Real-time monitoring method for curing rate field in polymer composite material

Abstract

The invention provides a method for monitoring a curing rate field in a polymer composite material in real time, which comprises the following steps: obtaining an interference spectrum of a cross section of a material to be detected through an optical coherence tomography system, and obtaining a wrapping phase difference of the cross section; predicting a phase gradient according to the wrapped phase difference; according to the wrapping phase difference and the phase gradient, obtaining a curing rate field of the research polymer curing process; and real-time imaging is performed according to the curing rate field, and the imaging quality is autonomously evaluated, so that real-time monitoring of the curing rate field in the curing process is realized. According to the curing rate field real-time monitoring method disclosed by the invention, full-field measurement of the polymer curing rate can be completed, and the breakthrough of curing rate monitoring from point to field is realized.

Description

Technical Field

[0001] The present disclosure relates to the field of functional imaging technology, and in particular to a method for real-time monitoring of a curing rate field inside a polymer composite material. Background Art

[0002] Curing rate field monitoring is an important means to optimize polymer composites processing.

[0003] Traditional curing rate monitoring methods, such as dielectric curing monitoring methods and differential scanning calorimetry, can only monitor the curing rate of a certain point inside the polymer material. Therefore, the imaging of these methods does not have spatial distribution characteristics and it is difficult to fully reflect the curing rate field of polymer composite materials under anisotropic characteristics. At the same time, since it is difficult to obtain prior information such as material structure, external loading environment, and measurement interference, the imaging quality of traditional monitoring methods cannot be quantitatively evaluated, making the existing curing rate monitoring results insufficiently confident.

[0004] Therefore, it is of great significance to develop a real-time monitoring method that can monitor the internal curing rate field of polymer composites and perform autonomous quantitative evaluation of imaging quality. Summary of the invention

[0005] The purpose of the present disclosure is to provide a real-time monitoring method for the curing rate field inside a polymer composite material, so as to at least solve one technical problem in the prior art.

[0006] The technical solution disclosed in the present invention is:

[0007] A method for real-time monitoring of a curing rate field inside a polymer composite material, comprising:

[0008] Obtaining an interference spectrum of a cross section of the material to be tested by an optical coherence tomography (OCT) system, and obtaining a wrapping phase difference of the cross section;

[0009] predicting a phase gradient based on the wrapped phase difference;

[0010] Acquiring a curing rate field of the curing process of the material to be tested according to the wrapping phase difference and the phase gradient;

[0011] Real-time imaging is performed based on the curing rate field, and the imaging quality is independently evaluated, thereby achieving real-time monitoring of the curing rate field during the curing process.

[0012] The method of obtaining the interference spectrum of the cross section of the material to be tested by an optical coherence tomography system (OCT) and obtaining the wrapping phase difference of the cross section comprises:

[0013] Placing the material to be tested on the testing platform;

[0014] The interference spectrum of the cross section of the material to be tested is obtained by an optical coherence tomography (OCT) system, and a differential operation is performed on the phase-frequency characteristic diagram obtained by Fourier transform to obtain the wrapping phase difference of the cross section.

[0015] The predicting the phase gradient according to the wrapped phase difference comprises:

[0016] Set up the gradient prediction network;

[0017] Inputting the wrapped phase difference into the gradient prediction network and outputting the predicted gradient;

[0018] Comparing the predicted gradient with the true gradient, and optimizing the weights and biases of the gradient prediction network according to the loss function to obtain a trained gradient prediction network;

[0019] According to the gradient prediction network, the prediction from the wrapped phase difference Δφ(y, z, t) to the phase gradient Φ(y, z, t) is completed.

[0020] The gradient prediction network is set, comprising: at least one convolution block;

[0021] Any of the convolution blocks consists of residual connection and dense convolution;

[0022] The convolution layers in the dense convolution are all normalized and activated by Relu, and a concrete dropout layer is added after each convolution layer for regularization.

[0023] In the downsampling stage of the gradient prediction network, the input wrapped phase difference is first passed through the convolution block to increase the number of channels, and then the size is continuously reduced to achieve deep feature extraction;

[0024] In the upsampling stage of the gradient prediction network, the original image size is gradually restored through deconvolution, and the number of channels is reduced to the number of channels that wrap the original phase difference.

[0025] The predicted gradient is compared with the true gradient, and according to the loss function, the weight and bias of the gradient prediction network are optimized to obtain the trained gradient prediction network, including:

[0026] Generate a Gaussian random surface and use it as a phase difference map;

[0027] Wrapping the phase difference image and adding speckle noise to simulate the wrapped phase difference obtained in actual measurement;

[0028] Acquire a true gradient according to the wrapped phase difference;

[0029] Obtain a large amount of input and labeled data by varying the mean and variance of the Gaussian distribution as well as the noise level;

[0030] The weights and biases of the gradient prediction network are optimized by iterative back propagation using adaptive moment estimation to obtain a trained gradient prediction network.

[0031] The loss function includes:

[0032] The loss function is:

[0033] Among them, K refers to the total number of pixels in the image. represents the label value, y represents the predicted value, and σ represents the prediction variance.

[0034] The step of completing the prediction from the wrapped phase difference Δφ(y, z, t) to the phase gradient φ(y, z, t) according to the gradient prediction network comprises:

[0035] Keep the concrete dropout in the trained gradient prediction network turned on, and input the package phase difference to be predicted; and perform at least 3 forward propagations, such as 50 times, and after collecting the results of each forward propagation, obtain the predicted mean and use it as the final prediction of the gradient prediction network.

[0036] The final prediction of the gradient prediction network is:

[0037]

[0038] Among them, T represents the number of forward propagation, y (t) represents the result obtained by t times of forward propagation, and μ represents the mean of the results after multiple forward propagations.

[0039] The step of obtaining a solidification rate field of the solidification process of the material to be tested according to the wrapping phase difference and the phase gradient includes:

[0040] according to Obtaining a solidification rate field;

[0041] Among them, λ c represents the central wavelength of the light source of the OCT system, δt represents the interval time between two adjacent spectra, and Φ(y, z, t, δt) represents the phase gradient;

[0042] The wrapped phase difference Δφ(y, z, t, δt) is input into the trained gradient prediction network to obtain the predicted phase gradient Φ(y, z, t, δt).

[0043] The real-time imaging according to the curing rate field and autonomous evaluation of the imaging quality include:

[0044] The imaging quality of the curing rate field is reflected by the model uncertainty of the Bayesian neural network:

[0045]

[0046] Among them, T represents the number of forward propagation; y (t) represents the result obtained by t times of forward propagation; μ represents the mean of the results after multiple forward propagations.

[0047] The beneficial effects of the present disclosure include at least:

[0048] The real-time monitoring method of the curing rate field disclosed in the present invention first obtains the interference spectrum of the cross section of the material to be tested through an optical coherence tomography system (OCT), and obtains the package phase difference of the cross section; then obtains the prediction of the package phase and phase gradient according to the package phase difference; then, obtains the curing rate field of the curing process of the material to be tested according to the package phase difference and the phase gradient; finally, real-time imaging is performed according to the curing rate field, and the imaging quality is autonomously evaluated, thereby realizing real-time monitoring of the curing rate field during the curing process. The real-time monitoring method of the curing rate field disclosed in the present invention can complete the full-field measurement of the curing rate of the polymer, and realize the breakthrough of curing rate monitoring from point to field; at the same time, a Bayesian convolutional neural network is used for uncertainty learning. In addition to outputting the prediction results, the quality of the prediction results of the model is autonomously estimated by outputting the model uncertainty, which can quantitatively evaluate the imaging quality and improve the confidence of the curing rate monitoring results. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is a schematic diagram of using an OCT system to obtain interference spectra and package phase differences in the real-time monitoring method disclosed in the present invention;

[0050] Figure 2 It is a schematic diagram of the network model of the Bayesian convolutional neural network;

[0051] Figure 3 It is the training process of Bayesian convolutional neural network;

[0052] Figure 4 It is a real-time imaging diagram of the polymer droplet solidification rate field. DETAILED DESCRIPTION

[0053] The present disclosure is further described below in conjunction with the accompanying drawings.

[0054] Specific embodiment 1:

[0055] The present disclosure provides an embodiment:

[0056] A method for real-time monitoring of the curing rate field inside a polymer composite material, comprising the following specific steps:

[0057] S1. Obtain the interference spectrum of the cross section of the material to be tested as it changes over time, and obtain the wrapping phase difference of the cross section:

[0058] The polymer is placed on the test platform, and the optical coherence tomography system OCT is used to collect the interference spectrum of the polymer over time. Then, the phase-frequency characteristic diagram obtained by Fourier transform is differentially calculated to obtain the wrapped phase difference Δφ(y, z, t) of the cross section. The specific process is as follows: Figure 1 shown.

[0059] S2. Predicting the phase gradient according to the package phase difference:

[0060] By training a Bayesian convolutional neural network, the end-to-end mapping from the wrapped phase difference Δφ(y, z, t) to the phase gradient Φ(y, z, t) is directly realized. The network model of the Bayesian convolutional neural network is as follows: Figure 2 shown.

[0061] The network model is improved based on UNet++. Specifically, each convolution block is composed of three dense convolutions and residual connections. Among them, the convolution layer except the residual connection uses a 1*1 convolution kernel, and the other convolution layers use a 3*3 convolution kernel. In the dense convolution, the convolution layers are all normalized and Relu activation functions are used, and a concrete dropout layer is added after each layer for regularization. The dropout rate will be learned during the training process. In the downsampling stage, the image first increases the number of channels through the convolution block, and then uses the maximum pooling layer to continuously reduce the size to achieve deep feature extraction. In the upsampling stage, the original image size is gradually restored through deconvolution, and the number of channels is reduced to the number of channels of the original image. Then, the features of different deep layers are spliced ​​through dense jump connections to improve the prediction effect.

[0062] For this Bayesian neural network, the output of the network is considered to be a Gaussian distribution. Therefore, the output layer of the model is composed of two convolutional layers, one for outputting the predicted phase gradient, that is, the mean, and the other for outputting the variance of the predicted results.

[0063] The loss function uses the negative log-likelihood function of the Gaussian distribution:

[0064]

[0065] In formula (1), K refers to the total number of pixels in the image. represents the label value, t represents the predicted value, and σ represents the prediction variance.

[0066] like Figure 3 , train the Bayesian neural network, and in each training cycle, input the wrapped phase difference into the gradient prediction network, compare the output predicted gradient with the true gradient, and then calculate the loss function to optimize the weights and biases of the network. Here, the optimization is performed by iterative backpropagation using adaptive moment estimation ADAM. The training cycle is set to 200, the batch size is 20, and the initial learning rate is set to 10 -4 , and set the learning rate scheduler parameter factor to 0.1 and patience to 4. That is, when the loss does not decrease for 4 consecutive cycles, the learning rate will be multiplied by a factor of 0.1. In addition, the minimum learning rate is set to 10 -9 In addition, an early stopping mechanism is added to the network, and the number of cycles is set to 10.

[0067] It is worth noting that in order to train this network, a series of wrapped phase differences, i.e., input data, and their corresponding phase gradients, i.e., label data, are prepared. These data are all generated by Matlab simulation. The specific generation process is as follows: First, a large number of Gaussian random surfaces are generated and used as phase difference maps. Then, the phase difference map is wrapped and speckle noise is added to simulate the wrapped phase difference obtained in the actual measurement. At the same time, the basic facts of the phase gradient value are also calculated based on the generated phase difference map. Then, by changing the Gaussian distribution, that is, the pseudo-random number presents a Gaussian distribution, the mean and variance, and changing the noise level, a large amount of input and label data can be obtained. Finally, the size of the data set used is 50,000.

[0068] After the network training is completed, the prediction from the wrapped phase difference Δφ(y, z, t) to the phase gradient Φ(y, z, t) can be completed. Specifically, keep the concrete dropout turned on during the prediction, input the image to be predicted, and then perform multiple forward propagations. Collect the results of each forward propagation to calculate the prediction mean and use it as the final prediction of the network. The formula is as follows:

[0069]

[0070] In formula (2), T represents the number of forward propagation, y (t) represents the result obtained after t times of forward propagation, μ represents the mean of the results after t times of forward propagation; t = 1, 2, ...T.

[0071] S3. According to the wrapping phase difference and the phase gradient, the curing rate field of the curing process of the material to be tested is obtained:

[0072] The curing rate field α(y, z, t) of the polymer curing process can be expressed by the following formula:

[0073]

[0074] In formula (3), λ c Represents the central wavelength of the OCT system light source, δt represents the interval time between two adjacent spectra, and φ(y, z, t, δt) represents the phase gradient. The wrapped phase difference Δφ(y, z, t, δt) is input into the trained network to obtain the predicted phase gradient Φ(y, z, t, δt), and the curing rate field is calculated.

[0075] S4. Real-time imaging of the curing rate field and autonomous evaluation:

[0076] The model uncertainty of the Bayesian neural network reflects the imaging quality of the curing rate field and can be obtained by calculating the variance of the prediction results. Formula (4) is as follows:

[0077]

[0078] Based on step S1-4, during the curing process of the polymer composite material, the optical coherence tomography system continuously collects signals and inputs them into the trained neural network. The network then performs real-time imaging of the curing rate field and outputs the model uncertainty to independently evaluate the imaging quality, thereby realizing real-time monitoring of the curing rate field during the curing process.

[0079] Application examples:

[0080] To verify the practicability of the real-time monitoring method described in Specific Example I, real-time imaging of the solidification rate field of a polymer droplet was performed.

[0081] like Figure 4 As shown in the figure, during the curing process, the OCT system is used to collect the time-varying interference spectrum of the droplet cross section, and then after Fourier transform and differential calculation, it can be obtained. Figure 4 (a-1) to (a-5) in the figure are the wrapped phase differences Δφ(y, z, t, δt) at different times. The wrapped phase difference is then sent to the network for prediction to obtain the predicted phase gradient φ(y, z, t, δt), which is then substituted into formula (3) in specific embodiment I to obtain Figure 4 (b-1) to (b-5) correspond to the solidification rate field at different times.

[0082] in accordance with Figure 4 It can be directly concluded that at 3s, 6s, and 9s, i.e., the beginning stage of solidification, some areas on the droplet have a higher solidification rate, while at 12s and 15s, i.e., when the solidification is about to end, the solidification rate gradually decreases and finally approaches 0. Figure 4(c-1) to (c-5) are model uncertainty diagrams at different times. It can be directly concluded that: throughout the entire curing process, the model uncertainty remains at a low state, which shows that the quality of the curing rate field imaging obtained by the network in the experiment is high. Therefore, this experiment proves the effectiveness and practicality of the real-time imaging method of the curing rate field inside the polymer composite material. Compared with the prior art, this embodiment realizes the full-field measurement of the polymer curing rate, and achieves a breakthrough in curing rate monitoring from point to field; at the same time, a Bayesian convolutional neural network is used for uncertainty learning. In addition to outputting the prediction results, the imaging quality of the curing rate field is also autonomously estimated by outputting the model uncertainty.

[0083] Specific embodiment II:

[0084] The present disclosure also provides an embodiment:

[0085] An electronic device comprises: a storage medium and a processing unit; wherein the storage medium is used to store a computer program, and the processing unit exchanges data with the storage medium, and is used to execute the computer program through the processing unit when real-time monitoring of the curing rate field inside a polymer composite material is performed to perform the steps of the real-time monitoring method of the curing rate field inside a polymer composite material as described in specific embodiment I.

[0086] Specific embodiment III:

[0087] A computer-readable storage medium having a computer program stored therein; when the computer program is run, the steps of the method for real-time monitoring of the curing rate field inside a polymer composite material as described in Specific Example 1 are executed.

[0088] In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, apparatus, or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0089] The above disclosures are only several specific implementation scenarios of the present disclosure, but the present disclosure is not limited thereto, and any changes that can be thought of by a person skilled in the art should fall within the protection scope of the present disclosure. The above disclosure serial numbers are only for description and do not represent the advantages and disadvantages of the implementation scenarios.

Claims

1. A method for real-time monitoring of the curing rate field inside a polymer composite material, characterized in that: include: Obtaining an interference spectrum of a cross section of the material to be tested by an optical coherence tomography system, and obtaining a wrapping phase difference of the cross section; predicting a phase gradient based on the wrapped phase difference; Acquiring a curing rate field of the curing process of the material to be tested according to the wrapping phase difference and the phase gradient; Real-time imaging is performed based on the curing rate field, and the imaging quality is independently evaluated, thereby achieving real-time monitoring of the curing rate field during the curing process.

2. The method for real-time monitoring of the curing rate field inside a polymer composite material according to claim 1, characterized in that: The method of obtaining the interference spectrum of the cross section of the material to be tested by the optical coherence tomography system and obtaining the wrapping phase difference of the cross section includes: Placing the material to be tested on the testing platform; The interference spectrum of the cross section of the material to be tested is obtained by an optical coherence tomography system, and a differential operation is performed on the phase-frequency characteristic diagram obtained by Fourier transform to obtain the wrapping phase difference of the cross section.

3. The method for real-time monitoring of the curing rate field inside a polymer composite material according to claim 1, characterized in that: The predicting the phase gradient according to the wrapped phase difference comprises: Set up the gradient prediction network; Inputting the wrapped phase difference into the gradient prediction network and outputting the predicted gradient; Comparing the predicted gradient with the true gradient, and optimizing the weights and biases of the gradient prediction network according to the loss function to obtain a trained gradient prediction network; According to the gradient prediction network, the prediction from the wrapped phase difference Δφ(y, z, t) to the phase gradient Φ(y, z, t) is completed.

4. The method for real-time monitoring of the curing rate field inside a polymer composite material according to claim 3, characterized in that: The gradient prediction network is set, comprising: at least one convolution block; Any of the convolution blocks consists of residual connection and dense convolution; The convolution layers in the dense convolution are all normalized and activated by Relu, and a concrete dropout layer is added after each convolution layer for regularization. In the downsampling stage of the gradient prediction network, the input wrapped phase difference is first passed through the convolution block to increase the number of channels, and then the size is continuously reduced to achieve deep feature extraction; In the upsampling stage of the gradient prediction network, the original image size is gradually restored through deconvolution, and the number of channels is reduced to the number of channels that wrap the original phase difference.

5. The method for real-time monitoring of the curing rate field inside a polymer composite material according to claim 3, characterized in that: The predicted gradient is compared with the true gradient, and according to the loss function, the weight and bias of the gradient prediction network are optimized to obtain the trained gradient prediction network, including: Generate a Gaussian random surface and use it as a phase difference map; Wrapping the phase difference image and adding speckle noise to simulate the wrapped phase difference obtained in actual measurement; Acquire a true gradient according to the wrapped phase difference; Obtain a large amount of input and labeled data by varying the mean and variance of the Gaussian distribution as well as the noise level; The weights and biases of the gradient prediction network are optimized by iterative back propagation using adaptive moment estimation to obtain a trained gradient prediction network.

6. The method for real-time monitoring of the curing rate field inside a polymer composite material according to claim 3 or 5, characterized in that: The loss function includes: The loss function is: Among them, K refers to the total number of pixels in the image. represents the label value, y represents the predicted value, and σ represents the prediction variance.

7. The method for real-time monitoring of the curing rate field inside a polymer composite material according to claim 3, characterized in that: The step of completing the prediction from the wrapped phase difference Δφ(y, z, t) to the phase gradient Φ(y, z, t) according to the gradient prediction network comprises: Keep the concrete dropout in the trained gradient prediction network turned on, and input the package phase difference to be predicted; and perform at least 3 forward propagations, collect the results of each forward propagation, obtain the predicted mean, and use it as the final prediction of the gradient prediction network.

8. The method for real-time monitoring of the curing rate field inside a polymer composite material according to claim 7, characterized in that: The final prediction of the gradient prediction network is: Among them, T represents the number of forward propagations, y(t) represents the result obtained after t forward propagations, and μ represents the mean of the results after multiple forward propagations.

9. The method for real-time monitoring of the curing rate field inside a polymer composite material according to claim 1, characterized in that: The step of obtaining a solidification rate field of the solidification process of the material to be tested according to the wrapping phase difference and the phase gradient includes: according to Obtaining a solidification rate field; Among them, λ c represents the central wavelength of the light source of the OCT system, δt represents the interval time between two adjacent spectra, and Φ(y, z, t, δt) represents the phase gradient; The wrapped phase difference Δφ(y,z,t,δt) is input into the trained gradient prediction network to obtain the predicted phase gradient Φ(y,z,t,δt).

10. The method for real-time monitoring of the curing rate field inside a polymer composite material according to claim 1, characterized in that: The real-time imaging according to the curing rate field and autonomous evaluation of the imaging quality include: The imaging quality of the curing rate field is reflected by the model uncertainty of the Bayesian neural network: Among them, T represents the number of forward propagation; y (t) represents the result obtained after t times of forward propagation; μ represents the mean of the results after t times of forward propagation; t = 1, 2, ...T.

Images