Nuclear dynamic defect detection method based on hydrogen-containing composite material for automobile manufacturing

Abstract

The invention discloses a nuclear dynamic defect detection method based on a hydrogen-containing composite material for automobile manufacturing, and relates to the technical field of industrial solid waste treatment material management.The nuclear dynamic defect detection method comprises the following steps that an integrated detection system of a thermal imaging sensor, an acceleration sensor, a temperature sensor and a sound sensor is installed on the structure of the hydrogen-containing composite material for an automobile; sensor data is continuously collected, and the sensor data comprises a temperature distribution image, a vibration signal, temperature data and a sound signal; by fusing multi-modal data, such as a temperature distribution image and a vibration signal, the influence of environmental factors such as temperature and light on a detection result is reduced, so that a reliable detection result can still be stably output and the detection precision is improved under the complex working conditions of high-speed running of an automobile and day and night temperature difference, and the detection precision is improved by combining various sensor data. The defects of a single imaging detection method are overcome, and more comprehensive and more accurate material state evaluation is provided.

Description

Technical Field

[0001] The present invention relates to the technical field of defect detection of hydrogen-containing composite materials, and specifically relates to a nuclear dynamic defect detection method for hydrogen-containing composite materials used in automobile manufacturing. Background Art

[0002] With the development of new energy vehicles and hydrogen energy technology, the application of hydrogen-containing composite materials in automobile manufacturing is becoming more and more extensive, especially in hydrogen storage tanks and other key components. During use, these materials may undergo changes in the surface or internal microstructure due to the absorption or release of hydrogen, resulting in a decline in performance or even failure, thereby affecting the safety of automobile use. Therefore, dynamic defect detection of hydrogen-containing composite materials to timely detect and evaluate the internal structural changes and defects of the materials is crucial for ensuring the safety and reliability of automobiles.

[0003] Currently, traditional nuclear dynamic defect detection technologies for hydrogen-containing composite materials often use laboratory-level neutron diffraction and X-ray computed tomography equipment, and apply ultrasonic detection and X-ray detection technologies. Although they have high-precision detection performance, they have high cost, strict and complex operation procedures, pose risks to personnel safety, damage the environment, and are difficult to achieve online monitoring, thus limiting their wide application in large automobile parts.

[0004] The prior art has the following deficiencies: The currently adopted relatively low-cost and non-destructive thermal imaging detection method has expanded the application of hydrogen-containing composite materials in automobile manufacturing, but the current method still has limitations in several aspects. Firstly, when the imaging detection technology is used to detect hydrogen-containing composite materials, it is extremely susceptible to environmental factors such as temperature and light, resulting in affected detection accuracy and further affecting the accurate assessment of the performance of composite materials. In addition, during the actual operation of automobiles, it is difficult to continuously monitor hydrogen-containing composite material components, so it is difficult to detect and replace them in a timely manner at the initial stage of material performance degradation. Most importantly, excessive reliance on the results of imaging detection ignores the comprehensive assessment of the performance and safety of hydrogen-containing composite materials.

[0005] The above information disclosed in the background art section is only used to enhance the understanding of the background of the present disclosure, and thus it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0006] The present invention aims to provide a nuclear dynamic defect detection method for hydrogen-containing composite materials used in automobile manufacturing, by integrating a detection system in the structure of hydrogen-containing composite materials, using multi-modal data fusion, and long-term monitoring and prediction with an LSTM neural network model to solve the problems in the above background art.

[0007] To achieve the above object, the present invention provides the following technical solutions: A method for detecting nuclear dynamic defects in hydrogen-containing composite materials for automobile manufacturing, comprising the following steps:

[0008] S1. Install an integrated detection system of a thermal imaging sensor, an acceleration sensor, a temperature sensor and a sound sensor on the structure of the hydrogen-containing composite material for automobiles, and continuously collect sensor data. Among them, the sensor data includes a temperature distribution image, a vibration signal, temperature data and a sound signal;

[0009] S2. Establish an environmental impact correction model, respectively use the temperature data to correct the temperature distribution image and the sound signal to correct the vibration signal, eliminate the interference of external environmental factors, and obtain an accurate temperature distribution image and vibration signal after correction;

[0010] S3. Integrate the corrected temperature distribution image and vibration signal through a data fusion algorithm to generate fusion data, so as to detect and judge abnormal internal structures of the material and locate surface defects, and use a multi-parameter evaluation model to comprehensively evaluate the performance and safety of the hydrogen-containing composite material;

[0011] S4. Use the LSTM neural network model to perform long-term monitoring and learning on the fusion data, predict the change trend of the material performance and potential service life problems, and obtain a prediction result;

[0012] S5. Set performance degradation and defect alarm thresholds, compare and analyze according to the prediction results. When the detection and judgment results are abnormal, an alarm is issued in real time and maintenance suggestions are provided to ensure the safe use of the material.

[0013] Optionally, the integrated detection system includes a high-resolution thermal imaging camera, a high-sensitivity accelerometer, a high-precision temperature sensor, a high-fidelity microphone, a data acquisition device and a cloud server. Among them, the thermal imaging camera obtains the surface temperature distribution image of the hydrogen-containing composite material, calibrated as TDM; the accelerometer collects the vibration signal of the hydrogen-containing composite material, calibrated as VS; the temperature sensor collects the temperature data of the environment where the hydrogen-containing composite material is located, calibrated as TD; the microphone captures the sound signal emitted by the hydrogen-containing composite material in the operating environment, calibrated as SD; the data acquisition device is connected to the thermal imaging camera, accelerometer, temperature sensor and microphone, and collects, preprocesses, classifies, stores and transmits sensor data in real time, and uploads it to the cloud server for secure storage and subsequent analysis.

[0014] Optionally, the environmental impact correction model includes the following steps:

[0015] According to the temperature distribution image TDM and temperature data TD measured by the thermal imaging sensor and the temperature sensor respectively, a relationship model between the temperature distribution image and the ambient temperature is established, and the temperature distribution image is corrected by the ambient temperature data, that is, TDM1 = TDM-τ×(TD-T0).

[0016] Where, TDM represents the surface temperature distribution image of the hydrogen-containing composite material collected by the thermal imaging sensor, τ represents the correction coefficient of the temperature distribution image TDM affected by the ambient temperature TD, which is obtained by calibration under the ideal experimental environment, TDM1 represents the actual temperature distribution image after model correction, TD represents the temperature data of the environment in which the hydrogen-containing composite material is located, and T0 represents the standard ambient temperature, that is, T0 = 20 °C;

[0017] According to the vibration signal VS and sound signal SD measured by the acceleration sensor and sound sensor respectively, a relationship model between the vibration signal and the ambient sound is established, and the vibration signal is corrected by the ambient sound signal, that is, VS1 = VS-δ×SD.

[0018] Where VS1 represents the actual vibration signal after model correction, VS represents the vibration signal of the hydrogen-containing composite material collected by the acceleration sensor, δ represents the correction coefficient of the vibration signal VS affected by the environmental sound SD, which is obtained through calibration under an ideal experimental environment, and SD represents the sound signal emitted by the hydrogen-containing composite material in the operating environment.

[0019] Optionally, the acquisition logic of the fusion data is as follows:

[0020] The actual temperature distribution image TDM1 and the actual vibration signal VS1 obtained after correction by the environmental impact correction model are integrated using a multimodal neural network model;

[0021] After image enhancement of the temperature distribution image TDM1, CNN is used to extract the feature map to obtain the image sequence of the temperature distribution image. The vibration signal VS1 is filtered and denoised, and then the time series pattern is extracted using LSTM to obtain the time series data of the vibration signal. The calculation formula of the image sequence is F t (TDM1)=CNN(I t (TDM1); θ), where F t (TDM1) is represented by the CNN model from the enhanced image I t The feature map of the t-th frame image extracted from (TDM1), I t (TDM1) represents the enhanced image obtained by image enhancement processing of the temperature distribution image TDM1, and θ represents the CNN model parameter;

[0022] The calculation formula for time series data is: Where h t(VS1) is represented as the LSTM hidden state at the t-th time step extracted from the filtered and denoised signal V t (VS1), and V t (VS1) is represented as the filtered and denoised signal obtained after filtering and denoising the vibration signal VS1, and h t-1 is represented as the LSTM output state at the previous t - 1 time step, is represented as the LSTM model parameters;

[0023] In the multi-modal neural network CNN + LSTM model, first align the dimensions of the outputs of CNN and LSTM, and then input the image sequence F t (TDM1) and the time series data of the vibration signal h t (VS1), where the calculation formula for dimension alignment is and In the formula, is represented as the result of CNN feature alignment for the image sequence F t (TDM1), is represented as the time series data h t (VS1) taking the hidden state at the last T time steps in LSTM, and T is the total number of time steps;

[0024] Fuse the outputs of the convolutional neural network CNN and the long short-term memory network LSTM through a fully connected layer to generate a unified feature vector. The output formula for the unified feature vector is In the formula, Z t is represented as the unified feature vector output by the CNN + LSTM model, σ represents the activation function, w0 represents the weight matrix, represents vector concatenation, and b0 represents the bias term;

[0025] According to the unified feature vector Z t , and define the classification and regression loss functions of the CNN + LSTM model to predict the classification result of the fused data, judge the probability of the defect type, or predict the numerical result of the fused data and judge the degree of structural abnormality.

[0026] Optionally, according to the fused data output by the CNN + LSTM model, further calculate the temperature distribution uniformity through the mean square error MSE, calculate the vibration amplitude through the peak-to-peak value of the vibration signal, and calculate the area of the defect region through an image segmentation algorithm to obtain the parameter values of the temperature distribution uniformity, vibration amplitude, and defect region area;

[0027] The calculation formula for the temperature distribution uniformity is and Wherein, U T represents the temperature distribution uniformity, m and M respectively represent the m-th pixel and the total number of M pixels, and I t (TDM1) m represents the temperature distribution image I after enhancement t The temperature value of the m-th pixel in (TDM1); represents the temperature distribution image I after enhancement t The average temperature of (TDM1);

[0028] The calculation formula for the vibration amplitude is A V = max(V t (VS1)) - min(V t (VS1)), wherein, A V represents the vibration amplitude, max(V t (VS1)) represents the maximum value of the vibration sequence of the filtered and denoised signal V t (VS1), min(V t (VS1) represents the minimum value of the vibration sequence of the filtered and denoised signal V t (VS1);

[0029] The calculation formula for the area of the defect region is and δ m = 0 or 1, wherein, S D represents the area of the defect region, m and M respectively represent the m-th pixel and the total number of M pixels, and δ m equals 1 when the m-th pixel is identified as a defect, otherwise 0, and as represents the actual area of a unit pixel;

[0030] According to the parameter values of the temperature distribution uniformity U T , the vibration amplitude A V and the area S of the defect region D , using the multi-parameter evaluation model to calculate the structural health index of the hydrogen-containing composite material, then the calculation formula for the structural health index is and α + β + γ = 1, wherein, SHI represents the structural health index, and α, β, and γ respectively represent the corresponding temperature distribution uniformity U T , the vibration amplitude A V and the area S of the defect region D The weight coefficients of, U T,max , A V,max , S D,max represent the maximum values in the theoretical limits, and, and are used for normalization to [0, 1] such that SHI ∈ [0, 1];

[0031] Judge the health status of the hydrogen-containing composite material according to the Structural Health Index (SHI) value, and divide it into three levels for evaluating low risk, medium risk, and high risk.

[0032] Optionally, the logical steps for obtaining the prediction results are as follows:

[0033] The unified feature vector Z obtained by data fusion t , is divided into input-output pairs according to the time series for supervised learning. The sliding window segmentation includes an input historical k-step data window as X = [Z t-k , Z t-k+1 , Z t-k+2 , …, Z t-1 ;

[0034] The predicted change trend of the material performance is output as Y1 = [Z t+1 , Z t+2 , …, Z t+k , where Z t+k represents the feature vector value for the next k steps;

[0035] The predicted service life of the material is output as Y2 = RUL t , where RUL t represents the number of steps to failure, that is, the remaining service life of the material;

[0036] When constructing the LSTM neural network model, stack the LSTM layer and the fully connected layer to extract the temporal pattern from the sliding window segmentation, output the hidden state, and let the fully connected layer map to the prediction target;

[0037] Through the LSTM model training and defining the loss function of the prediction target, output the prediction result. Among them, the calculation formula for the loss function of the predicted change trend of the material performance Y1 is and In the formula, loss1 represents the loss function of the predicted change trend of the material performance Y1, Y1 j represents the output value of the LSTM model predicting the future trend after j iterations, represents the average value of calculating the trend prediction Y1 by the mean squared error MSE;

[0038] The calculation formula for the loss function of the predicted service life of the material as Y2 is In the formula, loss2 represents the loss function of the predicted service life of the material as Y2, Y2 j represents the output value of the LSTM model predicting the future life after j iterations, represents the average value of calculating the life output value Y2 by the mean squared error MSE; j ;

[0039] Define the failure threshold Zt+k , and calculate the remaining time steps RULt from the current state to failure. Then the end - of - life condition is Y2 = argmin(RULt≥Z t+k ).

[0040] Optionally, the performance degradation and defect alarm threshold setting logic is as follows:

[0041] Set the alarm threshold according to three levels divided by the structural health index SHI value of the hydrogen - containing composite material. When SHI≥0.7, trigger a high - risk alarm, that is, an emergency state;

[0042] When 0.3≤SHI<0.7, trigger a medium - risk early warning, that is, a buffer state;

[0043] When SHI<0.3, do not trigger an alarm, that is, a low - risk normal state;

[0044] Only when a medium - risk early warning is triggered, use the LSTM model to output the predicted change trend Y1 of the material performance and the predicted service life Y2 of the material for further analysis of risk evolution. The prediction of the risk trend analysis, that is, the calculation formula of the risk trend slope is In the formula, Ka represents the risk trend slope of the prediction results Y1 and Y2, t represents the prediction result at the current t time step, and k represents the prediction result at the future k time step;

[0045] The calculation formula of the slope of the predicted curve of the structural health index SHI value is In the formula, Kb represents the risk trend slope of the SHI value, SHI t+k represents the structural health index value at the future k time step, SHI t represents the structural health index value at the current t time step;

[0046] Compare and analyze the risk trend slope Ka in the prediction results and the risk trend slope Kb in the SHI value. When Ka≥Kb, while triggering a medium - risk early warning, increase the alarm level. When Ka<Kb, while triggering a medium - risk early warning, lower the alarm level.

[0047] A computer device, comprising: a memory and a processor; the memory stores a computer program, and when the processor executes the computer program, it implements the steps of the nuclear dynamic defect detection method for hydrogen - containing composite materials used in automobile manufacturing according to any one of claims 1 to 7.

[0048] A computer - readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the nuclear dynamic defect detection method for hydrogen - containing composite materials used in automobile manufacturing according to any one of claims 1 to 7.

[0049] In the above technical solution, the technical effects and advantages provided by the present invention are as follows:

[0050] By integrating multi-modal data such as temperature distribution images and vibration signals, the present invention reduces the influence of environmental factors such as temperature and light on the detection results, enabling stable output of reliable detection results and improving the detection accuracy under complex working conditions of high-speed vehicle driving and large temperature differences between day and night. Moreover, by combining data from multiple sensors, it makes up for the deficiencies of single imaging detection methods and provides a more comprehensive and accurate assessment of the material state;

[0051] Using the long-term monitoring and prediction of the LSTM neural network model to analyze long-time series data, detect the long-term change trend of material properties, ensure timely discovery at the initial stage of performance degradation, significantly improve the detection accuracy and reliability, achieve a comprehensive assessment of material properties and safety, reduce unnecessary downtime, and lower maintenance costs;

[0052] And it also provides real-time alarms and scientific maintenance suggestions, greatly enhancing the safety and economy of the materials;

[0053] In summary, through the "perception - analysis - decision" closed-loop, the defect detection of hydrogen-containing composite materials has achieved a leap from "passive response" to "active prevention", providing a key technical guarantee for automotive lightweighting and the safe application of hydrogen energy. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0055] Figure 1 It is a flowchart of a nuclear dynamic defect detection method for hydrogen-containing composite materials used in automotive manufacturing.

[0056] Figure 2 It is a schematic structural diagram of an environmental protection integrated detection system. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0057] Now, the exemplary embodiments will be described more comprehensively with reference to the accompanying drawings. However, the exemplary embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these exemplary embodiments are provided so that the present disclosure will be more complete and comprehensive, and will fully convey the concept of the exemplary embodiments to those skilled in the art.

[0058] Embodiment 1

[0059] The present invention provides as Figure 1-2The nuclear dynamic defect detection method for hydrogen-containing composite materials used in automobile manufacturing shown below includes the following steps:

[0060] S1. Install an integrated detection system of a thermal imaging sensor, an acceleration sensor, a temperature sensor, and a sound sensor on the structure of the hydrogen-containing composite material for automobiles, and continuously collect sensor data. Among them, the sensor data includes a temperature distribution image, a vibration signal, temperature data, and a sound signal;

[0061] S2. Establish an environmental impact correction model, use temperature data to correct the temperature distribution image and sound signal to correct the vibration signal respectively to eliminate the interference of external environmental factors, and obtain an accurate temperature distribution image and vibration signal after correction;

[0062] S3. Integrate the corrected temperature distribution image and vibration signal through a data fusion algorithm to generate fusion data, detect and judge internal structural abnormalities of the material and locate surface defects, and use a multi-parameter evaluation model to comprehensively evaluate the performance and safety of the hydrogen-containing composite material;

[0063] S4. Use the LSTM neural network model to perform long-term monitoring and learning on the fusion data, predict the change trend of material performance and potential service life problems, and obtain a prediction result;

[0064] S5. Set performance degradation and defect alarm thresholds, compare and analyze according to the prediction results. When the detection and judgment results are abnormal, issue an alarm in real time and provide maintenance suggestions to ensure the safe use of the material.

[0065] Specifically, the integrated detection system includes a high-resolution thermal imaging camera, a high-sensitivity accelerometer, a high-precision temperature sensor, a high-fidelity microphone, a data acquisition device, and a cloud server. Among them, the thermal imaging camera acquires the surface temperature distribution image of the hydrogen-containing composite material, calibrated as TDM, to identify the hot spot area, timely detect overheating problems, and then identify the defect area on the surface of the hydrogen-containing composite material; the accelerometer collects the vibration signal of the hydrogen-containing composite material, calibrated as VS, to identify abnormal vibrations and evaluate the internal structure changes of the hydrogen-containing composite material; the temperature sensor collects the temperature data of the environment where the hydrogen-containing composite material is located, calibrated as TD, to detect the environmental temperature, assist in correcting the thermal imaging image, and improve the accuracy of temperature distribution evaluation; the microphone captures the sound signal emitted by the hydrogen-containing composite material in the operating environment, calibrated as SD, to monitor the operating noise, assist in correcting the vibration signal, and improve the judgment of the health status of the hydrogen-containing composite material; the data acquisition device is connected to the thermal imaging camera, accelerometer, temperature sensor, and microphone, and real-time collects, preprocesses, classifies, stores, and transmits sensor data, and uploads it to the cloud server for secure storage and subsequent analysis. During the acquisition and transmission of sensor data, by configuring the sampling rate and data transmission protocol of the data acquisition device, the real-time, synchronization, and accuracy of the data are ensured.

[0066] Furthermore, when the data acquisition device preprocesses the sensor data, through the preliminary processing of denoising and enhancing the contrast of the temperature distribution image, the accuracy of subsequent image analysis is improved. The vibration signal is filtered to remove irrelevant noise interference. The sound data is denoised, and the sound features related to the material state are extracted as the sound signal. Reliable and accurate sensor data is initially obtained, enabling the integrated detection system to provide multi-dimensional material state information, be able to monitor the state changes of the material in real time, detect material defects and performance degradation early, take timely measures to avoid potential safety accidents, and the integrated detection system is non-destructive and has low installation and maintenance costs, suitable for long-term monitoring.

[0067] Specifically, the environmental impact correction model includes the following steps:

[0068] According to the temperature distribution image TDM and temperature data TD measured by the thermal imaging sensor and temperature sensor respectively, a relationship model between the temperature distribution image and the environmental temperature is established, and the temperature distribution image is corrected by the environmental temperature data, that is, TDM1 = TDM - τ×(TD - T0).

[0069] In the formula, TDM represents the surface temperature distribution image of the hydrogen-containing composite material collected by the thermal imaging sensor, τ represents the correction coefficient of the temperature distribution image TDM affected by the ambient temperature TD, which is obtained by calibration in an ideal experimental environment, TDM1 represents the actual temperature distribution image after model correction, TD represents the temperature data of the environment where the hydrogen-containing composite material is located, and T0 represents the standard ambient temperature, that is, T0 = 20 °C;

[0070] According to the vibration signal VS and the sound signal SD measured by the acceleration sensor and the sound sensor respectively, a relationship model between the vibration signal and the ambient sound is established, and the vibration signal is corrected by the ambient sound signal, that is, VS1 = VS - δ × SD,

[0071] In the formula, VS1 represents the actual vibration signal after model correction, VS represents the vibration signal of the hydrogen-containing composite material collected by the acceleration sensor, δ represents the correction coefficient of the vibration signal VS affected by the ambient sound SD, which is obtained by calibration in an ideal experimental environment, and SD represents the sound signal emitted by the hydrogen-containing composite material in the operating environment.

[0072] Furthermore, by establishing the environmental impact correction model, the influence of temperature and noise on the hydrogen-containing composite material in the environment is corrected, which can effectively eliminate the interference of external environmental factors on the thermal imaging sensor and the acceleration sensor. Furthermore, accurate temperature distribution images and vibration signals can be obtained. The corrected temperature distribution image can more accurately identify the temperature anomaly area on the material surface, timely detect overheating problems, and prevent thermal damage. The corrected vibration signal can more accurately identify the structural anomalies of the material, such as looseness and damage, and take measures for repair in a timely manner, thereby improving the accuracy, precision and reliability of defect detection. After eliminating the influence of environmental factors, the detection results are more stable and reliable. Under different environmental conditions, the correction model can maintain the consistency of sensor data, making the long-term monitored data more meaningful, facilitating trend analysis, reducing the possibility of false detection and missed detection. Accurate temperature distribution images and vibration signals can help formulate a more reasonable maintenance plan, reduce unnecessary maintenance costs, and improve maintenance efficiency.

[0073] Specifically, the acquisition logic of the fused data is as follows:

[0074] Integrate the actual temperature distribution image TDM1 and the actual vibration signal VS1 obtained after correction by the environmental impact correction model using a multi-modal neural network model;

[0075] After image enhancement of the temperature distribution image TDM1, use CNN to extract the feature map, obtain the image sequence of the temperature distribution image, filter and denoise the vibration signal VS1, and then use LSTM to extract the temporal pattern to obtain the temporal data of the vibration signal. Among them, the calculation formula of the image sequence is F t (TDM1) = CNN(I t (TDM1); θ), where F t (TDM1) represents the feature map of the t-th frame image extracted from the enhanced image I t (TDM1) by the CNN model, I t (TDM1) represents the enhanced image obtained after the temperature distribution image TDM1 undergoes image enhancement processing, and θ represents the CNN model parameters;

[0076] The calculation formula of the temporal data is where h t (VS1) represents the LSTM hidden state at the t-th time step extracted from the filtered and denoised signal V t (VS1) by the LSTM model, V t (VS1) represents the filtered and denoised signal obtained after the vibration signal VS1 undergoes filtering and denoising processing, and h t-1 represents the LSTM output state at the previous t - 1 time step, represents the LSTM model parameters;

[0077] In the multi-modal neural network CNN + LSTM model, first align the dimensions of the outputs of CNN and LSTM, and then input the image sequence F t (TDM1) and the temporal data of the vibration signal h t (VS1), where the calculation formula for dimension alignment is and where represents the result of CNN feature alignment of the image sequence F t (TDM1), represents the hidden state of the temporal data h t (VS1) at the last T time steps in LSTM, and T is the total number of time steps;

[0078] Fuse the outputs of the convolutional neural network CNN and the long short-term memory network LSTM through the fully connected layer to generate a unified feature vector. Then the output formula of the unified feature vector is

[0079] where Z tDenoted as the unified feature vector output by the CNN+LSTM model, σ is denoted as the activation function, and w0 is denoted as the weight matrix. Denoted as vector concatenation, and b0 is denoted as the bias term.

[0080] According to the unified feature vector Z t , and define the classification and regression loss functions of the CNN+LSTM model to predict the classification result of the output fusion data, judge the probability of the defect type, or predict the numerical result of the output fusion data and judge the degree of structural abnormality.

[0081] Specifically, according to the fusion data output by the CNN+LSTM model, further calculate the temperature distribution uniformity through the mean square error MSE, calculate the vibration amplitude through the peak-to-peak value of the vibration signal, and calculate the area of the defect region through the image segmentation algorithm to obtain the parameter values of the temperature distribution uniformity, vibration amplitude, and defect region area.

[0082] Then the calculation formula for the temperature distribution uniformity is

[0083] And In the formula, U T Denotes the temperature distribution uniformity, m and M respectively denote the mth pixel and the total number of M pixels, and I t (TDM1) m Denotes the temperature distribution image I t (TDM1) after enhancement, and the temperature value of the mth pixel in it. Denotes the temperature distribution image I t (TDM1) after enhancement, and the average temperature of it.

[0084] The calculation formula for the vibration amplitude is A V = max(V t (VS1)) - min(V t (VS1)), in the formula, A V Denotes the vibration amplitude, max(V t (VS1)) denotes the maximum value of the vibration sequence of the filtered and denoised signal V t (VS1), and min(V t (VS1) denotes the minimum value of the vibration sequence of the filtered and denoised signal V t (VS1);

[0085] The calculation formula for the area of the defect region is And δ m = 0 or 1, in the formula, S D Denotes the area of the defect region, m and M respectively denote the mth pixel and the total number of M pixels, and δ mIt is denoted as 1 when the m-th pixel is recognized as a defect and 0 otherwise. as represents the actual area of a unit pixel;

[0086] According to the temperature distribution uniformity U T , vibration amplitude A V and the area S of the defect region D Based on the parameter values of, using the multi-parameter evaluation model to calculate the structural health index of the hydrogen-containing composite material, the calculation formula of the structural health index is and α + β + γ = 1. In the formula, SHI represents the structural health index, and α, β, and γ respectively represent the corresponding temperature distribution uniformity U T , vibration amplitude A V and the area S of the defect region D weight coefficients of, U Tmax , A Vmax , S Dmax represent the maximum values in the theoretical limits, and, and are used for normalization to [0, 1] such that SHI ∈ [0, 1];

[0087] Based on the value of the structural health index SHI, judge the health state of the hydrogen-containing composite material and divide it into three levels for evaluating low risk, medium risk, and high risk. The specific division is shown in the following table:

[0088] SHI value Risk level Health status Countermeasures 0 ≤ SHI < 0.3 Low risk Good Regular monitoring 0.3 ≤ SHI < 0.7 Medium risk Slight degradation Strengthen inspection and maintenance 0.7 ≤ SHI ≤ 1 High risk Severe damage Stop use for inspection and replacement .

[0089] Furthermore, through the multi-modal neural network CNN + LSTM model, the features of the temperature distribution image and the vibration signal can be effectively integrated to obtain more comprehensive material information, improve the accuracy and reliability of defect detection, and the CNN extracts image features, and the LSTM captures temporal features with strong expressive ability, which can automatically learn and extract complex defect features, reducing the workload of manual feature engineering. Then, according to the multi-parameter characteristics of comprehensively evaluating temperature, vibration, and defect area, comprehensively evaluate the performance and safety of the hydrogen-containing composite material, provide a scientific basis for maintenance decision-making, reduce the misjudgment rate, and at the same time can significantly improve the accuracy and sensitivity of defect detection, and discover tiny or hidden defects.

[0090] Specifically, the logical steps for obtaining the prediction result are as follows:

[0091] The unified feature vector Z obtained by data fusion t , is divided into input-output pairs according to the time series for supervised learning. The sliding window segmentation includes an input historical k-step data window as X = [Z t-k , Z t-k+1 , Z t-k+2 , …, Z t-1 ;

[0092] The predicted change trend of the output material performance is Y1 = [Z t+1 , Z t+2 , …, Z t+k , where Z t+k represents the feature vector value for the next k steps;

[0093] The predicted service life of the output material is Y2 = RUL t , where RUL t represents the number of steps to failure, that is, the remaining service life of the material;

[0094] When constructing the LSTM neural network model, stack the LSTM layer and the fully connected layer to extract the temporal pattern from the sliding window segmentation, output the hidden state, and let the fully connected layer map to the prediction target;

[0095] Through the LSTM model training and defining the loss function of the prediction target, the prediction result is output. Among them, the calculation formula of the loss function for the predicted change trend of the material performance Y1 is and In the formula, loss1 represents the loss function for the predicted change trend of the material performance Y1, and Y1 j represents the output value of the predicted future trend by the LSTM model after j iterations, represents the average value of the trend prediction Y1 calculated by the mean squared error MSE;

[0096] The calculation formula of the loss function for the predicted service life of the material Y2 is In the formula, loss2 represents the loss function for the predicted service life of the material Y2, and Y2 j represents the output value of the predicted future life by the LSTM model after j iterations, represents the average value of the life output value Y2 calculated by the mean squared error MSE j ;

[0097] Define the failure threshold Z t+k , and calculate the remaining time steps RULt from the current state to failure. Then the life termination condition is Y2 = argmin(RULt ≥ Z t+k ). Through the prediction results of the LSTM model for the change trend Y1 and service life Y2 of the material, the performance trend and life of the hydrogen-containing composite material can be judged, and then the dynamic early warning and life management of the material performance degradation can be realized, supporting the preventive maintenance decision-making.

[0098] Furthermore, by leveraging the memory capacity of LSTM, the fused data is further predicted and optimized. Combining dynamic data input, the prediction results are updated in real time to quickly respond to changes in the material state, capture the long-term change trends of the properties of hydrogen-containing composite materials, and monitor the service life of hydrogen-containing composite materials in real time. By adopting the long-term monitoring and learning technology of the LSTM model, the change trends of material properties and potential service life can be effectively predicted, providing guarantees for the safety and reliability of hydrogen-containing composite materials used in automobile manufacturing. In addition, based on the data-based prediction results, scientific bases can be provided for the selection, maintenance, and replacement of hydrogen-containing composite materials, improving the reliability of decision-making, and further ensuring the safety and efficiency of the automobile manufacturing process.

[0099] Specifically, the logic for setting the performance degradation and defect alarm thresholds is as follows:

[0100] Alarm thresholds are set according to the three levels divided by the Structural Health Index (SHI) value of the hydrogen-containing composite material. When SHI ≥ 0.7, a high-risk alarm is triggered, i.e., an emergency state, and the hydrogen-containing composite material used in the vehicle is immediately shut down, the defective area is comprehensively inspected, and the material is replaced or the structure is repaired.

[0101] When 0.3 ≤ SHI < 0.7, a medium-risk early warning is triggered, i.e., a buffer state, and the monitoring period for the performance and defects of the hydrogen-containing composite material is shortened, local non-destructive testing is carried out, and the workload is optimized.

[0102] When SHI < 0.3, no alarm is triggered, i.e., a low-risk normal state. Regular inspections of the hydrogen-containing composite material are required, data trends are recorded, and no immediate intervention is needed to ensure the intact service life and performance of the hydrogen-containing composite material.

[0103] Only when a medium-risk early warning is triggered, the LSTM model is used to output the predicted change trend Y1 of the material properties and the predicted service life Y2 of the material to further analyze the risk evolution and predict the risk trend analysis. That is, the calculation formula for the risk trend slope is In the formula, Ka represents the risk trend slope of the prediction results Y1 and Y2, t represents the prediction result at the current t time step, and k represents the prediction result at the future k time step.

[0104] The calculation formula for the slope of the prediction curve of the Structural Health Index (SHI) value is In the formula, Kb represents the risk trend slope of the SHI value, SHI t+k represents the Structural Health Index value at the future k time step, and SHI t represents the Structural Health Index value at the current t time step.

[0105] Compare and analyze the risk trend slope Ka in the prediction result and the risk trend slope Kb in the SHI value. When Ka ≥ Kb, trigger a medium-risk warning and increase the alarm level. When Ka < Kb, trigger a medium-risk warning and decrease the alarm level.

[0106] Furthermore, by setting the performance degradation and defect alarm thresholds, through comparative analysis based on the prediction results, implementing preventive maintenance measures based on the prediction results, extending the service life of the material, reducing unexpected shutdowns, intelligent warning and precise maintenance of the health status of hydrogen-containing composites can be achieved, ensuring the material is used within a safe range, reducing accident risks, ensuring the safety of personnel and equipment, avoiding major damages and reducing emergency repairs, reducing maintenance costs, ensuring safety and economy, and continuously collecting and analyzing data, optimizing the model, improving the prediction accuracy, and continuously improving the maintenance strategy.

[0107] Embodiment 2

[0108] This is the second embodiment of the present invention. The difference between this embodiment and the previous one is:

[0109] If the said function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs, etc., which can store program codes.

[0110] The logic and / or steps represented in the flowchart or described in other ways herein, for example, can be considered as a definite sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or in combination with these instruction execution systems, apparatus, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device.

[0111] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection (electronic device) having one or more wirings, a portable computer diskette (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable media can even be paper or other suitable media on which the program can be printed, as the program can be obtained electronically, for example, by optically scanning the paper or other media, then editing, interpreting, or otherwise processing it as appropriate, and then storing it in a computer memory.

[0112] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, the multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or a combination of the following techniques known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0113] As described above, the above are only specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed in the present application can easily think of changes or substitutions, which should all be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for detecting nuclear dynamic defects of hydrogen-containing composite materials for automobile manufacturing, characterized in that, It includes the following steps: S1. Install an integrated detection system of a thermal imaging sensor, an acceleration sensor, a temperature sensor, and a sound sensor on the structure of the hydrogen-containing composite material for automotive use, and continuously collect sensor data. Among them, the sensor data includes a temperature distribution image, a vibration signal, temperature data, and a sound signal; S2. Establish an environmental impact correction model, respectively use the temperature data to correct the temperature distribution image and the sound signal to correct the vibration signal, eliminate the interference of external environmental factors, and obtain an accurate temperature distribution image and vibration signal after correction; S3. Integrate the corrected temperature distribution image and vibration signal through a data fusion algorithm to generate fusion data, so as to detect and judge the internal structure abnormality of the material and locate the surface defect, and use a multi-parameter evaluation model to comprehensively evaluate the performance and safety of the hydrogen-containing composite material; S4. Use the LSTM neural network model to perform long-term monitoring and learning on the fusion data, predict the change trend of the material performance and potential service life problems, and obtain a prediction result; S5. Set the performance degradation and defect alarm thresholds, compare and analyze according to the prediction results. When the detection and judgment results are abnormal, send an alarm in real time and provide maintenance suggestions to ensure the safe use of the material.

2. The nuclear dynamic defect detection method for hydrogen-containing composite materials used in automobile manufacturing according to claim 1, characterized in that The integrated detection system includes a high-resolution thermal imaging camera, a high-sensitivity accelerometer, a high-precision temperature sensor, a high-fidelity microphone, a data acquisition device, and a cloud server. Among them, the thermal imaging camera obtains the surface temperature distribution image of the hydrogen-containing composite material, calibrated as TDM; the accelerometer collects the vibration signal of the hydrogen-containing composite material, calibrated as VS; the temperature sensor collects the temperature data of the environment where the hydrogen-containing composite material is located, calibrated as TD; the microphone captures the sound signal emitted by the hydrogen-containing composite material in the operating environment, calibrated as SD; the data acquisition device is connected to the thermal imaging camera, accelerometer, temperature sensor, and microphone, and collects, preprocesses, classifies, stores, and transmits the sensor data in real time, and uploads it to the cloud server for secure storage and subsequent analysis.

3. The nuclear dynamic defect detection method for hydrogen-containing composite materials used in automobile manufacturing according to claim 2, wherein, The environmental impact correction model includes the following steps: According to the temperature distribution image TDM and temperature data TD measured by the thermal imaging sensor and temperature sensor respectively, establish a relationship model between the temperature distribution image and the environmental temperature, and correct the temperature distribution image through the environmental temperature data, that is, TDM1 = TDM - τ×(TD - T0), In the formula, TDM represents the surface temperature distribution image of the hydrogen-containing composite material collected by the thermal imaging sensor, τ represents the correction coefficient of the temperature distribution image TDM affected by the environmental temperature TD, obtained by calibration in an ideal experimental environment, TDM1 represents the actual temperature distribution image after model correction, TD represents the temperature data of the environment where the hydrogen-containing composite material is located, and T0 represents the standard environmental temperature, that is, T0 = 20°C; According to the vibration signal VS and sound signal SD measured by the acceleration sensor and sound sensor respectively, establish a relationship model between the vibration signal and the environmental sound, and correct the vibration signal through the environmental sound signal, that is, VS1 = VS - δ×SD, Wherein, VS1 represents the actual vibration signal after model calibration, VS represents the vibration signal of the hydrogen-containing composite material collected by the acceleration sensor, δ represents the calibration coefficient of the vibration signal VS affected by the environmental sound SD, which is obtained by calibration in an ideal experimental environment, and SD represents the sound signal emitted by the hydrogen-containing composite material in the operating environment.

4. The nuclear dynamic defect detection method based on the hydrogen-containing composite material for automobile manufacturing according to claim 3, wherein, The acquisition logic of the fusion data is as follows: Integrate the actual temperature distribution image TDM1 and the actual vibration signal VS1 obtained after calibration by the environmental impact correction model using a multi-modal neural network model; After image enhancement of the temperature distribution image TDM1, CNN is used to extract the feature map to obtain the image sequence of the temperature distribution image. For the vibration signal VS1, filtering and denoising are performed, and then LSTM is used to extract the temporal pattern to obtain the temporal data of the vibration signal. Among them, the calculation formula of the image sequence is F t (TDM1) = CNN(I t (TDM1); θ), where F t (TDM1) represents the feature map of the t-th frame image extracted from the enhanced image I t (TDM1) by the CNN model. I t (TDM1) represents the enhanced image obtained after the temperature distribution image TDM1 undergoes image enhancement processing, and θ represents the CNN model parameters; The calculation formula for the time series data is where h t (VS1) represents the LSTM hidden state at the t-th time step extracted from the filtered and denoised signal V t (VS1), V t (VS1) represents the filtered and denoised signal obtained after the vibration signal VS1 is filtered and denoised, h t-1 represents the LSTM output state at the previous t-1 time step, represents the LSTM model parameters; In the multimodal neural network CNN+LSTM model, first align the dimensions of the outputs of CNN and LSTM, and then input the time series data of the image sequence F t (TDM1) and the vibration signal h t (VS1), where the calculation formula for dimension alignment is and In the formula, represents the result of CNN feature alignment for the image sequence F t (TDM1), represents the hidden state of the time series data h t (VS1) at the last T time steps in LSTM, where T is the total number of time steps; Fuse the outputs of the convolutional neural network (CNN) and the long short-term memory network (LSTM) through a fully connected layer to generate a unified feature vector. The output formula of the unified feature vector is In the formula, Z t represents the unified feature vector output by the CNN+LSTM model, σ represents the activation function, and w0 represents the weight matrix, represents vector concatenation, and b0 represents the bias term; According to the unified feature vector Z t , and define the classification and regression loss functions of the CNN+LSTM model to predict the classification results of the output fusion data, judge the probability of defect types, or predict the numerical results of the output fusion data and judge the degree of structural abnormality.

5. The nuclear dynamic defect detection method based on the hydrogen-containing composite material for automobile manufacturing according to claim 4, wherein, According to the fusion data output by the CNN+LSTM model, further calculate the temperature distribution uniformity through the mean square error MSE, calculate the vibration amplitude through the peak-to-peak value of the vibration signal, and calculate the area of the defect region through the image segmentation algorithm to obtain the parameter values of the temperature distribution uniformity, vibration amplitude, and defect region area; The calculation formula for the temperature distribution uniformity is as follows and where U T represents the temperature distribution uniformity, m and M respectively represent the m-th pixel and the total number of M pixels, and I t (TDM1) m represents the temperature distribution image I t (TDM1) after enhancement, and the temperature value of the m-th pixel in represents the temperature distribution image I t (TDM1) after enhancement; The calculation formula for the vibration amplitude is A V = max(V t (VS1)) - min(V t (VS1)), where A V represents the vibration amplitude, max(V t (VS1)) represents the maximum value of the vibration sequence of the filtered and denoised signal V t (VS1), and min(V t (VS1) represents the minimum value of the vibration sequence of the filtered and denoised signal V t (VS1); The calculation formula for the area of the defective area is and δ m = 0 or 1. In the formula, S D represents the area of the defective area, m and M respectively represent the m-th pixel and the total number of M pixels, δ m is equal to 1 when the m-th pixel is recognized as a defect, otherwise 0, and as represents the actual area of a unit pixel; According to the temperature distribution uniformity U T , vibration amplitude A V and the area S of the defect region D parameter values, using the multi-parameter evaluation model to calculate the structural health index of the hydrogen-containing composite material, the calculation formula of the structural health index is and α + β + γ = 1, where SHI represents the structural health index, and α, β, and γ respectively represent the corresponding temperature distribution uniformity U T , vibration amplitude A V and the area S of the defect region D weight coefficients, U T,max , A V,max , S D,max represent the maximum values in the theoretical limit, and and are used for normalization to [0, 1] so that SHI ∈ [0, 1]; Judge the health status of the hydrogen-containing composite material according to the structural health index SHI value, and divide it into three levels for evaluating low risk, medium risk, and high risk.

6. The nuclear dynamic defect detection method for hydrogen-containing composite materials used in automobile manufacturing according to claim 5, characterized in that, The acquisition logic steps of the prediction result are as follows: The unified feature vector Z obtained by data fusion t , is divided into input-output pairs according to the time series for supervised learning. The sliding window segmentation includes an input historical k-step data window as X = [Z t-k , Z t-k+1 , Z t-k+2 , …, Z t-1 ; The predicted change trend of the output material performance is Y1 = [Z t+1 , Z t+2 , …, Z t+k , where Z t+k represents the feature vector value for the next k steps; The predicted service life of the output material is Y2 = RUL t , RUL t is expressed as the number of steps to failure, that is, the remaining service life of the material; When constructing the LSTM neural network model, stack the LSTM layer and the fully connected layer to extract the time series pattern from the sliding window segmentation, output the hidden state, and let the fully connected layer map to the prediction target; By training the LSTM model and defining the loss function of the prediction target, the prediction result is output. Among them, the calculation formula of the loss function for predicting the change trend of material properties Y1 is and In the formula, loss1 represents the loss function for predicting the change trend of material properties Y1, and Y1 j represents the output value of predicting the future trend by iterating the LSTM model for j steps, represents the average value of calculating the trend prediction Y1 by the mean square error MSE; The calculation formula of the loss function for predicting the service life of the material as Y2 is In the formula, loss2 represents the loss function for predicting the service life of the material as Y2, and Y2 j represents the output value of predicting the future life by iterating j steps of the LSTM model, represents the average value of calculating the life output value Y2 by the mean square error MSE j ; Define the failure threshold Z t+k , and calculate the remaining time steps RULt from the current state to failure. Then the end-of-life condition is Y2 = argmin(RULt ≥ Z t+k ).

7. The nuclear dynamic defect detection method for hydrogen-containing composite materials used in automobile manufacturing according to claim 6, wherein The setting logic of the performance degradation and defect alarm threshold is as follows: Set the alarm threshold according to the three levels divided by the structural health index SHI value of the hydrogen-containing composite material. When SHI≥0.7, trigger a high-risk alarm, that is, an emergency state; When 0.3≤SHI<0.7, trigger a medium-risk early warning, that is, a buffer state; When SHI<0.3, no alarm is triggered, that is, a low-risk normal state; Only when the medium-risk early warning is triggered, use the LSTM model to output the predicted change trend Y1 of material properties and the predicted service life Y2 of materials for further analysis of risk evolution and prediction of risk trend analysis. That is, the calculation formula for the risk trend slope is In the formula, Ka represents the risk trend slope of the prediction results Y1 and Y2, t represents the prediction result at the current t time step, and k represents the prediction result at the future k time step; The calculation formula for the slope of the prediction curve of the structural health index SHI value is In the formula, Kb represents the risk trend slope of the SHI value, and SHI t+k represents the structural health index value at the future k time steps, and SHI t represents the structural health index value at the current t time step; Conduct a comparative analysis on the risk trend slope Ka in the prediction result and the risk trend slope Kb in the SHI value. When Ka≥Kb, while triggering a medium-risk early warning, increase the alarm level. When Ka<Kb, while triggering a medium-risk early warning, lower the alarm level.

8. A computer device, comprising: A memory and a processor; the memory stores a computer program, characterized in that: when the processor executes the computer program, the steps of the nuclear dynamic defect detection method based on the hydrogen-containing composite material for automobile manufacturing according to any one of claims 1 to 7 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, the steps of the nuclear dynamic defect detection method based on the hydrogen-containing composite material for automobile manufacturing according to any one of claims 1 to 7 are implemented.

Images