Method for generating mechanical cloud picture of AI-assisted mechanoluminescent material

Abstract

The invention relates to the field of intelligent sensing and computer vision, in particular to a mechanical cloud picture generation method of an AI auxiliary force-induced luminescent material, and aims to provide the mechanical cloud picture generation method of the AI auxiliary force-induced luminescent material. The method mainly solves the problem that the afterglow effect in the mechanoluminescence image limits the accurate description of the mechanical state of the component. The AI-assisted high-precision mechanical cloud picture generation model provided by the invention has the beneficial effects that the mechanical state of a part can be described more accurately by utilizing mechanoluminescence sensing data, and the average RMSE error is as low as 3.8368%.

Description

Technical Field

[0001] The present invention relates to the field of intelligent sensing and computer vision technology, and specifically to an AI-assisted method for generating a mechanical cloud map of a mechanoluminescent material. Background Art

[0002] As a special piezoelectric luminescent material, mechanoluminescent materials exhibit the characteristics of luminescence under stress through their piezoelectric effect or triboelectric effect, and are therefore widely used in the field of intelligent sensing. This type of material is particularly suitable for monitoring the mechanical behavior of structural components and detecting environmental changes, and is suitable for the development of various stress and strain sensor devices. However, a major problem with mechanoluminescent materials is the afterglow effect of luminescence, which makes it difficult for sensor devices to accurately describe the mechanical conditions of components under complex load conditions, thereby limiting their wide application in practical engineering.

[0003] With the advancement of deep learning and computer vision technology, AI (Artificial Intelligence)-assisted image generation methods have been widely used in many aspects such as image generation, image style transfer, and image restoration and reconstruction. Such methods have shown significant advantages in image analysis and error correction, and can solve the challenges faced by mechanoluminescence sensing technology, thereby improving the performance and application scope of sensor devices. Due to the afterglow effect in mechanoluminescence images, the accurate description of the mechanical state of components is limited, and it is urgent to develop an AI-assisted method to convert experimental luminescence images into accurate mechanical cloud maps. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides an AI-assisted method for generating mechanical cloud images of mechanoluminescent materials, which mainly solves the problem that the luminous afterglow in the mechanoluminescent image affects the accurate description of the mechanical state of the components.

[0005] In order to achieve the above purpose, the technical solutions provided are as follows:

[0006] An AI-assisted method for generating a mechanical cloud map of a mechanoluminescent material, characterized by comprising the following steps:

[0007] Step 1: Make a mechanoluminescence sensing component and collect a mechanoluminescence video of the entire loading process of the component;

[0008] Step 2: Use the finite element analysis method to perform finite element simulation on the entire process of component loading, and export the mechanical cloud map video of the entire process of component loading;

[0009] Step 3: extract frames from the luminous video obtained in step 1 and the mechanical cloud image video obtained in step 2 according to the time step increment; and perform preprocessing such as image cropping, scaling, and spatiotemporal registration on the obtained frames;

[0010] Step 4: Use the pre-processed luminescent images and cloud images obtained in step 3 to establish a proprietary image dataset of the mechanical state of components based on the mechanoluminescent device; and divide the dataset into a training set, a validation set, and a test set according to a certain ratio;

[0011] Step 5: Establish a network architecture of the model, wherein the network architecture is based on the VGG16-UNet image generation model;

[0012] Step 6: Set the loss function, optimizer, and training hyperparameters for model training;

[0013] Step 7: Input the training set images into the model for iterative training; after each round of training, use the validation set to evaluate the model performance; when the training and validation losses converge consistently, the final weight model is obtained;

[0014] Step 8: The network loads the final weight model; the luminous image in the test set is input into the model to generate the corresponding mechanical cloud map; and the error analysis of the generated cloud map is further performed.

[0015] Furthermore, in step one, the manufacturing of the mechanoluminescent sensing component includes fixing the sensor on the surface of the component to be tested by pasting and nesting technology, requiring good mechanical and optical contact between the sensor and the surface of the component to ensure accurate transmission and detection of the mechanoluminescent signal; the mechanoluminescent data acquisition includes capturing the mechanoluminescent signal emitted by the sensor by a high-speed camera, and the entire process from loading to unloading of the component should be recorded by video, and the collected data includes luminescence intensity, duration and spectral characteristics.

[0016] Furthermore, in step two, the finite element simulation of the entire component loading process includes: establishment of a finite element model of the component, setting of boundary conditions and loading conditions, meshing and finite element analysis, and detailed analysis of the simulation results to verify the consistency of the data displayed in the cloud map with the experimental and theoretical analysis. If necessary, the model parameters are adjusted according to the analysis results and the simulation is repeated to improve the accuracy and reliability of the simulation. The cloud map video export requires the use of a finite element visualization tool to export the dynamic cloud map into a video format.

[0017] Furthermore, in step three, the video frame extraction includes frame extraction of mechanoluminescence video and stress cloud map video, and the two types of video frames are extracted synchronously according to the set time step increment to ensure accurate time correspondence between the two for subsequent comparison and analysis; the image preprocessing includes: image cropping, image scaling and spatiotemporal alignment to ensure that the processed mechanoluminescence image and stress cloud map have uniform resolution and spatiotemporal alignment accuracy.

[0018] Furthermore, in step four, the image data set is established by integrating the luminescent image and mechanical cloud map preprocessed in step three, and a pair of images at each time step is used as a data sample to ensure that each luminescent image and the corresponding mechanical cloud map correctly correspond in the time series; then the data set is divided into a training set, a validation set, and a test set according to a ratio of 7:2:1.

[0019] Furthermore, in step five, the VGG16-UNet image generation model consists of a backbone feature extraction part, a feature fusion part and a feature decoding part.

[0020] Furthermore, the backbone feature extraction part adopts VGG16; the feature decoding part adopts the UNet decoder part; and the feature fusion part directly connects the feature maps of different layers of VGG16 to the corresponding upsampling layer of UNet.

[0021] Furthermore, in step six, the model training loss function adopts Huber Loss and SSIM Loss; the optimizer is: Adam (β1=0.90; β2=0.999); the training hyperparameters include: total number of training rounds; training batch parameters; regularization coefficient; initial learning rate; learning rate reduction strategy; learning rate reduction factor; number of learning rate step-by-step attenuation rounds; verification frequency; weight attenuation coefficient; gradient threshold.

[0022] Furthermore, in step seven, the consistent convergence of the training and validation losses indicates that the training loss and the validation loss have a consistent downward trend with the training process and approach a stable minimum value.

[0023] Furthermore, in step eight, the cloud map generation error indicator is RMSE.

[0024] The beneficial effect of the present invention is that the AI-assisted high-precision mechanical cloud map generation model provided by the present invention can utilize mechanoluminescence sensing data to achieve a more accurate description of the mechanical state of components, with an average RMSE error as low as 3.8368%. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is a flow chart of the present invention;

[0026] Figure 2 Four types of components of the sensor film arranged in the embodiment of the present invention;

[0027] Figure 3 are finite element models of four types of components in the embodiment of the present invention;

[0028] Figure 4 The preprocessed luminescence image and the corresponding stress cloud map in the embodiment of the present invention;

[0029] Figure 5 are the training, verification and test sets in the embodiments of the present invention;

[0030] Figure 6 It is a diagram of the VGG16-UNet network model architecture in an embodiment of the present invention;

[0031] Figure 7 is a graph showing the relationship between the training loss and the number of training rounds in an embodiment of the present invention;

[0032] Figure 8 5 luminescence images and the generated stress cloud map in an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0034] An AI-assisted method for generating mechanical cloud images of mechanoluminescent materials, such as Figure 1 The process diagram shown in FIG. 1 includes the following specific steps:

[0035] Step 1: For the arrangement of sensor components, a thin film pasting method is used to obtain components with sensing functions, including a cylindrical epoxy resin component, a defect-free stainless steel component, a defective stainless steel component, and a stainless steel beam, respectively. Figure 2 (a), (b), (c) and (d).

[0036] Step 2: For the finite element simulation of the whole process of loading of the four types of components mentioned in step 1, the finite element analysis software Abaqus is used to conduct simulation experiments and conclusion analysis. The constructed cylindrical epoxy resin component uniaxial compression model, defect-free stainless steel component uniaxial tension model, defective stainless steel component uniaxial tension model and stainless steel beam three-point bending model are as follows: Figure 3 (a), (b), (c) and (d).

[0037] Step 3: For the frame extraction and image preprocessing of the luminescence video and cloud image video obtained in step 1 and step 2, relevant codes are written using Python packages such as OpenCV-Python, Pillow, and os to implement data preprocessing. The luminescence images of each experimental condition obtained after processing are as follows: Figure 4 (a1), (b1), (c1) and (d1), the resulting stress cloud diagrams are shown as follows: Figure 4As shown in (a2), (b2), (c2) and (d2).

[0038] Step 4: Use the pre-processed luminescence images and stress cloud images obtained in step 3 to establish a proprietary image dataset, and divide the dataset into training set, validation set and test set according to a certain ratio, as follows: Figure 5 (a), (b) and (c).

[0039] Step 5: For the construction of the model network architecture, a network architecture based on VGG16-UNet is used. Among them, VGG16 is the backbone feature extraction part, and the UNet decoder part is the feature decoding part. The specific model architecture is as follows Figure 6 shown.

[0040] Step 6: Adjust and set the model's training loss function, optimization algorithm, and training hyperparameters. The specific parameters are shown in Table 1.

[0041] Table 1 Model training parameter configuration

[0042]

[0043] Step 7: Use the training set to iteratively train the model, and use the validation set to validate the model according to the model validation frequency until the training loss and validation loss converge to obtain the final weight model; the changes of training and validation losses with the number of training rounds are as follows: Figure 7 shown.

[0044] Step 8. The corresponding stress cloud map can be generated by using the luminous image obtained from the experiment and VGG16-UNet loaded with the final weight model.

[0045] like Figure 8 As shown in the figure, corresponding stress cloud maps are generated for four randomly selected luminescence images. Figure 8 (a) is the luminescence image; Figure 8 (b) is the corresponding true stress contour; Figure 8 (c) is the generated stress contour.

[0046] It can be found that the generated results are extremely accurate, achieving high-precision stress cloud map generation based on mechanoluminescence sensing images.

[0047] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. An AI-assisted method for generating mechanical cloud images of mechanoluminescent materials, characterized in that: The steps include: Step 1: Make a mechanoluminescence sensing component and collect a mechanoluminescence video of the entire loading process of the component; Step 2: Use the finite element analysis method to perform finite element simulation on the entire process of component loading, and export the mechanical cloud map video of the entire process of component loading; Step 3: extract frames from the luminous video obtained in step 1 and the mechanical cloud image video obtained in step 2 according to the time step increment; and perform image cropping, scaling, and spatiotemporal registration preprocessing on the obtained frames; Step 4: Use the pre-processed luminescent images and cloud images obtained in step 3 to establish a proprietary image dataset of the mechanical state of components based on the mechanoluminescent device; and divide the dataset into a training set, a validation set, and a test set according to a certain ratio; Step 5: Establish a network architecture of the model, wherein the network architecture is based on the VGG16-UNet image generation model; Step 6: Set the loss function, optimizer, and training hyperparameters for model training; Step 7: Input the training set images into the model for iterative training; after each round of training, use the validation set to evaluate the model performance; when the training and validation losses converge consistently, the final weight model is obtained; Step 8: The network loads the final weight model; The luminous images in the test set are input into the model to generate the corresponding mechanical cloud map; and the error analysis of the generated cloud map is further performed.

2. The method for generating a mechanical cloud image of an AI-assisted mechanoluminescent material according to claim 1, characterized in that: In step one, the manufacturing of the mechanoluminescent sensing component includes fixing the sensor on the surface of the component to be tested by pasting and nesting technology, requiring good mechanical and optical contact between the sensor and the surface of the component to ensure accurate transmission and detection of the mechanoluminescent signal; the mechanoluminescent data acquisition includes capturing the mechanoluminescent signal emitted by the sensor by a high-speed camera, and the entire process from loading to unloading of the component should be recorded by video, and the collected data includes luminescence intensity, duration and spectral characteristics.

3. The method for generating a mechanical cloud image of an AI-assisted mechanoluminescent material according to claim 1, characterized in that: In step two, the finite element simulation of the entire component loading process includes: establishment of a finite element model of the component, setting of boundary conditions and loading conditions, meshing and finite element analysis, detailed analysis of the simulation results, verification of the consistency of the data displayed in the cloud map with the experimental and theoretical analysis, and, if necessary, adjustment of the model parameters based on the analysis results and repeated simulation to improve the accuracy and reliability of the simulation; the cloud map video export requires the use of a finite element visualization tool to export the dynamic cloud map into a video format.

4. The method for generating a mechanical cloud image of an AI-assisted mechanoluminescent material according to claim 1, characterized in that: In step three, the video frame extraction includes frame extraction of mechanoluminescence video and stress cloud map video, and the two types of video frames are extracted synchronously according to the set time step increment to ensure accurate time correspondence between the two for subsequent comparison and analysis; the image preprocessing includes: image cropping, image scaling and spatiotemporal alignment to ensure that the processed mechanoluminescence image and stress cloud map have uniform resolution and spatiotemporal alignment accuracy.

5. The method for generating a mechanical cloud image of an AI-assisted mechanoluminescent material according to claim 1, characterized in that: In step four, the image data set is established by integrating the luminescent image and mechanical cloud map preprocessed in step three, and a pair of images at each time step is used as a data sample to ensure that each luminescent image and the corresponding mechanical cloud map correctly correspond in the time series; then the data set is divided into a training set, a validation set, and a test set according to a ratio of 7:2:

1.

6. The method for generating a mechanical cloud image of an AI-assisted mechanoluminescent material according to claim 1, characterized in that: In step five, the VGG16-UNet image generation model consists of a backbone feature extraction part, a feature fusion part and a feature decoding part.

7. The method for generating a mechanical cloud image of an AI-assisted mechanoluminescent material according to claim 1, characterized in that: The backbone feature extraction part adopts VGG16; the feature decoding part adopts the UNet decoder part; the feature fusion part directly connects the feature maps of different layers of VGG16 to the corresponding upsampling layer of UNet.

8. The method for generating a mechanical cloud image of an AI-assisted mechanoluminescent material according to claim 1, characterized in that: In step 6, the model training loss function uses Huber Loss and SSIM Loss; the optimizer is: Adam (β1 = 0.90; β2=0.999); the training hyperparameters include: total number of training rounds; training batch parameter; regularization coefficient; initial learning rate; learning rate reduction strategy; learning rate reduction factor; number of learning rate step-by-step attenuation rounds; verification frequency; weight attenuation coefficient; gradient threshold.

9. The method for generating a mechanical cloud image of an AI-assisted mechanoluminescent material according to claim 1, characterized in that: In step seven, the consistent convergence of the training and validation losses indicates that the training loss and the validation loss have a consistent downward trend with the training process and approach a stable minimum value.

10. The method for generating a mechanical cloud image of an AI-assisted mechanoluminescent material according to claim 1, characterized in that: In step eight, the cloud map generation error indicator is RMSE.

Images