Model-based multi-source detection data fusion evaluation and evolution prediction method

Abstract

In order to improve the accuracy of performance detection and evaluation and improve the effectiveness of service state reliability evaluation, the invention aims to solve the problems that the existing evaluation method cannot obtain the evolution condition of a complex working condition, the prediction evaluation accuracy is poor, and the influence of a real state is not comprehensively considered. According to the model-based multi-source detection data fusion evaluation and evolution prediction method provided by the invention, the evaluation model is frequently corrected and trained by using the corresponding relationship between the actual physical test detection data and the load working condition, so that the strong association between the actual quality state parameters and the performance prediction evaluation is realized; evolutionary prediction of key parameters such as stress state, internal defect and morphology precision of a large-scale structure is realized, and accurate evaluation of service performance of a product with a complex structure is realized.

Description

Technical Field

[0001] This invention relates to a model-based method for multi-source detection data fusion evaluation and evolution prediction, belonging to the field of reliability evaluation and prediction. Background Technology

[0002] With increasingly stringent requirements for technical specifications, timelines, and costs from deep space exploration, manned spaceflight programs, next-generation remote sensing platforms, and next-generation communication satellite platforms, the need for ease of use and usability is becoming more urgent. This places higher demands on improving product testing capabilities and the accuracy of product performance evaluation and prediction. How to accurately assess structural performance and predict structural reliability under various load conditions using test results and actual service conditions during the ground development phase is crucial to ensuring the high-quality development of aerospace products.

[0003] Existing traditional testing and evaluation methods mainly verify the environmental adaptability of products by conducting environmental tests on the tested structure and conduct preliminary analysis of the structural performance evolution by conducting tests before and after the test. However, these methods cannot obtain information on the performance evolution of structural products outside of the test conditions, and the correlation between various test results is limited. Furthermore, there is a lack of effective analysis of the influence between various parameters under the actual state of the structural product, and the overall state of the structure cannot be comprehensively and accurately assessed. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art. The present invention provides a model-based multi-source detection data fusion evaluation and evolution prediction method to achieve accurate prediction and evaluation of key parameter changes such as the evolution of complex structural defects, stress and strain distribution changes, and surface accuracy changes, thereby greatly improving the accuracy of reliability evaluation and the effectiveness of stress assessment.

[0005] The technical solution adopted in this invention is: a model-based multi-source detection data fusion evaluation and evolution prediction method, comprising: Step 1: Select the three-dimensional feature markers on the structure to be measured, and obtain the coordinate information of the feature markers in the digital three-dimensional model of the structure to be measured; Step 2: Obtain the three-dimensional coordinate information of the global topography 3D model and solid features of the structure under test in the coordinate system of the topography measurement equipment; establish a correlation between the three-dimensional coordinate information obtained by the measurement system and the coordinate information of the feature markers in the digital 3D model, obtain the transformation matrix K, and realize the transformation between the coordinate system of the topography measurement equipment and the digital modeling coordinate system; Step 3: Detect the internal defect status of the structure under test, obtain defect information at all locations of the structure under test, set feature targets at the identified defect locations, and simultaneously record the size and depth direction information of the corresponding defects at each defect location. Step 4: Use the transformation matrix K to transform the coordinate information of the feature target at the defect location to obtain the coordinate information of the corresponding defect in the digital modeling coordinate system; at the same time, associate the defect size and depth information with the defect coordinates in the digital modeling coordinate system; further, use the transformation matrix K to register and fuse the obtained global topography 3D model of the measured structure with the digital 3D model and place it in the digital modeling coordinate system. In the digital modeling coordinate system, the digital 3D model is corrected based on the global topography 3D model of the actual detection; according to the defect coordinate information and the parameters of defect size and depth direction, the defect model is established in the corrected model, realizing the modeling optimization based on the actual state of the tested structure, and obtaining the real state model A of the tested structure. Step 5: Perform simulation analysis on the actual state model A of the tested structure. Apply the service environment conditions of the tested structure to the actual state model A of the tested structure, obtain simulation data of various performance parameters of the tested structure under different working conditions, including defects, morphology and stress-strain distribution. Establish the evolution relationship Z0 between various performance parameters and load condition F, and use the evolution relationship Z0 as training sample to construct a primary prediction model S0 of structural performance based on the convolutional neural network machine learning model. Step 6: Conduct a small-scale load environment pre-test on the structure under test (load magnitude is 10% of the maximum magnitude), and obtain the performance parameters and load condition F of the structure under test before and after the pre-test and during the test. z Response relationship S1; Step 7: Compare the acquired performance parameters with the load condition F z The response relationship S1 and the corresponding load condition F predicted in the primary predictive model S0. z The performance parameter Z is below Y Comparative analysis should be conducted to compare whether the deviations of the measured structural state parameters under the same working conditions and the same action time are all less than the predicted demand deviation β. The comparative working conditions and action time should adopt the arithmetic sampling method, and the number of samples for comparison should not be less than 5 groups. If the deviation of the state parameters of the tested structure is less than the expected demand deviation β of the tested structure under the same working conditions and the same action time, then proceed to step 9; if the deviation of the state parameters of the tested structure is not less than the expected demand deviation β of the tested structure under the same working conditions and the same action time, then proceed to step 8. Step 8: Analyze and correct the prediction model based on the deviation between the response relationship S1 and the current prediction model S0, using the maximum deviation |S1-Z| between the actual detection result and the prediction model result. Y | MAX The correction factor is the ratio F / F of the preload condition load parameters to the load parameters of the predicted model. z To adjust the scale, the results of the predictive model were corrected: ZYnew =|S1-Z Y | MAX ×F / F z + Z Y ; The response relationship data S1 obtained from the actual preliminary experiment was used as the ground truth data in the training data set and input into the training model to revise and obtain a new predictive model S. new And repeat step 7; Step 9: Perform load simulation on the structure under test, conduct a full-condition simulation verification test, and collect the performance parameters of the structure under test and the load condition F before, during, and after the test. ALL Response relationship S ALL ; Step 10: Compare the acquired performance parameters with the load condition F ALL Response relationship S ALL With predictive model S new The corresponding load condition F predicted by the analysis ALL The performance parameter Z is below YALL Comparative analysis should be conducted to compare whether the deviations of the structural state parameters of the test subjects under the same working conditions and the same action time are all less than the predicted demand deviation β. The comparative working conditions and action time should adopt the arithmetic sampling method, and the number of samples for comparison should not be less than 5 groups. If the deviation of the test structure state parameters is less than the product's predicted demand deviation β under the same working conditions and the same action time, proceed to step 12; if the deviation of the test structure state parameters is not less than the predicted demand deviation β under the same working conditions and the same action time, proceed to step 11. Step 11: Based on the response relationship S ALL Compared with the current predictive model S new The deviation is used to proportionally correct the predictive model, based on the ratio S between the actual detection results and the predictive model results. ALL / S new To correct the coefficients, the results of the performance parameter analysis for all operating conditions of the predicted model are corrected: Z YALLnew = S ALL / S new ×Z YALL ; The response relationship data S obtained from the actual preliminary experiment ALL The ground truth data, used as the training data set, is input into the training model to refine and obtain a new predictive model S. newf Considering the risk of fatigue damage to the tested structure due to frequent loading, the modified model is used as the final predictive evaluation model and proceeds to step 12. Step 12: The current predictive model is the final model, which predicts the actual physical state evolution of the structure under test. Under service conditions, the corresponding structural performance parameters are indexed in the predictive model according to the current service condition parameters.

[0006] The advantages of this invention compared to the prior art are: 1. This invention is a performance evaluation prediction method that takes into account the actual object characteristics of the structure under test. It can solve the problems of deviation between existing simulation analysis results and actual state, and the inability of detection and evaluation methods to be completely equivalent to service conditions. It can correct the analysis model according to the actual physical state of the structure and improve the accuracy of analysis.

[0007] 2. This invention adopts a predictive evaluation method based on actual test data. It uses the correspondence between actual physical test data and load conditions to frequently correct and train the evaluation model, thereby achieving a strong correlation between actual quality state parameters and performance predictive evaluation. This effectively solves the problems of large deviation between traditional simulation analysis predictive models and actual conditions, and insufficient accuracy.

[0008] 3. This invention adopts a predictive analysis model optimization and correction method of "small-scale pre-test loading cyclic training + large-scale test ratio correction". It effectively improves the predictive model through batch small-scale test results and avoids the risk of abnormal failure of products due to frequent high load tests by using a large-scale result correction method. It achieves effective convergence of predictive model deviation while ensuring the quality of the tested structural product. Attached Figure Description

[0009] Figure 1 This is a flowchart of the model-based multi-source detection data fusion evaluation and evolution prediction method of the present invention. Detailed Implementation

[0010] The present invention will be further described below with reference to the accompanying drawings.

[0011] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments disclosed herein will be described in further detail below with reference to the accompanying drawings.

[0012] like Figure 1 A model-based method for multi-source detection data fusion evaluation and evolution prediction is proposed, and the specific implementation steps are as follows: Step 1: Select the three-dimensional feature marks on the structure to be measured and obtain the coordinate information of the feature marks in the digital 3D model of the structure. Generally, these are protruding objects (such as surface screws, connecting terminals, etc.) or obviously recessed features (such as mounting holes, through holes, etc.). If there are no obvious three-dimensional feature marks, external marks can be used (such as pasting three-dimensional targets, etc.). The extended marks should be selected at more than 4 non-coplanar positions.

[0013] Step 2: Use topographic measurement equipment (such as a structured light scanning detector, a vision measurement system, etc.) to acquire the global topographic 3D model of the structure under test and the 3D coordinate information of its features in the topographic measurement system. Establish a correlation between the 3D coordinate information acquired by the measurement system and the coordinate information of the feature markers in the digital 3D model, obtain the transformation matrix K, and realize the transformation between the measurement coordinate system and the digital modeling coordinate system.

[0014] Step 3: Use defect detection equipment (such as infrared non-destructive testing instruments, ultrasonic non-destructive testing instruments, etc.) to detect the internal defect status of the structure under test, obtain defect information at all locations of the structure under test, set feature targets (attach optical targets or target mirrors, etc.) at the identified defect locations, and simultaneously record the size and depth direction of the corresponding defects at each defect location.

[0015] Step 4: Use the transformation matrix K to transform the coordinate information of the feature target at the defect location to obtain the corresponding defect coordinate information in the digital modeling coordinate system. Simultaneously, associate the defect size and depth information with the defect coordinates in the digital modeling coordinate system. Further, use the transformation matrix K to register and fuse the obtained global 3D model of the measured structure with the digital 3D model, placing them in the same coordinate system (digital modeling coordinate system).

[0016] In the digital modeling coordinate system, the digital model is corrected based on the actual detected 3D topographic model. Simultaneously, a defect model is established in the corrected model according to the defect coordinate information and parameters of defect size and depth direction. This realizes the modeling optimization based on the actual state of the tested structure and obtains the true state model A of the tested structure.

[0017] Step 5: Based on the actual state model A of the tested structure, perform simulation analysis, apply the service environment conditions of the test specimen (such as temperature environment simulation, mechanical load simulation, etc.) to the model, obtain simulation data of performance parameters such as defects, morphology and stress-strain distribution of the test specimen under different working conditions, establish the evolution relationship Z0 between various performance parameters and load condition F, and use the evolution relationship Z0 as training sample to construct a primary predictive model S0 of structural performance based on the convolutional neural network machine learning model.

[0018] Step 6: Conduct a small-scale load environment pre-test on the test structural specimen (the load magnitude is usually 10% of the maximum magnitude), and obtain the performance parameters and load condition F of the test structure before and after the pre-test and during the test. z The response relationship S1.

[0019] Step 7: Compare the acquired performance parameters with the load condition F z The response relationship S1 and the corresponding load condition F predicted in the primary predictive model S0. z The performance parameter Z is belowY Conduct comparative analysis to check whether the deviations of the structural product state parameters under the same working conditions and duration are all less than the product's predicted demand deviation β (usually 5% of the measured structure). The comparative working conditions and durations should use an arithmetic progression sampling method, and the number of comparative samples should be no less than 5 groups.

[0020] If all values ​​are less than the product's projected demand deviation β, proceed to step 9; otherwise, proceed to step 8.

[0021] Step 8: Analyze and correct the prediction model based on the deviation between the response relationship S1 and the current prediction model S0, using the maximum deviation between the actual detection result and the prediction model result (|S1-Z) as the reference. Y | MAX The correction factor is the ratio of the preload condition load parameters to the load parameters of the predicted model (F / F). z To adjust the scaling ratio, the results of the predictive model are corrected. Z Ynew =|S1-Z Y | MAX ×F / F z + Z Y ; Simultaneously, the response relationship data S1 obtained from the actual pre-experiment is used as the ground truth data in the training data set and input into the training model to revise and obtain a new predictive model S. new And repeat step 7; Step 9: Simulate loading the structure under normal load conditions and conduct a full-condition simulation verification test. Collect the performance parameters of the structure under test and the load condition F before, during, and after the test. ALL Response relationship S ALL .

[0022] Step 10: Compare the acquired performance parameters with the load condition F ALL Response relationship S ALL With predictive model S new The corresponding load condition F predicted by the analysis ALL The performance parameter Z is below YALL Conduct comparative analysis to check whether the deviations of the structural product state parameters under the same working conditions and duration are all less than the product's predicted demand deviation β (usually 5% of the measured structure). The comparative working conditions and durations should use an arithmetic progression sampling method, and the number of comparative samples should be no less than 5 groups.

[0023] If all values ​​are less than the product's projected demand deviation β, proceed to step 12; otherwise, proceed to step 11.

[0024] Step 11: Based on the response relationship S ALL Compared with the current predictive model S newThe deviation is used to proportionally correct the prediction model, based on the ratio of the actual detection results to the prediction model results (S). ALL / S new The factor is a correction factor used to adjust the performance parameter analysis results for all operating conditions of the predicted model. Z YALLnew = S ALL / S new ×Z YALL ; Simultaneously, the response relationship data S obtained from the actual preliminary experiment will be used. ALL The ground truth data, used as the training data set, is input into the training model to refine and obtain a new predictive model S. newf Considering the risk of fatigue damage to the tested structure due to frequent loading, the modified model is used as the final predictive evaluation model and proceeds to step 12. Step 12: The current prediction model is the final model, which can effectively predict the actual physical state evolution of the tested structure. Under service conditions, the corresponding structural performance parameters can be indexed in the prediction model according to the current operating conditions parameters.

[0025] The parts of this invention not described in detail are well-known to those skilled in the art.

Claims

1. A model-based method for multi-source detection data fusion evaluation and evolution prediction, characterized in that, include: S1: Select the three-dimensional feature markers on the structure under test and obtain the coordinate information of the three-dimensional feature markers in the digital three-dimensional model of the structure under test; S2: Obtain the three-dimensional coordinate information of the global topography 3D model and solid feature marks of the structure under test in the coordinate system of the topography measurement equipment; establish a correlation between the three-dimensional coordinate information of the solid feature marks in the coordinate system of the topography measurement equipment and the coordinate information of the solid feature marks in the digital 3D model, and obtain the transformation matrix K; S3: Detect the internal defect status of the structure under test, obtain defect information at all locations of the structure under test, set feature targets at the defect locations, and simultaneously record the size and depth direction of the corresponding defects at each defect location. S4: Perform modeling optimization based on the actual state of the structure under test to obtain the true state model A of the structure under test; S5: Construct a structural performance prediction model S0 based on a convolutional neural network machine learning model; S6: Conduct a small-scale load environment pre-test on the structure under test, and obtain the performance parameters and load condition F of the structure under test before and after the pre-test and during the test. z Response relationship S1; S7: Compare the acquired performance parameters with load condition F z The response relationship S1 and the corresponding load condition F predicted in the primary predictive model S0. z The performance parameter Z is below Y A comparative analysis was conducted to compare whether the deviations of the measured structural state parameters under the same working conditions and the same duration of action were all less than the predicted demand deviation β. S8. If the deviation of the state parameters of the measured structure is less than the predicted demand deviation β of the measured structure under the same working conditions and the same action time, then proceed to S9. S9: Perform load simulation on the structure under test, conduct a full-condition simulation verification test, and collect the performance parameters and load condition F of the structure under test before, during, and after the test. ALL Response relationship S ALL ; S10: Compare the acquired performance parameters with load condition F ALL Response relationship S ALL With predictive model S new The corresponding load condition F predicted by the analysis ALL The performance parameter Z is below YALL A comparative analysis was conducted to compare whether the deviations of the state parameters of the tested structures under the same working conditions and the same duration of action were all less than the predicted demand deviation β. S11. If the deviation of the state parameters of the tested structure is less than the product's predicted demand deviation β under the same working conditions and the same action time, then proceed to S12. S12: The current predictive model is the final model, predicting the actual physical state evolution of the tested structure. Under service conditions, the corresponding structural performance parameters are indexed in the predictive model based on the current operating condition parameters.

2. The model-based multi-source detection data fusion evaluation and evolution prediction method according to claim 1, characterized in that, In step S4, the step of performing modeling optimization based on the actual state of the structure under test to obtain the true state model A of the structure under test includes: The coordinate information of the feature target at the defect location is transformed using the transformation matrix K to obtain the coordinate information of the corresponding defect in the digital modeling coordinate system; the size and depth information of the defect are associated with the defect coordinates in the digital modeling coordinate system. The obtained three-dimensional model of the global topography of the measured structure is registered and fused with the digital three-dimensional model by the transformation matrix K, and then placed in the digital modeling coordinate system. In the digital modeling coordinate system, the digital 3D model is corrected based on the global topography 3D model of the actual detection; according to the defect coordinate information and the parameters of defect size and depth direction, a defect model is established in the corrected model, realizing the modeling optimization based on the actual state of the tested structure, and obtaining the real state model A of the tested structure.

3. The model-based multi-source detection data fusion evaluation and evolution prediction method according to claim 2, characterized in that, In step S5, constructing a structural performance prediction model S0 based on a convolutional neural network machine learning model includes: Simulation analysis is performed on the actual state model A of the tested structure. Service environment conditions of the tested structure are applied to the actual state model A of the tested structure. Simulation data of various performance parameters of the tested structure under different working conditions are obtained. The evolution relationship Z0 between various performance parameters and load condition F is established. Using the evolution relationship Z0 as training sample, a primary prediction model S0 of structural performance is constructed based on a convolutional neural network machine learning model.

4. The model-based multi-source detection data fusion evaluation and evolution prediction method according to claim 3, characterized in that, The performance parameters include defects, morphology, and stress-strain distribution.

5. The model-based multi-source detection data fusion evaluation and evolution prediction method according to claim 4, characterized in that, In S6, the load magnitude in the small-scale load environment loading pre-test is 10% of the maximum magnitude.

6. The model-based multi-source detection data fusion evaluation and evolution prediction method according to claim 5, characterized in that, In S7, the comparison of working conditions and action time adopts the arithmetic sampling method, and the number of comparison samples should be no less than 5 groups.

7. The model-based multi-source detection data fusion evaluation and evolution prediction method according to claim 6, characterized in that, S8 also includes: If the deviations of the state parameters of the measured structure under the same working conditions and the same duration are not all less than the predicted deviation β of the measured structure, then: The prediction model is revised based on the deviation between the response relationship S1 and the current prediction model S0, using the maximum deviation |S1-Z| between the actual detection result and the prediction model result. Y | MAX The correction factor is the ratio F / F of the preload condition load parameters to the load parameters of the predicted model. z To adjust the proportions, the results of the predictive model were corrected; The response relationship data S1 obtained from the actual preliminary experiment was used as the ground truth data in the training data set and input into the training model to revise and obtain a new predictive model S. new , and repeat S7.

8. The model-based multi-source detection data fusion evaluation and evolution prediction method according to claim 7, characterized in that, The formula for correcting the results of the predictive model is: Z Ynew =|S1-Z Y | MAX ×F / F z + Z Y 。 9. The model-based multi-source detection data fusion evaluation and evolution prediction method according to claim 8, characterized in that, In step S10, the comparison of working conditions and action time adopts the arithmetic sampling method, and the number of comparison samples should be no less than 5 groups.

10. The model-based multi-source detection data fusion evaluation and evolution prediction method according to claim 9, characterized in that, S11 also includes: If the deviations of the state parameters of the tested structure under the same working conditions and the same duration are not all less than the predicted demand deviation β, then: According to the response relationship S ALL Compared with the current predictive model S new The deviation is used to proportionally correct the predictive model, based on the ratio S between the actual detection results and the predictive model results. ALL / S new To correct the coefficients, the results of the performance parameter analysis for all operating conditions of the predicted model are corrected: Z YALLnew = S ALL / S new ×Z YALL ; The response relationship data S obtained from the actual preliminary experiment ALL The ground truth data, used as the training data set, is input into the training model to refine and obtain a new predictive model S. newf The revised model is used as the final predictive evaluation model, and then proceeds to S12.

Images