An aircraft critical location damage prediction method, system, device, medium, and product

Abstract

The application discloses an aircraft key part damage prediction method, system, device, medium and product, relates to the field of aircraft life prediction, and the method comprises the steps of: generating a plurality of sample double-channel stress histories randomly for an aircraft key part; calculating a sample damage value history corresponding to each sample double-channel stress history by adopting a multi-axial fatigue calculation method based on a critical plane method according to material performance parameters; training a deep learning network model based on a time domain convolution network by taking the sample double-channel stress history as input and taking the sample damage value history as a label to obtain a fatigue damage prediction model; obtaining a measured double-channel stress history of the aircraft key part in a service process; inputting the measured double-channel stress history into the fatigue damage prediction model to obtain a corresponding damage value history, and realizing damage prediction of the aircraft key part. The application can improve damage value calculation efficiency, thereby improving aircraft residual life prediction efficiency.

Description

Technical Field

[0001] This application relates to the field of aircraft life prediction, and in particular to a method, system, device, medium and product for predicting damage to critical parts of an aircraft. Background Technology

[0002] Fatigue failure is a major form of aircraft structural failure. In the history of aviation development, apart from structural damage caused by human factors and natural disasters, structural failure caused by fatigue cracks generated under alternating loads is the core cause of aircraft structural damage, posing a serious threat to the safety and reliability of aircraft structures.

[0003] Single-aircraft life tracking technology accurately tracks the structural damage status of each aircraft by considering the load characteristics and service history differences of various aircraft. Based on this, it determines the damage value and remaining life of each aircraft and develops targeted inspection and maintenance plans. This technology can improve structural safety and reduce maintenance costs. Therefore, life management methods based on single-aircraft tracking have become the mainstream development direction in the field of aircraft life management in various countries.

[0004] When conducting single-aircraft structural life tracking, it is typically necessary to calculate damage values ​​based on the acquired multiaxial stress history of key components, and then convert these values ​​into the remaining structural life using appropriate formulas. Currently, the conventional process for calculating damage values ​​and life based on the stress history of key components is as follows: first, rainflow counting is performed on the stress time series; then, cyclic-level damage value calculation and damage accumulation are completed sequentially, finally obtaining the damage value and converting it into remaining life. This process often requires significant time and computing power, making it difficult to meet the engineering application requirements of life tracking for high-frequency updates. Summary of the Invention

[0005] The purpose of this application is to provide a method, system, device, medium, and product for predicting damage to critical parts of an aircraft, which can improve the efficiency of damage value calculation and thus improve the efficiency of aircraft remaining life prediction.

[0006] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides a method for predicting damage to critical components of an aircraft, including: For critical parts of the aircraft, multiple sample dual-channel stress histories are randomly generated; each sample dual-channel stress history includes a sample shear stress history in the shear stress channel and a sample normal stress history in the normal stress channel. Based on the material property parameters, the sample damage history corresponding to the dual-channel stress history of each sample is calculated using a multiaxial fatigue calculation method based on the critical plane method. Using the dual-channel stress history of the sample as input and the damage value history of the sample as label, a deep learning network model based on a temporal convolutional network is trained to obtain a fatigue damage prediction model. Obtain the measured dual-channel stress history of key components of the aircraft during service; The measured dual-channel stress history is input into the fatigue damage prediction model to obtain the corresponding damage value history, thereby realizing damage prediction of key parts of the aircraft.

[0007] Secondly, this application provides a damage prediction system for critical parts of an aircraft, including: The sample dual-channel stress history generation module is used to randomly generate multiple sample dual-channel stress histories for key parts of an aircraft; each sample dual-channel stress history includes a sample shear stress history for the shear stress channel and a sample normal stress history for the normal stress channel. The sample damage history calculation module is used to calculate the sample damage history corresponding to the dual-channel stress history of each sample based on the material performance parameters and the multiaxial fatigue calculation method based on the critical plane method. The training module is used to train a deep learning network model based on a temporal convolutional network with the dual-channel stress history of the sample as input and the damage value history of the sample as label, so as to obtain a fatigue damage prediction model. The measured dual-channel stress history acquisition module is used to acquire the measured dual-channel stress history of key parts of the aircraft during service. The damage history prediction module is used to input the measured dual-channel stress history into the fatigue damage prediction model to obtain the corresponding damage history, thereby realizing damage prediction of key parts of the aircraft.

[0008] Thirdly, this application 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 computer program to implement the above-described method for predicting damage to critical parts of an aircraft.

[0009] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned method for predicting damage to critical parts of an aircraft.

[0010] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned method for predicting damage to critical parts of an aircraft.

[0011] According to the specific embodiments provided in this application, this application has the following technical effects: This application trains a deep learning model based on a temporal convolutional network by combining randomly generated sample dual-channel stress histories with damage value histories calculated using material performance parameters and multi-axis fatigue calculation methods. Finally, the measured dual-channel stress histories are input into the trained model (i.e., the fatigue damage prediction model) to quickly obtain the damage value histories. The fatigue damage prediction model maps the entire physical calculation chain (including rainflow counting, multi-axis criteria, and cumulative theory) end-to-end into a data-driven fast prediction function. After the measured dual-channel stress histories are input, the model directly outputs the damage value histories without the need for explicit rainflow counting and loop-by-loop calculations. This significantly improves the computational efficiency of damage prediction for critical parts of aircraft and meets the engineering application requirements of high-frequency, low-latency damage assessment in single-aircraft life tracking. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 A flowchart illustrating a method for predicting damage to critical parts of an aircraft, provided as an embodiment of this application; Figure 2 A detailed flowchart illustrating a method for predicting damage to critical components of an aircraft, provided as an embodiment of this application; Figure 3 A schematic diagram illustrating the process of shear stress and normal stress; Figure 4 This is a schematic diagram of a multi-axis rainflow counting method; Figure 5 This is a schematic diagram of the network structure of a deep learning network model based on a temporal convolutional network. Figure 6 This is a schematic diagram of the TCN convolutional activation module; Figure 7 This is a schematic diagram of the TCN module. Figure 8 This is a structural diagram of the weighted splicing module; Figure 9 This is a schematic diagram of the Film modulation module; Figure 10 This is a schematic diagram of the dual-channel stress history; Figure 11The diagram shows the prediction effect of the fatigue damage prediction model on the training set and the test set for the cumulative damage value; where (a) is the prediction effect of the fatigue damage prediction model on the training set for the cumulative damage value, and (b) is the prediction effect of the model on the test set for the cumulative damage value. Figure 12 The diagrams show a comparison between the actual and predicted damage values ​​for some training set samples; (a) is a comparison between the actual and predicted damage values ​​for training set sample number 208; (b) is a comparison between the actual and predicted damage values ​​for training set sample number 269; (c) is a comparison between the actual and predicted damage values ​​for training set sample number 319; and (d) is a comparison between the actual and predicted damage values ​​for training set sample number 413. Figure 13 The following are schematic diagrams comparing the actual damage value history and the predicted damage value history of some test set samples: (a) is a schematic diagram comparing the actual damage value history and the predicted damage value history of test set sample number 48; (b) is a schematic diagram comparing the actual damage value history and the predicted damage value history of test set sample number 61; (c) is a schematic diagram comparing the actual damage value history and the predicted damage value history of test set sample number 135; and (d) is a schematic diagram comparing the actual damage value history and the predicted damage value history of test set sample number 198. Figure 14 This is a diagram showing the time consumption comparison of the three calculation processes. Detailed Implementation

[0014] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0015] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0016] In one exemplary embodiment, such as Figures 1-2 As shown, a method for predicting damage to critical parts of an aircraft is provided. This method is executed by computer equipment, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is described using a server as an example, and includes the following steps S1 to S55.

[0017] S1: For critical parts of the aircraft, multiple sample dual-channel stress histories are randomly generated; each sample dual-channel stress histories includes the sample shear stress histories of the shear stress channel and the sample normal stress histories of the normal stress channel.

[0018] S2: Based on the material performance parameters, the sample damage history corresponding to the dual-channel stress history of each sample is calculated using a multiaxial fatigue calculation method based on the critical plane method.

[0019] S3: Using the dual-channel stress history of the sample as input and the damage value history of the sample as label, train the deep learning network model based on the temporal convolutional network to obtain the fatigue damage prediction model.

[0020] S4: Obtain the measured dual-channel stress history of key parts of the aircraft during service.

[0021] S5: Input the measured dual-channel stress history into the fatigue damage prediction model to obtain the corresponding damage value history, thereby realizing damage prediction of key parts of the aircraft.

[0022] By implementing steps S1 to S5 above, a fatigue damage prediction model is constructed. Low-dimensional and fast data-driven mapping is used to directly approximate the calculation chain from stress history to damage value of key parts, reducing or partially replacing explicit rainflow counting and cycle-by-cycle accumulation processes, thereby improving the calculation speed of damage value prediction.

[0023] In one specific embodiment, step S1 specifically includes the following steps.

[0024] S11: The stress range is obtained by Latin hypercube sampling within the preset peak and valley extreme value sampling intervals.

[0025] For both shear stress and normal stress channels, a peak extreme value sampling interval and a valley extreme value sampling interval are pre-set. The peak extreme value and valley extreme value of the two channels of the dual-channel stress history of each sample are generated by Latin hypercube sampling. For any channel of any sample, its stress range is [the valley extreme value of the channel of the sample and the peak extreme value of the channel of the sample]. The stress value of the channel of the sample at all times in the entire process will be limited to this range.

[0026] S12: Based on the preset peak-valley value interval sampling range, randomly generate a peak-valley value sequence within the stress range.

[0027] First, set a base sampling frequency, then take the reciprocal to get the interval. For any channel, if the current value is a peak or trough, the next peak or trough may appear after h intervals. h is obtained by randomly sampling an integer within the preset peak-trough interval sampling range for that channel. Each time a peak or trough is generated, h is sampled. Therefore, the process of generating the peak-trough sequence for a specific channel of a sample is as follows: The first point is generated from [the extreme trough value of the sample and the extreme peak value of the sample and the channel] as the peak value at time 0. An integer is randomly selected from the preset peak-valley interval sampling range of the channel. ,exist Always A second point is randomly generated within the inner circle as... Valley of moment An integer is randomly selected from the preset peak-valley interval sampling range of the channel. ,exist Always A second point is randomly generated within the inner circle as... The peak value at each time point is recorded, and so on, alternating between peaks and troughs until the number of data points in that channel equals the preset number of data points. The peak-trough value sequence of two channels for each sample is obtained through the above method.

[0028] S13: Generate multiple sample dual-channel stress histories based on the peak-valley value sequence.

[0029] For any two channels in a sample, multiple random monotonic data points are added between the two peaks and troughs of any one channel. The number of these random monotonic data points is equal to the number of data points in the other channel within the time period covered by these two peaks and troughs. For example... Figure 3 As shown, Time and The normal stress appears at a peak and trough value at a certain moment, but the internal shear stress channel is in [the following context]. If a peak value occurs at a certain moment, then the normal stress channel needs to... Each time point is also supplemented with a data point. Therefore, for any recorded data point, both channels have recorded data points at that time, and at least one channel has a peak or trough value at that time. Data points are supplemented for both channels of all samples in the above manner. Then, based on the preset number of channel data points, the two channels are truncated to obtain multiple sample dual-channel stress histories containing shear stress histories and normal stress histories that meet the preset length.

[0030] The process of randomly generating the dual-channel stress history of the sample includes, but is not limited to, the method described in step S13 of generating the peak-valley sequence of the two channels and then randomly adding points to align the time axes of the two channels. Alternatively, it can be that within the stress range of the two channels of the sample, data points are generated in a completely random manner according to the preset sampling frequency to generate the dual-channel stress history of the sample.

[0031] In one specific embodiment, step S2 specifically includes: The damage value at a specific moment in the damage history of a given sample is equal to the damage value obtained by multiaxial fatigue calculation of the dual-channel stress history of that sample up to and including that moment. By calculating the damage value at all moments across all samples using the above method, the damage history of all samples can be obtained. The specific steps are as follows: S21: Suppose the sample dual-channel stress history contains n time series data points, corresponding to times t0 to t... n− 1. The damage history of the samples to be generated is a time series of the same dimension. S22: Set the damage value at time t0 to 0. tk At time t0, extract t0 to t k The sample dual-channel stress history is obtained by first counting rainflows and identifying the start and end times of all load cycles, and then extracting the mean and amplitude of the sample dual-channel stress within the corresponding time period of each load cycle; where k=1,2……n-1.

[0032] like Figure 4 The image shows an example of multi-axis rainflow counting, focusing on the main parameters. Rainflow counting yielded two load cycles. and , with load cycle For example, its corresponding auxiliary parameter history is CDEFG, and its amplitude is... This provides all the information for the load cycle. , This represents the range of shear stress variation during the first load cycle. This represents the range of normal stress variation for the first load cycle. Similarly, information for the second load cycle can be obtained. , The range of shear stress variation for the second load cycle. This represents the range of normal stress variation during the second load cycle.

[0033] S23: Substitute the mean and amplitude into Wang's multiaxial fatigue criterion formula, and calculate the single-cycle damage value for each load cycle in combination with material performance parameters.

[0034] The formula for Wang's multiaxial fatigue criterion is as follows: in, and These are the shear stress amplitude and the normal stress amplitude, respectively. and These are the mean shear stress and the mean normal stress, respectively. , , and These are the material's symmetrical torsional cyclic fatigue limit, symmetrical tensile-compressive cyclic fatigue limit, shear fatigue strength coefficient, and shear fatigue strength index, respectively. All four parameters are material performance parameters (obtainable through material-level fatigue tests). Fatigue life is the expected fatigue failure of a structure after N cycles of load applied according to the parameters used in the calculation.

[0035] S24: Based on Miner's linear cumulative damage theory, the damage values ​​of each single cycle are accumulated in chronological order to obtain t. k Damage value at any given time.

[0036] It assumes that the damage accumulation of the material is linear within a certain range, that is, the damage from each load level can be simply added together. It also assumes that the fatigue life of the structure under the i-th load level is... Under this level of load Damage value caused by each cycle It can be represented as If the structure is subjected to multiple levels of load, Miner's rule states that the total damage value is equal to the sum of the damage values ​​of each cyclic load, that is: in, This represents the total damage value, i.e., the damage value at the corresponding time point. Indicates the number of load levels.

[0037] S25: Move t0 to t n−1 Damage values ​​at each time point are integrated according to the time series to obtain a sample damage value history that matches the sample dual-channel stress history.

[0038] For any dual-channel stress history, all cycle information is obtained using the multi-axis rainflow counting method. Then, the damage value history corresponding to the dual-channel stress history can be calculated through steps S22 and S24.

[0039] In a specific embodiment, step S3 specifically includes: S31: Normalize the data of both channels and the damage value history of all samples respectively, and randomly divide all samples into training set, validation set and test set according to the ratio of 8:1:1, and record the parameters used for normalization, namely the maximum and minimum values ​​of the dual-channel stress history and damage value history.

[0040] S32: Build a deep learning network model based on temporal convolutional networks.

[0041] like Figure 5As shown, the deep learning network model based on temporal convolutional networks includes an input layer, multiple TCN convolutional activation modules, a weighted concatenation module, a Flim modulation module, and an output layer.

[0042] Deep learning network models based on temporal convolutional networks, through... Encoding semantics and The encoded semantics are weighted and concatenated, then input into TCN convolutional activation module 3 (i.e., the third TCN convolutional activation module) to obtain a gating signal, simulating the process of shear stress dominating multi-axis counting. This is achieved through the use of... Encoding semantic pairs The encoded semantics are modulated to obtain a stress-modulated signal, which is used to simulate the process of calculating the equivalent stress by performing function calculations on the two channel signals using the Wang multiaxial fatigue criterion.

[0043] The input layer data has a shape of (batch_size, T, 2), where batch_size is the preset training batch size, T represents the number of data points in two channels, and "2" represents the data feature dimension, i.e., there are two channels. It is then decomposed into... Channels and The data in both channels has a shape of (batch_size, T, 1). After passing through TCN convolutional activation module 1 (i.e., the first TCN convolutional activation module) and TCN convolutional activation module 2 (i.e., the second TCN convolutional activation module), the data is output respectively. Encoding semantics and The encoding semantics are as follows: the data shape is (batch_size, T, base_filters_1), where base_filters_1 represents the number of convolutional kernels in the same convolutional layers of the preset TCN convolutional activation modules 1, 2, and 4 (i.e., the fourth TCN convolutional activation module). The weighted concatenation result in a data shape of (batch_size, T, 2). In the TCN convolutional activation module 3, the number of kernels in the same convolutional layer is 1, so the shape of the gated signal is (batch_size, T, 1). The shapes of the preliminary modulation features after Flim modulation and the stress modulation signal are both (batch_size, T, base_filters_1). The shape of the output data after multiplying the gated signal and the stress modulation signal is (batch_size, T, base_filters_1). The shape of the fused feature after concatenation is (batch_size, T, 2). The base_filters output layer has a data shape of (batch_size, T, 1), representing the normalized damage value history.

[0044] The structure of the TCN convolutional activation module is as follows: Figure 6 As shown. Data is fed into a same convolutional layer after passing through multiple TCN modules, and then into a sigmoid activation layer. The structure of the TCN module is as follows: Figure 7 As shown, it is consistent with the publicly available TCN network structure with residual connections, including dilated convolutional layers, weight normalization layers, ReLU activation layers, Dropout layers, and same convolutional layers. The specific meaning and implementation method of each layer can be directly queried.

[0045] The structure of the weighted splicing module is as follows: Figure 8 As shown. Encoding semantics and The encoded semantics are multiplied by a constant weight and then directly concatenated.

[0046] The structure of the film modulation module is as follows: Figure 9 As shown, the modulation process can be represented by the following formula: in, and for The modulation function signal obtained through the same convolutional layer This represents the initial modulation characteristics output by the Film modulation module. for Encoding semantics, for Encoding semantics.

[0047] In the network structure of the deep learning network model based on temporal convolutional networks, the number of kernels in the same convolutional layer 3 is 1 to ensure that the shape of the output layer data is (batch_size, T, 1). The number of kernels in the same convolutional layer 1 is 16, and the number of kernels in the same convolutional layer 1 is 1.

[0048] S33: Set the loss function during training.

[0049] The loss function is: in, The total loss of the model, The mean square error at each time step. This is the incremental mean square error. The mean square error of the final value. , All are preset weight coefficients. When calculating the total loss of the model, only the model output value and the normalized damage value history from the training or validation set data are used; the dual-channel stress history is not used.

[0050] The mean squared error of the model is calculated at each time step for each data point. The mean squared error is then averaged over time to obtain the mean squared error of that data point. The mean squared error of the model is obtained by averaging the mean squared errors of all data points in the dataset. This term is used to control the overall conformity of the damage value history.

[0051] The incremental mean squared error of the model is calculated as follows: for a data point, the damage value at the next time step is subtracted from the damage value at the current time step to transform the damage value history into an incremental damage value history. Then, the mean squared error of the incremental damage value history at each time step is calculated, which is the incremental mean squared error of the data point. The average of the incremental mean squared errors of all data points is the incremental mean squared error of the model. This term is used to control the consistency of the growth trend of the damage value history.

[0052] The final mean squared error of the model is calculated by taking the damage values ​​of all data at the last moment to form a set, and then calculating the mean squared error of this set. This term is used to control the prediction accuracy of the damage values ​​at the last moment.

[0053] S34: Using the training set divided in step S31, the normalized sample shear stress history and the normalized sample normal stress history are used as inputs, and the normalized damage value history is used as output. The loss function constructed in step S33 is used as the loss function in the training process to train the deep learning network model based on the temporal convolutional network constructed in step S32 to obtain the fatigue damage prediction model.

[0054] In one specific embodiment, step S4 specifically includes the following steps.

[0055] S41: Normalize the measured dual-channel stress history to obtain normalized dual-channel stress data.

[0056] S42: Input the normalized dual-channel stress data into the input layer, and split the normalized dual-channel stress data into... Channel data and Channel data.

[0057] S43: The above The channel data and σ-channel data are respectively fed into the first TCN convolutional activation module and the second TCN convolutional activation module to generate... Encoding semantics and Encoding semantics.

[0058] S44: The above Encoding semantics and the stated The encoded semantics are input to the weighted concatenation module for weighted concatenation, and then input to the third TCN convolutional activation module to output the gated signal.

[0059] S45: [The sentence is incomplete and requires further context.] Encoding semantics and the stated The encoded semantics are input to the Film modulation module for Film modulation to generate preliminary modulation features.

[0060] S46: Input the preliminary modulation features into the fourth TCN convolution activation module and output the stress modulation signal.

[0061] S47: Multiply the stress modulation signal by the gating signal, and then multiply it by the... Semantic encoding is concatenated to form fused features.

[0062] S48: Input the fused features into the backbone TCN module, and output a high-order temporal representation after multi-layer temporal modeling.

[0063] S49: The higher-order temporal representation is mapped through the output layer to generate a damage value history.

[0064] The output layer outputs a normalized damage value history. The normalized damage value history is denormalized using the damage value history normalization parameters recorded in step S31 to obtain the damage value history in the physical domain. This damage value history can be used for aircraft life prediction and safety assessment.

[0065] The methods described above provided in this application are illustrated below with specific embodiments.

[0066] In this embodiment, using the settings shown in Table 1, 2000 dual-channel stress histories are generated, and the curve of one of the dual-channel stress histories is as follows. Figure 10 As shown.

[0067] Table 1

[0068] The 2000 generated dual-channel stress histories are calculated using the multiaxial fatigue calculation method described in step S2 to obtain damage value histories of equal length to the dual-channel stress histories, thus constructing a "dual-channel stress history - damage value history" dataset. The value at any point in the damage value history is the damage value calculated using the multiaxial calculation method for the stress histories preceding that point (including that point). Calculating the damage value at all time points in all dual-channel stress histories yields the corresponding damage value histories for all dual-channel stress histories.

[0069] The material property parameters used in the multiaxial fatigue calculation process are those of 7075-T651, and the values ​​are shown in Table 2.

[0070] Table 2

[0071] The deep learning network model based on a temporal convolutional network was constructed according to the method described in step S3, and the loss function was set during training. The parameters of the constructed deep learning network model and the parameter settings during training are shown in Table 3.

[0072] Table 3

[0073] The data in the "dual-channel stress history-damage value history" dataset were normalized, and the normalized data samples were divided into training set, validation set and test set in a ratio of 8:1:1.

[0074] The fatigue damage prediction model is obtained by training the deep learning network model built according to the parameters in Table 3, using the normalized shear stress history and normal stress history as inputs and the normalized damage value history as outputs.

[0075] Each data point from the training and test sets is input into the fatigue damage prediction model to obtain the damage value history in the normalized domain. Then, the damage value history output in the normalized domain is denormalized to obtain the damage value history in the physical domain. The damage value at the last moment of the damage value history is called the cumulative damage value of the corresponding stress history. A scatter plot is drawn with the actual cumulative damage value on the x-axis and the predicted cumulative damage value on the y-axis. The model's prediction performance for the cumulative damage value on the training and test sets is shown below. Figure 11 As shown in (a) and (b) in the figure.

[0076] Randomly select several samples from the training and test sets, and plot the curves of the actual and predicted damage values ​​on a single graph, such as... Figure 12 (a) to (d) and Figure 13As shown in (a) to (d), in the selected samples, the predicted damage value trajectory and the actual damage value trajectory are basically consistent in overall trend and key inflection points: the segmented acceleration, plateau segment, and sudden increase interval caused by non-proportional loading can be reproduced well, and the final cumulative damage value matches the level at the end of the sequence. Local deviations mainly occur at the beginning of rapid growth (intervals with more drastic slope switching), but their amplitude is small and does not change the main shape of the sequence or the final cumulative result. The training set and the test set perform similarly, indicating that the fatigue damage prediction model can still effectively track the evolution of damage values ​​over time even on unseen samples. Overall, the model can not only accurately predict the final cumulative damage value, but also reproduce the characteristics of the damage value sequence under multiaxial stress conditions well.

[0077] Using the settings shown in Table 1, another 50,000 dual-channel stress histories were generated. The cumulative damage value or damage history of each sample was calculated using both the multiaxial fatigue calculation method and the fatigue damage prediction model. Figure 14 As shown, under the same condition of processing 50,000 dual-channel stress histories: the fatigue damage prediction model takes 11.556 s to predict the total damage history, the multi-axis fatigue analysis method takes 77 s to calculate only the final cumulative damage value, and the multi-axis fatigue analysis method takes 4299 s to calculate the complete damage value sequence. The equipment used in the comparison and verification of calculation speed was a GPU (NVIDIA GeForce RTX 4060 Laptop GPU) and a CPU (Intel(R) Core(TM) i7-14650HX) of comparable level. Therefore, we can conclude that: Compared to calculating only the final cumulative damage value: the fatigue damage prediction model is about 6.7 × 10⁻⁶ (77 / 11.556 ≈ 6.67).

[0078] Relative to complete sequence calculation: fatigue damage prediction model is approximately 372 × 10⁻⁶ (4299 / 11.556 ≈ 372).

[0079] In terms of unit sample time delay, the fatigue damage prediction model takes about 0.231 ms / sample, the multiaxial fatigue analysis only calculates the final value of about 1.54 ms / sample, and the complete sequence takes about 86 ms / sample.

[0080] In summary, the damage prediction method for critical parts of aircraft proposed in this application can significantly improve the calculation speed of "sequence-level" damage value prediction while maintaining high accuracy, and increase the calculation speed of the final cumulative damage value by several times. It has the potential to significantly improve the calculation speed of damage value history corresponding to ultra-long stress sequences or ultra-multiple stress sequence segments.

[0081] Based on the same inventive concept, this application also provides a system for implementing the aforementioned method for predicting damage to critical parts of an aircraft. The solution provided by this system is similar to the implementation described in the above method; therefore, the specific limitations in one or more embodiments of the aircraft critical part damage prediction system provided below can be found in the limitations of the aircraft critical part damage prediction method described above, and will not be repeated here.

[0082] In one exemplary embodiment, a damage prediction system for critical parts of an aircraft is provided, including the following modules.

[0083] The sample dual-channel stress history generation module is used to randomly generate multiple sample dual-channel stress histories for key parts of an aircraft; each sample dual-channel stress history includes a sample shear stress history for the shear stress channel and a sample normal stress history for the normal stress channel. The sample damage history calculation module is used to calculate the sample damage history corresponding to the dual-channel stress history of each sample based on the material performance parameters and the multiaxial fatigue calculation method based on the critical plane method.

[0084] The training module is used to train a deep learning network model based on a temporal convolutional network, using the dual-channel stress history of the sample as input and the damage value history of the sample as labels, to obtain a fatigue damage prediction model.

[0085] The measured dual-channel stress history acquisition module is used to acquire the measured dual-channel stress history of key parts of the aircraft during service.

[0086] The damage history prediction module is used to input the measured dual-channel stress history into the fatigue damage prediction model to obtain the corresponding damage history, thereby realizing damage prediction of key parts of the aircraft.

[0087] In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments. The computer device may be a server or a terminal. The computer device includes a processor, a memory, an input / output interface (I / O), and a communication interface. The processor, memory, and I / O are connected via a system bus, and the communication interface is connected to the system bus via the I / O interface. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device stores data to be processed. The I / O interface of the computer device is used for exchanging information between the processor and external devices. The communication interface of the computer device is used for communicating with an external terminal via a network connection. When the computer program is executed by the processor, it implements the steps in the above-described method embodiments.

[0088] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0089] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0090] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0091] Those skilled in the art will understand that all or part of the processes in 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. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, 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, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0092] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0093] 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.

[0094] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for predicting damage to critical components of an aircraft, characterized in that, include: For critical parts of the aircraft, multiple sample dual-channel stress histories are randomly generated; each sample dual-channel stress history includes a sample shear stress history in the shear stress channel and a sample normal stress history in the normal stress channel. Based on the material property parameters, the sample damage history corresponding to the dual-channel stress history of each sample is calculated using a multiaxial fatigue calculation method based on the critical plane method. Using the dual-channel stress history of the sample as input and the damage value history of the sample as label, a deep learning network model based on a temporal convolutional network is trained to obtain a fatigue damage prediction model. Obtain the measured dual-channel stress history of key components of the aircraft during service; The measured dual-channel stress history is input into the fatigue damage prediction model to obtain the corresponding damage value history, thereby realizing damage prediction of key parts of the aircraft.

2. The method for predicting damage to critical components of an aircraft according to claim 1, characterized in that, Multiple sample dual-channel stress histories were randomly generated, specifically including: The stress range is obtained by Latin hypercube sampling within the preset peak and valley extreme value sampling intervals. Based on a preset peak-valley value interval sampling range, a peak-valley value sequence is randomly generated within the stress range; Multiple sample dual-channel stress histories are generated based on the peak-valley value sequence.

3. The method for predicting damage to critical components of an aircraft according to claim 1, characterized in that, Based on the material property parameters, a multiaxial fatigue calculation method based on the critical plane method is used to calculate the sample damage history corresponding to the dual-channel stress history of each sample, specifically including: Suppose the sample dual-channel stress history contains n time series data points, corresponding to times t0 to t... n−1 The damage history of the samples to be generated is a time series of the same dimension; Set the damage value at time t0 to 0, and then set t... k At time t0, extract t0 to t k The sample dual-channel stress history is obtained by first counting rainflows and identifying the start and end times of all load cycles in the sample shear stress history, and then extracting the mean and amplitude of the sample dual-channel stress in the corresponding time period of each load cycle; where k=1,2……n-1; Substitute the mean and amplitude into Wang's multiaxial fatigue criterion formula, and calculate the single-cycle damage value for each load cycle in combination with material performance parameters. Based on Miner's linear cumulative damage theory, the damage values ​​of each single cycle are accumulated in chronological order to obtain t. k Damage value at any given time; From t0 to t n−1 Damage values ​​at each time point are integrated according to the time series to obtain a sample damage value history that matches the sample dual-channel stress history.

4. The method for predicting damage to critical components of an aircraft according to claim 1, characterized in that, The deep learning network model based on temporal convolutional networks includes an input layer, multiple TCN convolutional activation modules, a weighted concatenation module, a Flim modulation module, and an output layer.

5. The method for predicting damage to critical components of an aircraft according to claim 1, characterized in that, The loss function of the deep learning network model based on temporal convolutional networks during training is: in, The total loss of the model, The mean square error at each time step. This is the incremental mean square error. The mean square error of the final value. , All of these are preset weighting coefficients.

6. The method for predicting damage to critical components of an aircraft according to claim 4, characterized in that, The measured dual-channel stress history is input into the fatigue damage prediction model to obtain the corresponding damage value history, specifically including: The measured dual-channel stress history was normalized to obtain normalized dual-channel stress data. The normalized dual-channel stress data is input to the input layer, and the normalized dual-channel stress data is split into... Channel data and Channel data; The The channel data and σ channel data are respectively fed into the first TCN convolutional activation module and the second TCN convolutional activation module to generate... Encoding semantics and Encoding semantics; The Encoding semantics and the stated The encoded semantics are input to the weighted concatenation module for weighted concatenation, and then input to the third TCN convolutional activation module to output the gate signal; The Encoding semantics and the stated The encoded semantics are input to the Film modulation module for Film modulation to generate preliminary modulation features; The preliminary modulation features are input into the fourth TCN convolutional activation module, which outputs a stress modulation signal. Multiply the stress modulation signal by the gating signal, and then multiply it by the... Semantic encoding is concatenated to form fused features; The fused features are input into the backbone TCN module, and after multi-layer temporal modeling, a high-order temporal representation is output. The higher-order temporal representation is mapped through the output layer to generate a damage value history.

7. A damage prediction system for critical components of an aircraft, characterized in that, include: The sample dual-channel stress history generation module is used to randomly generate multiple sample dual-channel stress histories for key parts of an aircraft; each sample dual-channel stress history includes a sample shear stress history for the shear stress channel and a sample normal stress history for the normal stress channel. The sample damage history calculation module is used to calculate the sample damage history corresponding to the dual-channel stress history of each sample based on the material performance parameters and the multiaxial fatigue calculation method based on the critical plane method. The training module is used to train a deep learning network model based on a temporal convolutional network with the dual-channel stress history of the sample as input and the damage value history of the sample as label, so as to obtain a fatigue damage prediction model. The measured dual-channel stress history acquisition module is used to acquire the measured dual-channel stress history of key parts of the aircraft during service. The damage history prediction module is used to input the measured dual-channel stress history into the fatigue damage prediction model to obtain the corresponding damage history, thereby realizing damage prediction of key parts of the aircraft.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the aircraft critical component damage prediction method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the aircraft critical component damage prediction method as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the aircraft critical component damage prediction method as described in any one of claims 1-6.

Images