Aero-engine assembly quality anomaly detection and feature inference method based on auto-encoder

Abstract

The invention relates to the technical field of aero-engine performance optimization, in particular to an aero-engine assembly quality anomaly detection and feature inference method based on auto-encoders, which comprises the following steps: acquiring assembly test data sets of an aero-engine, and constructing a stacked auto-encoder corresponding to each training data set; each stacked auto-encoder is composed of an encoder and a decoder; according to each trained stacked auto-encoder, a Bagging stacked auto-encoder model is constructed; setting a confidence level to determine a dynamic threshold of the abnormal data; and carrying out anomaly detection on the assembly quality of the aero-engine on the basis of the dynamic threshold value according to the assembly test data of the aero-engine to be detected and the Bagging stacking auto-encoder model. According to the method, efficient low-dimensional representation of the aero-engine assembly quality data is realized, the data reconstruction error is effectively reduced, and the robustness and generalization ability of feature learning are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of aircraft engine performance optimization, and in particular to an aircraft engine assembly quality anomaly detection and feature inference method based on an autoencoder. Background Art

[0002] Aircraft engines are a highly complex and sophisticated type of mechanical equipment that provides aircraft flight power, involves many disciplines, and integrates many cutting-edge technologies. They have extremely strict requirements on reliability and environmental adaptability, resulting in high investment costs, high development risks, and long development cycles in their development process.

[0003] In actual production, the assembly and test run of aircraft engines are two crucial links in the manufacturing process. The assembly quality of aircraft engines directly determines the performance of aircraft engines. The performance of aircraft engines needs to be tested by test runs. Only aircraft engines that meet the test run performance specifications can be put into actual use. Otherwise, the aircraft engines will be disassembled and reassembled until they meet the test run test requirements. The assembly process of aircraft engines involves many parts. It not only requires accurate positioning of the parts, but also ensures that the relative positions of all parts are kept within the standard range. Moreover, the connection reliability and assembly accuracy must meet strict specifications, which makes the assembly process of aircraft engines extremely cumbersome. However, the assembly process of aircraft engines mainly relies on manual operation, and workers are often unable to accurately complete such high-precision assembly tasks. Even a small assembly feature error will lead to the degradation of the test run performance of the aircraft engine and affect the overall performance of the whole machine. At present, there is no set of systematic assembly specification guidance documents for the assembly process of aircraft engines. The assembly range of assembly quality features is generally determined based on the experience of workers. The assembly quality varies greatly, resulting in a large range of fluctuations in the test run performance results. Aircraft engines that do not meet the test run requirements need to be disassembled and reassembled, which seriously wastes production resources and increases production costs. In order to improve the accuracy and efficiency of aircraft engine assembly, from the perspective of test performance, complete the detection of aircraft engine assembly quality anomalies, identify assembly anomalies, and infer the quality characteristic assembly range corresponding to specific test performance indicators. Provide more detailed and scientific assembly guidance, thereby improving the first-time test pass rate, reducing aircraft engine production costs, and improving the market competitiveness of aircraft engine manufacturers.

[0004] The prior art discloses a method for detecting and inferring the quality anomaly of aircraft engine assembly based on autoencoders. First, the aircraft engine assembly and test data are collected and sorted, and data preprocessing is performed to ensure the neatness of the data; secondly, based on the core idea of ​​ensemble learning and based on autoencoders, a bagging stacked autoencoder model is constructed to learn the features of aircraft engine assembly data and minimize the data reconstruction error; then, based on the bagging stacked autoencoder model, a dynamic threshold strategy is proposed for anomaly detection, and the sample is identified as abnormal by comparing the reconstruction error of the sample with the dynamic threshold; finally, the loss function is optimized by matching the encoding space of the bagging stacked autoencoder model with the test performance value, and the specific test performance index is used as the input of the decoder to obtain the corresponding aircraft engine assembly quality features, and the quality feature assembly interval is inferred.

[0005] In the field of quality anomaly detection, although existing technologies have made certain progress, there are still inevitable problems and challenges. Traditional quality anomaly detection, such as statistical methods and control charts, relies on rules obtained by experience or manually selected features. They usually work well when the feature space is small and the data distribution is simple. However, when faced with complex high-dimensional data, the performance is often unsatisfactory. The aircraft engine assembly process involves a large number of complex and mutually influential features. Traditional methods are difficult to capture the relationship between the quality of aircraft engine assembly, resulting in low detection accuracy and poor robustness. With the popularization of machine learning, algorithms based on supervised learning and unsupervised learning have been widely introduced into anomaly detection tasks. For example, support vector machines and decision trees have outstanding anomaly detection effects in specific scenarios. However, these methods usually require a large amount of labeled data for training, while in actual scenarios, it is difficult and costly to obtain labels, and they are often difficult to obtain. In addition, supervised learning methods have poor generalization ability for abnormal samples and are easily affected by the distribution of training sets; while unsupervised learning relies on distance metrics or clustering assumptions, and is difficult to adapt to complex, nonlinear high-dimensional data of aircraft engine assembly. Based on the above research deficiencies, the autoencoder, as a deep learning algorithm, can effectively extract potential features by learning the low-dimensional feature representation of the input data and mapping it to the output, and accurately judge the anomaly by reconstructing the error. The autoencoder can adapt to the anomaly detection of complex high-dimensional data, and has strong generalization ability. It is widely used in the field of quality anomaly detection, but it is still susceptible to noise interference and overfitting. It needs to be optimized and improved to overcome the shortcomings, so as to improve the performance of the autoencoder in anomaly detection. Summary of the invention

[0006] The technical solution of the present invention aims at the problems existing in the above-mentioned background technology, and proposes a method for abnormal detection and feature inference of aircraft engine assembly quality based on autoencoders. Based on the stacked autoencoder model, the present invention combines the core idea of ​​ensemble learning, matches the encoding space of the stacked autoencoder with the test performance, and constructs a bagging stacked autoencoder model by optimizing the loss function, so as to realize efficient low-dimensional representation of aircraft engine assembly quality data, effectively reduce data reconstruction errors, and improve the robustness and generalization ability of feature learning.

[0007] The first object of the present invention is to provide a method for detecting and inferring the quality anomaly of an aircraft engine assembly based on an autoencoder, comprising:

[0008] Acquire an assembly and test data set of an aircraft engine, wherein the assembly and test data set includes an assembly quality data set and a corresponding test run performance data set;

[0009] Generate multiple training data sets by random sampling with replacement from the test data set;

[0010] Construct a stacked autoencoder corresponding to each training dataset; each stacked autoencoder consists of an encoder and a decoder;

[0011] Using each training data set to train the corresponding stacked autoencoder until convergence, each trained stacked autoencoder is obtained;

[0012] During the training process, the assembly quality data is used as the input of the stacked autoencoder; the test performance data is used as the encoded feature representation and as the input data of the decoder;

[0013] Build a Bagging stacked autoencoder model based on each trained stacked autoencoder;

[0014] Set confidence levels to determine dynamic thresholds for abnormal data;

[0015] The assembly and test data of the aircraft engine to be tested is used according to the Bagging stacked autoencoder model, and abnormality detection of the aircraft engine assembly quality is performed based on the dynamic threshold.

[0016] Preferably, the loss function for training each stacked autoencoder includes:

[0017] The error between the input data and the output of the stacked autoencoder is taken as the reconstruction error;

[0018] The error between the encoding of the stacked autoencoder and the test performance data is taken as the encoding error;

[0019] The sum of the reconstruction error and the encoding error is used as the loss function of the stacked autoencoder.

[0020] Preferably, the step of determining the dynamic threshold of abnormal data includes:

[0021] The test data set is used to generate a validation data set through random sampling with replacement; both the training set and the validation set contain normal data;

[0022] Use the validation set to reconstruct the Bagging stacked autoencoder model and obtain the reconstruction error of the validation set;

[0023] The dynamic threshold of abnormal data is determined by setting the confidence level according to the reconstruction error of the validation set.

[0024] Preferably, when constructing a bagging stacked autoencoder model, it includes:

[0025] The average of the output data of each trained stacked autoencoder is taken as the final output of the model.

[0026] Preferably, the method further comprises: using the set aircraft engine test performance as the input of the decoder of each stacked autoencoder in the Bagging stacked autoencoder model to infer the assembly interval of the aircraft engine assembly quality feature.

[0027] Preferably, obtaining the assembly interval inference of the aircraft engine assembly quality characteristics includes:

[0028] Discretize the set aircraft engine test performance to obtain multiple discrete values;

[0029] The discrete values ​​are used as the input of the decoders of each stacked autoencoder in the bagging stacked autoencoder model to reconstruct the corresponding assembly quality feature values;

[0030] Fitting the reconstructed corresponding assembly quality characteristic values ​​to obtain an assembly quality characteristic variation curve;

[0031] The assembly range of aircraft engine assembly quality characteristics is inferred based on the assembly quality characteristic change curve.

[0032] A second object of the present invention is to provide a computer program product, comprising a computer program, which, when executed by a processor, implements a method for detecting and inferring anomalous quality of aircraft engine assembly based on an autoencoder.

[0033] A third object of the present invention is to provide an electronic device, comprising:

[0034] Processor; and

[0035] A memory, configured to store executable instructions of the processor;

[0036] The processor is configured to execute the method of autoencoder-based aircraft engine assembly quality anomaly detection and feature inference by executing the executable instructions.

[0037] A fourth object of the present invention is to provide a system for detecting and inferring the quality anomaly of an aircraft engine assembly based on an autoencoder, comprising:

[0038] The data acquisition module is used to obtain an assembly and test data set of an aircraft engine, wherein the assembly and test data set includes an assembly quality data set and a corresponding test performance data set; the assembly and test data set is used to generate multiple training data sets by a random sampling method with replacement;

[0039] The encoder model construction module is used to construct a stacked autoencoder corresponding to each training data set; each stacked autoencoder consists of an encoder and a decoder; each training data set is used to train the stacked autoencoder corresponding to it until convergence, so as to obtain each trained stacked autoencoder; wherein, during the training process, the assembly quality data is used as the input of the stacked autoencoder; the test performance data is used as the feature representation of the encoding and as the input data of the decoder; a bagging stacked autoencoder model is constructed according to each trained stacked autoencoder;

[0040] The anomaly detection module is used to set the confidence level to determine the dynamic threshold of abnormal data; the assembly and test data of the aircraft engine to be detected is used according to the Bagging stacked autoencoder model, and the anomaly detection of the aircraft engine assembly quality is performed based on the dynamic threshold.

[0041] The present invention has at least the following beneficial effects:

[0042] The present invention provides a method for abnormality detection and feature inference of aircraft engine assembly quality based on an autoencoder. The present invention realizes abnormality detection and fault diagnosis of assembly quality from the perspective of test performance, and infers assembly-related information that can meet the test performance requirements; the present invention is based on a stacked autoencoder model, combines the core idea of ​​ensemble learning, matches the encoding space of the stacked autoencoder with the test performance, constructs a Bagging stacked autoencoder model by optimizing the loss function, realizes efficient low-dimensional representation of aircraft engine assembly quality data, effectively reduces data reconstruction errors, and improves the robustness and generalization ability of feature learning; the modeling process of the present invention is simple and efficient, and has good applicability to abnormality detection and feature inference methods in other manufacturing scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a schematic diagram of the structure of the autoencoder.

[0044] Figure 2Schematic diagram of the structure of stacked autoencoders.

[0045] Figure 3 This is a framework diagram of the bagging stacked autoencoder model of the present invention.

[0046] Figure 4 Framework diagram for aircraft engine assembly quality anomaly detection and feature inference based on Bagging stacked autoencoders.

[0047] Figure 5 Flowchart for aircraft engine assembly quality feature inference based on Bagging stacked autoencoders.

[0048] Figure 6 This is the structural diagram of the stacked autoencoder in this example.

[0049] Figure 7 This is the stacked autoencoder training error curve for this example.

[0050] Figure 8 This is a comparison chart of the dynamic thresholds of the stacked autoencoder and the bagging stacked autoencoder models in this example.

[0051] Fig. 9 This is a scatter plot of the reconstruction error of the Bagging stacked autoencoder model in the training set, validation set, and test set.

[0052] Fig.10 This is a comparison chart of the F1 scores of the stacked autoencoder and bagging stacked autoencoder models in this example.

[0053] Fig.11 This is the assembly interval inference diagram of the aircraft engine assembly quality characteristics in this example. DETAILED DESCRIPTION

[0054] In order to illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the following is a detailed description in conjunction with embodiments.

[0055] Based on the current status of anomaly detection research in the prior art, the method proposed in the present invention can better capture the potential characteristics of high-dimensional complex data, enhance the accuracy of data reconstruction, and thus more accurately identify anomalies; the model constructed based on the ensemble learning method can effectively reduce the impact of noise and improve the sensitivity and robustness of the model to anomalies; the proposed method embeds the test performance into the coding space, improves the generalization ability of the model, and realizes the inference of assembly quality characteristics; the method is suitable for assembly quality anomaly detection in various complex environments, and has excellent adaptability and application range.

[0056] The purpose of the present invention is to optimize the assembly quality of aircraft engines. The present invention provides an aircraft engine assembly quality anomaly detection and feature inference method based on autoencoders. The method integrates multiple stacked autoencoders using the Bagging method, matches the encoding space of the stacked autoencoders with the test performance, constructs a Bagging stacked autoencoder model, and learns the feature expression of the aircraft engine assembly quality; then, a dynamic threshold strategy is proposed to realize the detection of aircraft engine assembly quality anomalies; finally, by using the characteristics of the decoder, a specific test performance index is input to realize the inference of the assembly intervals of various quality features of the aircraft engine.

[0057] The main contents of the present invention include:

[0058] Obtain aircraft engine assembly quality and test performance data, and perform simple data preprocessing, such as outlier and missing value processing;

[0059] Generate multiple training sets through random sampling with replacement, use the bagging ensemble method, train multiple stacked autoencoder models with different performances according to different training sets, match the encoding space of the stacked autoencoder with the test performance, optimize the loss function of the stacked autoencoder, take the average of the reconstructed output data of all individual learners as the final output result, and build the bagging stacked autoencoder model;

[0060] According to the dynamic threshold strategy, the judgment threshold of assembly quality anomaly is obtained, and the task of assembly quality anomaly detection is achieved by comparing it with the reconstruction error of the sample;

[0061] The required aircraft engine test performance value is used as the input of the decoder in the model to realize the inference of the assembly interval of the aircraft engine quality characteristics.

[0062] In order to achieve the above purpose, a method for detecting and inferring the quality anomaly of aircraft engine assembly based on an autoencoder comprises:

[0063] S1. Acquire an assembly and test data set of an aircraft engine, wherein the assembly and test data set includes an assembly quality data set and a corresponding test run performance data set;

[0064] Generate multiple training data sets by random sampling with replacement from the test data set;

[0065] S2. Construct a stacked autoencoder corresponding to each training data set; each stacked autoencoder consists of an encoder and a decoder;

[0066] S3, using each training data set to train the corresponding stacked autoencoder until convergence, to obtain each trained stacked autoencoder;

[0067] The loss function for training each stacked autoencoder includes:

[0068] The error between the input data and the output of the stacked autoencoder is taken as the reconstruction error;

[0069] The error between the encoding of the stacked autoencoder and the test performance data is taken as the encoding error;

[0070] The sum of the reconstruction error and the encoding error is used as the loss function of the stacked autoencoder.

[0071] During the training process, the assembly quality data is used as the input of the stacked autoencoder; the test performance data is used as the encoded feature representation and as the input data of the decoder;

[0072] S4. Build a Bagging stacked autoencoder model based on each trained stacked autoencoder;

[0073] Set confidence levels to determine dynamic thresholds for abnormal data;

[0074] When building a bagging stacked autoencoder model, it includes:

[0075] The average of the output data of each trained stacked autoencoder is taken as the final output of the model.

[0076] The step of determining the dynamic threshold of abnormal data includes:

[0077] The test data set is used to generate a validation data set through random sampling with replacement; both the training set and the validation set contain normal data;

[0078] Use the validation set to reconstruct the Bagging stacked autoencoder model and obtain the reconstruction error of the validation set;

[0079] The dynamic threshold of abnormal data is determined by setting the confidence level according to the reconstruction error of the validation set.

[0080] S5. The assembly and test data of the aircraft engine to be tested is used according to the bagging stacked autoencoder model, and abnormality detection of the aircraft engine assembly quality is performed based on the dynamic threshold.

[0081] It also includes: using the set aircraft engine test performance as the input of the decoder of each stacked autoencoder in the Bagging stacked autoencoder model to infer the assembly interval of the aircraft engine assembly quality characteristics.

[0082] When obtaining the assembly interval inference of the aircraft engine assembly quality characteristics, it includes:

[0083] Discretize the set aircraft engine test performance to obtain multiple discrete values;

[0084] The discrete values ​​are used as the input of the decoders of each stacked autoencoder in the bagging stacked autoencoder model to reconstruct the corresponding assembly quality feature values;

[0085] Fitting the reconstructed corresponding assembly quality characteristic values ​​to obtain an assembly quality characteristic variation curve;

[0086] The assembly range of aircraft engine assembly quality characteristics is inferred based on the assembly quality characteristic change curve.

[0087] In order to illustrate the method for aircraft engine assembly quality anomaly detection and feature inference based on an autoencoder provided by the present invention, it is described in conjunction with the accompanying drawings.

[0088] In the aircraft engine manufacturing scenario, the aircraft engine assembly quality data includes the geometric dimensions of parts, assembly errors, etc., which are recorded as where x i represents the i-th assembly quality feature, and N represents the number of features contained in the aircraft engine assembly feature set. The aircraft engine test data refers to the state values ​​recorded by the sensor during the test, which is recorded as where y J Represents the jth aircraft engine test performance, and M represents the number of aircraft engine test performance indicators. The assembly data and test data collected from the production site are organized into an aircraft engine assembly and test data set, and duplicate data records in the data set are deleted, missing values ​​and outliers are processed, data normalization and normalization are achieved, data quality is improved, and data analysis difficulty is reduced. For an aircraft engine assembly and test data set containing P samples, the present invention combines the core idea of ​​ensemble learning, based on a stacked autoencoder model, and proposes an aircraft engine assembly quality anomaly detection and feature inference method based on an autoencoder, as shown below.

[0089] Ensemble learning is a method that integrates multiple weak learners into a strong learner according to a specific integration strategy. It can effectively improve the prediction accuracy and generalization ability of the model, and make up for the defect that a single model cannot fully capture data features. As one of the most classic methods of ensemble learning, the bagging method obtains T new data sets through random sampling with replacement on the original data set to construct T learners, and integrates the T results according to the combination strategy to output the final result. For classification problems, the combination strategy is generally the voting method, and the category with the most votes is the final output. For regression problems, the combination strategy is generally the average method or the weighted method, and the average or weighted average is calculated as the model output. The bagging method performs very well and has a strong generalization ability, which is of great significance in reducing the variance of model predictions.

[0090] An autoencoder is an unsupervised learning algorithm based on the data feature expression of a neural network architecture. It uses a neural network to obtain a low-dimensional expression of high-dimensional data, i.e., encoding, and then uses the encoding to reconstruct the data. An autoencoder consists of an encoder and a decoder, and its structure is as follows: Figure 1 As shown. The encoder is used to receive input data and map its value to a low-dimensional expression to complete the "encoding" of high-dimensional data. The decoder converts the code back to an approximate expression of the original data to achieve reconstruction of the input data. In order to improve the accuracy of data reconstruction, the network structure of the encoder and decoder is generally mirror-symmetric, and the loss function of the autoencoder is the difference between the input data and the reconstructed data. In the forward propagation process, the encoder uses the function h=g(x) to encode the input data, and the decoder uses the function r=f(h) to reconstruct the code into the output result. In the back propagation, the gradient descent algorithm is used to update the network weights to minimize the loss function so that the input data x and the reconstructed data r are close to each other. As a variant of the autoencoder, the stacked autoencoder is a deep learning model composed of multiple autoencoders stacked in series. Its structure is as shown Figure 2 As shown in the figure, the hidden layer of the autoencoder serves as the input layer of another autoencoder, and each layer learns features at different levels, thereby forming a multi-layer network structure, which makes the stacked autoencoder have better feature interpretation ability and can mine more effective information such as aircraft engine assembly quality features.

[0091] For complex aircraft engine assembly and test data, stacked autoencoders are used as the basic model for anomaly detection and feature inference. However, a single stacked autoencoder often lacks stable robustness and generalization ability, resulting in inaccurate results of anomaly detection and feature inference. Combining the advantages of ensemble learning, the bagging method is used to integrate the output results of multiple stacked autoencoders to reduce the error of data reconstruction. Therefore, this section proposes a bagging stacked autoencoder model based on ensemble learning to realize anomaly detection of aircraft engine assembly quality features, thereby further realizing the range inference of assembly quality features.

[0092] Let the sample of the aircraft engine test data set be represented as {X k ,Y k}, k = 1, ..., P, where Indicates assembly quality data, Represents the test performance data. The framework of the bagging stacked autoencoder model is as follows Figure 3 As shown in the figure, Q training data sets are generated from the original data set by random sampling with replacement; these training data sets are used to obtain the corresponding Q stacked autoencoders, namely AE1,…,AE m ,…,AE QThe stacked autoencoder model is used as the individual learner of the bagging method. The output of each stacked autoencoder can be expressed as Where m is 1,…,Q. The task of reconstructing the quality data of aircraft engine assembly is obviously a regression problem. The integration strategy is to calculate the average value of all stacked autoencoder output results as the final result and construct a Bagging stacked autoencoder model. The final output result of the model is shown in formula (1).

[0093]

[0094] When constructing a stacked autoencoder as an individual learner, in order to achieve the purpose of inferring the range of characteristic values ​​of aircraft engine assembly quality from aircraft engine test performance, the encoding dimension is kept consistent with the number of aircraft engine test performance indicators. The loss function of the stacked autoencoder is optimized as the sum of the reconstruction error and the encoding error. While realizing the stacked autoencoder to restore the input data, the test performance information is embedded in the encoding of the autoencoder to improve the feature reasoning ability. The calculation formula is shown in formula (2):

[0095]

[0096] in, Represents the reconstruction error, that is, the error between the input data and the output result;

[0097] represents the encoding error, that is, the error value between the encoding and the test performance of the stacked autoencoder.

[0098] The logical framework based on the constructed bagging stacked autoencoder model to realize aero-engine assembly quality anomaly detection and feature inference is as follows: Figure 4As shown. In the process of training the model, the data set is divided into a training set, a validation set and a test set by a random sampling method with replacement, wherein the training set and the validation set only contain normal data, and the test set contains normal data and abnormal data. The bagging stacked autoencoder can obtain the learning ability to reconstruct normal data through training with the training set, but it is difficult to achieve the reconstruction of abnormal data. According to the principle of the autoencoder, the reconstruction error of the abnormal data of the assembly quality of the aircraft engine will be significantly greater than the reconstruction error of the normal data. Therefore, setting an accurate error threshold is conducive to improving the efficiency of anomaly detection. The present invention proposes a dynamic threshold strategy for anomaly detection. The reconstruction error of the calculation model on the validation set obtained based on random sampling is set, and the confidence level is set to determine the dynamic threshold of the abnormal data. For example, if the reconstruction error of 99% of the normal data is less than d, then d is used as the judgment threshold of the abnormal data at the 99% confidence level. By adjusting the confidence level, different dynamic thresholds for distinguishing normal data from abnormal data are obtained. The confidence level depends on the actual situation and is generally between 98% and 99.5%. This ensures that the Bagging stacked autoencoder model can accurately identify abnormal data, avoid mistaking normal data for abnormal data, and improve the accuracy of abnormality detection in aircraft engine assembly quality.

[0099] Based on the Bagging stacked autoencoder model, the specified aircraft engine test performance indicators are used as the encoded feature representation and as the input data of the decoder in the stacked autoencoder to obtain the reconstructed output of each stacked autoencoder. The mean of all stacked autoencoder output results is obtained according to the integration method, which is the reconstruction result of the aircraft engine assembly quality of the Bagging stacked autoencoder model. The process is as follows: Figure 5 shown.

[0100] In summary, the present invention pre-processes the collected aircraft engine assembly data and test data and organizes them into a systematic aircraft engine assembly and test data set. Combining the core idea of ​​ensemble learning, stacked autoencoders are used as individual learners, the encoding of stacked autoencoders is represented as aircraft engine test performance, and the loss function of the autoencoder is optimized, thereby constructing a bagging stacked autoencoder model.

[0101] Then, a dynamic threshold strategy for anomaly detection is proposed. Based on the reconstruction error distribution of the validation set, the corresponding confidence level is set according to actual experience, so as to obtain the dynamic threshold under the confidence level, and compare the reconstruction error of the test sample with it, so as to realize the identification of abnormal data samples. The dynamic threshold under the set confidence level is obtained, and the reconstruction error of the test sample is compared with it, so as to realize the identification of abnormal data samples. It should be noted that the confidence level is set empirically according to the actual situation, and the range is generally small, between 98% and 99.5%. The present invention mainly selects the appropriate confidence value based on the detection effect of actual application.

[0102] Finally, the specific aircraft engine test performance is used as the input of the decoder of each stacked autoencoder in the model to realize the assembly interval inference of the corresponding aircraft engine assembly quality characteristics, thereby providing more refined guidance on assembly process standards, improving the assembly quality of aircraft engines, improving assembly efficiency and one-time test pass rate, and improving the efficiency and market competitiveness of aircraft engine manufacturers.

[0103] In order to further illustrate the method for aircraft engine assembly quality anomaly detection and feature inference based on an autoencoder provided by the present invention, it is described in conjunction with a specific embodiment.

[0104] This embodiment takes a certain aircraft engine manufacturer as the research object. The aircraft engine assembly data and test data are collected through field measurement, and the aircraft engine assembly and test data set obtained by sorting contains 1000 samples, each of which has two aircraft engine test performance indicators and 18 aircraft engine assembly quality characteristics. The data is divided according to the test performance. When the test performance index values ​​of the samples exceed the 80% threshold of the whole, the sample is marked as abnormal data, and there are 150 abnormal sample data in total. According to the random sampling method with replacement, the aircraft engine assembly and test data set is divided into a training set, a validation set and a test set, wherein the training set contains 800 normal data, the validation set contains 200 normal data, and the test set contains 50 normal data and 150 abnormal data. In the data preprocessing process, the maximum and minimum normalization method is used to normalize the data set samples.

[0105] In this example experiment, the stacked autoencoder is composed of three layers of autoencoders. The number of neurons in each linear layer and the activation function are as follows: Figure 6 As shown in Figure 1, the bagging stacked autoencoder is constructed by integrating 6 independent stacked autoencoders. In the hyperparameter setting, the number of training times is 500, the batch size of the input data is 32, and the learning rate is 0.005.

[0106] The error curves of each stacked autoencoder during training are as follows: Figure 7As shown. Since the training data set of each stacked autoencoder is generated by a random sampling method with replacement, all training data sets are different, so the stacked autoencoder shows different training error change curves. However, it can be seen from the figure that after nearly 500 rounds of training, the training error of the stacked autoencoder tends to be stable and no longer fluctuates violently, indicating that the stacked autoencoder has completed training, its performance has converged, and the parameters no longer change significantly. As can be seen from the figure, stacked autoencoder 4 shows the strongest data reconstruction ability, and its loss function is significantly lower than that of other encoders. In addition, the performance of stacked autoencoder 5 is poor, and the error change curve decreases the least.

[0107] According to the proposed dynamic threshold strategy for anomaly detection, this example sets the confidence level to 98%. According to the error distribution of the model in the validation set, the dynamic threshold for anomaly detection is 1.266. The anomaly detection thresholds of each stacked autoencoder at the 98% confidence level are all greater than the dynamic threshold. The comparison results are shown in Figure 2. Figure 8 The lower the dynamic threshold, the smaller the overall reconstruction error of the model. The bagging stacked autoencoder model can show better feature learning ability by integrating multiple stacked autoencoders. The overall error of reconstructing input data is significantly lower than that of a single stacked autoencoder, overcoming the problems of overfitting and underfitting of a single model, thus further verifying the practicality and effectiveness of the method proposed in this section in the detection of aero-engine assembly quality anomalies.

[0108] The bagging stacked autoencoder model calculates the reconstruction error of the test set samples to evaluate the accuracy of the model in anomaly detection. Fig. 9 The figure shows the scatter plot of reconstruction error of bagging stacked autoencoder in each data set. The blue dots in the figure are training set samples, the green dots represent validation set samples, the red dots represent test set samples, and the red straight line is the dynamic threshold line for abnormal detection with a confidence level of 98%. The reconstruction errors of normal data and validation set data in the test set are almost kept in the same range and are almost distributed below the dynamic threshold line, while the reconstruction error of abnormal data in the test set is significantly higher than the dynamic threshold line for abnormal detection, indicating that the bagging stacked autoencoder model constructed in this section has excellent results in assembly quality abnormality detection, and the proposed dynamic threshold strategy is very reasonable in application in abnormality detection, which can accurately identify abnormal data of assembly quality by discriminating reconstruction errors, and is scientific and effective.

[0109] F1 score is an important indicator used in statistics to evaluate the accuracy of classification models. Its calculation formula is shown in formula (3):

[0110]

[0111] Where Precision is the accuracy, which indicates the ratio of the number of samples correctly predicted as positive to the number of all samples predicted as positive. The calculation formula is shown in formula (4). Recall is the recall, which indicates the ratio of the number of samples correctly predicted as positive to the number of all samples predicted as positive. The calculation formula is shown in formula (5).

[0112]

[0113] Among them, TP, FP and FN are terms in the confusion matrix, which are the number of samples that are actually positive and predicted to be positive, the number of samples that are actually negative and predicted to be positive, and the number of samples that are actually positive and predicted to be negative. The F1 score, as the harmonic mean of precision and recall, can more comprehensively evaluate the comprehensive performance of the model. It has outstanding performance in various scenarios such as data imbalance and can be used for the analysis and evaluation of the abnormal detection results of aircraft engine assembly quality in this experiment.

[0114] The F1 score comparison of all stacked autoencoders and bagging stacked autoencoder models is shown below Fig.10 As shown in the figure, the F1 scores of each stacked autoencoder are 0.801, 0.925, 0.889, 0.948, 0.918, and 0.933, respectively, while the F1 score of the Bagging stacked autoencoder model is 0.962, indicating that the Bagging stacked autoencoder model based on ensemble learning can effectively improve the performance of a single stacked autoencoder, make full use of the output results of multiple stacked autoencoders, avoid the overfitting problem caused by a single stacked autoencoder, enhance the robustness of the ensemble model, and improve the accuracy of data reconstruction.

[0115] According to the actual needs of aircraft engine production, the test performance index values ​​of aircraft engines need to be maintained at a relatively low level. Assuming that the test performance index of aircraft engines is kept between the 10% and 50% quantiles of the original test performance distribution as much as possible, this example needs to infer the assembly range of the corresponding aircraft engine assembly quality characteristic value based on this standard. The decoder of the bagging stacked autoencoder model can only input a single encoding value. For example, the input encoding [0.1, 0.1] represents the test performance index value in the 10% quantile of the overall distribution, thereby obtaining the aircraft engine assembly quality characteristic value. In order to realize the feature inference of the 10% to 50% quantile test performance, the data is discretized, and the test performance range is divided into multiple discrete values ​​at intervals of 1%, namely [0.1, 0.1], [0.11, 0.11], …, [0.5, 0.5]. The discrete code is used as the input of the decoder to obtain the corresponding aircraft engine assembly quality characteristic value. The reconstruction result of the assembly quality feature is fitted to obtain the assembly quality feature change curve, such as Fig.11As shown, quality feature inference is achieved.

[0116] The present invention provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method for detecting abnormality of assembly quality of an aircraft engine and inferring characteristics based on an autoencoder is implemented.

[0117] The present invention provides an electronic device, comprising:

[0118] Processor; and

[0119] A memory, configured to store executable instructions of the processor;

[0120] Wherein, the processor is configured to execute the above-mentioned method of aircraft engine assembly quality anomaly detection and feature inference based on autoencoder by executing the executable instructions.

[0121] The present invention provides a system for detecting and inferring the quality anomaly of aircraft engine assembly based on an autoencoder, comprising:

[0122] The data acquisition module is used to obtain an assembly and test data set of an aircraft engine, wherein the assembly and test data set includes an assembly quality data set and a corresponding test performance data set; the assembly and test data set is used to generate multiple training data sets by a random sampling method with replacement;

[0123] The encoder model construction module is used to construct a stacked autoencoder corresponding to each training data set; each stacked autoencoder consists of an encoder and a decoder; each training data set is used to train the stacked autoencoder corresponding to it until convergence, so as to obtain each trained stacked autoencoder; wherein, during the training process, the assembly quality data is used as the input of the stacked autoencoder; the test performance data is used as the feature representation of the encoding and as the input data of the decoder; a bagging stacked autoencoder model is constructed according to each trained stacked autoencoder;

[0124] The anomaly detection module is used to set the confidence level to determine the dynamic threshold of abnormal data; the assembly and test data of the aircraft engine to be detected is used according to the Bagging stacked autoencoder model, and the anomaly detection of the aircraft engine assembly quality is performed based on the dynamic threshold.

[0125] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for detecting and inferring the quality anomaly of aircraft engine assembly based on an autoencoder, characterized in that: include: Acquire an assembly and test data set of an aircraft engine, wherein the assembly and test data set includes an assembly quality data set and a corresponding test run performance data set; Generate multiple training data sets by random sampling with replacement from the test data set; Construct a stacked autoencoder corresponding to each training dataset; each stacked autoencoder consists of an encoder and a decoder; Using each training data set to train the corresponding stacked autoencoder until convergence, each trained stacked autoencoder is obtained; During the training process, the assembly quality data is used as the input of the stacked autoencoder; the test performance data is used as the encoded feature representation and as the input data of the decoder; Build a Bagging stacked autoencoder model based on each trained stacked autoencoder; Set confidence levels to determine dynamic thresholds for abnormal data; The assembly and test data of the aircraft engine to be tested is used according to the Bagging stacked autoencoder model, and abnormality detection of the aircraft engine assembly quality is performed based on the dynamic threshold.

2. The method for detecting and inferring the quality anomaly of an aircraft engine assembly based on an autoencoder according to claim 1, characterized in that: The loss function for training each stacked autoencoder includes: The error between the input data and the output of the stacked autoencoder is taken as the reconstruction error; The error between the encoding of the stacked autoencoder and the test performance data is taken as the encoding error; The sum of the reconstruction error and the encoding error is used as the loss function of the stacked autoencoder.

3. The method for detecting and inferring the quality anomaly of an aircraft engine assembly based on an autoencoder according to claim 2, characterized in that: The step of determining the dynamic threshold of abnormal data includes: The test data set is used to generate a validation data set through random sampling with replacement; both the training set and the validation set contain normal data; Use the validation set to reconstruct the Bagging stacked autoencoder model and obtain the reconstruction error of the validation set; The dynamic threshold of abnormal data is determined by setting the confidence level according to the reconstruction error of the validation set.

4. The method for detecting and inferring anomalous quality of aircraft engine assembly based on an autoencoder according to claim 1, characterized in that: When building a bagging stacked autoencoder model, it includes: The average of the output data of each trained stacked autoencoder is taken as the final output of the model.

5. The method for detecting and inferring anomalous quality of aircraft engine assembly based on autoencoder according to claim 1, characterized in that: Also includes: The set aircraft engine test performance is used as the input of the decoder of each stacked autoencoder in the Bagging stacked autoencoder model to infer the assembly interval of the aircraft engine assembly quality characteristics.

6. The method for detecting and inferring anomalous quality of aircraft engine assembly based on autoencoder according to claim 5, characterized in that: When obtaining the assembly interval inference of the aircraft engine assembly quality characteristics, it includes: Discretize the set aircraft engine test performance to obtain multiple discrete values; The discrete values ​​are used as the input of the decoders of each stacked autoencoder in the bagging stacked autoencoder model to reconstruct the corresponding assembly quality feature values; Fitting the reconstructed corresponding assembly quality characteristic values ​​to obtain an assembly quality characteristic variation curve; The assembly range of aircraft engine assembly quality characteristics is inferred based on the assembly quality characteristic change curve.

7. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for detecting and inferring anomalous quality of aircraft engine assembly based on an autoencoder according to any one of claims 1 to 6 is implemented.

8. An electronic device, characterized in that: include: processor; as well as A memory, configured to store executable instructions of the processor; The processor is configured to execute the method for aircraft engine assembly quality anomaly detection and feature inference based on an autoencoder according to any one of claims 1 to 6 by executing the executable instructions.

9. A system for detecting and inferring the quality anomaly of aircraft engine assembly based on an autoencoder, characterized in that: include: The data acquisition module is used to obtain an assembly and test data set of an aircraft engine, wherein the assembly and test data set includes an assembly quality data set and a corresponding test performance data set; the assembly and test data set is used to generate multiple training data sets by a random sampling method with replacement; The encoder model construction module is used to construct a stacked autoencoder corresponding to each training data set; each stacked autoencoder consists of an encoder and a decoder; each training data set is used to train the stacked autoencoder corresponding to it until convergence, so as to obtain each trained stacked autoencoder; wherein, during the training process, the assembly quality data is used as the input of the stacked autoencoder; the test performance data is used as the feature representation of the encoding and as the input data of the decoder; a bagging stacked autoencoder model is constructed according to each trained stacked autoencoder; The anomaly detection module is used to set the confidence level to determine the dynamic threshold of abnormal data; the assembly and test data of the aircraft engine to be detected is used according to the Bagging stacked autoencoder model, and the anomaly detection of the aircraft engine assembly quality is performed based on the dynamic threshold.

Images