A method for predicting fatigue life of a hybrid joint structure of a composite material

Abstract

The application provides a fatigue life prediction method of a composite material mixed connection structure. It relates to the technical field of materials. The method is used to solve the problem that the existing evidence theory damage identification is difficult to process large-scale acoustic emission data due to the exponential growth of the number of focal element combinations with the number of categories, resulting in high computational complexity. The method comprises the following steps: obtaining an acoustic emission feature matrix of the structure to be tested under cyclic loading; determining the number of damage categories based on an adaptive clustering evaluation index; calculating a membership matrix by a fast trust division algorithm to represent the uncertain trust allocation level of each sample to different damage mode focal elements; calculating a mechanism embedded cumulative quantity according to the membership matrix to represent the physical degradation channel of the coupling of multiple damage modes; fusing the time series statistical features and the mechanism embedded cumulative quantity to construct a physical driving feature vector, inputting the physical driving feature vector into a hybrid prediction network, and outputting a remaining useful life prediction value. The method can consider the uncertainty representation and computational efficiency of multiple damage modes, and improve the prediction accuracy and real-time performance.

Description

Technical Field

[0001] This invention relates to the field of materials technology, and in particular to a method for predicting the fatigue life of composite material hybrid connection structures. Background Technology

[0002] Composite material hybrid connection structures are increasingly used in aerospace, automotive and other fields, and their fatigue life prediction is a core part of structural integrity assessment. Acoustic emission monitoring technology has become an important non-destructive testing method in this field because it can capture transient signals during the damage evolution process in real time.

[0003] Damage pattern recognition based on evidence theory is a mainstream technique for handling uncertainties in acoustic emission signals. Its core principle is to introduce a trust function framework, assigning acoustic emission feature samples not only to single damage categories but also to combined focal elements composed of multiple categories, thereby characterizing the overlap and uncertainty of different damage patterns in the feature space. This method achieves soft classification of various damage modes, such as matrix cracking and fiber fracture, by optimizing an objective function containing trust constraints and iteratively solving the trust membership matrix of samples to each focal element.

[0004] However, the core mechanism of this method requires the allocation and optimization of quality functions for all possible multi-class combinations. As the number of damage mode categories increases, the number of combined focal elements grows exponentially, leading to a sharp expansion of the solution space for the objective function and an exponential increase in computational complexity. This computational bottleneck makes this method unsuitable for large-scale, high-sampling-rate acoustic emission datasets generated in practical engineering, and it cannot meet the real-time requirements of fatigue life prediction for data processing efficiency, thus limiting its engineering application in complex service environments. Summary of the Invention

[0005] Therefore, it is necessary to provide a fatigue life prediction method for composite material hybrid connection structures to address the above-mentioned technical problems.

[0006] The following technical solution is adopted in this specification: This specification provides a method for predicting the fatigue life of composite material hybrid connection structures, including: Obtain the acoustic emission characteristic matrix of the composite hybrid connection structure under test under cyclic loading; Based on the preset number of damage categories and cluster center parameters, a fast trust partitioning is performed on the acoustic emission feature matrix to obtain a fast trust membership matrix. The fast trust membership matrix represents the degree of trust allocation of each acoustic emission signal sample to different combinations of damage modes. The trust assignment results of each damage mode in the fast trust membership matrix are used as weights to perform weighted aggregation on the cluster center parameters, thereby obtaining the mechanism embedding accumulation quantity that reflects the law of multi-source damage coupling accumulation. Based on the acoustic emission feature matrix, an equivalent cumulative damage index is constructed to characterize the damage energy release level within the current time window. The temporal statistical features, mechanism embedding accumulation, and equivalent cumulative damage index in the acoustic emission feature matrix are fused to construct a physical driving feature vector; The physical-driven feature vector is input into a pre-trained prediction network, which outputs the predicted remaining service life of the composite hybrid connection structure under test.

[0007] Furthermore, the step of performing fast trust partitioning on the acoustic emission feature matrix based on a preset number of damage categories and cluster center parameters to obtain a fast trust membership matrix includes: Based on the evidence theory framework, the focal element set is composed of a power set of the pre-defined number of damage categories; The focal center of each non-empty focal element is determined by arithmetic mean based on the center of the basic cluster contained in each non-empty focal element in the focal element set. A fast trust partitioning objective function is constructed, and the membership matrix is ​​iteratively optimized by minimizing the objective function so that each acoustic emission signal sample is assigned to the corresponding focal element with different trust levels, thus obtaining the fast trust membership matrix.

[0008] Furthermore, the step of using the trust assignment results of each damage mode in the fast trust membership matrix as weights to perform weighted aggregation on the cluster center parameters to obtain the mechanism embedding accumulation quantity reflecting the cumulative law of multi-source damage coupling includes: The trust assignment results of each damage mode combination in the fast trust membership matrix are used as weight coefficients, and the cluster center parameters of each damage mode combination are weighted and summed to obtain the mode coupling damage amount in each discrete time window. The mode coupling damage amount within each discrete time window is accumulated along the time series to obtain the cumulative amount of mechanism embedding.

[0009] Furthermore, the construction of an equivalent cumulative damage index characterizing the damage energy release level within the current time window based on the acoustic emission feature matrix includes: The original damage energy within each discrete time window is determined based on the acoustic emission energy parameters in the acoustic emission feature matrix. Based on the historical maximum damage energy statistics determined during the pre-training phase of the prediction network, the original damage energy is normalized, and the obtained strictly monotonically increasing cumulative damage index is used as the equivalent cumulative damage index.

[0010] Furthermore, the process of fusing the temporal statistical features, mechanistic embedding accumulation, and equivalent cumulative damage index in the acoustic emission feature matrix to construct a physical driving feature vector includes: Extract the temporal statistical features of the acoustic emission feature matrix within the current time window. The temporal statistical features include the mean and extreme values ​​of the acoustic emission features. By fusing temporal statistical features, mechanistic embedded cumulative amount, and equivalent cumulative damage index, a physical driving feature vector is constructed.

[0011] Furthermore, the pre-trained prediction network includes an input layer, a convolutional feature extraction layer, a bidirectional long short-term memory layer, and an output layer connected in sequence; The convolutional feature extraction layer is used to extract local features from the physical driving feature vector to obtain a high-dimensional feature representation; A bidirectional long short-term memory layer is used to mine the temporal dependencies of high-dimensional feature representations in both forward and backward directions, and to obtain the hidden state output that integrates bidirectional temporal information. The output layer is used to map the hidden state output to the remaining lifetime prediction value.

[0012] Furthermore, the preset number of damage categories is adaptively determined in the following manner: Cluster quality evaluation indicators were constructed based on the silhouette coefficient and the Davidson-Bolding index. Based on the extreme points of the clustering quality evaluation index as a function of the number of clusters, the optimal number of damage clusters is determined as the preset number of damage clusters.

[0013] This specification provides a fatigue life prediction device for composite material hybrid connection structures, comprising: The data acquisition module is used to acquire the acoustic emission characteristic matrix of the composite material hybrid connection structure under test under cyclic loading; The trust partitioning module is used to perform fast trust partitioning on the acoustic emission feature matrix based on the preset number of damage categories and cluster center parameters, and obtain a fast trust membership matrix. The fast trust membership matrix represents the degree of trust allocation of each acoustic emission signal sample to different combinations of damage modes. The mechanism embedding module is used to use the trust assignment results of each damage mode in the fast trust membership matrix as weights to perform weighted aggregation on the cluster center parameters, so as to obtain the mechanism embedding accumulation quantity that reflects the law of multi-source damage coupling accumulation. The cumulative damage extraction module is used to construct an equivalent cumulative damage index that characterizes the damage energy release level within the current time window based on the acoustic emission feature matrix. The feature fusion module is used to fuse the temporal statistical features, mechanism embedding accumulation, and equivalent cumulative damage index in the acoustic emission feature matrix to construct a physical driving feature vector. The lifetime prediction module is used to input the physical driving feature vector into the pre-trained prediction network and output the predicted value of the remaining lifetime of the composite material hybrid connection structure under test.

[0014] This specification provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for predicting the fatigue life of composite material hybrid connection structures.

[0015] This specification provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the fatigue life prediction method for the composite material hybrid connection structure described above.

[0016] The above-mentioned technical solutions adopted in this specification can achieve the following beneficial effects: This invention performs an efficient and fast trust partitioning on the acoustic emission feature matrix based on a preset number of damage categories and cluster center parameters. This directly obtains a fast trust membership matrix characterizing the degree of trust allocation for each sample to different damage mode combinations, significantly reducing the solution space. Furthermore, by using the trust allocation results in this membership matrix as weights to weightedly aggregate the cluster center parameters, a mechanism embedding accumulation reflecting the coupling law of multi-source damage is generated, avoiding redundant calculations of enumerating all combined focal elements one by one. Simultaneously, a physical driving feature vector is constructed by combining equivalent cumulative damage indices and temporal statistical features, and a pre-trained network is used to output the remaining lifetime prediction value. Overall, while maintaining the ability to model the coupling of complex damage modes, the computational complexity is significantly reduced, meeting the real-time processing requirements of large-scale, high-sampling-rate acoustic emission data and improving computational efficiency. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0018] Figure 1 This is a flowchart illustrating a method for predicting the fatigue life of a composite material hybrid connection structure provided in this specification. Figure 2 This is a schematic diagram showing the dimensions of a composite material glue-screw hybrid connection structure specimen provided in this specification; Figure 3 A schematic diagram of the fatigue hysteresis curve of a composite material glue-screw hybrid connection structure provided in this specification; Figure 4 This is a schematic diagram illustrating the fatigue life prediction results of a composite material glue-screw hybrid connection structure provided in this specification under high load levels. Figure 5 This is a schematic diagram illustrating the fatigue life prediction results of a composite material glue-screw hybrid connection structure provided in this specification under low load levels. Figure 6 This is a schematic diagram illustrating a comparison of prediction results based on the coefficient of determination, as provided in this specification. Figure 7 This is a schematic diagram of the characteristic contribution distribution of a composite material glue-screw hybrid connection structure based on SHAP values, as provided in this specification. Figure 8 A schematic diagram showing the importance ranking of SHAP features in a composite material glue-screw hybrid connection structure provided in this specification; Figure 9 This is a structural schematic diagram of a fatigue life prediction device for a composite material hybrid connection structure provided in this specification. Figure 10 This is a schematic diagram of a computer device provided for this specification. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort are within the scope of protection of this application.

[0020] The technical solution provided by this invention can be applied to fatigue damage monitoring and remaining service life prediction of composite hybrid connection structures under cyclic loading. During fatigue, composite hybrid connection structures exhibit multi-damage mode coupling evolution, and acoustic emission signals are heavily affected by noise interference and feature overlap, making it difficult for existing data-driven methods to extract stable degradation patterns, resulting in insufficient prediction accuracy and robustness. To address these issues, this application provides a method for predicting the fatigue life of composite materials based on adaptive trust mechanism embedding.

[0021] The fatigue life prediction method for the composite material hybrid connection structure of the present invention is described below with reference to the accompanying drawings.

[0022] Figure 1 This is a flowchart illustrating a fatigue life prediction method for a composite material hybrid connection structure provided in this specification, as shown below. Figure 1 As shown, the method includes the following: S100: Obtain the acoustic emission characteristic matrix of the composite material hybrid connection structure under cyclic loading.

[0023] For example, composite material screw-and-plastic hybrid connection test specimens can be prepared according to ASTM standards, and the ultimate load can be determined by quasi-static tension (e.g., tensile rate of 1.00 mm / min). Subsequently, the force control range, stress ratio, and loading frequency for fatigue testing are determined based on the ultimate load, and a protective displacement (e.g., 4 mm) or the ultimate cycle number is set as the stopping condition. During fatigue testing, acoustic emission (AE) instruments and sensors are used for real-time damage monitoring, collecting multiple parameter signals including peak amplitude (PA), duration (TD), rise time (RT), ringing counts (RC), ascending counts (AC), energy (EEnergy), root mean square (RMS), average signal level (ASL), number of impacts (H), impact rate (HR), center of mass frequency (GF), and peak frequency (PF). The above signals are statistically analyzed according to discrete time windows (such as intervals per second), and the mean and extreme values ​​of each parameter are extracted to form an acoustic emission feature matrix that represents the time series set of multi-parameter damage signals of the structure within the discrete time window.

[0024] S200. Based on the preset number of damage categories and cluster center parameters, perform fast trust partitioning on the acoustic emission feature matrix to obtain a fast trust membership matrix. The fast trust membership matrix represents the degree of trust allocation of each acoustic emission signal sample to different combinations of damage modes.

[0025] For example, the preset number of damage categories and cluster center parameters are obtained through the pre-training phase. During the pre-training phase, an adaptive computation model is constructed by comprehensively evaluating the intra-cluster compactness (such as the silhouette coefficient) and inter-cluster separation (such as the Davidson-Bolding index) of the clustering results. This model directly outputs the optimal number of damage categories, avoiding subjective bias caused by human intervention in the discussion. Based on this optimal number of categories, the cluster center parameters are obtained by iteratively updating the objective function of minimizing the fast trust partition.

[0026] In the prediction phase, a fast trust partitioning is performed directly on the current acoustic emission feature matrix based on the aforementioned preset parameters. Compared to traditional evidence theory, which requires traversing all multi-class combination focal elements, leading to an exponential increase in computational complexity, this step utilizes preset anchor points to drive fast mapping, avoiding blind searching of the entire combination space and significantly reducing computational complexity. This efficiently obtains a fast trust membership matrix representing the uncertain attribution of samples to single or combined damage modes such as matrix cracking, fiber breakage, and interlaminar debonding.

[0027] S300. Using the trust assignment results of each damage mode in the fast trust membership matrix as weights, the cluster center parameters are weighted and aggregated to obtain the mechanism embedding accumulation quantity that reflects the cumulative law of multi-source damage coupling.

[0028] For example, the membership values ​​of samples to different damage pattern combinations in the fast trust membership matrix are extracted as weights, and these weights are then weighted and summed with the cluster center parameters of the corresponding damage pattern combinations (i.e., the centroid vector characterizing the feature intensity of the pattern). This process replaces cumbersome iterative optimization with instantaneous algebraic operations, mapping the acoustic emission signal from the overlapping and ambiguous signal space to the decoupled damage mechanism space. This allows for the rapid extraction of the mechanism embedding accumulation quantity reflecting the coupling and accumulation law of multi-source damage, achieving deep embedding of the physical mechanism.

[0029] S400: Based on the acoustic emission feature matrix, construct an equivalent cumulative damage index that characterizes the damage energy release level within the current time window.

[0030] For example, firstly, the equivalent damage energy within the current time window is extracted from the acoustic emission feature matrix to characterize the current absolute damage release level of the structure. Then, the historical degradation benchmark extreme value determined during the pre-training phase (i.e., the maximum equivalent damage energy calculated from the training set) is introduced as a reference scale. The current equivalent damage energy is mapped to a relative ratio relative to this historical degradation benchmark extreme value, and a damage evolution index is introduced to ensure its strict monotonically increasing, thereby constructing an equivalent cumulative damage index. This construction method utilizes historical statistical extreme values ​​for relative mapping, effectively avoiding the future information leakage problem caused by lifetime extreme value normalization, and providing the model with a reliable physical degradation path.

[0031] S500: The temporal statistical features, mechanism embedding accumulation, and equivalent cumulative damage index in the acoustic emission feature matrix are fused to construct a physical driving feature vector.

[0032] For example, the mean and extreme values ​​of the acoustic emission feature matrix within each time window are extracted as temporal statistical features. At preset time intervals, the mean, extreme values, cumulative mechanism embedding, and equivalent cumulative damage index within the same time window are concatenated and fused to construct a physically driven feature vector that simultaneously incorporates data statistical semantics and physical degradation semantics. This vector not only covers the surface statistical fluctuations of the AE signal but also deeply integrates the coupling mechanism and energy degradation law of multiple damage modes, significantly enhancing the physical interpretability of the features.

[0033] S600: Input the physical driving feature vector into the pre-trained prediction network and output the predicted value of the remaining service life of the composite material hybrid connection structure to be tested.

[0034] For example, the pre-trained prediction network includes a Convolutional Neural Network (CNN) module and a Bidirectional Long Short-Term Memory (BiLSTM) module. The CNN module is used to extract local spatial features from the physics-driven feature vector, capturing the abrupt changes in the damage state. The BiLSTM module is used to model the temporal dependencies of the extracted spatial features from both forward and reverse time directions to capture the nonlinear evolution of slow onset and accelerated degradation of fatigue damage. Furthermore, during the pre-training phase, the prediction network uses a plateau-type Remaining Useful Life (RUL) function to construct data-driven supervision labels. This function maintains a constant value in the early stages of fatigue damage to represent slow degradation onset, and monotonically decreases over time in the accelerated damage stage to represent rapid performance degradation, thus matching the nonlinear fatigue degradation physical process of composite materials. After inputting the physics-driven feature vector into this network, the current remaining useful life prediction value can be output with high accuracy and robustness.

[0035] Furthermore, this application provides a specific embodiment: Step 1: Prepare composite material connection structure test specimens according to ASTM standards.

[0036] Step Two: First, subject the specimen to quasi-static loading at a tensile rate of 1.00 mm / min to determine the ultimate load. Then, determine the fatigue test force control range and stress ratio based on the specimen's ultimate load. Loading frequency The protective displacement was set to 4 mm before loading. The test was stopped when the specimen was stretched beyond the protective displacement or the number of cycles exceeded 10⁵. During the fatigue test, acoustic emission instruments and sensors were used for real-time damage monitoring. The monitored data are shown in Table 1.

[0037] Table 1: Acoustic emission signal parameters monitored during the experiment Step 3: Due to the uncertainty superposition of multiple damage signals in the AE signal, deep learning models struggle to directly learn stable degradation patterns. First, a mechanism embedding from the "signal space" to the "damage mechanism space" is implemented to improve learning capabilities. The AE damage matrix of each test piece obtained from the acoustic emission device can be represented as: ,in, For the first Acoustic emission damage matrix corresponding to each time window The total number of samples of the acoustic emission signal, where each sample is a A real eigenvector of dimension 1. It can be 13.

[0038] Assume the number of clusters is The clustering label set when using the evidence framework is Its power set constitutes the focal element set. ,in, This serves as an identification framework for evidence theory, encompassing all injury patterns and combinations. yes The first in Each damage status label; It is the first A non-empty focal element represents the set of "all possible subsets of states supported by model evidence" in evidence theory.

[0039] Step 4: Using the AE damage matrix as input, calculate the number of damage categories using the silhouette coefficient SI and the Davidson-Bolding index DB. For a given Let the sample silhouette coefficient be... Evaluation indicators of overall clustering quality Represented as:

[0040] ; ; In the formula, The total number of samples, For the first The silhouette coefficient of each sample. This represents the average distance between the corresponding sample and other samples within the same cluster. It is the average distance to its nearest neighbor cluster sample. The clustering density and separation are measured, with values ​​ranging from -1 to 1. The closer the value is to 1, the more reasonably the samples belong to the current cluster, and the better the clustering effect.

[0041] Davidson-Borgin Din Index Defined as: ; In the formula, and For clusters and cluster The average dispersion of the in-samples For clusters center with cluster center The Euclidean distance is used to measure the intra-cluster compactness and inter-cluster separation of clustering results. A smaller value indicates more compact intra-cluster clusters and greater inter-cluster separation, resulting in better clustering performance. However, existing literature mainly calculates different... The final SI and DB numbers will be determined through discussion. To address the problem of determining the value, an adaptive approach is proposed. The calculation method is defined as follows:

[0042] ; In the formula, The total number of candidate values ​​for the tested categories. For the first Number of candidate categories For the number of categories The average profile coefficient at that time, For the number of categories The Davidson-Boldt index at that time. The formula first applies the silhouette coefficient. and- Normalization was performed separately to eliminate dimensional differences. Then, the two normalized indicators were weighted and fused with a weight ratio of 1:0.5 to obtain each candidate index. The final score is the maximum score. This serves as the final number of categories. The final number can be directly obtained using this formula. This allows for the automation of the entire process, avoiding human intervention and discussion.

[0043] Step 5: Based on each non-empty focal element center Determined by the arithmetic mean of the centers of the basic clusters it contains: ; ; In the formula, For Jiao Yuan The base number, For basic clusters The center The total number of basic clusters, i.e., adaptively determined. , Instruction No. Does each basic cluster belong to the focal element? .

[0044] The objective function for fast trust partitioning can be defined as: ; In the formula, For the sample To Jiao Yuan The Euclidean distance at the center For this To Jiao Yuan Membership degree; For the focal element cardinality, and These are the cardinality weighting coefficient and membership weighting coefficient for the focal element.

[0045] By minimizing the objective function The membership matrix can be calculated as follows: ; In the formula, For the first The feature vector of each sample.

[0046] In subsequent iterations, the centroid vector used to represent the meta-clustering is... Defined as: ; In the formula, ,otherwise and During the iteration process, the updated basic cluster center is used. The centroid vector of each focal element is recalculated, implementing an iterative loop of fundamental cluster update → focal element center update. In the final update, the membership matrix is ​​defined as:

[0047] ; In the formula, For the first The unassigned confidence quality of a sample, i.e., the uncertainty that a sample does not belong to any known focal element. For Jiao Yuan , For the sample To Jiao Yuan The Euclidean distance to the center. Once the accurate cluster center is obtained, the ability to complete damage data pattern recognition in one step is achieved. At this point, although the algorithm complexity is reduced from... Reduce to However, it still retains its ability to divide trust.

[0048] Step Six: Construct the cumulative damage features based on the membership matrix as follows: ; ; In the formula, For the first Class damage to The cumulative number of times. Index for the current time window, The first cluster obtained by RCCM (Rapid Credibility Clustering Method) Damage mode labels corresponding to each time window As an indicator function, when time The damage mode is The value is 1 if the time is right, and 0 otherwise. This formula calculates the time from the initial moment to the current moment. , No. The cumulative number of occurrences of a damage pattern transforms the dynamic damage pattern identification results into a monotonically non-decreasing cumulative feature over time, intuitively reflecting the evolution process of a specific damage pattern.

[0049] After completing the mechanism-based semantic embedding, to enhance the physical interpretability of RUL tags, an equivalent damage evolution index based on AE is further introduced. The equivalent damage energy is defined. ,in, For the first Acoustic emission signal sequences collected within a time window This represents the discrete time window index.

[0050] Step 7: To avoid the problem of future information leakage caused by normalization of the maximum value over the entire lifetime, normalization is performed using training set statistics to construct an equivalent cumulative damage variable: ; ; in, The damage evolution index, For the first Equivalent damage energy per time step The maximum equivalent damage energy calculated from the training set. A cumulative damage index that indicates a strictly monotonically increasing trend. The cumulative damage index in the training set data The maximum value, For normalized cumulative damage.

[0051] Step 8: Construct the physical drive RUL tag as follows: ; in, This represents the critical damage threshold. Finally, the [number]th [damage threshold] is constructed according to a time interval of one second. Feature vectors for each time window:

[0052] ; in and Indicates that AE is in the Mean and extreme values ​​of AE features within seconds. Indicates the first Within a time window, damage pattern The mechanism of embedded accumulation, For the first The normalized cumulative damage index for each time window represents the physical degradation path calculated based on historical information.

[0053] Step 9: Build data-driven tags.

[0054] Because fatigue damage exhibits a slow initiation and accelerated degradation evolution, a platform-type RUL function is used to construct data-driven labels. The total specimen life time window is set as follows: The platform ratio coefficient is The platform stage length is The corresponding normalized data-driven RUL is defined as follows:

[0055] ; The purpose of this formula is: early When the damage is in a stable phase, the RUL tag remains at 1; when the damage enters an accelerated degradation phase... The RUL label decreases linearly to 0 over time, balancing physical realism and model training stability.

[0056] Step 10: RCCM-CNN-BiLSTM model construction.

[0057] To simultaneously extract local features and long-term temporal dependencies of acoustic emission signals, an RCCM-CNN-BiLSTM model was constructed. RCCM handles rapid clustering, CNN captures local pattern features of the signal, and BiLSTM models the temporal evolution of damage, achieving multi-dimensional feature fusion. The final model consists of an input layer, an RCCM module, a CNN module, a Bidirectional Long Short-Term Memory (BiLSTM) module, and an output layer.

[0058] Step 11: Model training and optimization.

[0059] To train the network, mean squared error (MSE) is used as the loss function to measure the error between the predicted RUL and the true RUL label. The Adam optimizer is used to adaptively adjust the model parameters to minimize the loss function. The MSE loss function is:

[0060] ; in, The total number of samples, For the first The total number of time windows for each sample The RUL value predicted by the model. This is a genuine RUL label.

[0061] The optimizer used is Adam: ; in, For model parameters, For learning rate, and Indicates the number of iterations. and This represents the bias correction estimates for the first and second moments. is the numerical stability constant.

[0062] To verify the model's early warning capability in practical engineering, an acceptable early warning range was defined, and multi-dimensional indicators were used to evaluate the model's performance, ensuring the reliability and engineering applicability of the prediction results. Actual trigger time and predicted trigger time:

[0063] ; ; in, A pre-set warning threshold is set, and an early warning is triggered when the remaining lifespan falls below a certain percentage. For the actual trigger moment, The trigger time predicted by the model. Find the time point when the warning conditions are first met, and use it as the trigger time for the structural failure warning.

[0064] Construct an acceptable warning range: ; in, To allow for an acceptable early warning error, This represents the allowable delay error for early warning. The formula defines an engineering-acceptable range for early warning time; if the predicted trigger time falls within this range, the early warning result is considered valid.

[0065] Finally, the error metrics for the training and test sets are calculated.

[0066] During the fatigue test, the actual remaining life is assumed to be... The predicted remaining lifespan is The total number of samples is The indicators include the root mean square error (RMSE), the mean absolute error (MAE), and the coefficient of determination. The accuracy of the test results is based on the above indicators.

[0067] To further enhance the model's generalization ability in actual service environments, an iterative update mechanism based on continuous monitoring data is proposed. This mechanism expands the training sample library by continuously accumulating operational data, thereby achieving dynamic optimization of the model. The core logic of this step is: to conduct continuous monitoring during actual service, continuously accumulate operational data and expand the training sample library, and then iteratively update and optimize the model based on this data, thereby further improving the accuracy and reliability of the connection structure lifetime prediction.

[0068] Based on the above specific embodiments, this application also provides an experimental verification result of the above solution: Figure 2This is a schematic diagram showing the dimensions of a composite material glue-screw hybrid connection structure specimen provided in this specification, according to... Figure 2 Composite material joint specimens with a hybrid glue-screw connection were prepared. An MTS electro-hydraulic servo-hydraulic fatigue testing machine with a maximum load of 100 kN was used in the experiments. First, quasi-static loading was applied to the specimens at a tensile rate of 1.00 mm / min to determine the ultimate load. Then, the fatigue test force control range and stress ratio were determined based on the specimen's ultimate load. Loading frequency The protective displacement was set to 4 mm before loading. The test was stopped when the specimen was stretched beyond the protective displacement or the number of cycles exceeded 10⁵. The final test results for a certain specimen are as follows. Figure 3 As shown, Figure 3 This is a schematic diagram of the fatigue hysteresis curve of a composite material glue-bolt hybrid connection structure provided in this specification. Before 1.5×104 cycles, the hysteresis area of ​​the lap joint structure is small and the mechanical response mode is close to that of the glue layer bearing. After about 2×104 cycles, the hysteresis area increases significantly, and a change from the glue layer to the bolt bearing response mode occurs. Finally, it fails near 1×105 cycles.

[0069] Based on acoustic emission monitoring data, after training the RCCM-CNN-BiLSTM model, the prediction results of representative samples are as follows: Figure 4 and Figure 5 As shown, Figure 4 This diagram illustrates the fatigue life prediction results of a composite material glue-screw hybrid connection structure provided in this specification under high load levels. Figure 5 This diagram illustrates the fatigue life prediction results of a composite material glue-screw hybrid connection structure provided in this specification under low load levels. Overall, after incorporating mechanistic information, the model's output life curve maintains high consistency with the actual remaining life throughout the entire service life, and can relatively continuously reflect the nonlinear evolution process from initial slow degradation to later rapid failure. For Figure 4 While some deviations exist in the short-life samples during the local damage fluctuation phase, the model still accurately portrays the overall degradation trend and rate of change. For medium- and long-life samples, the model shows no significant error accumulation over a long period and can still capture the mid-to-late-stage performance degradation characteristics well. Regarding early warning performance, the predicted trigger locations are generally within the confidence interval and close to the actual failure threshold. Furthermore, the 95% confidence interval gradually shrinks as the failure approach occurs, indicating that prediction uncertainty is effectively suppressed as damage information continues to accumulate. In addition, this method can identify the phased characteristics of the transition from adhesive-dominated to bolt-dominated load-bearing mechanisms, thereby reducing prediction lag.

[0070] Figure 6This diagram illustrates a comparison of prediction results based on the coefficient of determination (R²) provided in this specification. From the overall evaluation results, Figure 6 The predicted and actual values ​​are roughly distributed along the diagonal, and the R² level remains high and stable, especially in the later stages, indicating that the RCCM module has a stronger ability to characterize the rapid degradation process. Figure 7 This is a schematic diagram of the characteristic contribution distribution based on SHAP values ​​for a composite material glue-screw hybrid connection structure provided in this specification. Figure 7 The study reveals the intrinsic link between key features and damage mechanisms. Features contributing significantly include PF-Wmean (mean), PF-Wmax (mean), and AC-Cum (cumulative value). PF-type indices primarily reflect the energy release characteristics of high-frequency damage events such as crack propagation, fiber fracture, and interface failure; their mean and peak values ​​correspond to progressive and sudden damage, respectively. The AC-cumulative value describes the cumulative process of damage events, characterizing the evolution of microcrack density and corresponding to the monotonic decay trend of lifetime. Furthermore, several of the constructed mechanistic features ranked highly in importance and demonstrated significant contributions to the SHAP analysis. Figure 8 This is a schematic diagram illustrating the importance ranking of SHAP features in a composite material glue-screw hybrid connection structure provided in this specification, as shown below. Figure 8 As shown, this demonstrates that the relevant mechanistic information has been fully utilized by the model. Specifically, Delamination-Inf corresponds to the interlayer damage propagation behavior, Adhesive-Inf reflects the degradation of the adhesive interface properties, and Frictional-Inf characterizes the energy dissipation features during contact and friction. These variables with physical semantics complement the data features, significantly enhancing the model's ability to identify complex degradation behaviors.

[0071] The ablation test results (see Table 2) further demonstrate that the proposed method outperforms the baseline model and other comparative methods in all evaluation metrics, achieving a 30.12% performance improvement compared to the baseline. Simultaneously, the results show significantly reduced fluctuations, exhibiting better stability and robustness. After removing the mechanistic semantic module, the model's R² decreases while the RMSE increases, indicating that relying solely on statistical features is insufficient to fully describe the degradation process under multi-source damage coupling. This phenomenon aligns with the conclusion of the importance of mechanistic features in SHAP analysis, further validating its crucial role in RUL prediction. In contrast, the R² values ​​of purely data-driven models such as BiLSTM, GRU, and Transformer are mostly concentrated in the 0.75–0.80 range with greater dispersion, indicating limited generalization ability in damage evolution modeling. Models based on a single physical quantity show even more significant errors. Therefore, the higher fitting accuracy and lower prediction error achieved by this method are mainly attributed to the effective characterization of the damage stage transformation process by mechanistic information. In summary, the RCCM-CNN-BiLSTM model, which integrates mechanistic information, significantly improves performance compared to the baseline and exhibits better stability, demonstrating the important value of mechanistic semantic fusion in fatigue life prediction.

[0072] Table 2: Comparison Results of Ablation Tests This invention addresses the challenges of highly coupled multi-source damage signals, difficulty in effectively utilizing mechanistic information, and limited lifetime prediction accuracy in composite hybrid bonding structures during fatigue service. It proposes a modeling method integrating "acoustic emission monitoring—damage pattern recognition—damage mechanism embedding—life prediction." By performing damage pattern recognition on acoustic emission signals and introducing uncertainty characterization techniques, the overlapping phenomenon of different damage modes in the feature space is effectively mitigated, enabling clearer differentiation of typical damages such as matrix cracking, fiber breakage, and interlaminar debonding. Thus, the input features are no longer limited to traditional statistics but are transformed into mechanistic parameters with clear physical meaning, laying a more reliable foundation for subsequent analysis.

[0073] Building upon this foundation, the identified damage patterns are embedded as key intermediate information into the life prediction model, establishing an intrinsic link between the damage evolution process and the remaining life of the structure. Compared to modeling methods that rely solely on data-driven approaches, this method better reflects the phased characteristics of the fatigue process, reduces interference from noise and data fluctuations, and thus maintains high prediction accuracy and stability even under complex operating conditions. Furthermore, the model possesses stronger interpretability, capable of identifying the dominant damage behaviors at different stages, providing support for structural performance evaluation and optimized design.

[0074] Furthermore, by optimizing the overall modeling process, the method balances computational efficiency and engineering applicability while ensuring recognition accuracy, enabling it to meet the needs of large-scale data processing and online monitoring in actual service environments. This not only improves the reliability of fatigue life prediction but also provides an effective means to reduce structural design redundancy, lower maintenance costs, and extend service life, demonstrating promising application prospects and economic value in large composite material load-bearing structures.

[0075] The fatigue life prediction device for composite material hybrid connection structure provided by the present invention is described below. The fatigue life prediction device for composite material hybrid connection structure described below can be referred to in correspondence with the fatigue life prediction method for composite material hybrid connection structure described above.

[0076] Figure 9 This is a schematic diagram of a fatigue life prediction device for a composite material hybrid connection structure provided by the present invention. For example, please refer to [link to schematic diagram]. Figure 9 As shown, the fatigue life prediction device for the composite material hybrid connection structure may include: The data acquisition module is used to acquire the acoustic emission characteristic matrix of the composite material hybrid connection structure under test under cyclic loading.

[0077] The trust partitioning module is used to perform fast trust partitioning on the acoustic emission feature matrix based on the preset number of damage categories and cluster center parameters, and obtain a fast trust membership matrix. The fast trust membership matrix represents the degree of trust allocation of each acoustic emission signal sample to different combinations of damage modes.

[0078] The mechanism embedding module is used to use the trust assignment results of each damage mode in the fast trust membership matrix as weights to perform weighted aggregation on the cluster center parameters, thereby obtaining the mechanism embedding accumulation quantity that reflects the cumulative law of multi-source damage coupling.

[0079] The cumulative damage extraction module is used to construct an equivalent cumulative damage index that characterizes the damage energy release level within the current time window based on the acoustic emission feature matrix.

[0080] The feature fusion module is used to fuse the temporal statistical features, mechanism embedding accumulation, and equivalent cumulative damage index in the acoustic emission feature matrix to construct the physical driving feature vector.

[0081] The lifetime prediction module is used to input the physical driving feature vector into the pre-trained prediction network and output the predicted value of the remaining lifetime of the composite material hybrid connection structure under test.

[0082] Specific limitations regarding the fatigue life prediction device for composite material hybrid connection structures can be found in the above-mentioned limitations on fatigue life prediction for composite material hybrid connection structures, and will not be repeated here. Each module in the aforementioned fatigue life prediction device for composite material hybrid connection structures can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0083] This specification also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1 A method for predicting the fatigue life of composite material hybrid connection structures is provided.

[0084] This instruction manual also provides Figure 10 The schematic diagram of the computer device shown is as follows: Figure 10 At the hardware level, the computer device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then runs it to achieve the above-mentioned functions. Figure 1 A method for predicting the fatigue life of composite material hybrid connection structures is provided.

[0085] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0086] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A method for predicting the fatigue life of a composite material hybrid connection structure, characterized in that, include: Obtain the acoustic emission characteristic matrix of the composite hybrid connection structure under test under cyclic loading; Based on the preset number of damage categories and cluster center parameters, a fast trust partitioning is performed on the acoustic emission feature matrix to obtain a fast trust membership matrix. The fast trust membership matrix represents the degree of trust allocation of each acoustic emission signal sample to different combinations of damage modes. The trust assignment results of each damage mode in the fast trust membership matrix are used as weights to perform weighted aggregation on the cluster center parameters, thereby obtaining the mechanism embedding accumulation quantity that reflects the law of multi-source damage coupling accumulation. Based on the acoustic emission feature matrix, an equivalent cumulative damage index is constructed to characterize the damage energy release level within the current time window. The temporal statistical features, mechanism embedding accumulation, and equivalent cumulative damage index in the acoustic emission feature matrix are fused to construct a physical driving feature vector; The physical-driven feature vector is input into a pre-trained prediction network, which outputs the predicted remaining service life of the composite hybrid connection structure under test.

2. The fatigue life prediction method for composite material hybrid connection structures according to claim 1, characterized in that, The method of performing fast trust partitioning on the acoustic emission feature matrix based on the preset number of damage categories and cluster center parameters to obtain a fast trust membership matrix includes: Based on the evidence theory framework, the focal element set is composed of a power set of the pre-defined number of damage categories; The focal center of each non-empty focal element is determined by arithmetic mean based on the center of the basic cluster contained in each non-empty focal element in the focal element set. A fast trust partitioning objective function is constructed. By minimizing the objective function, the membership matrix is ​​iteratively optimized so that each acoustic emission signal sample is assigned to the corresponding focal element with different trust levels, thus obtaining the fast trust membership matrix.

3. The fatigue life prediction method for composite material hybrid connection structures according to claim 2, characterized in that, The step of using the trust assignment results of each damage mode in the fast trust membership matrix as weights to perform weighted aggregation of the cluster center parameters to obtain the mechanism embedding accumulation quantity reflecting the law of multi-source damage coupling accumulation includes: The trust assignment results of each damage mode combination in the fast trust membership matrix are used as weight coefficients, and the cluster center parameters of each damage mode combination are weighted and summed to obtain the mode coupling damage amount in each discrete time window. The mode coupling damage amount within each discrete time window is accumulated along the time series to obtain the cumulative amount of mechanism embedding.

4. The fatigue life prediction method for composite material hybrid connection structures according to claim 1, characterized in that, The equivalent cumulative damage index, constructed based on the acoustic emission feature matrix, characterizing the damage energy release level within the current time window, includes: The original damage energy within each discrete time window is determined based on the acoustic emission energy parameters in the acoustic emission feature matrix. Based on the historical maximum damage energy statistics determined during the pre-training phase of the prediction network, the original damage energy is normalized, and the obtained strictly monotonically increasing cumulative damage index is used as the equivalent cumulative damage index.

5. The fatigue life prediction method for composite material hybrid connection structures according to claim 1, characterized in that, The process of fusing temporal statistical features, mechanistic embedding accumulation, and equivalent cumulative damage index from the acoustic emission feature matrix to construct a physical driving feature vector includes: Extract the temporal statistical features of the acoustic emission feature matrix within the current time window. The temporal statistical features include the mean and extreme values ​​of the acoustic emission features. By fusing temporal statistical features, mechanistic embedded cumulative amount, and equivalent cumulative damage index, a physical driving feature vector is constructed.

6. The fatigue life prediction method for composite material hybrid connection structures according to claim 1, characterized in that, The pre-trained prediction network comprises an input layer, a convolutional feature extraction layer, a bidirectional long short-term memory layer, and an output layer connected in sequence. The convolutional feature extraction layer is used to extract local features from the physical driving feature vector to obtain a high-dimensional feature representation; A bidirectional long short-term memory layer is used to mine the temporal dependencies of high-dimensional feature representations in both forward and backward directions, and to obtain the hidden state output that integrates bidirectional temporal information. The output layer is used to map the hidden state output to the remaining lifetime prediction value.

7. The fatigue life prediction method for composite material hybrid connection structures according to claim 1, characterized in that, The preset number of damage categories is adaptively determined in the following manner: Cluster quality evaluation indicators were constructed based on the silhouette coefficient and the Davidson-Bolding index. Based on the extreme points of the clustering quality evaluation index as the number of clusters changes, the determined optimal number of damage clusters is used as the preset number of damage clusters.

8. A fatigue life prediction device for composite material hybrid connection structures, characterized in that, include: The data acquisition module is used to acquire the acoustic emission characteristic matrix of the composite material hybrid connection structure under test under cyclic loading. The trust partitioning module is used to perform fast trust partitioning on the acoustic emission feature matrix based on the preset number of damage categories and cluster center parameters, and obtain a fast trust membership matrix. The fast trust membership matrix represents the degree of trust allocation of each acoustic emission signal sample to different combinations of damage modes. The mechanism embedding module is used to use the trust assignment results of each damage mode in the fast trust membership matrix as weights to perform weighted aggregation on the cluster center parameters, so as to obtain the mechanism embedding accumulation quantity that reflects the law of multi-source damage coupling accumulation. The cumulative damage extraction module is used to construct an equivalent cumulative damage index that characterizes the damage energy release level within the current time window based on the acoustic emission feature matrix. The feature fusion module is used to fuse the temporal statistical features, mechanism embedding accumulation, and equivalent cumulative damage index in the acoustic emission feature matrix to construct the physical driving feature vector. The lifetime prediction module is used to input the physical driving feature vector into the pre-trained prediction network and output the predicted value of the remaining lifetime of the composite material hybrid connection structure under test.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the fatigue life prediction method for composite material hybrid connection structures as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the fatigue life prediction method for composite material hybrid connection structures as described in any one of claims 1 to 7.

Images