, and biointerface engineering due to their excellent biocompatibility, flexibility, and high water content. In practical applications, these materials typically need to achieve stable and controllable adhesion to skin or other moist substrates; therefore, accurate prediction of their adhesion properties is of great significance.

[0003] In existing technologies, research on the adhesion properties of hydrocolloids mainly relies on two types of methods. One type is based on physical models of contact mechanics and interfacial thermodynamics, such as the JKR model or the DMT model. However, these methods are mostly designed for ideal dry interfaces and are difficult to effectively describe the coupled effects of multiple factors such as water layer, viscoelastic dissipation, and interfacial roughness in actual humid environments, resulting in limited prediction accuracy. The other type of method is a data-driven approach based on machine learning, which trains models using experimental data to predict adhesion properties. Although this can improve prediction accuracy to some extent, it suffers from poor model generalization ability due to the lack of physical constraints and is highly dependent on the scale of the data.

[0004] Furthermore, while some existing technologies attempt to combine physical models with data-driven methods, most employ simple cascading or residual correction approaches, failing to achieve a deep integration of physical mechanisms and intelligent algorithms. Simultaneously, existing methods typically neglect the crucial role of interface states in the adhesion process, failing to differentiate and model different interface conditions such as dry, wet transition, and liquid bridge states, resulting in insufficient model stability under complex operating conditions. Summary of the Invention

[0005] To address the aforementioned problems, the present invention aims to provide a method and system for predicting the adhesion performance of weakly viscous hydrocolloids, which can comprehensively consider the differences in interface states, introduce physical constraints, and combine artificial intelligence modeling to achieve high-precision and high-generalization prediction.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A method for predicting the adhesion properties of weakly viscous hydrocolloids includes the following steps: S1: Construct a multi-source information fusion data acquisition system to obtain hydrocolloid material parameters and interface state information and preprocess them to obtain standardized structured data and raw interface multimodal data; S2: Based on the original interface multimodal data, identify the status labels through a classification model, and align the status labels with standardized structured data to form an extended dataset with status labels; S3: Construct a segmented physical model system driven by interface state, and obtain state-adaptive physical prediction results based on an extended dataset with state labels; S4: Use the state-adaptive physical prediction results as a physical constraint embedded neural network structure to build a prediction model, and obtain the model prediction results based on an extended dataset with state labels; S5: A multi-model collaborative fusion mechanism is adopted, which comprehensively considers the output of the state-driven physical model and the model prediction results to obtain the final prediction value through weighted fusion. S6: Using the final prediction results and model internal parameters as input, perform feature contribution analysis, identify key influencing factors, and output parameter optimization suggestions.

[0007] Furthermore, a multi-source information fusion data acquisition system is constructed, covering material bulk parameters, interface characteristic parameters, and environmental condition parameters. The material bulk parameters include moisture content, storage modulus, loss modulus, crosslinking density, and surface energy. The interface characteristic parameters cover substrate surface energy, surface roughness, and wettability. The environmental condition parameters include temperature, relative humidity, applied pressure, and contact time. Simultaneously, multi-modal information acquisition of interface states is introduced, including contact area microscopic imaging, infrared detection of interface moisture distribution, and dynamic contact angle time series, forming a comprehensive data source where structured data and unstructured image data coexist.

[0008] Further preprocessing is performed as follows: For numerical data, outlier removal, missing value completion, and standardization are performed; for image and time series data, denoising, cropping, resolution unification, and keyframe extraction are performed; and data from different sources are aligned according to sample number to establish a one-to-one correspondence between material parameters, interface parameters, and interface state data.

[0009] Furthermore, based on the original interface multimodal data, state labels are identified using a classification model, and these state labels are aligned with standardized structured data to form an extended dataset with state labels, as detailed below: Taking the raw interface multimodal data I0 and the normalized structured data X0 output in step S1 as input, the raw interface multimodal data I0 includes the contact area image I img Moisture distribution map I water and dynamic contact angle sequence I angle (t); Based on contact area image I img Extract feature parameters that characterize the interface morphology, including: Contact area percentage: Where, Nc N represents the number of pixels in the contact area and the total number of pixels in the area. total The total number of pixels in the area is equal to the number of pixels in the contact area. Boundary complexity: ; Where P is the perimeter of the contact boundary; Gray scale uniformity: ; Where, σ g and μ g These are the standard deviation and mean of grayscale, respectively. Based on moisture distribution map I water Extract relevant parameters of the interface water layer: Average water layer thickness: ; Moisture distribution dispersion: ; Where, w i Pixel water intensity; Based on contact angle time series I angle (t), extracting dynamic features: Average contact angle: ; Contact angle change rate: ; Stability metrics: ; The above features are combined into a state discrimination feature vector: ; Building a classification model based on support vector machines: ; The identified status label S is matched with the structured data X0 according to the sample index.

[0010] Furthermore, a segmented physical model system driven by interface state is constructed. Based on an extended dataset with state labels, state-adaptive physical prediction results are obtained, as follows: Define the state selection function: ; in These correspond to adhesion physical models under different interface states; S1, S2, and S3 represent the dry contact state, the wet transition state, and the liquid bridge-dominated state, respectively. When S=S1, the adhesion force is calculated using the JKR model based on contact mechanics: ; Where R is the equivalent radius of curvature; W is the interface work; Considering the viscoelasticity correction of the material, and introducing the modulus effect term, the corrected adhesion force is obtained: ; in, This is an empirical coefficient; Energy storage modulus; For loss modulus; When S=S2, a thin water layer exists at the interface, which can be modified to obtain effective interface work. ; Among them, h w β is the average water layer thickness at the interface; β is the attenuation coefficient. The adhesion force is then expressed as: ; Also considering the contribution of viscoelastic dissipation: ; Where η is the equivalent viscosity; v is the separation rate; The corrected adhesion force is obtained by combining the results: ; When S=S3, a liquid bridge structure is formed at the interface, and the adhesion force is mainly determined by capillary force. ; Where, γ l θ is the surface tension of the liquid; θ is the contact angle; ΔP is the Laplace pressure. Considering the influence of interface roughness, a correction factor is introduced: ; Where λ is the roughness attenuation coefficient; Based on the state selection function, the output is a state-adaptive physical prediction result: .

[0011] Furthermore, the state-adaptive physical prediction results are used as a physical constraint embedded neural network structure to construct a prediction model. Based on an extended dataset with state labels, the model prediction results are obtained, as follows: The extended dataset X with state labels obtained in step S2 ext=(X0,S) is used as the model input, where X0 is the standardized structured feature and S is the interface state label, to construct a feedforward neural network model f. θ (·), the output is the predicted value of hydrocolloid adhesion performance: Y AI =f θ (X) ext ); Where θ represents the network parameters; the network structure adopts a multi-layer fully connected form and uses a nonlinear activation function to achieve complex nonlinear mapping; The physical model is embedded as a constraint term in the neural network training process, and the state obtained in step S3 drives the physical prediction results. As a physical reference, physical consistency constraints are introduced; Construct the joint loss function: ; ; ; ; in, , and These represent the data error term, the physical consistency constraint term, and the state smoothness constraint term, respectively; λ1 and λ2 are weighting coefficients; N is the number of samples. A state modulation mechanism is introduced to conditionally control the intermediate layers of the network, and a state modulation factor is introduced at the k-th layer: ; Among them, h k For the output of the k-th layer, γ(S) and β(S) are the scaling and bias functions generated by the state labels, and ⊙ represents element-wise multiplication; Finally, by iteratively optimizing the model parameters θ by minimizing the joint loss function described above, the optimal model parameters are obtained: ; After training is complete, the dataset X will be expanded. ext The input model corresponds to the output prediction of adhesion performance. .

[0012] Furthermore, a multi-model collaborative fusion mechanism is adopted, which comprehensively considers the output of the state-driven physical model and the model prediction results, and obtains the final prediction value through a weighted fusion method, as follows: The state-adaptive physics prediction results obtained in step S3 and the model prediction result Y obtained in step S4 AIAs input, consistency processing is performed on the outputs of the two types of models, including unit unification and numerical normalization, to eliminate the influence of differences in the output scales of different models on the fusion results. Let the results after normalization be as follows: and ; A nonlinear fusion approach is introduced to enhance expressive power, resulting in the final prediction result: ; in, φ is the fusion weight; φ(·) is the nonlinear mapping function.

[0013] Furthermore, using the final prediction results and the model's internal parameters as input, feature contribution analysis is performed to identify key influencing factors and output parameter optimization suggestions, as follows: Based on a pre-trained physical constraint embedded neural network model The model's response to input features is extracted, a feature contribution analysis mechanism is constructed, and a gradient-based method is used to calculate the sensitivity of each input feature to the prediction result. ; Among them, C j Let x represent the contribution of the j-th feature. j For each input feature, Y represents the model prediction result; and the Shapley value method is used to evaluate the contribution of the feature in the global scope, thereby obtaining the feature importance ranking. Based on the contribution of each feature, key influencing factors are identified and classified. A threshold τ is set according to the contribution magnitude to select those that meet C. j The characteristics of τ are taken as the dominant factor, and combined with its physical meaning, it is divided into material parameter class, interface parameter class and state-related parameter class. Based on the contribution and direction of key features, a parameter optimization strategy is constructed. By analyzing the sign and trend of the partial derivatives of the model output with respect to key parameters, the direction of parameter adjustment is determined.

[0014] A system for predicting the adhesion performance of weakly viscous hydrocolloids includes a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it specifically performs the steps in the method for predicting the adhesion performance of weakly viscous hydrocolloids as described above.

[0015] A computer storage medium storing a plurality of instructions adapted for loading by a processor and executing the method steps described above.

[0016] The present invention has the following beneficial effects: 1. This invention constructs a segmented physical model system driven by interface state, and models the adhesion behavior under complex interface conditions in a segmented state, so that different physical mechanisms can be expressed in a targeted manner under corresponding states. It can accurately characterize the different mechanisms under various interface conditions such as dry contact, wet transition and liquid bridge dominance, thereby significantly improving the applicability and prediction accuracy of physical modeling. 2. This invention embeds the physical model into the neural network in the form of constraints, forming a physical constraint embedded learning framework, which realizes the deep integration of the mechanism model and the data-driven model. On the one hand, it uses the nonlinear fitting ability of the data-driven model to improve the prediction accuracy, and on the other hand, it ensures that the prediction results conform to the objective mechanism law through physical constraints, thereby effectively avoiding the distortion problem of pure data model under small sample or cross-condition conditions, and improving the generalization ability and stability of the model. 3. This invention introduces a multi-model collaborative fusion and feature contribution analysis mechanism, which not only obtains high-precision prediction results but also enhances the interpretability of the model. By identifying and analyzing the direction of action of key influencing factors, it can further output parameter optimization suggestions with guiding significance, forming a closed-loop system of "prediction-interpretation-optimization". This provides a reliable basis for the design and performance control of hydrocolloid materials and has high engineering application value. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0018] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments: refer to Figure 1 In this embodiment, a method for predicting the adhesion properties of weakly viscous hydrocolloids is provided, comprising the following steps: S1: Construct a multi-source information fusion data acquisition system to obtain hydrocolloid material parameters and interface state information and preprocess them to obtain standardized structured data and raw interface multimodal data; S2: Based on the original interface multimodal data, identify the status labels through a classification model, and align the status labels with standardized structured data to form an extended dataset with status labels; S3: Construct a segmented physical model system driven by interface state, and obtain state-adaptive physical prediction results based on an extended dataset with state labels; S4: Use the state-adaptive physical prediction results as a physical constraint embedded neural network structure to build a prediction model, and obtain the model prediction results based on an extended dataset with state labels; S5: A multi-model collaborative fusion mechanism is adopted, which comprehensively considers the output of the state-driven physical model and the model prediction results to obtain the final prediction value through weighted fusion. S6: Using the final prediction results and model internal parameters as input, perform feature contribution analysis, identify key influencing factors, and output parameter optimization suggestions.

[0019] In this embodiment, a multi-source information fusion data acquisition system is constructed, covering material bulk parameters, interface characteristic parameters, and environmental condition parameters. The material bulk parameters include key physical properties such as moisture content, storage modulus, loss modulus, crosslinking density, and surface energy, which are obtained through standard methods such as rheological testing, DSC analysis, and contact angle measurement. The interface characteristic parameters include substrate surface energy, surface roughness, and wettability, which are measured using equipment such as surface energy analyzers, atomic force microscopes, and roughness meters. The environmental condition parameters include external variables such as temperature, relative humidity, applied pressure, and contact time, which are monitored and recorded in real time by an environmental control system. Simultaneously, multimodal information acquisition of interface states is introduced, including contact area microscopic imaging, infrared detection of interface moisture distribution, and dynamic contact angle time series, forming a comprehensive data source where structured data and unstructured image data coexist.

[0020] In this embodiment, the preprocessing is as follows: For numerical data, outlier removal, missing value completion, and standardization are performed to make data of different dimensions comparable; for image and time series data, denoising, cropping, resolution unification, and keyframe extraction are performed to improve the stability and accuracy of subsequent analysis; and data from different sources are aligned according to sample number to establish a one-to-one correspondence between material parameters, interface parameters, and interface state data.

[0021] In this embodiment, based on the original interface multimodal data, a classification model is used to identify state labels, and the state labels are aligned with standardized structured data to form an extended dataset with state labels, as detailed below: Taking the raw interface multimodal data I0 and the normalized structured data X0 output in step S1 as input, the raw interface multimodal data I0 includes the contact area image I img Moisture distribution map I water and dynamic contact angle sequence I angle (t); Based on contact area image I img Extract feature parameters that characterize the interface morphology, including: Contact area percentage: Where, N c N represents the number of pixels in the contact area and the total number of pixels in the area. total The total number of pixels in the area is equal to the number of pixels in the contact area. Boundary complexity: ; Where P is the perimeter of the contact boundary; Gray scale uniformity: ; Where, σ g and μ g These are the standard deviation and mean of grayscale, respectively. Based on moisture distribution map I water Extract relevant parameters of the interface water layer: Average water layer thickness: ; Moisture distribution dispersion: ; Where, w i Pixel water intensity; Based on contact angle time series I angle (t), extracting dynamic features: Average contact angle: ; Contact angle change rate: ; Stability indicators (degree of volatility): ; The above features are combined into a state discrimination feature vector: ; Building a classification model based on support vector machines: ; The identified status label S is matched with the structured data X0 according to the sample index.

[0022] In this embodiment, a segmented physical model system driven by interface state is constructed. Based on an extended dataset with state identifiers, state-adaptive physical prediction results are obtained, as follows: Define the state selection function: ; in These correspond to adhesion physical models under different interface states; S1, S2, and S3 represent the dry contact state, the wet transition state, and the liquid bridge-dominated state, respectively. When S=S1, the adhesion force is calculated using the JKR model based on contact mechanics: ; Where R is the equivalent radius of curvature; W is the interface work; Considering the viscoelasticity correction of the material, and introducing the modulus effect term, the corrected adhesion force is obtained: ; in, This is an empirical coefficient; Energy storage modulus; For loss modulus; When S=S2, a thin water layer exists at the interface, which can be modified to obtain effective interface work. ; Among them, h w β is the average water layer thickness at the interface; β is the attenuation coefficient. The adhesion force is then expressed as: ; Also considering the contribution of viscoelastic dissipation: ; Where η is the equivalent viscosity; v is the separation rate; The corrected adhesion force is obtained by combining the results: ; When S=S3, a liquid bridge structure is formed at the interface, and the adhesion force is mainly determined by capillary force. ; Where, γ l θ is the surface tension of the liquid; θ is the contact angle; ΔP is the Laplace pressure. Considering the influence of interface roughness, a correction factor is introduced: ; Where λ is the roughness attenuation coefficient; Based on the state selection function, the output is a state-adaptive physical prediction result: .

[0023] In this embodiment, the state-adaptive physical prediction results are used as the physical constraint embedded neural network structure to construct a prediction model. Based on an extended dataset with state labels, the model prediction results are obtained, as follows: The extended dataset X with state labels obtained in step S2 ext =(X0,S) is used as the model input, where X0 is the standardized structured feature and S is the interface state label, to construct a feedforward neural network model f. θ (·), the output is the predicted value of hydrocolloid adhesion performance: Y AI =f θ (X) ext ); Where θ represents the network parameters; the network structure adopts a multi-layer fully connected form and uses a nonlinear activation function (such as ReLU) to achieve complex nonlinear mapping; The physical model is embedded as a constraint term in the neural network training process, and the state obtained in step S3 drives the physical prediction results. As a physical reference, physical consistency constraints are introduced to ensure that the model prediction results approximate the experimental values ​​without deviating from the physical mechanism. Construct the joint loss function: ; ; ; ; in, , and These represent the data error term, the physical consistency constraint term, and the state smoothness constraint term, respectively; λ1 and λ2 are weighting coefficients; N is the number of samples. To enhance the model's adaptability to different interface states, a state modulation mechanism can be introduced to conditionally control the intermediate layers of the network. A state modulation factor is introduced in the k-th layer: ; Among them, h k For the output of the k-th layer, γ(S) and β(S) are scaling and bias functions generated by the state labels, and ⊙ represents element-wise multiplication, thereby realizing adaptive adjustment of feature response under different states; Finally, by iteratively optimizing the model parameters θ by minimizing the joint loss function described above, the optimal model parameters are obtained: ; After training is complete, the dataset X will be expanded. ext The input model corresponds to the output prediction of adhesion performance. .

[0024] In this embodiment, a multi-model collaborative fusion mechanism is adopted. The final predicted value is obtained by weighted fusion of the output of the state-driven physical model and the model prediction results, as follows: The state-adaptive physics prediction results obtained in step S3 and the model prediction result Y obtained in step S4 AI As input, consistency processing is performed on the outputs of the two types of models, including unit unification and numerical normalization, to eliminate the influence of differences in the output scales of different models on the fusion results. Let the results after normalization be as follows: and ; A nonlinear fusion approach is introduced to enhance expressive power, resulting in the final prediction result: ; in, φ is the fusion weight; φ(·) is the nonlinear mapping function.

[0025] In this embodiment, the final prediction results and model internal parameters are used as input to perform feature contribution analysis, identify key influencing factors, and output parameter optimization suggestions, as follows: Based on a pre-trained physical constraint embedded neural network model The model's response to input features is extracted, a feature contribution analysis mechanism is constructed, and a gradient-based method is used to calculate the sensitivity of each input feature to the prediction result. ; Among them, C j Let x represent the contribution of the j-th feature. j For each input feature, Y represents the model prediction result; and the Shapley value method is used to evaluate the contribution of the feature in the global scope, thereby obtaining the feature importance ranking. Based on the contribution of each feature, key influencing factors are identified and classified. A threshold τ is set according to the contribution magnitude to select those that meet C. j The characteristics of τ are taken as the dominant factor, and combined with their physical meaning, they are divided into material parameters (such as modulus and water content), interface parameters (such as surface energy and roughness), and state-related parameters (such as water layer thickness and contact angle). At the same time, the variation law of the characteristic effect under different interface states is analyzed by combining the internal parameters of the model (such as state modulation coefficients γ(S) and β(S)), thereby revealing the coupling relationship between "parameter-state-performance". Based on the contribution and direction of key features, a parameter optimization strategy is constructed. By analyzing the sign and trend of the partial derivatives of the model output with respect to key parameters, the direction of parameter adjustment is determined.

[0026] For example, when a certain parameter satisfies: ; Increasing this parameter can enhance adhesion performance; conversely, decreasing it should help. Based on this, further constraints (such as material machinability or application safety limits) can be considered to determine the optimal parameter range. Furthermore, differentiated optimization suggestions were formulated for different interface states, such as prioritizing the control of surface tension and water layer thickness in the liquid bridge-dominated state, and focusing on optimizing material modulus and interfacial energy in the dry contact state.

[0027] A system for predicting the adhesion performance of weakly viscous hydrocolloids includes a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it specifically performs the steps in the method for predicting the adhesion performance of weakly viscous hydrocolloids as described above.

[0028] A computer storage medium storing a plurality of instructions adapted for loading by a processor and executing the method steps described above.

[0029] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0030] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. 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 flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0031] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0032] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0033] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for predicting the adhesion properties of weakly viscous hydrocolloids, characterized in that, Includes the following steps: S1: Construct a multi-source information fusion data acquisition system to obtain hydrocolloid material parameters and interface state information and preprocess them to obtain standardized structured data and raw interface multimodal data; S2: Based on the original interface multimodal data, identify the status labels through a classification model, and align the status labels with standardized structured data to form an extended dataset with status labels; S3: Construct a segmented physical model system driven by interface state, and obtain state-adaptive physical prediction results based on an extended dataset with state labels; S4: Use the state-adaptive physical prediction results as a physical constraint embedded neural network structure to build a prediction model, and obtain the model prediction results based on an extended dataset with state labels; S5: A multi-model collaborative fusion mechanism is adopted, which comprehensively considers the output of the state-driven physical model and the model prediction results to obtain the final prediction value through weighted fusion. S6: Using the final prediction results and model internal parameters as input, perform feature contribution analysis, identify key influencing factors, and output parameter optimization suggestions.

2. The method for predicting the adhesion properties of weakly viscous hydrocolloids according to claim 1, characterized in that, The constructed multi-source information fusion data acquisition system covers material bulk parameters, interface characteristic parameters, and environmental condition parameters. The material bulk parameters include moisture content, storage modulus, loss modulus, crosslinking density, and surface energy. The interface characteristic parameters cover substrate surface energy, surface roughness, and wettability. The environmental condition parameters include temperature, relative humidity, applied pressure, and contact time. Simultaneously, multimodal information acquisition of interface states is introduced, including contact area microscopic imaging, infrared detection of interface moisture distribution, and dynamic contact angle time series, forming a comprehensive data source where structured data and unstructured image data coexist.

3. The method for predicting the adhesion properties of weakly viscous hydrocolloids according to claim 2, characterized in that, The preprocessing is as follows: for numerical data, outlier removal, missing value completion, and standardization are performed; for image and time series data, denoising, cropping, resolution unification, and keyframe extraction are performed; and data from different sources are aligned according to sample number to establish a one-to-one correspondence between material parameters, interface parameters, and interface state data.

4. The method for predicting the adhesion properties of weakly viscous hydrocolloids according to claim 1, characterized in that, The process involves identifying state labels using a classification model based on the original interface multimodal data, and then aligning these state labels with standardized structured data to form an extended dataset with state identifiers, as detailed below: Taking the raw interface multimodal data I0 and the normalized structured data X0 output in step S1 as input, the raw interface multimodal data I0 includes the contact area image I img Moisture distribution map I water and dynamic contact angle sequence I angle (t); Based on contact area image I img Extract feature parameters that characterize the interface morphology, including: Contact area percentage: ； Where, N c N represents the number of pixels in the contact area and the total number of pixels in the area. total The total number of pixels in the area is equal to the number of pixels in the contact area. Boundary complexity: ； Where P is the perimeter of the contact boundary; Gray scale uniformity: ； Where, σ g and μ g These are the standard deviation and mean of grayscale, respectively. Based on moisture distribution map I water Extract relevant parameters of the interfacial water layer: Average water layer thickness: ； Moisture distribution dispersion: ； Where, w i Pixel water intensity; Based on contact angle time series I angle (t), extracting dynamic features: Average contact angle: ; Contact angle change rate: ; Stability metrics: ; The above features are combined into a state discrimination feature vector: ； Building a classification model based on support vector machines: ； The identified status label S is matched with the structured data X0 according to the sample index.

5. The method for predicting the adhesion properties of weakly viscous hydrocolloids according to claim 1, characterized in that, The construction of a segmented physical model system driven by interface state, based on an extended dataset with state labels, obtains state-adaptive physical prediction results, as detailed below: Define the state selection function: ； in These correspond to adhesion physical models under different interface states; S1, S2, and S3 represent the dry contact state, the wet transition state, and the liquid bridge-dominated state, respectively. When S=S1, the adhesion force is calculated using the JKR model based on contact mechanics: ； Where R is the equivalent radius of curvature; W is the interface work; Considering the viscoelasticity correction of the material, and introducing the modulus effect term, the corrected adhesion force is obtained: ； in, This is an empirical coefficient; Energy storage modulus; For loss modulus; When S=S2, a thin water layer exists at the interface, which can be corrected to obtain effective interface work. ； Among them, h w β is the average water layer thickness at the interface; β is the attenuation coefficient. The adhesive force is then expressed as: ； Also considering the contribution of viscoelastic dissipation: ； Where η is the equivalent viscosity; v is the separation rate; The corrected adhesion force is obtained by combining the results: ； When S=S3, a liquid bridge structure is formed at the interface, and the adhesion force is mainly determined by capillary force. ； Where, γ l θ is the surface tension of the liquid; θ is the contact angle; ΔP is the Laplace pressure. To account for the influence of interface roughness, a correction factor is introduced: ； Where λ is the roughness attenuation coefficient; Based on the state selection function, the output is a state-adaptive physical prediction result: 。 6. The method for predicting the adhesion properties of weakly viscous hydrocolloids according to claim 5, characterized in that, The process involves using the state-adaptive physical prediction results as a physical constraint embedded neural network structure to construct a prediction model. Based on an extended dataset with state labels, the model prediction results are obtained, as detailed below: The extended dataset X with state labels obtained in step S2 ext =(X0,S) is used as the model input, where X0 is the standardized structured feature and S is the interface state label, to construct a feedforward neural network model f. θ (·), the output is the predicted value of hydrocolloid adhesion performance: Y AI =f θ ( X ext ); Where θ represents the network parameters; the network structure adopts a multi-layer fully connected form and uses a nonlinear activation function to achieve complex nonlinear mapping; The physical model is embedded as a constraint term in the neural network training process, and the state obtained in step S3 drives the physical prediction results. As a physical reference, physical consistency constraints are introduced; Construct the joint loss function: ； ； ； ； in, , and These represent the data error term, the physical consistency constraint term, and the state smoothness constraint term, respectively; λ1 and λ2 are weighting coefficients; N is the number of samples. A state modulation mechanism is introduced to conditionally control the intermediate layers of the network, and a state modulation factor is introduced at the k-th layer: ； Among them, h k For the output of the k-th layer, γ(S) and β(S) are the scaling and bias functions generated by the state labels, and ⊙ represents element-wise multiplication; Finally, by iteratively optimizing the model parameters θ by minimizing the joint loss function described above, the optimal model parameters are obtained: ； After training is complete, the dataset X will be expanded. ext The input model corresponds to the output prediction of adhesion performance. 。 7. The method for predicting the adhesion properties of weakly viscous hydrocolloids according to claim 6, characterized in that, The multi-model collaborative fusion mechanism is adopted, which comprehensively considers the output of the state-driven physical model and the model prediction results, and obtains the final prediction value through a weighted fusion method, as follows: The state-adaptive physics prediction results obtained in step S3 and the model prediction result Y obtained in step S4 AI As input, consistency processing is performed on the outputs of the two types of models, including unit unification and numerical normalization, to eliminate the influence of differences in the output scales of different models on the fusion results. Let the results after normalization be as follows: and ; A nonlinear fusion approach is introduced to enhance expressive power, resulting in the final prediction result: ； in, φ is the fusion weight; φ(·) is the nonlinear mapping function.

8. The method for predicting the adhesion properties of weakly viscous hydrocolloids according to claim 1, characterized in that, The process involves using the final prediction results and model internal parameters as input to perform feature contribution analysis, identify key influencing factors, and output parameter optimization suggestions, as detailed below: Based on a pre-trained physical constraint embedded neural network model The model's response to input features is extracted, a feature contribution analysis mechanism is constructed, and a gradient-based method is used to calculate the sensitivity of each input feature to the prediction result. ； Among them, C j Let x represent the contribution of the j-th feature. j For each input feature, Y represents the model prediction result; and the Shapley value method is used to evaluate the contribution of the feature in the global scope, thereby obtaining the feature importance ranking. Based on the contribution of each feature, key influencing factors are identified and classified. A threshold τ is set according to the contribution magnitude to select those that meet C. j The characteristics of τ are taken as the dominant factor, and combined with its physical meaning, it is divided into material parameter class, interface parameter class and state-related parameter class. Based on the contribution and direction of key features, a parameter optimization strategy is constructed. By analyzing the sign and trend of the partial derivatives of the model output with respect to key parameters, the direction of parameter adjustment is determined.

9. A system for predicting the adhesion properties of weakly viscous hydrocolloids, characterized in that, It includes a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it specifically performs the steps in the method for predicting the adhesion performance of a weakly viscous hydrocolloid as described in any one of claims 1-8.

10. A computer storage medium, characterized in that, The computer storage medium stores a plurality of instructions adapted for loading by a processor and executing the method steps as claimed in any one of claims 1-8.

Images