Aviation fuel pump fault diagnosis method based on adaptive weighting under data imbalance

Abstract

The invention relates to the technical field of engineering, and discloses an aviation fuel pump fault diagnosis method based on adaptive weighting under data imbalance, comprising the following steps: acquiring an aviation fuel pump operation data sample as a training set; adaptively acquiring distribution information of various types of operation condition samples in the training set; determining the mixing density of each data sample in all categories according to the distribution information, determining a corresponding sample weight according to the proportion of the mixing density, and generating a weight matrix; weighting the training set through the weight matrix, training a neural network model by using the weighted training set, and performing fault diagnosis; according to the method, the diagnosis accuracy of the model under the condition of data imbalance and multi-fault classification of the aviation fuel pump can be improved.

Description

Technical Field

[0001] The present invention relates to the field of engineering technology, and more specifically, to a fault diagnosis method for an aviation fuel pump based on adaptive weighting under data imbalance. Background Art

[0002] The aviation fuel system plays an important role in ensuring the safe and stable navigation of an aircraft, and the aviation fuel pump is a key component of the fuel system, playing a crucial role in the safe operation of the fuel system. As a key component of the aircraft fuel system, the stable operation of the aviation fuel pump is directly related to flight safety. However, due to its high-reliability design and extremely low failure rate, the healthy state data is much more than the failure data. This data imbalance problem poses a great challenge to the fault diagnosis of the aviation fuel pump.

[0003] Currently, there are three main strategies for dealing with the data imbalance problem: 1 undersampling strategy, 2 oversampling strategy, and 3 data generation strategy. Based on the undersampling strategy, by reducing the data volume of the majority class samples (usually healthy state samples) to match that of the minority class samples (fault state samples), the model training time and computing resources can be significantly reduced. Based on the oversampling strategy, by increasing the number of minority class samples to match that of the majority class samples. The common method is to copy the minority class samples or use algorithms to generate new minority class samples, thereby retaining the information of all original data. The data generation strategy mainly generates new samples based on existing samples through mathematical modeling, statistical methods, or data-driven algorithms to expand the dataset and alleviate the data imbalance problem.

[0004] However, the above methods still have their deficiencies, which may lead to data loss, affect the generalization performance of the model, and increase the risk of overfitting in the case of data imbalance. The Broad Learning System (BLS) has shown significant advantages in dealing with the data imbalance problem.

[0005] Therefore, how to improve the accuracy of fault diagnosis of aviation fuel pumps under unbalanced data conditions is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides a fault diagnosis method for an aviation fuel pump based on adaptive weighting under data imbalance, which can construct an adaptive weight matrix between categories, assign higher weights to the few-category data, and improve the model's recognition ability for the few-category fault data.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] A fault diagnosis method for an aviation fuel pump based on adaptive weighting under data imbalance, comprising the following steps:

[0009] Obtain the operation data samples of the aviation fuel pump as the training set;

[0010] Adaptively obtain the distribution information of various operation conditions in the training set;

[0011] Confirm the mixture density of each data sample in all categories according to the distribution information, and confirm the corresponding sample weights according to the proportion of the mixture density, and generate a weight matrix;

[0012] Weight the training set through the weight matrix, and use the weighted training set to train a neural network model for fault diagnosis.

[0013] Preferably, each weight in the weight matrix is confirmed by the following method:

[0014]

[0015] Among them, is the weight of the i-th data sample, is the mixture density of the i-th data sample, n is the total amount of data samples, is the mixture density of the j-th data sample.

[0016] Preferably, adaptively obtain the distribution information of the training set and generate a weight matrix. The steps include:

[0017] Traverse each data sample in the training set;

[0018] For the current data sample during the traversal process , confirm the same-class samples and different-class samples of the current data sample;

[0019] Calculate the local proximity of the current data sample among the same-class samples and among the different-class samples respectively;

[0020] Confirm the mixture density of the current data sample according to the local proximity;

[0021] Calculate the weights of each data sample according to the mixture density of each data sample.

[0022] Preferably, the method for confirming the local proximity includes:

[0023]

[0024]

[0025] Among them, represents the local proximity among the same-class samples; denote The local proximity in other types of samples, where k is the number of samples in the sample set.

[0026] Preferably, the calculation method of the mixed density is as follows:

[0027]

[0028] where is the mixed density, is the trade-off factor.

[0029] Preferably, the neural network model includes a width learning model, and the enhancement layer in the width learning model is used to output class probabilities.

[0030] Preferably, the data processing method of the enhancement layer is specifically:

[0031]

[0032] where is the softmax d function, which is used to add weights to the output of the sample; is the weight parameter matrix of the sample in the feature space, is the diagonal matrix used to handle class imbalance between samples, and the middle element in the diagonal matrix is the weight value corresponding to each sample .

[0033] Preferably, the neural network model further includes an SVM model, and the SVM model is used to further classify according to the class probability to confirm the final class.

[0034] An aviation fuel pump fault diagnosis system based on adaptive weighting under data imbalance includes a data acquisition module, a distribution information and weight generation module, and a model training and prediction module.

[0035] The data acquisition module is used to acquire the operation data sample set of the aviation oil pump; the distribution information and weight generation module is used to acquire the operation sample set and adaptively generate a weight matrix according to the distribution information of each working condition sample in the sample set; the model training set prediction module is used to weight the samples in the training set according to the weight matrix, and train the neural network model according to the weighted training samples; and is used to predict the fuel pump fault according to the trained neural network model.

[0036] A computer-readable storage medium stores a computer program, and when the computer program is executed, the above-mentioned aviation fuel pump fault diagnosis method is implemented.

[0037] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a fault diagnosis method for an aviation fuel pump based on adaptive weighting under data imbalance, which can calculate the weights of each sample according to the class distribution of the samples, form a class weight matrix at the sample level, and realize assigning higher weights to the less-class data by weighting each sample, improve the recognition ability of the model for the less-class fault data, and can improve the diagnostic accuracy of the model under data imbalance conditions and for multi-fault classification of aviation fuel pumps. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts.

[0039] Figure 1 It is a schematic diagram of a fault diagnosis method for an aviation fuel pump based on adaptive weighting under data imbalance provided by the present invention.

[0040] Figure 2 It is a schematic diagram of the data analysis effect of the ablation experiment in the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0042] Embodiment 1

[0043] The embodiment of the present invention discloses a fault diagnosis method for an aviation fuel pump based on adaptive weighting under data imbalance, including the following steps:

[0044] S1: Obtain the operation data samples of the aviation fuel pump as the training set. Among them, the collected training samples are imbalanced.

[0045] S2: Construct an adaptive weight generation matrix. Adaptively obtain the distribution information of various operating conditions in the training set, confirm the mixed density of each data sample in all classes according to the distribution information, and confirm the corresponding sample weights according to the proportion of the mixed density, and generate a weight matrix.

[0046] S3: Weight the training set using a weight matrix and train a neural network model using the weighted training set; for example, establish an improved width learning system based on the weight matrix, and construct a fault diagnosis model based on the width learning system for training.

[0047] S4: Use the trained fault diagnosis model for fault diagnosis and output the fault diagnosis result.

[0048] In this embodiment, weighting adjusts the loss and gradient during training, enabling the model to "pay more attention" to data of minority classes, thereby solving the data imbalance problem and improving the model's performance on minority classes. The weight matrix affects loss calculation, gradient update, and the final model's prediction output through weighting during training. For example, in a fault diagnosis task, a width learning network (BLS) is used for training. The weights of each sample in the training set are adjusted in an adaptive weighting manner, enabling important samples to have a greater impact on the model's learning during training. Then, an extended layer is used for feature mapping to obtain a richer feature representation.

[0049] Exemplarily, if the cross-entropy loss function is adopted, its formula is:

[0050]

[0051] where is the loss value, N is the number of samples, is the true label of sample i, is the predicted probability of sample i.

[0052] In the case of data imbalance, there are samples of the minority class and N samples of the majority class. Without weighting, the total loss is:

[0053]

[0054] where is the total loss under data balance.

[0055] When sample weights are introduced, the weighted total loss is:

[0056]

[0057] where is the total loss under sample weighting in the case of data imbalance, is the sample weight.

[0058] Through this weighting, the proportion of the loss of minority-class samples in the total loss increases. The present invention can provide sample-level weights, and each sample has a corresponding weight value. In addition, sample categories can be further considered on the basis of samples, and the weights of samples of the same category are aggregated to confirm the weights between categories. At this time, each category corresponds to a weight value.

[0059] To further implement the above technical solution, the specific data collected in S1 is the sound data when the aviation fuel pump is running. A handheld sound acquisition device is used to collect the sound signal of the running aviation fuel pump, and then the collected sound signal is sampled to construct an imbalanced dataset. Exemplarily, during the collection process, the signal sampling interval is set to 21 ms. A total of four working conditions of sound data are collected, with 200 samples for each working condition and the length of each sample being 1200.

[0060] Further, the steps for constructing the imbalanced dataset include:

[0061] First, signal processing is performed. The data under the four working conditions collected are discretized at a sampling interval of 21 ms, covering a total duration of 1050 s. Subsequently, the signal is normalized to more clearly extract time-domain features such as the kurtosis, deviation, variance, and root mean square value of the signal.

[0062] Then, the signal processing results under the four working conditions are sampled to construct training data and test data; the four working conditions include 1 healthy bearing working condition and 3 different fault category working conditions of real damage. For the training data, the imbalanced dataset is respectively composed of 100 healthy samples and 60 fault state samples. For the test data, the total number of data samples for the 4 working conditions is 100.

[0063] To further implement the above technical solution, in S2, the distribution information of the training set is adaptively obtained and a weight matrix is generated. The steps include:

[0064] S21: Traverse each data sample in the training set.

[0065] S22: For the current data sample during the traversal , identify the same-class samples and different-class samples of the current data sample; calculate the local proximity of the current data sample among the same-class samples and among the different-class samples respectively.

[0066] Specifically, according to the same-class sample set and different-class sample set corresponding to the current data sample, calculate the local proximity:

[0067]

[0068]

[0069] Among them, represents the local proximity in the same - type samples; represents the local proximity in other - type samples; e is its average value.

[0070] S23: Confirm the mixed density of the current data sample according to the local proximity.

[0071] Specifically, combining the proximity between the same - type and other - type samples, and then calculating its mixed density through its (trade - off factor) as follows: as shown in the following formula:

[0072]

[0073] Among them, in this case, only consider the proximity of the same - type, completely ignoring the influence of different - type neighbors; : at this time, only consider the different - type proximity, completely ignoring the proximity of the same - type neighbors; : this means that both the same - type and different - type contribute.

[0074] S24: Calculate the weights of each data sample according to the mixed proximity of each data sample.

[0075] Specifically, normalize the mixed proximity of each sample as shown in the following formula:

[0076]

[0077] Then, calculate the data weight as shown in the following formula:

[0078]

[0079] To further implement the above - mentioned technical solution, the neural network model adopted is a hybrid model of a Broad Learning System model (BLS) and a Support Vector Machine model (SVM); input the original data into the BLS model, and output the class probabilities of the working conditions through the improved BLS model. The classes include healthy (normal) and various fault types. After obtaining the class probabilities, perform a further classification through the SVM model to confirm the final class. Among them, the improvement of the BLS model lies in: to face the multi - classification problem, optimize the non - linear activation function of the enhancement layer to the softmax function, softmaxWeighted is represented as and weight the output of each of its samples as shown in the following formula:

[0080]

[0081] Among them, is the activation function softmax, is adjusted by giving weights to the softmax function; is the weight matrix, which is a diagonal matrix and will adjust the imbalanced data.

[0082] Specifically, the data processing steps of the BLS model are as follows:

[0083] S31: Obtain the original sample data, and map the original sample data through randomly generated parameters to obtain the mapped features and form a feature layer.

[0084] S32: Enhance the features of the feature layer using the softmax function and form an enhanced layer. Among them, the enhanced layer is the predicted probability output for each category.

[0085] In this embodiment, the contraction parameter s of the width learning system is set to 0.8, the regularization parameter is set to 2−30, the number of same-class and different-class neighbors and h are both set to 5. The mixed weight parameter is set to 0.5. The BLS structure parameters, namely the number of feature nodes 1, the number of windows 2, and the number of enhanced nodes 𝑁3, are set to 10, 5, and 50 respectively. For SVM, the penalty factor and the parameter of the radial basis function kernel are set to 10.0 and 0.01 respectively.

[0086] Furthermore, an SVM classifier is used for classification, and the steps are as follows:

[0087] S33: Construct and into the feature layer extracted by BLS, and the way to construct the feature set is shown in the following formula:

[0088] S34: For each working condition category, train an SVM model respectively, and use one kind of data as the positive class and other data as the negative class for binary classification. Each model will generate a decision function , as shown in the following formula:

[0089]

[0090] Among them, is the confidence of each input value ; is the weight vector of category , and each element in it represents the weight of a specific feature under this category, that is, the influence of the feature on the classification decision of this category; is the bias term, is the input feature vector.

[0091] S35: Train the model, and the process is shown as follows:

[0092]

[0093] where \(w\) and \(b\) are the weight vector and bias, \(\lambda\) is the regularization parameter; \(\xi\) is the slack variable.

[0094] S36: Use the trained model for classification, and the predicted class is the maximum class of , as shown in the following formula:

[0095]

[0096] Furthermore, after the model training is completed, use the test set to verify the accuracy of the model and conduct a comparative analysis.

[0097] Specifically, use the test set to verify the accuracy of the model and conduct a comparative analysis. An ablation experiment was carried out, and KNN, random forest, multi-layer perceptron (MLP), original BLS, AWBLS (not optimized by SVM), and the proposed AW-IBLS model in this paper were selected for comparative analysis to verify the effects of each model in dealing with imbalanced data.

[0098] Calculate the precision:

[0099]

[0100] where, \(TP\) is the number of true positives of class \(c\), \(FP\) is the number of negative examples mispredicted as class \(c\).

[0101] Calculate the recall:

[0102]

[0103] where, \(FN\) is the number of actual class \(c\) but mispredicted by the model.

[0104] Calculate the F1 score, which is the harmonic mean of the precision and recall:

[0105]

[0106] To more intuitively represent the superiority of the algorithm, a confusion matrix was used to visually analyze the diagnostic results as Figure 2As shown, it can be seen from the confusion matrix that the method proposed by the present invention can achieve a diagnostic accuracy of 99% under data imbalance.

[0107] Embodiment 2

[0108] Based on the same inventive concept, an embodiment of the present invention discloses a fault diagnosis system for an aviation fuel pump based on adaptive weighting under data imbalance, which adopts the aviation fuel pump fault diagnosis method in Embodiment 1, and includes a data acquisition module, a distribution information and weight generation module, and a model training and prediction module.

[0109] The data acquisition module is used to acquire the operation data sample set of the aviation oil pump; the distribution information and weight generation module is used to acquire the operation sample set and adaptively generate a weight matrix according to the distribution information of each working condition sample in the sample set; the model training and prediction module is used to weight the samples in the training set according to the weight matrix, and train a neural network model according to the weighted training samples; and is used to predict the faults of the fuel pump according to the trained neural network model.

[0110] Embodiment 3

[0111] Based on the same inventive concept, an embodiment of the present invention discloses a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed, the aviation fuel pump fault diagnosis method in Embodiment 1 is implemented.

[0112] In this specification, the various embodiments are described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts among the various embodiments can be referred to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0113] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but will be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for fault diagnosis of aviation fuel pump based on adaptive weighting under data imbalance, characterized in that: The following steps are involved: Obtain aviation fuel pump operation data samples as a training set; Adaptively obtaining distribution information of operating condition samples of each category in the training set; Determine the mixed density of each data sample in all categories according to the distribution information, and determine the corresponding sample weights according to the proportion of the mixed density to generate a weight matrix; The training set is weighted by the weight matrix, and the weighted training set is used to train a neural network model to perform fault diagnosis.

2. The method for diagnosing aviation fuel pump faults based on adaptive weighting under data imbalance according to claim 1 is characterized in that: Each weight in the weight matrix is ​​confirmed by the following method: ； in, is the weight of the ith data sample, For the The mixing density of data samples, n is the total number of data samples, For the The mixture density of the data samples.

3. The method for diagnosing aviation fuel pump faults based on adaptive weighting under data imbalance according to claim 1 or 2, characterized in that: Adaptively obtaining the distribution information of the training set and generating a weight matrix, the steps include: Traversing each data sample in the training set; For the current data sample during the traversal , confirm the samples of the same category and different category as the current data sample; Calculate the local proximity of the current data sample in the same type of samples and in different types of samples respectively; Confirm the mixture density of the current data sample based on local proximity; The weight of each data sample is calculated according to its mixture density.

4. The method for diagnosing aviation fuel pump faults based on adaptive weighting under data imbalance according to claim 3 is characterized in that: The local proximity confirmation method includes: ； ； in, express Local proximity among similar samples; express The local proximity among samples of other classes, k is the number of samples in the sample set.

5. The method for diagnosing aviation fuel pump faults based on adaptive weighting under data imbalance according to claim 4, characterized in that: The mixed density is calculated as: ； in, is the mixed density, is a trade-off factor.

6. The method for diagnosing aviation fuel pump faults based on adaptive weighting under data imbalance according to claim 1, characterized in that: The neural network model includes a width learning model, in which an enhancement layer is used to output category probabilities.

7. The method for diagnosing aviation fuel pump faults based on adaptive weighting under data imbalance according to claim 6, characterized in that: The data processing method of the enhancement layer is specifically as follows: ； in, is softmax d function, used to add weight to the output of the sample; is the weight parameter matrix of the sample in the feature space, is a diagonal matrix used to process category imbalance between samples, that is, the weight matrix.

8. The method for diagnosing aviation fuel pump faults based on adaptive weighting under data imbalance according to claim 6 or 7, characterized in that: The neural network model also includes an SVM model, and the SVM model is used to further classify according to the category probability to confirm the final category.

9. An aviation fuel pump fault diagnosis system based on adaptive weighting under data imbalance, characterized in that: The aviation fuel pump fault diagnosis method according to any one of claims 1 to 8 is adopted, comprising a data acquisition module, a distribution information and weight generation module, and a model training and prediction module; The data acquisition module is used to acquire a sample set of aviation fuel pump operation data; The distribution information and weight generation module is used to obtain the operating sample set and adaptively generate a weight matrix according to the distribution information of each working condition sample in the sample set; The model training set prediction module is used to weight the samples in the training set according to the weight matrix, and train the neural network model according to the weighted training samples; and is used to predict the fuel pump failure according to the trained neural network model.

10. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, which, when executed, implements the aviation fuel pump fault diagnosis method according to any one of claims 1 to 8.

Images