Civil aircraft bulkhead structure crack repair decision method fusing physical information and case data

Abstract

The invention discloses a civil aircraft bulkhead structure crack repair decision-making method fusing physical information and case data, and solves the problem that bulkhead structure crack repair is difficult to make decisions due to the fact that a structure repair manual is not mature enough and repair cases are not accumulated enough at the initial stage of civil aircraft operation. A civil aircraft bulkhead structure crack repair historical case library is constructed through integrated learning of bulkhead structure crack repair cases of current mainstream aircrafts from manufacturers, operators and maintenance units. Selecting a test piece made of the same material as the civil aircraft bulkhead structure, performing a constant-amplitude fatigue crack propagation experiment according to the actual operation load spectrum, and recording the crack size and the corresponding load cycle number in detail. And selecting a PINN deep learning model, adding a Paris law as a model constraint, and embedding physical information into the network through a composite loss function to realize the prediction of the residual life of the crack. Structured data, text data and image data are analyzed by using pre-training models of PINN, BERT and ViT respectively, on this basis, multi-modal and multi-dimensional similarity among cases is analyzed by using a twin architecture, and a similar case repair scheme is matched for new bulkhead structure crack damage, so that a case-based civil aircraft bulkhead structure crack repair decision is realized. According to the method, accurate and reliable repair case matching is realized, the civil aircraft bulkhead structure repair decision is supported, and the decision efficiency and the commercial competitiveness are improved.

Description

Technical Field

[0001] The present invention relates to the field of civil aircraft operation support technology, and in particular to a civil aircraft bulkhead structure crack repair decision method integrating physical information and case data. Background Art

[0002] At present, the repair of civil aircraft bulkhead structures is mainly carried out according to the structural repair manual. However, after the newly developed aircraft is put into use, the manual usually has a long maturity period. Therefore, in the initial operation stage, there will be a large number of bulkhead structure crack repair tasks that exceed the manual, and engineers need to comprehensively consider the aircraft design and material properties, damage assessment, aircraft operation records, operating environment, maintenance resources and capabilities to formulate repair plans. The repair plan for this type of repair task is difficult to formulate, highly dependent on experience, and has a long plan approval cycle, which seriously affects the efficiency of repair decision-making and leads to low availability of civil aircraft in operation. Therefore, it is urgent to develop a method that can fully utilize domain knowledge and current mainstream aircraft bulkhead structure crack repair cases to achieve rapid response and efficient decision-making for civil aircraft bulkhead structure crack repair.

[0003] Deep learning has become an important research tool in the field of engineering technology in recent years due to its good knowledge embedding and data mining capabilities. This makes it possible to use physical information and case data of structural repair to make decisions on crack repairs in civil aircraft structures. The physical information neural network model is a deep learning model that imposes physical information constraints in model training. It can embed fracture mechanics physical information on the basis of making full use of case structured data to accurately predict the remaining life of cracks in bulkhead structures. At the same time, the pre-trained model of text and images can fully extract the characteristics of case text and image data when the sample size is limited. By processing and analyzing physical information and case data, repair plans are matched for bulkhead structure crack repair tasks, supporting civil aircraft bulkhead structure crack repair decisions, and improving decision-making efficiency and commercial competitiveness of civil aircraft. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a decision-making method for crack repair of a civil aircraft bulkhead structure by integrating physical information and case data, so as to shorten the repair plan formulation cycle and improve the efficiency of repair decision-making.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A civil aircraft bulkhead structure crack repair decision method integrating physical information and case data, characterized in that it comprises the following steps:

[0007] Step 1: Integrate and learn the crack repair cases of bulkhead structures of current mainstream aircraft models to build a historical case library of crack repair of bulkhead structures of civil aircraft;

[0008] Step 2: According to the civil aircraft bulkhead structure, select the same material crack tensile test piece, simulate the fatigue load spectrum in actual operation, conduct a constant amplitude fatigue crack growth experiment, and record the crack length and corresponding load cycle number in detail;

[0009] Step 3: Select the PINN deep learning model, input the experimental data recorded in step 2 and the case structured data collected in step 1 into the model, add the Paris law as a model constraint, embed the physical information into the network through the composite loss function, and realize the prediction of the remaining life of the crack;

[0010] Step 4: Select the pre-trained BERT model, input the case text data collected in step 1 into the model for fine-tuning, and extract the features of the case text data;

[0011] Step 5: Select the pre-trained ViT model, input the case image data collected in step 1 into the model for fine-tuning, and extract the features of the case image data;

[0012] Step 6. Based on the analysis results of steps 3, 4, and 5, use the twin architecture to implement multi-modal and multi-dimensional similarity analysis between cases, match similar case repair plans for new bulkhead structure crack damage, and make case-based civil aircraft bulkhead structure crack repair decisions.

[0013] In step 1 of the present invention, the case data of crack repair of bulkhead frame structures of current mainstream aircraft models are integrated and learned from manufacturers, operators and maintenance units, and the independence and correlation analysis between indicators is performed through SVM, representative indicators are retained, and redundant data are eliminated, so as to construct a case library of crack repair of bulkhead frame structures of civil aircraft containing structured data, text data and image data;

[0014] The data learned from the manufacturer is mainly structured data, which is the initial data of various civil aircraft determined by design and manufacturing, including: design and manufacturing data, geometry and size data, flight performance data, structural health monitoring data, material property data, damage tolerance and fatigue analysis data, and technical publications;

[0015] The data learned from the operator includes structured data, image data and text data, which are data generated by the operator during the operation of the aircraft, including: aircraft operation data, aircraft repair records, specific damage information, damage location information, damage discovery records, and damage location images;

[0016] The data learned from the maintenance unit is mainly text data, involving specific repair information of crack damage in the bulkhead structure, including: description of the repair method, specific repair plan, repair materials, bonding process, surface treatment and protective measures, and inspection requirements after the repair is completed.

[0017] In step 2 of the present invention, a test piece made of the same material as the civil aircraft bulkhead structure is selected, the applied load is determined according to the actual operating environment and frequency, a fatigue crack growth experiment is performed, and the crack length and the corresponding number of load cycles are recorded.

[0018] In step 3 of the present invention, a fully connected PINN neural network model is established, the Paris law of crack propagation in fracture mechanics is used as a model constraint, and physical information is embedded in the network through a composite loss function; experimental record data and structured data in repair cases are used as input, and the output is the remaining life of the crack, and the relationship between the crack damage and fatigue cycle of the civil aircraft bulkhead structure is established, and the network gradient is solved by automatic differentiation to adjust the network weight;

[0019] The Paris law of crack extension is specifically expressed as:

[0020]

[0021] Where a is the crack length, n is the number of load cycles, and C and m are material-related parameters;

[0022] The composite loss function is specifically expressed as:

[0023] Loss = w Data Loss Data +w Physics Loss Physics

[0024] In the formula, Loss is the PINN network loss function, Loss Data For the structured data loss part of the case, Loss Physics is the physical information loss part, w Data and w Physics are the weight coefficients of the structured data loss part and the physical information loss part of the case respectively.

[0025] In step 4 of the present invention, a BERT neural network model is established, and the model is initialized using the bert-base-chinese pre-trained weights provided by Google. In the training stage, the MLM training task is constructed using the text data in the case of crack repair of the civil aircraft frame structure, and the model is fine-tuned to obtain a language model with the characteristics of the field of crack repair of the civil aircraft frame structure; in the inference stage, the BERT model is used to vectorize the input text and extract the characteristics of the case text data;

[0026] The MLM training task is an unsupervised learning task. The model replaces the segmentation units in the text with a probability of 15%, 80% of the units are replaced by masks, 10% of the units are replaced by any other unit, and the remaining 10% remain unchanged.

[0027] Step 5 of the present invention is to establish a ViT neural network model, and initialize the model using the training weights obtained by training on a civil aircraft bulkhead structure crack detection image dataset. In the inference stage, the ViT model is used to perform multi-level feature extraction on the input image to extract case image data features.

[0028] In step 6 of the present invention, a twin architecture model with two branch networks is established, and each branch network is composed of the three neural network models in steps 4, 5, and 6; the two branches are used to perform multi-modal and multi-dimensional processing and analysis on the new bulkhead structure crack damage cases and the cases in the historical case library, respectively, and the two branch outputs are subjected to similarity analysis; the new bulkhead structure crack damage cases are subjected to similarity analysis with the cases in the historical case library one by one, and the repair plan of the most similar case is matched to make a case-based decision on the repair of the civil aircraft bulkhead structure cracks.

[0029] Aiming at practical engineering problems that need to be solved urgently in the initial operation stage of civil aircraft, the present invention discloses an accurate, reliable and highly interpretable system solution, which uses deep learning models to embed fracture mechanics knowledge and mine repair case data, and integrates physical information and case data to realize multi-modal and multi-dimensional similarity analysis between repair cases, achieves accurate matching of repair cases, effectively shortens the repair plan formulation cycle, improves the decision-making efficiency of crack repair of civil aircraft bulkhead structure, and effectively supports the commercial operation of civil aircraft. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A schematic flow chart of a method for making decisions on crack repair in a civil aircraft bulkhead structure that integrates physical information and case data provided by the present invention. DETAILED DESCRIPTION

[0031] The following will be combined with the accompanying drawings of the present invention to clearly and completely describe the technical solution of the present invention. However, this description should not be considered as a limitation of the present invention, but should be understood as a more detailed description of certain features and technical solutions of the present invention.

[0032] like Figure 1 As shown, it is a schematic diagram of a process flow of a decision-making method for repairing cracks in a civil aircraft bulkhead structure that integrates physical information and case data provided by the present invention. The method works in the following steps:

[0033] S1. Integrate and learn the current mainstream aircraft frame structure crack repair case data from manufacturers, operators and maintenance units, and use SVM to analyze the independence and correlation between indicators, retain representative indicators, eliminate redundant data, and build a civil aircraft frame structure crack repair case library containing structured data, text data and image data. The specific steps include:

[0034] S11. The data learned from manufacturers are mainly structured data, which are the initial data of various civil aircraft determined by design and manufacturing, including: design and manufacturing data, geometry and size data, flight performance data, structural health monitoring data, material property data, damage tolerance and fatigue analysis data, and technical publications;

[0035] S12. The data learned from the operator includes structured data, image data and text data, which are data generated by the operator during the operation of the aircraft, including: aircraft operation data, aircraft repair records, specific damage information, damage location information, damage discovery records, and damage location images;

[0036] S13. The data learned from the maintenance unit is mainly text data, involving specific repair information of crack damage in the bulkhead structure, including: description of the repair method, specific repair plan, repair materials, bonding process, surface treatment and protective measures, and inspection requirements after the repair is completed;

[0037] S14. Use a binary form to represent the repair case of cracks in the frame structure of a civil aircraft. The specific definition is:

[0038] Case = {Problem, Solution};

[0039] Case represents a complete civil aircraft bulkhead structure crack repair case, C = {C1, C2, C3, ..., C i , …, C n} represents a collection of repair cases, that is, a case library of crack repairs for civil aircraft bulkhead structures. Among them, n represents the number of cases, C i represents the i-th case in the case library C, and i∈{1, 2, 3, …, n};

[0040] Problem is a description of the repair case represented by Case. i ={S i1 , S i2 , S i3 , …, S ij , …, S im} represents the index set of the i-th repair case, where m represents the number of case indicators, S ij represents the jth indicator in the i-th repair case, and j∈{1, 2, 3, …, m};

[0041] Solution is the specific repair plan of the repair case represented by Case, A={A1,A2,A3,…,A i , …, A n} represents a set of repair solutions. i Represents the repair solution for the i-th case in the case library C.

[0042] S2. According to the civil aircraft bulkhead structure, the same material crack tensile test piece was selected to simulate the fatigue load spectrum in actual operation, and the equal amplitude fatigue crack growth test was carried out. The crack length and the corresponding load cycle number were recorded, and 200 data points were recorded for each experiment. Among them, the selected test piece was 50 mm wide and 12 mm thick, and a sinusoidal load was used for loading, with a maximum load of 4.5 kN, a minimum load of 0.9 kN, and a frequency of 20 Hz.

[0043] S3. Establish a fully connected PINN neural network model. The input is the experimental record data and the structured data in the repair case. The output is the remaining life of the crack. The number of fully connected layers of the neural network is set to [15, 30, 30, 1], which means that the network structure of PINN is 5 layers, and the number of nodes in each layer is 15, 30, 30, and 1 respectively. The Paris law of crack propagation in fracture mechanics is used as a model constraint. The physical information is embedded in the network through a composite loss function. The Paris law of crack propagation is specifically expressed as:

[0044]

[0045] Where a is the crack length, n is the number of load cycles, and C and m are material-related parameters;

[0046] The composite loss function is specifically expressed as:

[0047] Loss = w Data Loss Data +w Physics Loss Physics ;

[0048] Loss is the PINN network loss function, Loss Data For the structured data loss part of the case, Loss Physics is the physical information loss part, w Data and w Physics are the weight coefficients of the structured data loss part and the physical information loss part of the case respectively.

[0049] S4. Establish a BERT neural network model to extract case text data features. The specific steps include:

[0050] S41. Preprocessing the text data of the crack repair case of the civil aircraft bulkhead structure, including removing useless symbols, deleting irrelevant symbols, spaces, numbers, labels, letters and punctuation marks in the text, removing stop words and segmenting the words;

[0051] S41. Build a BERT neural network model with 12 Transformer modules, 768 hidden layers, 12 heads of multi-head attention modules, and 64 maximum input sentences.

[0052] S42. Use the publicly available BERT model weights pre-trained on a large amount of Chinese text to initialize the local model parameters. The model weights are the bert-base-chinese pre-trained weights provided by Google.

[0053] S43. Use the pre-processed civil aircraft bulkhead structure crack repair case text data to construct an MLM training task. The MLM training task is an unsupervised learning task. The model replaces the segmentation units in the text with a probability of 15%, 80% of the units are replaced with masks, 10% of the units are replaced with any other units, and the remaining 10% remain unchanged; use the constructed MLM training data set to fine-tune the pre-trained model to obtain a language model with the characteristics of the civil aircraft bulkhead structure crack repair field;

[0054] S44. Use the fine-tuned model for reasoning, input the case text into the model, extract the case text data features, and obtain a feature vector sequence.

[0055] S5. Establish a ViT neural network model to extract case image data features. The specific steps include:

[0056] S51. Preprocess the image data of the civil aircraft bulkhead structure crack repair case, rotate and scale the image to keep the crack acquisition scale and angle consistent; remove image noise through mean filtering to eliminate obvious noise points; locally enhance the image to highlight the outline of the bulkhead structure crack in the image and improve the contrast.

[0057] S52. Establish a ViT neural network model, use the training weights obtained by training on a civil aircraft bulkhead structure crack detection image dataset to initialize the model, and use the ViT model to extract features from the input image.

[0058] S6. Use the twin architecture to analyze the multi-modal and multi-dimensional similarities between cases, match similar case repair solutions for new bulkhead structure crack damage, and thus realize case-based civil aircraft bulkhead structure crack repair decision-making. The specific steps include:

[0059] S61. Establish a twin architecture network model Siamese-PINN-BERT-ViT with two branch networks. Each branch network is composed of three neural network models in S4, S5, and S6 in parallel. The two branch networks share the model structure and weight parameters.

[0060] S62. Input new crack damage cases of bulkhead structures and cases in the historical case library into the two branch networks respectively. Each branch network splices the extracted different modal features. The features extracted by the two branch networks are calculated by cosine similarity to obtain the similarity analysis results between the cases. The specific similarity calculation method is expressed as:

[0061]

[0062] X and Y are the feature vectors output by the two branch networks respectively;

[0063] S63. Perform similarity analysis on new cases of bulkhead structure crack damage with cases in the historical case library one by one, match the repair plan of the most similar case, and repair the civil aircraft bulkhead structure with reference to the repair plan of the matched case, so as to realize the repair decision of the civil aircraft bulkhead structure crack based on physical information and case data.

[0064] The present invention discloses a crack repair decision method for a civil aircraft bulkhead structure that integrates physical information and case data. The above is only a preferred implementation of the present invention. It should be pointed out that for ordinary technicians in this technical field, several deformations and combinations can be made without departing from the principle of the present invention. These deformations and combinations should also be regarded as the protection scope of the present invention.

Claims

1. A decision-making method for crack repair of civil aircraft bulkhead structure integrating physical information and case data, characterized in that The following steps are involved: Step 1: Integrate and learn the crack repair cases of bulkhead structures of current mainstream aircraft models to build a historical case library of crack repair of bulkhead structures of civil aircraft; Step 2: According to the civil aircraft bulkhead structure, select the same material crack tensile test piece, simulate the fatigue load spectrum in actual operation, conduct a constant amplitude fatigue crack growth experiment, and record the crack length and corresponding load cycle number in detail; Step 3: Select the PINN deep learning model, input the experimental data recorded in step 2 and the case structured data collected in step 1 into the model, add the Paris law as a model constraint, embed the physical information into the network through the composite loss function, and realize the prediction of the remaining life of the crack; Step 4: Select the pre-trained BERT model, input the case text data collected in step 1 into the model for fine-tuning, and extract the features of the case text data; Step 5: Select the pre-trained ViT model, input the case image data collected in step 1 into the model for fine-tuning, and extract the features of the case image data; Step 6. Based on the analysis results of steps 3, 4, and 5, use the twin architecture to implement multi-modal and multi-dimensional similarity analysis between cases, match similar case repair plans for new bulkhead structure crack damage, and make case-based civil aircraft bulkhead structure crack repair decisions.

2. The civil aircraft bulkhead structure crack repair decision method integrating physical information and case data according to claim 1 is characterized in that: The step 1 integrates and learns the current mainstream aircraft frame structure crack repair case data from manufacturers, operators and maintenance units, and uses SVM to perform independence and correlation analysis between indicators, retain representative indicators, and eliminate redundant data, thereby building a civil aircraft frame structure crack repair case library containing structured data, text data and image data; The data learned from the manufacturer is mainly structured data, which is the initial data of various civil aircraft determined by design and manufacturing, including: design and manufacturing data, geometry and size data, flight performance data, structural health monitoring data, material property data, damage tolerance and fatigue analysis data, and technical publications; The data learned from the operator includes structured data, image data and text data, which are data generated by the operator during the operation of the aircraft, including: aircraft operation data, aircraft repair records, specific damage information, damage location information, damage discovery records, and damage location images; The data learned from the maintenance unit is mainly text data, involving specific repair information of crack damage in the bulkhead structure, including: description of the repair method, specific repair plan, repair materials, bonding process, surface treatment and protective measures, and inspection requirements after the repair is completed.

3. The civil aircraft bulkhead structure crack repair decision method integrating physical information and case data according to claim 1 is characterized in that: In step 2, according to the civil aircraft bulkhead structure, a crack tensile test piece of the same material is selected to simulate the fatigue load spectrum in actual operation, and a constant amplitude fatigue crack growth experiment is performed to record the crack length and the corresponding load cycle number, and 200 data points are recorded for each experiment.

4. The civil aircraft bulkhead structure crack repair decision method integrating physical information and case data according to claim 1 is characterized in that: The step 3 is to establish a fully connected PINN neural network model, take the Paris law of crack extension in fracture mechanics as a model constraint, and embed physical information into the network through a composite loss function; take the experimental record data and the structured data in the repair case as input, output the crack remaining life, establish the relationship between the crack damage and fatigue cycle of the civil aircraft bulkhead structure, solve the network gradient with the help of automatic differentiation, and adjust the network weight; The Paris law of crack extension is specifically expressed as: Where a is the crack length, n is the number of load cycles, and C and m are material-related parameters; The composite loss function is specifically expressed as: Loss=w Data Loss Data +w Physics Loss Physics In the formula, Loss is the PINN network loss function, Loss Data For the structured data loss part of the case, Loss Physics is the physical information loss part, W Data and w Physics are the weight coefficients of the structured data loss part and the physical information loss part of the case respectively.

5. The civil aircraft bulkhead structure crack repair decision method integrating physical information and case data according to claim 1 is characterized in that: In step 4, a BERT neural network model is established, and the model is initialized using the bert-base-chinese pre-trained weights provided by Google. In the training stage, the MLM training task is constructed using the text data in the civil aircraft bulkhead structure crack repair case, and the model is fine-tuned to obtain a language model with the characteristics of the civil aircraft bulkhead structure crack repair field; in the inference stage, the BERT model is used to vectorize the input text and extract the case text data features; The MLM training task is an unsupervised learning task. The model replaces the segmentation units in the text with a probability of 15%, 80% of the units are replaced by masks, 10% of the units are replaced by any other unit, and the remaining 10% remain unchanged.

6. The civil aircraft bulkhead structure crack repair decision method integrating physical information and case data according to claim 1 is characterized in that ,The step 5 establishes a ViT neural network model, and initializes the model using the training weights obtained from ,the civil aircraft bulkhead structure crack detection image dataset. ,In the inference stage, the ViT model is used to perform multi-level feature extraction on the ,input image and extract the case image data features.

7. The civil aircraft bulkhead structure crack repair decision method integrating physical information and case data according to claim 1 is characterized in that: The step 6 is to establish a twin architecture model with two branch networks, each branch network is composed of the three neural network models in claim 4, claim 5, and claim 6; the two branches are used to perform multi-modal and multi-dimensional processing and analysis on the new bulkhead structure crack damage cases and the cases in the historical case library, respectively, and perform similarity analysis on the outputs of the two branches; the new bulkhead structure crack damage cases are analyzed for similarity with the cases in the historical case library one by one, and the repair plan of the most similar case is matched to make a case-based decision on the repair of the civil aircraft bulkhead structure cracks.

Images