Low temperature resistance evaluation method for composite material

Abstract

The invention provides a low-temperature-resistant performance evaluation method for a composite material, and belongs to the technical field of material performance determination.The method comprises the steps that firstly, in a low-temperature environment box, a temperature sensor is used for accurately controlling the temperature of a sample, and meanwhile, a high-frequency strain sensor is used for collecting data; then, wavelet transform is adopted to carry out denoising and decomposition on the strain signals, and key low-temperature characteristic parameters are extracted; and constructing a performance evaluation model based on deep learning, inputting the temperature field, the strain characteristics and the material basic parameters into the model, and predicting the low-temperature strength. A prediction result is verified through a standard test, a prediction error is calculated, an error distribution matrix is established, and an improved back propagation algorithm is adopted to optimize the model. Finally, a fuzzy comprehensive evaluation method is applied, indexes such as strength retention rate and strain stability are comprehensively considered, and the low temperature resistance grade of the composite material is scientifically evaluated. The method solves the problem that the actual mechanical properties of the composite material in the low-temperature environment are difficult to comprehensively reflect in the prior art.

Description

Technical Field

[0001] This invention belongs to the field of material performance testing technology, and more specifically, relates to a method for evaluating the low-temperature performance of composite materials. Background Technology

[0002] Composite materials are widely used in aerospace, shipbuilding, and rail transportation due to their excellent specific strength, specific stiffness, and lightweight structure. Among them, high-performance composite materials such as carbon fiber reinforced epoxy resin (CFRP) and glass fiber reinforced polyester resin (GFRP) have become indispensable key materials in these fields. These composite materials often encounter various extreme environments during use, especially low-temperature environments.

[0003] At low temperatures, the mechanical properties of composite materials undergo significant changes. On the one hand, the matrix resin becomes brittle, and the interfacial bonding strength decreases; on the other hand, differences in the coefficients of thermal expansion between different components lead to uneven distribution of internal stress and strain. These factors combined cause a decline in the strength, stiffness, and toughness of the composite material, severely impacting its service life and reliability. Therefore, accurately assessing the mechanical behavior of composite materials under low-temperature conditions is crucial for ensuring their safe use in harsh environments.

[0004] Current methods for evaluating the low-temperature performance of composite materials mainly rely on standard tensile and flexural tests. However, this approach has the limitation of failing to comprehensively reflect the actual mechanical properties of composite materials under low-temperature conditions. Summary of the Invention

[0005] In view of this, the present invention provides a method for evaluating the low-temperature performance of composite materials, which can solve the problem that existing methods cannot fully reflect the actual mechanical properties of composite materials in low-temperature environments.

[0006] This invention is implemented as follows: This invention provides a method for evaluating the low-temperature performance of composite materials, comprising the following steps: placing a composite material sample in a low-temperature environment chamber and monitoring the sample surface temperature by constructing a temperature field distribution matrix using multiple temperature sensors; installing multiple strain sensors on the sample surface to construct a strain measurement network and collecting strain signals and temperature field data of the sample at a preset test temperature; performing wavelet transform denoising and multivariate hybrid decomposition on the strain signals to establish a strain-temperature response matrix and extracting low-temperature strain characteristic parameters; inputting the temperature field distribution characteristics, strain characteristic parameter set, and basic material property parameters into a pre-trained deep learning low-temperature performance evaluation model to obtain the predicted low-temperature strength value and its confidence interval of the sample; conducting a low-temperature tensile test on the sample, recording the error between the actual low-temperature strength value and the predicted value, and performing error analysis; constructing a model based on the error distribution to optimize the objective function, and using an improved backpropagation algorithm to optimize the low-temperature performance evaluation model; using the optimized model to evaluate the low-temperature performance of new composite material samples; and establishing a performance grading standard based on fuzzy comprehensive evaluation to determine the low-temperature performance level of the new composite material samples.

[0007] The step of placing the composite material sample in the low-temperature environment chamber specifically involves: collecting the length, width, and thickness dimensions of the composite material sample; grinding the surface of the composite material sample until the surface roughness is less than 0.5 micrometers; placing the processed composite material sample in the center of the low-temperature environment chamber; and uniformly arranging eight high-precision temperature sensors along the length and width directions on the surface of the composite material sample.

[0008] The temperature sensor has a measurement accuracy of 0.1 degrees Celsius. An 8×8 temperature field distribution matrix is ​​constructed using the temperature data collected by the temperature sensor. Temperature control is completed when the temperature fluctuation value of each measurement point in the temperature field distribution matrix does not exceed ±0.5 degrees Celsius and reaches the preset test temperature.

[0009] Specifically, the step of installing multiple strain sensors on the sample surface to construct a strain measurement network involves arranging 16 resistance strain sensors at predetermined positions along the length and width directions on the surface of the composite material sample to construct a strain measurement network. The strain sensors are connected to a 24-bit high-precision data acquisition system through a 6-channel signal conditioning circuit.

[0010] Specifically, the step of collecting strain signals and temperature field data of the sample at a preset test temperature involves setting the sampling frequency of the data acquisition system to 200Hz and the acquisition duration to 180s, and simultaneously collecting dynamic strain signal data and temperature field data of the composite material sample under stable preset test temperature conditions.

[0011] Specifically, the steps of performing wavelet transform denoising and multivariate hybrid decomposition on the strain signal are as follows: performing fourth-order wavelet transform denoising on the acquired strain signal data, using empirical mode decomposition to decompose the denoised strain signal into four intrinsic mode components, and extracting the steady-state strain component and the dynamic strain component.

[0012] Specifically, the step of establishing the strain-temperature response matrix involves: establishing a 16×16 strain-temperature response matrix based on the steady-state strain components and temperature field data; and extracting the steady-state strain value, strain fluctuation coefficient, strain-temperature sensitivity coefficient, and strain anisotropy index from the strain-temperature response matrix as low-temperature strain characteristic parameters.

[0013] Specifically, the step of inputting the temperature field distribution features, strain feature parameter set, and material basic property parameters into the pre-trained deep learning low-temperature performance evaluation model involves inputting the 8×8 temperature field distribution feature matrix, the 16-dimensional strain feature parameter vector, and the elastic modulus, Poisson's ratio, and fiber mass fraction of the composite material into the pre-trained 5-layer deep neural network model.

[0014] The hidden layer of the deep neural network model uses a rectified linear unit activation function, and the output layer provides the predicted low-temperature strength value and prediction confidence interval of the composite material sample.

[0015] The specific steps of conducting a low-temperature tensile test on the sample are as follows: under the preset test temperature conditions, the composite material sample is loaded into a low-temperature universal testing machine, and a uniaxial tensile test is conducted at a loading rate of 1 mm / min. A high-precision extensometer is used to record the stress-strain curve of the composite material sample in real time.

[0016] Specifically, the error analysis steps are as follows: calculating the relative error between the predicted low-temperature strength value and the actual low-temperature strength value; constructing an 8×8 error distribution matrix that includes temperature field error components, strain characteristic error components, and material parameter error components; and using principal component analysis to analyze the weight distribution of error sources.

[0017] Specifically, the step of constructing a model optimization objective function based on the error distribution involves: constructing a model optimization objective function containing a mean squared error term and a regularization term based on an 8×8 error distribution matrix; optimizing the weight parameters of the low-temperature performance evaluation model using a stochastic gradient descent algorithm with a driving term; and introducing [variables] during the optimization process. Norm regularization terms prevent overfitting.

[0018] The step of establishing a performance grading standard based on fuzzy comprehensive evaluation specifically involves: establishing a performance grading standard based on fuzzy hierarchical analysis, setting the weighting coefficient for low-temperature strength retention rate to 0.4, the weighting coefficient for strain response stability to 0.3, the weighting coefficient for temperature sensitivity to 0.2, and the weighting coefficient for prediction reliability to 0.1, and dividing the evaluation results into four performance levels: special grade, first grade, second grade, and third grade.

[0019] Specifically, the steps included in this method are described in detail below: S10. After measuring the dimensions and treating the surface of the composite material sample, place it in a low-temperature environment chamber. Use multiple temperature sensors to construct a temperature field distribution matrix to monitor the surface temperature of the sample in real time. When the temperature fluctuation at each point in the temperature field distribution matrix does not exceed ±0.5℃ and reaches the preset test temperature, the temperature control is completed. S20. Install multiple strain sensors at preset positions on the sample surface to construct a strain measurement network. Connect the network to a high-precision data acquisition system through a signal conditioning circuit. Set the sampling frequency to be no less than 100Hz and the acquisition time to be no less than 60s. Simultaneously acquire the strain signal and temperature field data of the sample at the preset test temperature. S30. Perform wavelet transform denoising and multivariate hybrid decomposition on the strain signal to decompose the strain signal into steady-state components and dynamic components, and establish a strain-temperature response matrix to extract low-temperature strain characteristic parameters including steady-state strain value, strain fluctuation coefficient, strain-temperature sensitivity coefficient, and strain anisotropy index. S40. Input the temperature field distribution characteristics, strain characteristic parameter set and material basic property parameters into a pre-trained deep learning-based low-temperature performance evaluation model, and use a multilayer perceptron structure and ReLU activation function to obtain the predicted low-temperature strength value and its confidence interval of the sample. S50. Perform a low-temperature tensile test on the specimen according to the standard test method, record the stress-strain curve throughout the entire process, extract mechanical parameters including the actual low-temperature strength value and elastic modulus, and establish an experimental database containing complete test process data. S60. Calculate the relative error between the predicted and measured values ​​of the low-temperature intensity, establish an error distribution matrix to analyze the sources of error, calculate the reliability index of the prediction model, and evaluate the applicability of the model's prediction results. S70. Based on the error distribution matrix, construct a model to optimize the objective function, and use an improved backpropagation algorithm and regularization method to optimize the weight parameters of the low temperature performance evaluation model. Iterate the optimization until the accuracy requirements are met to obtain the optimized low temperature performance evaluation model. S80. The optimized low-temperature performance evaluation model is used to evaluate the low-temperature performance of the new composite material sample, the confidence level of the prediction results is given, and a complete evaluation report containing all key parameters is output to obtain the low-temperature strength evaluation value of the new composite material sample. S90. Establish a performance grading standard based on fuzzy comprehensive evaluation, comprehensively considering indicators such as low-temperature strength retention rate, strain response stability, temperature sensitivity and prediction reliability, and determine the low-temperature performance level of the new composite material sample based on the low-temperature strength evaluation value.

[0020] The following are some typical composite materials suitable for this evaluation method: carbon fiber reinforced epoxy resin composites (CFRP), glass fiber reinforced polyester resin composites (GFRP), aramid fiber reinforced polypropylene composites, carbon nanotube / graphene modified epoxy resin composites, basalt fiber reinforced polyamide composites, short-cut carbon fiber / long carbon fiber hybrid reinforced PEEK composites, natural fiber reinforced bio-based composites, ceramic fiber reinforced metal matrix composites, 3D braided carbon fiber reinforced composites, and hybrid fiber reinforced composites. Among them: Carbon fiber reinforced epoxy resin composite (CFRP) is mainly composed of carbon fiber and epoxy resin matrix. It has the characteristics of high strength, high stiffness and low density. In low temperature environment, the epoxy resin matrix may become brittle and the fiber / matrix interface bonding strength may decrease. It is widely used in low temperature environment such as aerospace and deep sea equipment. Glass fiber reinforced polyester resin composite (GFRP) is composed of glass fiber and unsaturated polyester resin. It has good corrosion resistance and electrical insulation. However, the toughness of the resin matrix decreases significantly at low temperatures, so its low-temperature embrittlement characteristics need to be evaluated. It is often used in industrial equipment such as low-temperature pipelines and storage tanks. Aramid fiber reinforced polypropylene composite material, composed of aramid fiber and polypropylene matrix, has excellent impact resistance and fatigue resistance. Low temperature environment will affect the crystallinity and toughness of polypropylene, making it suitable for low temperature protective equipment and sports equipment. Carbon nanotube / graphene modified epoxy resin composites are made by adding nanomaterials such as carbon nanotubes or graphene to epoxy resin to improve the thermal conductivity and mechanical properties of the material. The interface effect of the nanofillers is more significant at low temperatures, and they are used in thermal protection systems for aerospace equipment and low-temperature electronic packaging. Basalt fiber reinforced polyamide composite material, prepared by basalt fiber and polyamide resin, has good temperature resistance and dimensional stability. The changes in matrix toughness and fiber / matrix interface bonding strength at low temperature need to be closely monitored. It is applied to low-temperature industrial equipment and building components. Short-cut carbon fiber / long carbon fiber hybrid reinforced PEEK composite material is composed of short-cut carbon fiber, continuous carbon fiber and PEEK resin, which combines the advantages of short fiber and long fiber. The PEEK matrix maintains good toughness at low temperature and is used for low temperature bearings and transmission components. Natural fiber reinforced bio-based composite materials use natural fibers such as flax and jute and bio-based resins. They are environmentally friendly and biodegradable, have a low coefficient of thermal expansion, and their low-temperature performance is significantly affected by the moisture content of the fibers. They are used in low-temperature insulation materials and packaging containers. Ceramic fiber reinforced metal matrix composites, composed of ceramic fibers and metal matrices such as aluminum and magnesium, have high strength and good thermal conductivity. The thermal expansion mismatch effect between the matrix and fibers at low temperatures needs to be evaluated. They are used in low-temperature heat exchangers and structural components. 3D braided carbon fiber reinforced composite material uses three-dimensional braided carbon fiber preforms and resin matrix. It has high interlaminar strength and good anti-delamination performance. However, the anisotropic changes under low temperature environment need to be carefully evaluated. It is used in low temperature pressure vessels and structural components. Hybrid fiber reinforced composites combine various reinforcing agents such as carbon fiber, glass fiber, and aramid fiber to achieve complementary properties and improve overall performance. The interfacial behavior between different fibers and the matrix is ​​complex at low temperatures, making them suitable for cryogenic equipment requiring multifunctional performance.

[0021] The steps for obtaining the training dataset for the low-temperature performance evaluation model specifically include: Step 1: Select five different fiber contents (30%, 40%, 50%, 60%, 70%) of carbon fiber reinforced epoxy resin composite materials, and prepare 20 standard samples for each fiber content. Step 2: Select five glass fiber reinforced polyester resin composites with different fiber contents (30%, 40%, 50%, 60%, 70%), and prepare 20 standard samples for each fiber content. Step 3: Select 5 different fiber contents (30%, 40%, 50%, 60%, 70%) of aramid fiber reinforced polypropylene composite materials, and prepare 20 standard samples for each fiber content. Step 4: Collect temperature field data and strain response data for each sample at 5 preset temperature points (0 degrees Celsius, -20 degrees Celsius, -40 degrees Celsius, -60 degrees Celsius, -80 degrees Celsius); Step 5: Perform a standard tensile test on each specimen at the corresponding temperature point to obtain the actual low-temperature strength value; Step 6: Combine the collected temperature field data, strain response data, material parameters, and actual strength values ​​into training sample pairs, totaling 1500 sets of data; Step 7: Divide the dataset into training and validation sets in an 8:2 ratio.

[0022] The training steps for the low-temperature performance evaluation model specifically include: Step 1: Construct a 5-layer deep neural network. The number of nodes in the input layer is equal to the feature dimension of the training samples. The number of nodes in the hidden layer are 128, 256, 256, and 128, respectively. The number of nodes in the output layer is 1. Step 2: Initialize network parameters, weights are adopted The normal distribution is randomly initialized, and the bias term is initialized to 0; Step 3: Set training parameters, including the learning rate. Batch size is 32, training epochs are 1000, and regularization coefficient is... ; Step 4: Standardize the training data, transforming each feature to the [0,1] interval: ; Step 5: Train the model using batch stochastic gradient descent. The loss function for each batch is: ; Step 6: Update model parameters using the Adam optimizer: ; ; ; ; ; In the formula, This is the current gradient; For first-order and second-order momentum; The momentum decay rate; It is a numerically stable term; Step 7: Calculate model performance metrics on the validation set, including mean relative error and coefficient of determination: Mean relative error: ; Coefficient of determination: ; Step 8: When the rate of change of the loss function on the validation set is less than 0.1% for 10 consecutive rounds, stop training and save the model parameters.

[0023] During the above training process: 1. The dataset was constructed considering multiple variables such as material type, fiber content, and temperature, and has good representativeness; 2. The network structure adopts a "narrow-wide-narrow" form, which is beneficial for feature extraction and representation; 3. The Adam optimizer combines momentum and adaptive learning rate to accelerate convergence. 4. Monitoring of multiple performance metrics ensures the reliability of model training.

[0024] Compared with existing technologies, this invention provides a method for evaluating the low-temperature performance of composite materials. First, a precisely temperature-controlled low-temperature test environment is constructed through precision measurement and surface treatment. Within this environment, multi-point temperature sensors and strain sensors are used to monitor the temperature field distribution and strain response characteristics of the composite material sample in real time under low-temperature conditions. This lays the foundation for subsequent performance analysis.

[0025] Secondly, signal processing techniques such as wavelet transform and empirical mode decomposition are used to extract steady-state strain components and dynamic strain components from the dynamic strain signal, establish a strain-temperature response matrix, and extract key low-temperature strain characteristic parameters, such as steady-state strain value, strain fluctuation coefficient, and strain-temperature sensitivity coefficient. These parameters can accurately characterize the strain response properties of composite materials under low-temperature conditions.

[0026] Furthermore, the temperature field distribution characteristics, strain characteristic parameters, and fundamental material performance parameters are input into a pre-trained deep learning model. This model can predict the low-temperature strength of composite materials based on multi-source information. Compared to traditional single mechanical tests, this data-driven prediction model has higher prediction accuracy and reliability.

[0027] Finally, a performance grading standard based on fuzzy comprehensive evaluation is established. This standard considers multiple key indicators such as low-temperature strength retention rate, strain response stability, temperature sensitivity, and prediction reliability, which can more comprehensively evaluate the low-temperature performance of composite materials.

[0028] Compared with the prior art, the method of the present invention has the following advantages: 1) A precise and controllable low-temperature test environment was established, which can more accurately obtain the mechanical response data of composite materials at low temperatures, laying the foundation for subsequent analysis; 2) Advanced signal processing technology was used to analyze the strain characteristics of composite materials under low-temperature conditions, providing richer parameters for performance evaluation; 3) Deep learning-based prediction models can fully utilize multi-source information, improving the accuracy and reliability of low-temperature intensity prediction; 4) The performance grading standard takes into account multiple factors, and evaluates the low-temperature resistance of composite materials more comprehensively.

[0029] In summary, this invention solves the problem that existing methods cannot fully reflect the actual mechanical properties of composite materials in low-temperature environments. Attached Figure Description

[0030] Figure 1A flowchart of the method provided by the present invention. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0032] like Figure 1 The diagram shown is a flowchart of a method for evaluating the low-temperature resistance of composite materials provided by this invention. This method includes the following steps: S10. After measuring the dimensions and treating the surface of the composite material sample, place it in a low-temperature environment chamber. Use multiple temperature sensors to construct a temperature field distribution matrix to monitor the surface temperature of the sample in real time. When the temperature fluctuation at each point in the temperature field distribution matrix does not exceed ±0.5℃ and reaches the preset test temperature, the temperature control is completed. S20. Install multiple strain sensors at preset positions on the sample surface to construct a strain measurement network. Connect the network to a high-precision data acquisition system through a signal conditioning circuit. Set the sampling frequency to be no less than 100Hz and the acquisition time to be no less than 60s. Simultaneously acquire the strain signal and temperature field data of the sample at the preset test temperature. S30. Perform wavelet transform denoising and multivariate hybrid decomposition on the strain signal to decompose the strain signal into steady-state components and dynamic components, and establish a strain-temperature response matrix to extract low-temperature strain characteristic parameters including steady-state strain value, strain fluctuation coefficient, strain-temperature sensitivity coefficient, and strain anisotropy index. S40. Input the temperature field distribution characteristics, strain characteristic parameter set and material basic property parameters into the pre-trained deep learning-based low temperature performance evaluation model, and use a multilayer perceptron structure and ReLU activation function to obtain the low temperature strength prediction value and its confidence interval of the sample. S50. Perform low-temperature tensile tests on the specimens according to standard test methods, record the stress-strain curves throughout the entire process, extract mechanical parameters including actual low-temperature strength values ​​and elastic modulus, and establish an experimental database containing complete test process data. S60. Calculate the relative error between the predicted and measured values ​​of low-temperature intensity, establish an error distribution matrix to analyze the sources of error, calculate the reliability index of the prediction model, and evaluate the applicability of the model's prediction results. S70. Based on the error distribution matrix, construct a model to optimize the objective function. Use an improved backpropagation algorithm and regularization method to optimize the weight parameters of the low-temperature performance evaluation model. Iterate the optimization until the accuracy requirements are met to obtain the optimized low-temperature performance evaluation model. S80. The optimized low-temperature performance evaluation model is used to evaluate the low-temperature performance of the new composite material specimen, the confidence level of the prediction results is given, a complete evaluation report containing all key parameters is output, and the low-temperature strength evaluation value of the new composite material specimen is obtained. S90. Establish a performance grading standard based on fuzzy comprehensive evaluation, taking into account indicators such as low-temperature strength retention rate, strain response stability, temperature sensitivity and prediction reliability, and determine the low-temperature performance level of new composite material samples based on the low-temperature strength evaluation value.

[0033] The specific implementation methods of the above steps are described in detail below: The specific implementation of step S10 involves using precision measuring instruments to measure the dimensional parameters of the composite material sample, including length, width, and thickness. To ensure the accuracy of the temperature field distribution, the sample surface needs to be polished to achieve a surface roughness of less than 0.5 micrometers. The treated sample is placed in the center of the low-temperature environment chamber, and eight high-precision temperature sensors with a measurement accuracy of 0.1 degrees Celsius are evenly arranged along the length and width of the sample surface. The data collected by these temperature sensors can be used to construct an 8x8 temperature field distribution matrix. Temperature control is considered complete only when the temperature fluctuation at each measurement point in the temperature field distribution matrix does not exceed ±0.5 degrees Celsius and reaches the preset test temperature. The main purpose of this step is to establish a precise and controllable low-temperature environment, providing a foundation for subsequent strain testing and performance prediction.

[0034] In step S20, 16 resistance strain sensors are arranged at predetermined positions along the length and width of the composite material sample surface to construct a strain measurement network. These strain sensors are connected to a 24-bit high-precision data acquisition system via a 6-channel signal conditioning circuit. The sampling frequency of the data acquisition system is set to 200 Hz, and the acquisition duration is 180 seconds. Under the stable preset test temperature conditions established in step S10, dynamic strain signal data and temperature field data of the sample are acquired simultaneously. The purpose of this step is to obtain strain response characteristic data of the composite material at low temperatures, providing basic data for subsequent strain analysis and performance prediction.

[0035] The specific implementation of step S30 involves first performing fourth-order wavelet transform denoising on the acquired strain signal data to eliminate noise interference. Then, the Empirical Mode Decomposition (EMD) method is used to decompose the denoised strain signal into four intrinsic mode components. Steady-state strain components and dynamic strain components are extracted from these intrinsic mode components. A 16x16 strain-temperature response matrix is ​​established based on the steady-state strain components and temperature field data. Four low-temperature strain characteristic parameters are extracted from this response matrix: steady-state strain value, strain fluctuation coefficient, strain-temperature sensitivity coefficient, and strain anisotropy index. The purpose of this step is to extract key parameters characterizing the low-temperature strain response properties of the composite material.

[0036] The specific implementation of step S40 involves inputting the 8x8 temperature field distribution feature matrix constructed in step S10, the 16-dimensional strain feature parameter vector extracted in step S30, and basic parameters such as the elastic modulus, Poisson's ratio, and fiber mass fraction of the composite material into a pre-trained low-temperature performance evaluation model. This low-temperature performance evaluation model employs a 5-layer deep neural network structure, with Rectified Linear Units (ReLU) used as the activation function in the hidden layers. The output layer of the model provides the predicted low-temperature strength of the composite material sample and its confidence interval. The purpose of this step is to utilize a deep learning model to integrate multi-source information such as temperature field distribution, strain characteristics, and basic material parameters to predict the low-temperature strength of the composite material.

[0037] The specific implementation of step S50 involves placing the composite material sample into a low-temperature universal testing machine under a preset test temperature condition and conducting a uniaxial tensile test at a loading rate of 1 mm per minute. A high-precision extensometer is used to record the stress-strain curve of the sample in real time. The actual low-temperature strength, elastic modulus, and fracture strain are extracted from this stress-strain curve. All temperature field data, strain data, and mechanical parameters collected throughout the test are stored in the experimental database. The purpose of this step is to obtain the actual mechanical properties of the composite material at low temperatures, providing a benchmark for subsequent model evaluation and optimization.

[0038] The specific implementation of step S60 involves first calculating the relative error between the predicted low-temperature strength value obtained in step S40 and the actual low-temperature strength value obtained in step S50. Then, an 8x8 error distribution matrix is ​​constructed, containing three components: temperature field error component, strain characteristic error component, and material parameter error component. Principal component analysis is used to analyze this error distribution matrix to determine the influence weights of different factors on the accuracy of the prediction results. Finally, based on the error analysis results, the reliability index of the prediction model under different temperature ranges and strain levels is calculated to clarify the applicability of the model's prediction results. The purpose of this step is to comprehensively evaluate the performance of the current prediction model and provide a basis for subsequent model optimization.

[0039] The specific implementation of step S70 involves first constructing a model optimization objective function containing a mean squared error term and a regularization term based on the 8x8 error distribution matrix obtained in step S60. Then, a stochastic gradient descent algorithm with a driving term is used to optimize the weight parameters of the low-temperature performance evaluation model. An L2 norm regularization term is introduced during the optimization process to prevent overfitting. When the relative error change rate is less than 0.1% after 10 consecutive iterations, optimization is stopped and the optimized low-temperature performance evaluation model is saved. The purpose of this step is to address the existing error problems in the current model by optimizing the model parameters to improve prediction accuracy and ensure better generalization ability of the model.

[0040] The specific implementation of step S80 involves collecting temperature field data and strain characteristic parameters of the new composite material sample according to the methods in steps S10 to S40, and then inputting them into the low-temperature performance evaluation model optimized in step S70. The model will provide prediction results and their 95% confidence intervals. Finally, a complete evaluation report is generated, including temperature field distribution, strain response characteristics, material parameters, predicted strength values, and confidence levels, outputting the low-temperature strength evaluation value of the new composite material sample. The purpose of this step is to use the optimized model to predict and comprehensively evaluate the low-temperature performance of the new composite material sample, providing a basis for the low-temperature application of composite materials.

[0041] The specific implementation of step S90 involves first establishing a performance grading standard based on fuzzy hierarchical analysis. The weighting coefficients for low-temperature strength retention are set to 0.4, strain response stability to 0.3, temperature sensitivity to 0.2, and prediction reliability to 0.1. Then, based on the low-temperature strength assessment values ​​obtained in step S80, a fuzzy comprehensive evaluation of the low-temperature resistance of the new composite material samples is performed, and the evaluation results are divided into four performance levels: Special Grade, Grade 1, Grade 2, and Grade 3. The purpose of this step is to comprehensively consider multiple performance indicators to score and grade the low-temperature resistance of the composite material, providing a reference for subsequent material selection and application.

[0042] Specifically, the principle of this invention is to establish a prediction model based on multi-source information fusion and combine it with a comprehensive performance grading standard to achieve a comprehensive evaluation of the mechanical behavior of composite materials in low-temperature environments.

[0043] First, a precise and controllable low-temperature testing environment was constructed through meticulous measurement and surface treatment. Within this environment, multi-point temperature and strain sensors were used to monitor the temperature field distribution and strain response characteristics of the composite material samples in real time. This allows for more accurate acquisition of mechanical behavior data of the composite material under low-temperature conditions, laying the foundation for subsequent performance analysis.

[0044] Secondly, advanced signal processing techniques such as wavelet transform and empirical mode decomposition were used to extract steady-state and dynamic strain components from the obtained strain response data. These strain characteristic parameters can more comprehensively reflect the strain response characteristics of composite materials under low-temperature conditions, providing key basis for performance evaluation.

[0045] Furthermore, multi-source information, including temperature field distribution characteristics, strain characteristic parameters, and fundamental material performance parameters, is input into a pre-trained deep learning model. This model, through a multilayer perceptron structure and the ReLU activation function, effectively extracts the complex mapping relationship between input features and low-temperature intensity, thus providing accurate prediction results and their confidence intervals. Compared to traditional single mechanical tests, this data-driven prediction model fully utilizes multi-faceted information, significantly improving prediction accuracy and reliability.

[0046] Finally, this invention establishes a performance grading standard based on fuzzy hierarchical analysis. This standard comprehensively considers multiple key indicators such as low-temperature strength retention rate, strain response stability, temperature sensitivity, and prediction reliability, enabling a more comprehensive evaluation of the low-temperature resistance of composite materials. Based on the evaluation results, the low-temperature resistance of composite materials can be classified into different grades, such as premium, first-grade, second-grade, and third-grade, providing a basis for material selection and application.

[0047] In summary, the technical principle of the method of this invention is as follows: 1) Establish a precise and controllable low-temperature test environment to obtain accurate mechanical response data of composite materials at low temperatures; 2) Advanced signal processing technology is used to deeply analyze the low-temperature strain characteristics of composite materials; 3) A deep learning-based prediction model comprehensively utilizes multi-source information to predict low-temperature intensity; 4) Establish a comprehensive performance grading standard to fully evaluate the low-temperature resistance of composite materials.

[0048] This multi-level, multi-dimensional evaluation method can more accurately reflect the actual mechanical behavior of composite materials in low-temperature environments, providing reliable technical support for the low-temperature application of composite materials.

[0049] The following provides a specific embodiment 1 of the present invention. The specific implementation of the steps in this embodiment 1 is described in detail below: The specific implementation of step S10 is as follows: First, the dimensional parameters of the composite material sample are measured using precision measuring instruments. , representing the length, width, and thickness of the sample, respectively. To ensure the accuracy of the temperature field distribution, the sample surface also needs to be polished to reduce its surface roughness. Less than 0.5 micrometers.

[0050] The treated sample was placed in the center of the low-temperature environment chamber, and eight high-precision temperature sensors with a measurement accuracy of 0.1 degrees Celsius were evenly arranged on the sample surface along its length and width. The data collected by these temperature sensors can be used to construct an 8x8 temperature field distribution matrix. ,in In the temperature field distribution matrix, when the temperature fluctuation value at each measurement point... satisfy And the preset test temperature is reached. Temperature control is considered complete only when the temperature reaches a certain level. The main purpose of this step is to establish a precise and controllable low-temperature environment, providing a foundation for subsequent strain testing and performance prediction.

[0051] The specific implementation method of step S20 is as follows: Sixteen resistance strain sensors were arranged at predetermined positions along the length and width directions on the surface of the composite material sample. A strain measurement network was constructed. These strain sensors were connected to a 24-bit high-precision data acquisition system via a 6-channel signal conditioning circuit. The sampling frequency of the data acquisition system... Set to 200 Hz, sampling duration The duration is 180 seconds. The stable preset test temperature established in step S10... Under these conditions, dynamic strain signal data of the specimen are acquired synchronously. and temperature field data The purpose of this step is to obtain strain response characteristic data of composite materials at low temperatures, providing basic data for subsequent strain analysis and performance prediction.

[0052] The specific implementation method of step S30 is as follows: First, the collected strain signal data... Denoising using a 4th-order wavelet transform can be expressed as: ; In the formula, and These are the wavelet scaling function and the wavelet function, respectively. and These are the corresponding wavelet coefficients. This helps eliminate noise interference.

[0053] Then, the Empirical Mode Decomposition (EMD) method is used to denoise the strain signal. Decomposed into 4 intrinsic mode components Steady-state strain components are extracted from these intrinsic modal components. and dynamic strain components Based on steady-state strain components and temperature field data Establish a 16x16 strain-temperature response matrix : ; From this response matrix The following four low-temperature strain characteristic parameters were extracted: steady-state strain value ; strain fluctuation coefficient ;

[0054] Strain-temperature sensitivity coefficient ;

[0055] Strain anisotropy index ; The purpose of this step is to extract key parameters that characterize the low-temperature strain response properties of composite materials.

[0056] The specific implementation method of step S40 is as follows: The 8x8 temperature field distribution feature matrix constructed in step S10 The 16-dimensional strain feature parameter vector extracted in step S30 and the elastic modulus of composite materials Poisson's ratio Fiber mass fraction Basic parameters as input The input is fed into a pre-trained low-temperature performance evaluation model. This low-temperature performance evaluation model uses a 5-layer deep neural network structure, where the hidden layers use Rectified Linear Units (ReLU) as the activation function. The model's output layer provides predicted low-temperature strength values ​​for the composite material specimen. and its confidence interval .

[0057] Specifically, the structure of this neural network model can be represented as follows: Input layer: ; Hidden layer 1: ; Hidden layer 2: ; Hidden layer 3: ; Hidden layer 4: ; Output layer: ; The purpose of this step is to use a deep learning model to integrate multi-source information such as temperature field distribution, strain characteristics, and basic material parameters to predict the low-temperature strength of composite materials.

[0058] The specific implementation method of step S50 is as follows: At the preset test temperature Under these conditions, the composite material specimens were loaded into a low-temperature universal testing machine and subjected to a loading rate of 1 mm per minute. Uniaxial tensile tests were conducted. A high-precision extensometer was used to record the stress-strain curves of the specimen in real time. The actual low-temperature strength value is extracted from this stress-strain curve. Elastic modulus Fracture strain Isomechanical parameters. Temperature field data collected throughout the experiment. Strain data All mechanical parameters are stored in the experimental database. The purpose of this step is to obtain the actual mechanical property data of the composite material at low temperatures, providing a benchmark for subsequent model evaluation and optimization.

[0059] The specific implementation method of step S60 is as follows: First, calculate the predicted low-temperature intensity value obtained in step S40. Compared with the actual low-temperature strength value obtained in step S50 relative error between Then construct an 8x8 error distribution matrix. This includes temperature field error components. Strain characteristic error components and material parameter error components (in represent The error distribution matrix is ​​composed of three parts (e.g., principal component analysis). To determine the weights of different factors on the accuracy of prediction results. Finally, based on the error analysis results, the prediction model was calculated for different temperature ranges. Different strain levels Reliability index This step clarifies the applicability of the model's prediction results. The purpose of this step is to comprehensively evaluate the performance of the current prediction model and provide a basis for subsequent model optimization.

[0060] The specific implementation method of step S70 is as follows: First, based on the 8x8 error distribution matrix obtained in step S60... Construct a model optimization objective function that includes a mean squared error term and a regularization term: ; In the formula, For model parameters, is the regularization coefficient.

[0061] Then, the stochastic gradient descent algorithm with driving term is used to adjust the weight parameters of the low-temperature performance evaluation model. Optimize: ; ; ; In the formula, For first and second momentum, The momentum decay rate, This is a numerically stable term. An L2 norm regularization term is introduced during optimization to prevent overfitting. When the relative error change rate is less than 0.1% after 10 consecutive iterations, optimization is stopped and the optimized low-temperature performance evaluation model is saved. The purpose of this step is to address the existing error issues in the current model by optimizing the model parameters to improve prediction accuracy and ensure better generalization ability.

[0062] The specific implementation method of step S80 is as follows: Temperature field data were collected from the new composite material sample according to steps S10 to S40. and strain characteristic parameters Then, input the data into the optimized low-temperature performance evaluation model from step S70. The model will then provide prediction results. and its 95% confidence interval Finally, a complete evaluation report is generated, including temperature field distribution, strain response characteristics, material parameters, predicted strength values, and confidence levels, outputting the low-temperature strength assessment values ​​for the new composite material specimens. The purpose of this step is to use the optimized model to predict and comprehensively evaluate the low-temperature performance of new composite material samples, providing a basis for the low-temperature application of composite materials.

[0063] The specific implementation method of step S90 is as follows: First, a performance grading standard based on fuzzy hierarchical analysis is established. This standard includes the low-temperature strength retention rate. Weighting coefficients Set to 0.4, strain response stability Weighting coefficients Set to 0.3, temperature sensitivity Weighting coefficients Set to 0.2, prediction reliability Weighting coefficients Set it to 0.1. Then, based on the low-temperature strength evaluation value obtained in step S80... Fuzzy comprehensive evaluation of the low-temperature resistance of new composite material samples:

[0064] ; The evaluation results Classified as Special Grade ( ), Level 1 ( ), Level 2 ( Level 3 Four performance levels. The purpose of this step is to comprehensively consider multiple performance indicators to score and classify the low-temperature resistance of the new composite material, providing a reference for subsequent material selection and application.

[0065] To better understand and implement this invention, Example 2, a specific application scenario, is provided below: A shipbuilding company is developing a novel carbon nanotube / graphene-modified epoxy resin composite material for ship protection. This composite material not only possesses excellent mechanical properties but also needs to maintain good mechanical stability at low temperatures. To comprehensively evaluate the mechanical properties of this composite material under low-temperature conditions, the company decided to use the evaluation method proposed in this invention for testing and analysis.

[0066] First, the researchers precisely measured the dimensional parameters of the composite material specimens. The length of the specimens... It is 150 mm wide. It is 50 mm thick. The diameter is 5 mm. To ensure the accuracy of the temperature field distribution, the researchers also polished the sample surface to reduce its surface roughness. It should be controlled below 0.3 micrometers.

[0067] The treated sample was placed in the center of the low-temperature environment chamber. Four high-precision temperature sensors were arranged on the sample surface along both the length and width directions to ensure measurement accuracy. The temperature was 0.1 degrees Celsius. Using data collected from these eight temperature sensors, researchers constructed an 8x8 temperature field distribution matrix. As shown in Table 1.

[0068] Table 1 Temperature field distribution matrix ;

[0069] In the temperature field distribution matrix, the temperature fluctuation values ​​at each measurement point All temperatures were less than ±0.5 degrees Celsius and reached the preset test temperature of -60 degrees Celsius, so the experimenters believed that temperature control had been completed.

[0070] Next, the researchers placed 16 resistance strain sensors at predetermined positions along the length and width directions on the surface of the composite material sample, constructing a strain measurement network. These strain sensors were connected to a 24-bit high-precision data acquisition system via a 6-channel signal conditioning circuit. The researchers adjusted the sampling frequency of the data acquisition system. Set to 200 Hz, sampling duration The test duration was set to 180 seconds. Under a stable test temperature of -60 degrees Celsius, the data acquisition system simultaneously recorded the dynamic strain signal data of the composite material specimens. and temperature field data .

[0071] For the collected strain signal data After denoising using fourth-order wavelet transform, the researchers decomposed it into four intrinsic mode components using the Empirical Mode Decomposition (EMD) method. Steady-state strain components are extracted from these intrinsic modes. and dynamic strain components Based on steady-state strain components and temperature field data The researchers established a 16x16 strain-temperature response matrix. As shown in Table 2.

[0072] Table 2 Strain-Temperature Response Matrix ;

[0073] From this response matrix In the experiment, the researchers extracted the following four low-temperature strain characteristic parameters: steady-state strain value ; strain fluctuation coefficient ; Strain-temperature sensitivity coefficient ; Strain anisotropy index ; The above temperature field distribution feature matrix strain characteristic parameter vector and the elastic modulus of composite materials Poisson's ratio and fiber mass fraction Basic parameters as input The input is fed into a pre-trained low-temperature performance evaluation model. This model employs a 5-layer deep neural network structure, with the hidden layers using the ReLU activation function. The output layer provides the predicted low-temperature intensity value. and its 95% confidence interval .

[0074] After completing the above predictions, the researchers also conducted a standard low-temperature tensile test on the composite material specimen. The test was conducted at a temperature of -60 degrees Celsius, with a loading rate of 1 mm / min. The specimen was subjected to uniaxial tension. The stress-strain curves were recorded in real time. And extract the actual low-temperature strength value from it. Elastic modulus and fracture strain Parameters such as these are stored in the experimental database.

[0075] Next, the researchers calculated the predicted low-temperature intensity value. Compared with measured values relative error between An 8x8 error distribution matrix was constructed. This includes temperature field error components. Strain characteristic error components and material parameter error components Through principal component analysis, the researchers determined the weights of different error sources as follows: , and Finally, based on the error analysis results, the reliability index of the prediction model under -60 degrees Celsius conditions was calculated. This indicates that the model's predictions within this temperature range are reliable.

[0076] Next, the researchers used an 8x8 error distribution matrix... A model optimization objective function containing a mean squared error term and a regularization term was constructed. Furthermore, a stochastic gradient descent algorithm with a driving term was used to adjust the weight parameters of the low-temperature performance evaluation model. Optimization was performed. During the optimization process, an L2 norm regularization term was introduced to prevent overfitting. The optimization process was stopped when the relative error change rate was less than 0.1% after 10 consecutive iterations, and the optimized low-temperature performance evaluation model parameters were saved.

[0077] Next, the researchers collected temperature field data on the newly prepared carbon nanotube / graphene modified epoxy resin composite material sample following the aforementioned steps. and strain characteristic parameters The values ​​are then input into the optimized low-temperature performance evaluation model. The model provides predicted low-temperature strength values ​​for the new specimen. and its 95% confidence interval .

[0078] Finally, the researchers established a performance grading standard based on fuzzy hierarchical analysis. This standard included the low-temperature strength retention rate. Weighting coefficients Set to 0.4, strain response stability Weighting coefficients Set to 0.3, temperature sensitivity Weighting coefficients Set to 0.2, prediction reliability Weighting coefficients Set it to 0.1. Based on the comprehensive performance evaluation indicators. Researchers rated this novel carbon nanotube / graphene modified epoxy resin composite material as having excellent low-temperature resistance.

[0079] Through the above specific implementation, it can be seen that the method of the present invention can comprehensively and reliably evaluate the mechanical properties of composite materials under low-temperature environments. First, a precise and controllable low-temperature test environment was constructed through precision measurement and surface treatment, laying the foundation for subsequent data acquisition. Second, advanced signal processing technology was used to extract key strain characteristic parameters of the composite material at low temperatures, providing richer evidence for performance analysis. Third, a deep learning-based prediction model comprehensively utilizes multi-source information to provide accurate low-temperature strength prediction results and their confidence intervals. Finally, the established performance grading standard comprehensively evaluates the low-temperature resistance of the composite material, providing a reliable basis for material selection and application.

[0080] It should be noted that the relevant variables of this invention are explained in detail in Table 3: Table 3. Variable Explanation Table

[0081] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for evaluating the low-temperature resistance of composite materials, characterized in that, include: The composite material sample was placed in a low-temperature environment chamber, and the surface temperature of the sample was monitored by constructing a temperature field distribution matrix using multiple temperature sensors. Multiple strain sensors are installed on the sample surface to construct a strain measurement network, and strain signals and temperature field data of the sample are collected at a preset test temperature. Wavelet transform denoising and multivariate hybrid decomposition are performed on the strain signals to establish a strain-temperature response matrix and extract low-temperature strain characteristic parameters. The temperature field distribution characteristics, strain characteristic parameter set and material basic property parameters are input into a pre-trained deep learning low-temperature performance evaluation model to obtain the predicted low-temperature strength value of the sample and its confidence interval. Low-temperature tensile tests were conducted on the specimens, and the error between the actual low-temperature strength value and the predicted value was recorded. Error analysis was then performed. The objective function is optimized by constructing a model based on the error distribution, and the low-temperature performance evaluation model is optimized by an improved backpropagation algorithm. The optimized model is used to evaluate the low-temperature performance of new composite material samples. A performance grading standard based on fuzzy comprehensive evaluation is established to determine the low-temperature performance level of new composite material samples.

2. The method according to claim 1, characterized in that, The step of placing the composite material sample in the low-temperature environment chamber specifically involves: collecting the length, width, and thickness dimensions of the composite material sample; grinding the surface of the composite material sample until the surface roughness is less than 0.5 micrometers; placing the processed composite material sample in the center of the low-temperature environment chamber; and uniformly arranging eight high-precision temperature sensors along the length and width directions on the surface of the composite material sample.

3. The method according to claim 2, characterized in that, The temperature sensor has a measurement accuracy of 0.1 degrees Celsius. An 8×8 temperature field distribution matrix is ​​constructed using the temperature data collected by the temperature sensor. Temperature control is completed when the temperature fluctuation value of each measurement point in the temperature field distribution matrix does not exceed ±0.5 degrees Celsius and reaches the preset test temperature.

4. The method according to claim 3, characterized in that, The step of installing multiple strain sensors on the sample surface to construct a strain measurement network specifically involves: arranging 16 resistance strain sensors at predetermined positions along the length and width directions on the surface of the composite material sample to construct a strain measurement network; the strain sensors are connected to a 24-bit high-precision data acquisition system through a 6-channel signal conditioning circuit.

5. The method according to claim 4, characterized in that, The steps for collecting strain signals and temperature field data of the sample at a preset test temperature are as follows: the sampling frequency of the data acquisition system is set to 200Hz and the acquisition duration is set to 180s. Dynamic strain signal data and temperature field data of the composite material sample are collected synchronously under stable preset test temperature conditions.

6. The method according to claim 5, characterized in that, The steps of performing wavelet transform denoising and multivariate hybrid decomposition on the strain signal are as follows: performing fourth-order wavelet transform denoising on the collected strain signal data, using empirical mode decomposition to decompose the denoised strain signal into four intrinsic mode components, and extracting steady-state strain components and dynamic strain components.

7. The method according to claim 6, characterized in that, The steps for establishing the strain-temperature response matrix are as follows: a 16×16 strain-temperature response matrix is ​​established based on the steady-state strain components and temperature field data, and the steady-state strain value, strain fluctuation coefficient, strain-temperature sensitivity coefficient, and strain anisotropy index are extracted from the strain-temperature response matrix as low-temperature strain characteristic parameters.

8. The method according to claim 7, characterized in that, The steps of inputting the temperature field distribution features, strain feature parameter set and material basic property parameters into the pre-trained deep learning low-temperature performance evaluation model are as follows: inputting the 8×8 temperature field distribution feature matrix, the 16-dimensional strain feature parameter vector and the elastic modulus, Poisson's ratio and fiber mass fraction of the composite material into the pre-trained 5-layer deep neural network model.

9. The method according to claim 8, characterized in that, The hidden layer of the deep neural network model uses a rectified linear unit activation function, and the output layer provides the predicted low-temperature strength value and prediction confidence interval of the composite material sample.

10. The method according to claim 9, characterized in that, The steps for conducting a low-temperature tensile test on the specimen are as follows: under the preset test temperature conditions, the composite material specimen is loaded into a low-temperature universal testing machine, and a uniaxial tensile test is conducted at a loading rate of 1 mm / min. A high-precision extensometer is used to record the stress-strain curve of the composite material specimen in real time.

Images