CN-122016556-A - Intelligent inversion method and system for rheological characteristics of single-component alkali-activated mortar

CN122016556ACN 122016556 ACN122016556 ACN 122016556ACN-122016556-A

Abstract

The invention discloses a single-component alkali-activated mortar rheological property intelligent inversion method and system, which utilize a global shutter camera and a temperature and humidity sensor to synchronously acquire a dynamic image sequence and environmental temperature and humidity data of a mortar slump process, introduce polar coordinate transformation in data processing to expand a circular mortar expansion image into a rectangular panorama, remarkably improve edge recognition precision, construct a multi-mode fusion network, extract static texture features through improvement ResNet of an embedding CBAM attention module, capture dynamic rheological features of the slump process by utilizing a CNN-LSTM network, map environmental parameters into high-dimensional vectors by utilizing MLP, finally fuse the three features to invert fluidity diameter, recognize abnormal states (such as bleeding and segregation) and output a water supplementing suggestion of a production line. The invention effectively solves the detection problems of strong environmental sensitivity and complex micro texture of the single-component alkali-activated material, and realizes high-precision non-contact intelligent detection.

Inventors

MA SHUQI
TIAN SONG
BAI JIANBIAO
ZHANG WEIGUANG
LI YANHUI
WANG GUANGHUI
CHEN JIAZHENG

Assignees

安徽理工大学

Dates

Publication Date: 20260512
Application Date: 20251229

Claims (10)

1. An intelligent inversion method for rheological properties of single-component alkali-activated mortar is characterized by comprising the following steps: synchronously collecting a dynamic image sequence of a single-component alkali-activated mortar slump process and temperature and humidity data of a current environment under an annular shadowless light source and a polarized filtering environment; performing polar coordinate transformation on the acquired image, and mapping the image containing the circular mortar expansion surface into a rectangular panoramic expansion map; Constructing and training a space-time double-flow neural network for image feature extraction, wherein a space flow branch adopts a deep convolution network embedded with CBAM attention modules to extract static textures and edge defect features, and a time flow branch adopts a time circulation neural network to analyze optical flow changes of an image sequence and extract dynamic features representing viscosity; And splicing and fusing the space-time visual feature vector extracted by the space-time double-flow neural network and the encoded temperature and humidity feature vector, and outputting a fluidity predicted value and rheological state evaluation through full-connection layer regression.
2. The intelligent inversion method for rheological properties of single-component alkali-activated mortar according to claim 1, wherein dynamic image sequences of the single-component alkali-activated mortar slump process and temperature and humidity data of the current environment are synchronously acquired under the environment of an annular shadowless light source and polarization filtration, and the method specifically comprises the following steps: and acquiring a dynamic image sequence of the initial few seconds of the single-component alkali-activated mortar slump process by using a global shutter industrial camera, and synchronously sampling temperature and humidity data through a high-precision sensor and image acquisition.
3. The intelligent inversion method for rheological properties of single-component alkali-activated mortar according to claim 1, wherein the method for intelligently inverting rheological properties of single-component alkali-activated mortar is characterized by performing polar coordinate transformation on the acquired image and mapping the image containing a circular mortar expansion surface into a rectangular panoramic expansion map, and specifically comprises the following steps: Extracting ROI, namely automatically cutting out a region containing mortar by using a background difference method; Polar coordinate transformation, namely setting the center of an image as a pole O (x 0 , y 0 ), and mapping the image in a Cartesian coordinate system (x, y) to a polar coordinate system (r, θ), wherein the formula is as follows: After polar coordinate transformation, the irregular circular mortar expansion surface is changed into a rectangular strip chart; The center point location of the polar coordinate transformation is automatically obtained by adopting a centroid method.
4. The intelligent inversion method of rheological properties of single-component alkali-activated mortar according to claim 1, wherein the CBAM attention module comprises a channel attention sub-module and a space attention sub-module which are connected in series, and the space attention sub-module generates a space weight map by using maximum pooling and average pooling operation, automatically weights a high-frequency texture area of the edge of the mortar, and suppresses background noise.
5. The intelligent inversion method for rheological properties of single-component alkali-activated mortar according to claim 1, wherein the input of the spatial flow branch is a rectangular panorama expansion diagram obtained after polar coordinate transformation, and the input of the temporal flow branch is a multi-frame differential image sequence before the single-component alkali-activated mortar slump process.
6. The intelligent inversion method for rheological properties of single-component alkali-activated mortar according to claim 5, wherein the method comprises the steps of splicing and fusing the space-time visual feature vector extracted by the space-time double-flow neural network with the encoded temperature and humidity feature vector, and outputting a fluidity predicted value and rheological state evaluation by full-connection layer regression, and specifically comprises the following steps: and mapping scalar data into high-dimensional feature vectors by the temperature and humidity data through a multi-layer perceptron, splicing the high-dimensional feature vectors with two feature vectors output by a space flow branch and a time flow branch in a channel dimension, and outputting a fluidity predicted value and an abnormal state confidence through a full-connection-layer regressor.
7. The intelligent inversion method for rheological properties of single-component alkali-activated mortar according to claim 6, further comprising: The reverse compensation calculation is carried out, namely, based on the predicted fluidity deviation and the current environment temperature and humidity, the recommended water supplementing amount in the single-component mortar production process is output through a trained nonlinear regression model, wherein the fluidity deviation is the difference value between a preset standard target fluidity value of mortar and the fluidity predicted value output by the full-connection layer regressor, and the nonlinear regression model takes the fluidity deviation and the current environment temperature and humidity as input and takes the water supplementing amount required by eliminating the corresponding fluidity deviation as output.
8. The intelligent inversion method for rheological properties of single-component alkali-activated mortar according to claim 1, wherein the spatial stream branch adopts ResNet-50 as a backbone network, and a CBAM attention module is inserted after each residual block.
9. The intelligent inversion method for rheological properties of single-component alkali-activated mortar according to claim 1, wherein the time flow branch adopts a CNN-LSTM network, and the mortar expansion acceleration characteristic is extracted by processing the differential characteristic of an image sequence to characterize the viscosity property.
10. An intelligent inversion system for rheological properties of single-component alkali-activated mortar, which is characterized by comprising: the multi-mode data acquisition module is used for synchronously acquiring a dynamic image sequence of the single-component alkali-excited mortar slump process and temperature and humidity data of the current environment under the annular shadowless light source and the polarized filtering environment; The geometric space transformation module is used for carrying out polar coordinate transformation on the acquired image and mapping the image containing the circular mortar expansion surface into a rectangular panoramic expansion map; The space-time feature extraction module is used for constructing and training a space-time double-flow neural network for extracting image features, wherein a space flow branch adopts a deep convolution network embedded with a CBAM attention module to extract static textures and edge defect features, and a time flow branch adopts a cyclic neural network to analyze optical flow changes of an image sequence and extract dynamic features representing viscosity; The multi-mode fusion inversion is used for splicing and fusing the space-time visual feature vector extracted by the space-time double-flow neural network and the encoded temperature and humidity feature vector, and outputting a fluidity predicted value and rheological state evaluation through full-connection layer regression.

Description

Intelligent inversion method and system for rheological characteristics of single-component alkali-activated mortar Technical Field The invention relates to the technical field of single-component alkali-activated mortar rheological property detection, in particular to an intelligent inversion method and system for single-component alkali-activated mortar rheological property. Background The single-component alkali-activated material has the potential of replacing the traditional cement due to the low-carbon characteristic and the convenience of water adding. However, this material system has complex rheological properties: (1) The environment sensitivity is strong, the alkali excitation reaction is exothermic, and the reaction rate is greatly influenced by the environment temperature and humidity. The internal viscosity change is not completely reflected by the image appearance alone. (2) The microscopic features are complex, namely that if the solid excitant is insufficiently dissolved, microscopic bleeding or aggregate segregation phenomenon can be caused, and the fine texture differences are difficult to capture by traditional machine vision (such as SVM) or basic deep learning (such as common ResNet). (3) Static limitations existing image detection techniques only measure the final spread diameter (static value), ignoring the "spread rate" of the mortar from cone slump to stop flow. The spreading rate is directly related to the yield stress (YIELD STRESS) and plastic viscosity (Plastic Viscosity) of the material and is a more critical index for judging the workability. Therefore, the prior art lacks an intelligent detection means capable of combining the time sequence dynamic characteristics and the environment physical parameters. Disclosure of Invention The invention provides an intelligent inversion method for rheological properties of single-component alkali-activated mortar, which aims to solve the problem that the prior art lacks an intelligent detection means for rheological properties of single-component alkali-activated mortar which can combine time sequence dynamic characteristics and environmental physical parameters. According to a first aspect, in one embodiment, there is provided a method for intelligent inversion of rheological properties of a one-component alkali-activated mortar, the method comprising: synchronously collecting a dynamic image sequence of a single-component alkali-activated mortar slump process and temperature and humidity data of a current environment under an annular shadowless light source and a polarized filtering environment; performing polar coordinate transformation on the acquired image, and mapping the image containing the circular mortar expansion surface into a rectangular panoramic expansion map; Constructing and training a space-time double-flow neural network for image feature extraction, wherein a space flow branch adopts a deep convolution network embedded with CBAM attention modules to extract static textures and edge defect features, and a time flow branch adopts a time circulation neural network to analyze optical flow changes of an image sequence and extract dynamic features representing viscosity; And splicing and fusing the space-time visual feature vector extracted by the space-time double-flow neural network and the encoded temperature and humidity feature vector, and outputting a fluidity predicted value and rheological state evaluation through full-connection layer regression. Further, under the environment of an annular shadowless light source and polarization filtration, synchronously acquiring a dynamic image sequence of a single-component alkali-excited mortar slump process and temperature and humidity data of the current environment, and specifically comprising the following steps: and acquiring a dynamic image sequence of the initial few seconds of the single-component alkali-activated mortar slump process by using a global shutter industrial camera, and synchronously sampling temperature and humidity data through a high-precision sensor and image acquisition. Further, performing polar coordinate transformation on the acquired image, and mapping the image containing the circular mortar expansion surface into a rectangular panorama expansion diagram, specifically comprising: Extracting ROI, namely automatically cutting out a region containing mortar by using a background difference method; Polar coordinate transformation, namely setting the center of an image as a pole O (x 0, y0), and mapping the image in a Cartesian coordinate system (x, y) to a polar coordinate system (r, θ), wherein the formula is as follows: After polar coordinate transformation, the irregular circular mortar expansion surface is changed into a rectangular strip chart; The center point location of the polar coordinate transformation is automatically obtained by adopting a centroid method. Further, the CBAM attention module comprises a channel attention sub-module and