Online intelligent prediction method for residual life of aircraft structure based on acoustic emission

Abstract

An on-line intelligent prediction method for the residual life of an aircraft structure based on acoustic emission comprises the following steps: firstly, carrying out a fatigue experiment on the aircraft structure to obtain original data of a plurality of full-life acoustic emission waveform flows; performing sliding window sampling feature extraction on the original data, screening a part with good monotonicity to construct a health index, randomly selecting a part as a test set, taking the rest as a training set, cutting a data set according to a 3 sigma principle, and searching an initial prediction point; and finally, in a test stage, inputting test set data into the trained life prediction model to realize online intelligent prediction of the residual life of the aircraft structure. According to the method, Densenet and comparative learning are combined, online prediction is carried out on the residual life of the aircraft structure through an intelligent algorithm, and a good prediction effect is achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of aircraft structure health monitoring, and in particular relates to an online intelligent prediction method for the remaining life of an aircraft structure based on acoustic emission. Background Art

[0002] Traditional aviation structural maintenance relies on regular manual inspection and component replacement, which is not only time-consuming and labor-intensive, but also sometimes involves unnecessary costs. Among various non-destructive testing technologies, acoustic emission signals can continuously monitor the actual health status of aircraft structures. By analyzing the characteristics of acoustic emission signals, such as signal frequency, energy distribution and waveform characteristics, researchers can establish health status monitoring models. Such models not only help identify existing damage, but can also be used to predict possible fatigue development in the future, thereby providing an important basis for the assessment of the remaining life of the structure; the introduction of this dynamic health monitoring technology can help aviation practitioners deeply analyze structural behavior and enhance their control over structural safety. However, in the process of fatigue damage of aircraft structural parts, due to the complex structure of the aircraft and the harsh working conditions, it is easily affected by many factors such as internal and external dynamic loads and atmospheric pressure forces, resulting in the weak acoustic emission signals generated by the initiation and expansion of fatigue cracks being submerged by massive background noise signals, and the quality of the collected acoustic emission signal data is low.

[0003] At present, the processing methods of acoustic emission signals can be mainly divided into two categories: parameter analysis and waveform analysis. Parameter analysis is a multi-parameter quantitative description of acoustic emission signals. This method extracts characteristic parameters of the signal, such as signal energy, burst number, signal frequency, etc., and analyzes it quickly, simply and intuitively; the advantage of parameter analysis is its rapidity and easy-to-understand characteristics, which enables engineers to quickly obtain information about the health status of the structure. However, this method usually requires a certain amount of prior knowledge, and in the presence of environmental noise or other interference, it may affect the accuracy of the analysis results. The waveform analysis method is to identify and evaluate the possible damage and defects in the material by performing a detailed analysis of the time domain waveform and frequency domain characteristics of the acoustic emission signal. This method mainly focuses on the waveform morphology, spectrum distribution, duration, amplitude and other characteristics of the signal; the advantage of waveform analysis is its strong anti-interference ability and is applicable to many different types of signals. However, due to its complexity and the need for a large amount of calculation and data processing, waveform analysis often requires a high level of technology and a long analysis time, so it may face challenges in large-scale engineering applications.

[0004] In recent years, data-driven structural health monitoring has shown superior performance in feature extraction and structural life prediction, providing a new idea for life prediction based on aircraft structural acoustic emission signals. This method does not rely on a specific physical model, but automatically extracts features from data through machine learning algorithms to establish a mapping relationship between data and structural health status. It has wider applicability and higher real-time performance; however, existing supervised learning methods often require a large amount of labeled data to establish monitoring models ([1] Huang Yong, Bao Yuequan, Li Hui. Research progress in machine learning methods for structural state identification and evaluation [J]. Advances in Mechanics, 2023, 53(04): 774-792.), which greatly increases the cost of the experiment.

[0005] As an unsupervised learning technique, contrastive learning does not rely on explicit labels, which enables it to learn in the absence of labeled data. This feature makes contrastive learning particularly suitable for processing large-scale unlabeled data sets, thereby reducing the cost and time of data annotation. Currently, there is no literature on the use of contrastive learning for data-driven structural health monitoring and life prediction. Summary of the invention

[0006] In order to overcome the defects of the above-mentioned prior art, the purpose of the present invention is to provide an online intelligent prediction method for the remaining life of an aircraft structure based on acoustic emission, which combines Densenet with contrastive learning to perform online prediction of the remaining life of the aircraft structure through an intelligent algorithm, and has a good prediction effect.

[0007] In order to achieve the above object, the technical solution adopted by the present invention is:

[0008] An online intelligent prediction method for the remaining life of an aircraft structure based on acoustic emission is firstly carried out on the aircraft structure through fatigue test to obtain the original data of multiple full-life acoustic emission waveform streams; then the original data is subjected to sliding window sampling feature extraction, and the part with good monotonicity is screened to construct the health index, and a part is randomly selected as the test set, and the rest is used as the training set. The data set is cut according to the 3σ principle to find the starting prediction point; finally, in the test stage, the test set data is input into the trained life prediction model to realize the online intelligent prediction of the remaining life of the aircraft structure.

[0009] An online intelligent prediction method for the remaining life of an aircraft structure based on acoustic emission comprises the following steps:

[0010] Step 1: Collect the original data of acoustic emission waveform flow of the whole life cycle of aviation structural parts from healthy state to failure process, denoted as O = {O 1 ,O 2 ,…,O k}, which contains k different data records;

[0011] Step 2: Perform sliding window sampling on the original data. During the sampling process, calculate 21 eigenvalues ​​in each window, including waveform index, variance, standard deviation, mean, median, peak-to-peak value, maximum value, minimum value, root mean square frequency, root mean square value, mean square frequency, mean square value, skewness, skewness index, spectrum variance, spectrum skewness, spectrum kurtosis, kurtosis, kurtosis index, pulse index, and margin index. Assume that the original data O contains L O data points, the window size of the sliding window sampling is w, the calculated eigenvalues ​​are combined into eigenvectors and arranged in chronological order, thus obtaining the full life cycle data set O e = {V k}, 1≤k≤l, where each eigenvector V consists of n eigenvalues, It indicates the length of the original data after feature extraction. The calculation formula of the original data feature is as follows:

[0012]

[0013] In the formula, y i represents the i-th point in each window, M is the mean, S 2 is the variance, PI is the pulse index, P is the peak index, G is the margin index, w is the sampling window size, and n is the number of eigenvalues;

[0014] Step 3: Draw a characteristic trend graph based on the 21 eigenvalues ​​obtained in step 2, and find features with good monotonicity to construct health indicators. Finally, a data set is obtained with eight features: pulse index, peak index, margin index, average amplitude, root amplitude, maximum value, standard deviation, and spectral energy. V k ={f 1 ,f 2 ,…f 8}, from the dataset A piece of data is randomly selected as the test set, and the remaining data is used as the training set;

[0015] Step 4: Construct a Health Index (HI) model. The basic network structure framework uses a DenseNet (Densely connected convolutional networks). The module in the DenseNet network consists of "DenseBlock + Transition". The HI model contains 2 Dense Blocks, 2 Transition Layers, and 2 fully connected layers. After inputting the training set obtained in Step 3 into a single-layer convolution, the data reduces the number of parameters through dimensionality reduction in the Dense Block, and then the size of the feature map is reduced by the Transition. Finally, a max-pooling operation with a size of 3×3 and a stride of 2 is performed.

[0016]

[0017] Obtain the model output result N represents the total number of sample points, and x n,k is the health index corresponding to the sample point, and y n,k is the label of the sample;

[0018] Step 5: Calculate the loss using contrastive learning. Randomly select two different sample points from the model output result Y k obtained in Step 4 and denote them as a and b, where a < b. The specific calculation formula is as follows:

[0019] L d = -θ × (x a,k - x b,k )

[0020] θ is the amplification factor used to amplify the distance loss L d value;

[0021] Step 6: Repeat Step 4 - Step 5, gradually reduce the distance loss L d in Step 5, iterate and update the model parameters until the maximum number of iterations is reached, complete the training, and obtain the final prediction model;

[0022] Step 7: Input the test set into the prediction model trained in Step 6 in chronological order to construct the health index HI, obtain the image of the health index HI, and take the mean M p of the first p% of the data as the health state value of the aircraft structure and calculate the standard deviation σ p of the first p% of the data, and calculate the health index critical value P h of the health state of the aircraft structure. The specific calculation formula is as follows:

[0023] P h = M p - 3σ p

[0024] Step 8: Take the first value below the health indicator threshold P h The sample point after the starting prediction point is used as the data set O′ e Input the trained prediction model in step 6, and let it enter the fully connected layer after passing through 2 Dense Blocks and 2 Transition Layers, and finally obtain the remaining life prediction of the damaged segment of the test set.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] The present invention proposes an online intelligent prediction method for the remaining life of an aircraft structure based on acoustic emission. First, the original data of multiple full-life acoustic emission waveform streams of the aircraft structure are feature extracted, and several features with good monotonicity are selected to construct health indicators. This greatly reduces the complexity of the original data and reduces the model training and testing time. The 3σ principle is used to find the starting prediction point, and the distance loss is constructed based on comparative learning, which improves the generalization of the model and reduces the need for professional knowledge. This method has a good effect on the remaining life prediction of aircraft structural parts under different environments, and can accurately determine the time when damage to the aircraft structure occurs, thereby reducing the accident rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a flow chart of an embodiment of the present invention.

[0028] Figure 2 Schematic diagram of a life prediction model according to an embodiment of the present invention.

[0029] Figure 3 This is a diagram showing the construction of the health index HI and the positioning of the starting prediction point according to an embodiment of the present invention.

[0030] Figure 4 This is a diagram showing the remaining life prediction effect of an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The present invention is further described in detail below with reference to the embodiments and drawings.

[0032] like Figure 1 As shown, an online intelligent prediction method for the remaining life of an aircraft structure based on acoustic emission includes the following steps:

[0033] Step 1: Collect the original data set of acoustic emission waveform flow of the whole life cycle of aviation structural parts from healthy state to failure process, denoted as O = {O 1 ,O 2 ,…,O k}, which contains k different data records;

[0034] Step 2: Perform sliding window sampling on the original data. During the sampling process, calculate 21 eigenvalues ​​in each window, including waveform index, variance, standard deviation, mean, median, peak-to-peak value, maximum value, minimum value, root mean square frequency, root mean square value, mean square frequency, mean square value, skewness, skewness index, spectrum variance, spectrum skewness, spectrum kurtosis, kurtosis, kurtosis index, pulse index, and margin index; specifically, assume that the original data O contains L O data points, the window size of the sliding window sampling is w, the calculated eigenvalues ​​are combined into eigenvectors and arranged in chronological order, thus obtaining the full life cycle data set O e = {V k}(1≤k≤l), where each eigenvector V consists of n eigenvalues, It indicates the length of the original data after feature extraction. The calculation formula of the original data feature is as follows:

[0035]

[0036] In the formula, y i represents the i-th point in each window, M is the mean, S 2 is the variance, PI is the pulse index, P is the peak index, G is the margin index, w is the sampling window size, and n is the number of eigenvalues;

[0037] Step 3: Draw a characteristic trend graph based on the 21 eigenvalues ​​obtained in step 2, and find features with good monotonicity to construct health indicators. Finally, a data set is obtained with eight features: pulse index, peak index, margin index, average amplitude, root amplitude, maximum value, standard deviation, and spectral energy. V k ={f 1 ,f 2 ,…f 8}, from the dataset A piece of data is randomly selected as the test set, and the remaining data is used as the training set;

[0038] Step 4: Construct a Health Index HI model. The basic network structure framework adopts the DenseNet (Densely connected convolutional networks). The module in the DenseNet network consists of "DenseBlock + Transition". The Health Index HI model contains 2 Dense Blocks, 2 Transition Layers, and 2 fully connected layers. After inputting the training set obtained in Step 3 into a single-layer convolution, the data reduces the number of parameters through dimensionality reduction in the Dense Block, and then the size of the feature map is reduced by the Transition. Finally, a max pooling operation with a size of 3×3 and a stride of 2 is performed.

[0039]

[0040] Obtain the model output result N represents the total number of sample points, and x n,k is the health index corresponding to the sample point, and y n,k is the label of the sample;

[0041] Step 5: Calculate the loss using contrastive learning. Randomly select two different sample points from the model output result Y k obtained in Step 4 and denote them as a and b (a < b). The specific calculation formula is as follows:

[0042] L d = -θ × (x a,k - x b,k )

[0043] θ is the amplification factor used to amplify the distance loss L d value;

[0044] Step 6: Repeat Step 4 - Step 5, gradually reduce the distance loss L d in Step 5, iterate and update the model parameters until the maximum number of iterations is reached, and the training is completed to obtain the final prediction model;

[0045] Step 7: Input the test set into the prediction model trained in Step 6 in chronological order to construct the Health Index HI, obtain the image of the Health Index HI, and take the mean M p of the first p% of the data as the health state value of the aircraft structure and calculate the standard deviation σ p of the first p% of the data, and calculate the health index critical value P h of the health state of the aircraft structure. The specific calculation formula is as follows:

[0046] P h = M p - 3σ p

[0047] Step 8: Take the first value below the health indicator threshold P h The sample point after the starting prediction point is used as the data set D′ e Input the trained prediction model in step 6, and let it enter the fully connected layer after passing through 2 Dense Blocks and 2 Transition Layers, and finally obtain the remaining life prediction of the damaged segment of the test set.

[0048] Embodiments The feasibility of the present invention is verified by taking an experiment in a laboratory environment in which an aluminum alloy plate changes from a healthy state to a fracture under a cyclic load as an example.

[0049] The experimental plan is to collect data from healthy to broken aluminum alloy plates under periodic loads, which are collected by two acoustic emission sensors placed at both ends of the alloy plates. The experiment was carried out four times, and a total of eight raw data of the full-life acoustic emission waveform flow of the aluminum alloy plates were obtained. The eight raw data were extracted and feature trend graphs were made. Eight features with good monotonicity were selected to form feature vectors to form a new feature data set. The data set was sampled with time overlap, and a total of 8400 training samples and 1200 test samples were obtained. Each sample includes 300 data points. One of them is randomly selected as the test set, and the remaining seven data are used as the training set; the training set is input into the life prediction model for training; the life prediction model includes 2 Dense Blocks, 2 Transition Layers, and 2 fully connected layers, such as Figure 2 As shown, the model training parameters are shown in Table 1, and the model structure parameters are shown in Table 2.

[0050] Table 1 Model training parameters

[0051]

[0052] Table 2 Model structure parameters

[0053]

[0054] After the model training is completed, the normalized data set is input into the model as a test set to obtain the acoustic emission information health index of the alloy plate, such as Figure 3 As shown by Figure 3 It can be seen that the aluminum alloy plate is in the transition period from the healthy state to the damaged state in the early stage. Since we pay more attention to the data of the aluminum alloy plate in the damaged period, we take the mean M of the first 40% of the data, and then use the training set data to cut the data set based on the 3σ principle. The cut data set is input into the trained Densenet network to predict the remaining life of the aircraft structure. The prediction results are as follows: Figure 4As shown, it can be seen that the prediction results are consistent with the overall trend of the real data, and the life prediction results have high accuracy, which verifies the superiority of this model.

Claims

1. An online intelligent prediction method for the remaining life of an aircraft structure based on acoustic emission, characterized by: First, fatigue tests are conducted on the aircraft structure to obtain the original data of multiple full-life acoustic emission waveform flows. Then, the original data is subjected to sliding window sampling feature extraction, and the parts with good monotonicity are screened to construct health indicators. Some parts are randomly selected as test sets, and the rest are used as training sets. The data set is cut according to the 3σ principle to find the starting prediction point. Finally, in the testing phase, the test set data is input into the trained life prediction model to achieve online intelligent prediction of the remaining life of the aircraft structure.

2. An online intelligent prediction method for the remaining life of an aircraft structure based on acoustic emission, characterized in that: The following steps are involved: Step 1: Collect the original data of acoustic emission waveform flow of the whole life cycle of aviation structural parts from healthy state to failure process, denoted as O = {O1, O2, …, O k }, which contains k different data records; Step 2: Perform sliding window sampling on the original data. During the sampling process, calculate 21 eigenvalues ​​in each window, including waveform index, variance, standard deviation, mean, median, peak-to-peak value, maximum value, minimum value, root mean square frequency, root mean square value, mean square frequency, mean square value, skewness, skewness index, spectrum variance, spectrum skewness, spectrum kurtosis, kurtosis, kurtosis index, pulse index, and margin index. Assume that the original data O contains L O data points, the window size of the sliding window sampling is w, the calculated eigenvalues ​​are combined into eigenvectors and arranged in chronological order, thus obtaining the full life cycle data set O e = {V k }, 1≤k≤l, where each eigenvector V consists of n eigenvalues, It indicates the length of the original data after feature extraction. The calculation formula of the original data feature is as follows: In the formula, y i represents the i-th point in each window, M is the mean, S 2 is the variance, PI is the pulse index, P is the peak index, G is the margin index, w is the sampling window size, and n is the number of eigenvalues; Step 3: Draw a characteristic trend graph based on the 21 eigenvalues ​​obtained in step 2, and find features with good monotonicity to construct health indicators. Finally, a data set is obtained with eight features: pulse index, peak index, margin index, average amplitude, root amplitude, maximum value, standard deviation, and spectral energy. From the dataset A piece of data is randomly selected as the test set, and the remaining data is used as the training set; Step 4: Construct the health index HI model. The basic network structure framework adopts the densely connected network DenseNet (Densely connected convolutional networks). The modules in the DenseNet network are composed of "DenseBlock+Transition". The health index HI model contains 2 Dense Blocks, 2 TransitionLayers, and 2 fully connected layers. After the training set obtained in step 3 is input into a single-layer convolution, the data is reduced in dimension in DenseBlock to reduce the number of parameters, and then the size of the feature map is reduced by Trasition. Finally, a maximum pooling operation with a size of 3×3 and a stride of 2 is performed. Get the model output results N represents the total number of sample points, x n,k is the health index corresponding to the sample point, y n,k is the label of the sample; Step 5: Calculate the loss using contrastive learning. Randomly select two different sample points from the model output result Y obtained in Step 4 and denote them as a and b, where a < b. The specific calculation formula is as follows: k in which the specific calculation formula is as follows: L d =-θ×(x a，k -x b，k ) θ is the amplification factor, which is used to amplify the distance loss L d The value of Step 6: Repeat steps 4 to 5, gradually reducing the distance loss L in step 5 d , iterate and update the optimization model parameters until the maximum number of iterations is reached, the training is completed, and the final prediction model is obtained; Step 7: Input the test set into the prediction model trained in step 6 in chronological order to construct the health index HI, obtain the image of the health index HI, and take the mean M of the first p% of the data p As the health status value of the aircraft structure and calculate the standard deviation σ of the first p% data p , calculate the critical value P of the health index of the aircraft structure health status h , the specific calculation formula is as follows: P h =M p -3σ p Step 8: Take the first value below the health indicator threshold P h The sample point after the starting prediction point is used as the data set O′ e Input the trained prediction model in step 6, and let it enter the fully connected layer after passing through 2 Dense Blocks and 2 Transition Layers, and finally obtain the remaining life prediction of the damaged segment of the test set.

Images