Intelligent inversion method and system for rheological properties of single-component alkali-activated mortar

Abstract

The invention discloses an intelligent inversion method and system for rheological characteristics of single-component alkali-activated mortar, and the method comprises the steps: synchronously collecting a dynamic image sequence and environment temperature and humidity data of a mortar collapse process through a global shutter camera and a temperature and humidity sensor; polar coordinate transformation is introduced in the aspect of data processing to unfold a circular mortar expansion image into a rectangular panorama, so that the edge recognition precision is remarkably improved; static texture features are extracted through an improved ResNet embedded into a CBAM attention module, dynamic rheological features in the collapse process are captured by using a CNN-LSTM network, and environmental parameters are mapped into high-dimensional vectors through MLP; and finally, fusing the three features to invert the fluidity diameter, identifying an abnormal state (such as bleeding and segregation) and outputting a production line water replenishing suggestion. According to the invention, the detection problems of strong environmental sensitivity and complex micro texture of a single-component alkali-activated material are effectively solved, and high-precision non-contact intelligent detection is realized.

Description

Technical Field

[0001] This invention relates to the field of rheological property testing technology for single-component alkali-activated mortar, specifically to an intelligent inversion method and system for the rheological properties of single-component alkali-activated mortar. Background Technology

[0002] Single-component alkali-activated materials have the potential to replace traditional cement due to their low-carbon properties and convenient "just add water" application. However, this material system exhibits complex rheological properties: (1) High environmental sensitivity: Alkali-induced reaction is an exothermic reaction, and the reaction rate is greatly affected by ambient temperature and humidity. The appearance of the image alone cannot fully reflect the changes in its internal viscosity.

[0003] (2) Complex microscopic features: If the solid activator is not fully dissolved, it will cause microscopic "bleeding" or "aggregate segregation" phenomena. Traditional machine vision (such as SVM) or basic deep learning (such as ordinary ResNet) can hardly capture these subtle texture differences.

[0004] (3) Static limitations: Existing image detection technology only measures the final spreading diameter (static value), ignoring the "spreading rate" of mortar during the process of mortar collapsing from the cone to stopping flow. However, the spreading rate is directly related to the material's yield stress and plastic viscosity, which are more critical indicators for judging construction performance.

[0005] Therefore, existing technologies lack an intelligent detection method that can simultaneously combine temporal dynamic characteristics and environmental physical parameters. Summary of the Invention

[0006] This invention provides an intelligent inversion method for the rheological properties of single-component alkali-activated mortar, in order to solve the problem that existing technologies lack intelligent detection methods for the rheological properties of single-component alkali-activated mortar that can simultaneously combine temporal dynamic characteristics and environmental physical parameters.

[0007] According to the first aspect, one embodiment provides a smart inversion method for the rheological properties of a single-component alkali-activated mortar, the method comprising: Under the conditions of a ring-shaped shadowless light source and polarization filter, dynamic image sequences of the collapse process of single-component alkali-activated mortar and the current temperature and humidity data of the environment were simultaneously acquired. The acquired images are transformed using polar coordinates, mapping the image containing the circular mortar extension surface into a rectangular panoramic unfolded image; A spatiotemporal dual-stream neural network for image feature extraction was constructed and trained. The spatial flow branch uses a deep convolutional network with embedded CBAM attention modules to extract static texture and edge defect features, while the temporal flow branch uses a temporal recurrent neural network to analyze the optical flow changes of the image sequence and extract dynamic features characterizing viscosity. The spatiotemporal visual feature vector extracted by the spatiotemporal dual-flow neural network is concatenated and fused with the encoded temperature and humidity feature vector, and the flowability prediction value and rheological state evaluation are output through fully connected layer regression.

[0008] Furthermore, under an annular shadowless light source and polarization filtering environment, dynamic image sequences of the collapse process of single-component alkali-activated mortar and the current ambient temperature and humidity data were simultaneously acquired, specifically including: A global shutter industrial camera was used to capture a dynamic image sequence of the first few seconds of the collapse process of single-component alkali-activated mortar. Temperature and humidity data were sampled synchronously with the image acquisition using a high-precision sensor.

[0009] Furthermore, the acquired images are subjected to polar coordinate transformation to map the image containing the circular mortar extension surface into a rectangular panoramic unfolded image, specifically including: ROI extraction: Automatically crop out areas containing mortar using background subtraction method; Polar coordinate transformation: Setting the image center as the pole O(x0, y0), the image in Cartesian coordinates (x, y) is mapped to polar coordinates (r, θ), as shown in the following formula:

[0010] After polar coordinate transformation, the irregular circular mortar extension surface becomes a rectangular band. The center point of the polar coordinate transformation is automatically obtained using the centroid method.

[0011] Furthermore, the CBAM attention module includes a channel attention submodule and a spatial attention submodule connected in series; the spatial attention submodule uses max pooling and average pooling operations to generate a spatial weight map, automatically weighting the high-frequency texture region of the mortar edge to suppress background noise.

[0012] Furthermore, the input to the spatial flow branch is a rectangular panoramic unfolded image obtained after polar coordinate transformation, and the input to the temporal flow branch is a sequence of differential images of the first multiple frames of the single-component alkali-activated mortar collapse process.

[0013] Furthermore, the spatiotemporal visual feature vector extracted by the spatiotemporal dual-stream neural network is concatenated and fused with the encoded temperature and humidity feature vector. The fluidity prediction and rheological state evaluation are then output through a fully connected layer regression, specifically including: Temperature and humidity data are mapped from scalar data to high-dimensional feature vectors using a multilayer perceptron. These feature vectors are then concatenated with the two feature vectors output from the spatial flow branch and the temporal flow branch in the channel dimension. Finally, a fully connected layer regressor outputs the predicted mobility value and the confidence level of the abnormal state.

[0014] Furthermore, the method also includes: Reverse compensation calculation: Based on the predicted flowability deviation and the current ambient temperature and humidity, the recommended water replenishment amount in the single-component mortar production process is output through a trained inverse mapping relationship; the flowability deviation refers to the difference between the preset standard target flowability value of mortar and the flowability prediction value output by the fully connected layer regressor; the nonlinear regression model takes the flowability deviation and the current ambient temperature and humidity as inputs, and the water replenishment amount required to eliminate the corresponding flowability deviation as output.

[0015] Furthermore, the spatial flow branch uses ResNet-50 as the backbone network, and inserts a CBAM attention module after each residual block.

[0016] Furthermore, the time-flow branch employs a CNN-LSTM network to extract mortar expansion acceleration features to characterize viscosity properties by processing the difference features of the image sequence.

[0017] According to a second aspect, one embodiment provides an intelligent inversion system for the rheological properties of a single-component alkali-activated mortar, the system comprising: The multimodal data acquisition module is used to simultaneously acquire dynamic image sequences of the collapse process of single-component alkali-activated mortar and the temperature and humidity data of the current environment under the conditions of a ring shadowless light source and polarization filter. The geometric space transformation module is used to perform polar coordinate transformation on the acquired images, mapping the image containing the circular mortar extension surface into a rectangular panoramic unfolded image. The spatiotemporal feature extraction module is used to construct and train a spatiotemporal dual-stream neural network for image feature extraction. The spatial flow branch uses a deep convolutional network with an embedded CBAM attention module to extract static texture and edge defect features, while the temporal flow branch uses a recurrent neural network to analyze the optical flow changes of the image sequence and extract dynamic features characterizing viscosity. Multimodal fusion inversion is used to concatenate and fuse the spatiotemporal visual feature vectors extracted by the spatiotemporal dual-stream neural network with the encoded temperature and humidity feature vectors, and output the predicted mobility value and rheological state evaluation through fully connected layer regression.

[0018] According to a third aspect, one embodiment provides an electronic device, the device comprising: a processor and a memory; The memory is used to store one or more program instructions; The processor is configured to run one or more program instructions to perform the steps of the intelligent inversion method for the rheological properties of a single-component alkali-activated mortar as described in any of the preceding claims.

[0019] According to a fourth aspect, one embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the intelligent inversion method for the rheological properties of a single-component alkali-activated mortar as described in any of the preceding claims.

[0020] This invention provides an intelligent inversion method and system for the rheological properties of single-component alkali-activated mortar. Utilizing a global shutter camera and temperature and humidity sensors, it simultaneously acquires dynamic image sequences of the mortar slump process and environmental temperature and humidity data. In terms of data processing, polar coordinate transformation is introduced to expand the circular mortar image into a rectangular panoramic view, significantly improving edge recognition accuracy. The system constructs a multimodal fusion network: static texture features are extracted using an improved ResNet with embedded CBAM attention modules; dynamic rheological features of the slump process are captured using a CNN-LSTM network; and environmental parameters are mapped into high-dimensional vectors using an MLP. Finally, the three features are fused to invert the flow diameter, identify abnormal states (such as bleeding and segregation), and output water replenishment suggestions for the production line. This invention effectively solves the detection challenges of single-component alkali-activated materials with high environmental sensitivity and complex microstructures, achieving high-precision non-contact intelligent detection. Attached Figure Description

[0021] Figure 1 A flowchart of an intelligent inversion method for the rheological properties of a single-component alkali-activated mortar provided in one embodiment of the present invention; Figure 2 A schematic diagram of the hardware composition of a smart inversion method for the rheological properties of a single-component alkali-activated mortar provided in an embodiment of the present invention; Figure 3 This is a diagram of the overall architecture of a multimodal fusion network in a smart inversion method for the rheological properties of a single-component alkali-activated mortar, provided in an embodiment of the present invention. Detailed Implementation

[0022] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings. Similar elements in different embodiments are referred to by associated similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of the invention. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to the present invention are not shown or described in the specification. This is to avoid obscuring the core parts of the invention with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0023] Furthermore, the features, operations, or characteristics described in the specification can be combined in any suitable manner to form various embodiments. At the same time, the steps or actions in the method description can be rearranged or adjusted in a manner obvious to those skilled in the art. Therefore, the various orders in the specification and drawings are only for the clear description of a particular embodiment and do not imply a necessary order, unless otherwise stated that a particular order must be followed.

[0024] The first embodiment of this invention provides an intelligent inversion method for the rheological properties of single-component alkali-activated mortar. This method utilizes computer vision combined with multi-sensor data fusion technology to perform non-contact, high-precision intelligent inversion and closed-loop control of the flow properties and rheological dynamics characteristics of single-component alkali-activated mortar. The following is a detailed description... Figure 1 Please provide a detailed explanation.

[0025] like Figure 1 As shown, in step S100, under the conditions of a ring-shaped shadowless light source and polarization filter, dynamic image sequences of the collapse process of single-component alkali-activated mortar and the temperature and humidity data of the current environment are simultaneously acquired.

[0026] The above steps specifically include: In this embodiment, not only are high-definition images of the mortar spreading captured, but also a video stream of the spreading process, as well as the current ambient temperature and relative humidity, are simultaneously captured. Figure 2 As shown, a multi-dimensional sensing terminal is constructed for this purpose: 1) Visual acquisition: A global shutter industrial camera with a frame rate set to 30fps was used to capture the dynamic process of the first 3 seconds of the collapse.

[0027] 2) Lighting Environment: Retain the high CRI LED ring shadowless light source design. Specific Improvements: The light source color temperature is set to 5000K cool white light, and a polarizing filter (CPL) is added in front of the lens to eliminate excessive specular reflection of the liquid film on the mortar surface and retain true texture details.

[0028] 3) Environmental perception: A DHT22 high-precision temperature and humidity sensor is integrated on the side of the testing station, and the data sampling frequency is synchronized with the image acquisition.

[0029] 4) Computing unit: Based on the NVIDIA Jetson Orin module, deploying the PyTorch inference engine.

[0030] like Figure 1 As shown, in step S200, the acquired image is transformed by polar coordinates to map the image containing the circular mortar extension surface into a rectangular panoramic unfolded image.

[0031] In this embodiment, a polar coordinate transformation is introduced to unfold the circular mortar image into a rectangular panoramic image. This transformation converts "edge roundness detection" into "linear boundary smoothness detection," significantly reducing the learning difficulty of the convolutional neural network and improving the edge recognition accuracy.

[0032] The above steps specifically include: S210, ROI extraction: Automatically crop out areas containing mortar using background subtraction method.

[0033] S220, Polar coordinate transformation (key innovation): Set the image center as the pole O(x0, y0), and map the image in the Cartesian coordinate system (x, y) to the polar coordinate system (r, θ).

[0034]

[0035] After transformation, the irregular circular mortar extension surface becomes a rectangular band image. Minor defects at the mortar edge (such as gaps and protrusions) are magnified into obvious peaks and troughs in the rectangular image, making it easier for the convolution kernel to extract features.

[0036] In this embodiment, the center point of the polar coordinate transformation is automatically obtained using the centroid method; the transformed rectangular image is used as the input of the convolutional neural network, giving the network natural robustness to the rotational invariance of the mortar expansion shape.

[0037] like Figure 1As shown, in step S300, a spatiotemporal dual-stream neural network for image feature extraction is constructed and trained. The spatial flow branch uses a deep convolutional network with embedded CBAM attention modules to extract static texture and edge defect features, while the temporal flow branch uses a temporal recurrent neural network to analyze the optical flow changes of the image sequence and extract dynamic features characterizing viscosity.

[0038] In this embodiment, the spatio-temporal dual-stream network architecture is as follows: 1) Spatial Stream: Analyzes single-frame images. A Convolutional Block Attention Module (CBAM) is embedded in the ResNet-50 backbone network. The CBAM attention module consists of cascaded channel attention and spatial attention sub-modules. The spatial attention sub-module uses max pooling and average pooling operations to generate a spatial weight map, automatically weighting high-frequency texture regions at the mortar edges to suppress background noise. This module includes channel attention and spatial attention, automatically suppressing background noise and focusing on the minute jagged edges of the mortar (representing segregation) and the glossy texture of the surface (representing the water-cement ratio).

[0039] 2) Temporal Stream: Utilizes a CNN-LSTM structure to process video frame sequences, extracts acceleration features of mortar edge expansion, and thus inverts the viscosity characteristics of the material.

[0040] like Figure 1 As shown, in step S400, the spatiotemporal visual feature vector extracted by the spatiotemporal dual-flow neural network is spliced ​​and fused with the encoded temperature and humidity feature vector, and the flowability prediction value and rheological state evaluation are output through the fully connected layer regression.

[0041] In this embodiment, physical parameter fusion and inversion are performed: before the fully connected layer, the extracted "visual feature vector" and the encoded "temperature and humidity environment vector" are concatenated, and the flowability prediction value and suggested water replenishment amount are output simultaneously through multi-task learning.

[0042] In summary, this invention constructs a multi-modal fusion network, such as... Figure 3 As shown: 1) Branch 1: Static texture feature extraction (based on improved ResNet) Input: The final image after polar coordinate transformation. Backbone: ResNet-50. Improvement: Insert a CBAM attention module after each residual block.

[0043] Channel attention: Determine which feature maps are more important to fluidity (e.g., texture feature map weight > color feature map weight).

[0044] Spatial attention: Focus on high-frequency areas in the image (i.e., mortar edges), ignoring flat areas in the center.

[0045] 2) Branch Two: Dynamic Rheological Feature Extraction (Based on LSTM) Input: A sequence of 10 differential images of the first 10 frames of the collapse process. Network: Geometric features of each frame are extracted using a lightweight CNN and input into a Long Short-Term Memory (LSTM) network. LSTM can remember the changes in the time series, thereby capturing the "expansion acceleration" of the mortar, a physical quantity that is strongly correlated with the yield stress of the mortar.

[0046] 3) Branch Three: Environmental Parameter Coding Input: Temperature (T) and humidity (H). Network: A 3-layer multilayer perceptron (MLP) maps scalar data into 128-dimensional high-dimensional feature vectors.

[0047] 4) Feature fusion and output: The feature vectors of the three branches are concatenated along the channel dimension and input into the fully connected layer regressor. Output 1: Flowability diameter prediction (mm). Output 2: Confidence of abnormal states (e.g., excessive viscosity, segregation, bleeding).

[0048] 5) In this embodiment, the model also includes a reverse compensation calculation module; based on the predicted flowability deviation and the current ambient temperature and humidity, the recommended water replenishment amount (ml) in the single-component mortar production process is output through the trained nonlinear regression model, that is, output 3: recommended water replenishment value for the production line (ml / kg).

[0049] Flowability deviation refers to the difference between the preset standard target flowability value of mortar and the flowability prediction value output by the fully connected layer regressor. In industrial production, qualified mortar products have a preset standard value. The difference between the standard value and the prediction value (e.g., 20mm) represents "how much drier the current mortar is than the ideal state". The system needs to know this gap in order to determine how much water to add in the next step.

[0050] The nonlinear regression model takes the fluidity deviation and the current ambient temperature and humidity as inputs, and the amount of water replenishment required to eliminate the corresponding fluidity deviation as output. It is trained using a large amount of experimental data.

[0051] On the same dataset, the results of comparing the present invention with the original scheme (ordinary ResNet) and the traditional SVM method are as follows: (1) Anti-interference: Under the condition of ambient temperature change (10℃ to 35℃), after the introduction of temperature and humidity compensation, the prediction error (MAE) of this system is stable within ±1.5mm, while the error of ordinary ResNet without the introduction of environmental parameters expands to ±5.2mm.

[0052] (2) Defect identification: Thanks to the CBAM attention mechanism, the accuracy of this system in identifying the "edge oozing" phenomenon reaches 98%, which is much higher than the 75% of ordinary deep learning models.

[0053] Corresponding to the above-disclosed intelligent inversion method for the rheological properties of single-component alkali-activated mortar, this invention also discloses an intelligent inversion system for the rheological properties of single-component alkali-activated mortar, which specifically includes: The multimodal data acquisition module is used to simultaneously acquire dynamic image sequences of the collapse process of single-component alkali-activated mortar and the temperature and humidity data of the current environment under the conditions of a ring shadowless light source and polarization filter. The geometric space transformation module is used to perform polar coordinate transformation on the acquired images, mapping the image containing the circular mortar extension surface into a rectangular panoramic unfolded image. The spatiotemporal feature extraction module is used to construct and train a spatiotemporal dual-stream neural network for image feature extraction. The spatial flow branch uses a deep convolutional network with an embedded CBAM attention module to extract static texture and edge defect features, while the temporal flow branch uses a recurrent neural network to analyze the optical flow changes of the image sequence and extract dynamic features characterizing viscosity. Multimodal fusion inversion is used to concatenate and fuse the spatiotemporal visual feature vectors extracted by the spatiotemporal dual-stream neural network with the encoded temperature and humidity feature vectors, and output the predicted mobility value and rheological state evaluation through fully connected layer regression.

[0054] It should be noted that for a detailed description of the intelligent inversion system for the rheological properties of a single-component alkali-activated mortar provided in the embodiments of the present invention, please refer to the relevant description of the intelligent inversion method for the rheological properties of a single-component alkali-activated mortar provided in the embodiments of the present invention, which will not be repeated here.

[0055] In addition, embodiments of the present invention also provide an electronic device, the device comprising: a processor and a memory; the memory being used to store one or more program instructions; the processor being used to execute one or more program instructions to perform the steps of the intelligent inversion method for the rheological properties of a single-component alkali-activated mortar as described in any of the preceding embodiments.

[0056] It should be noted that for a detailed description of the electronic device provided in the embodiments of the present invention, please refer to the relevant description of the intelligent inversion method for the rheological properties of a single-component alkali-activated mortar provided in the embodiments of this application, which will not be repeated here.

[0057] In addition, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the intelligent inversion method for the rheological properties of a single-component alkali-activated mortar as described in any of the preceding claims.

[0058] It should be noted that for a detailed description of the computer-readable storage medium provided in the embodiments of the present invention, please refer to the relevant description of the intelligent inversion method for the rheological properties of a single-component alkali-activated mortar provided in the embodiments of this application, which will not be repeated here.

[0059] Those skilled in the art will understand that all or part of the functions of the various methods in the above embodiments can be implemented by hardware or by computer programs. When all or part of the functions in the above embodiments are implemented by computer programs, the program can be stored in a computer-readable storage medium, which may include: read-only memory, random access memory, disk, optical disk, hard disk, etc., and the program is executed by a computer to achieve the above functions. For example, the program can be stored in the memory of a device, and when the program in the memory is executed by the processor, all or part of the above functions can be achieved. In addition, when all or part of the functions in the above embodiments are implemented by computer programs, the program can also be stored in a server, another computer, disk, optical disk, flash drive, or external hard drive, etc., and can be downloaded or copied to the memory of a local device, or the system of the local device can be updated. When the program in the memory is executed by the processor, all or part of the functions in the above embodiments can be achieved.

[0060] The above examples illustrate the present invention only to aid in understanding it and are not intended to limit the scope of the invention. Those skilled in the art can make various simple deductions, modifications, or substitutions based on the principles of this invention.

Claims

1. A smart inversion method for the rheological properties of single-component alkali-activated mortar, characterized in that, The method includes: Under the conditions of a ring-shaped shadowless light source and polarization filter, dynamic image sequences of the collapse process of single-component alkali-activated mortar and the current temperature and humidity data of the environment were simultaneously acquired. The acquired images are transformed using polar coordinates, mapping the image containing the circular mortar extension surface into a rectangular panoramic unfolded image; A spatiotemporal dual-stream neural network for image feature extraction was constructed and trained. The spatial flow branch uses a deep convolutional network with embedded CBAM attention modules to extract static texture and edge defect features, while the temporal flow branch uses a temporal recurrent neural network to analyze the optical flow changes of the image sequence and extract dynamic features characterizing viscosity. The spatiotemporal visual feature vector extracted by the spatiotemporal dual-flow neural network is concatenated and fused with the encoded temperature and humidity feature vector, and the flowability prediction value and rheological state evaluation are output through fully connected layer regression.

2. The intelligent inversion method for rheological properties of single-component alkali-activated mortar as described in claim 1, characterized in that, Under an annular shadowless light source and polarization filter environment, dynamic image sequences of the slump process of single-component alkali-activated mortar and the current ambient temperature and humidity data were simultaneously acquired, specifically including: A global shutter industrial camera was used to capture a dynamic image sequence of the first few seconds of the collapse process of single-component alkali-activated mortar. Temperature and humidity data were sampled synchronously with the image acquisition using a high-precision sensor.

3. The intelligent inversion method for the rheological properties of single-component alkali-activated mortar as described in claim 1, characterized in that, The acquired images are subjected to polar coordinate transformation to map the image containing the circular mortar extension surface into a rectangular panoramic unfolded image, specifically including: ROI extraction: Automatically crop out areas containing mortar using background subtraction method; Polar coordinate transformation: Setting the image center as the pole O(x0, y0), the image in Cartesian coordinates (x, y) is mapped to polar coordinates (r, θ), as shown in the following formula: After polar coordinate transformation, the irregular circular mortar extension surface becomes a rectangular band. The center point of the polar coordinate transformation is automatically obtained using the centroid method.

4. The intelligent inversion method for rheological properties of single-component alkali-activated mortar as described in claim 1, characterized in that, The CBAM attention module includes a channel attention submodule and a spatial attention submodule connected in series. The spatial attention submodule uses max pooling and average pooling operations to generate a spatial weight map, automatically weighting the high-frequency texture region at the edge of the mortar to suppress background noise.

5. The intelligent inversion method for rheological properties of single-component alkali-activated mortar as described in claim 1, characterized in that, The input to the spatial flow branch is a rectangular panoramic unfolded image obtained after polar coordinate transformation, and the input to the temporal flow branch is a sequence of differential images of the first multiple frames of the single-component alkali-activated mortar collapse process.

6. The intelligent inversion method for rheological properties of single-component alkali-activated mortar as described in claim 5, characterized in that, The spatiotemporal visual feature vector extracted by the spatiotemporal dual-flow neural network is concatenated and fused with the encoded temperature and humidity feature vector. The resulting flowability prediction and rheological state evaluation are then output via a fully connected layer regression. Specifically, this includes: Temperature and humidity data are mapped from scalar data to high-dimensional feature vectors using a multilayer perceptron. These feature vectors are then concatenated with the two feature vectors output from the spatial flow branch and the temporal flow branch in the channel dimension. Finally, a fully connected layer regressor outputs the predicted mobility value and the confidence level of the abnormal state.

7. The intelligent inversion method for rheological properties of single-component alkali-activated mortar as described in claim 6, characterized in that, The method further includes: Reverse compensation calculation: Based on the predicted fluidity deviation and the current ambient temperature and humidity, the recommended water replenishment amount in the production process of single-component mortar is output through a trained nonlinear regression model; the fluidity deviation refers to the difference between the preset standard target fluidity value of mortar and the fluidity prediction value output by the fully connected layer regressor; the nonlinear regression model takes the fluidity deviation and the current ambient temperature and humidity as inputs, and the water replenishment amount required to eliminate the corresponding fluidity deviation as output.

8. The intelligent inversion method for rheological properties of single-component alkali-activated mortar as described in claim 1, characterized in that, The spatial flow branch uses ResNet-50 as the backbone network, and inserts a CBAM attention module after each residual block.

9. The intelligent inversion method for rheological properties of single-component alkali-activated mortar as described in claim 1, characterized in that, The time-flow branch uses a CNN-LSTM network to extract mortar expansion acceleration features to characterize viscosity properties by processing the difference features of the image sequence.

10. A smart inversion system for the rheological properties of a single-component alkali-activated mortar, characterized in that, The system includes: The multimodal data acquisition module is used to simultaneously acquire dynamic image sequences of the collapse process of single-component alkali-activated mortar and the temperature and humidity data of the current environment under the conditions of a ring shadowless light source and polarization filter. The geometric space transformation module is used to perform polar coordinate transformation on the acquired images, mapping the image containing the circular mortar extension surface into a rectangular panoramic unfolded image. The spatiotemporal feature extraction module is used to construct and train a spatiotemporal dual-stream neural network for image feature extraction. The spatial flow branch uses a deep convolutional network with an embedded CBAM attention module to extract static texture and edge defect features, while the temporal flow branch uses a recurrent neural network to analyze the optical flow changes of the image sequence and extract dynamic features characterizing viscosity. Multimodal fusion inversion is used to concatenate and fuse the spatiotemporal visual feature vectors extracted by the spatiotemporal dual-stream neural network with the encoded temperature and humidity feature vectors, and output the predicted mobility value and rheological state evaluation through fully connected layer regression.

Images