An unknown fault diagnosis method, device and medium for a landing gear system

Abstract

The present application relates to the technical field of fault diagnosis, and in particular to a landing gear system unknown fault diagnosis method, device and medium, the present application method is based on an explainable feature decomposition method to enhance the model feature extraction capability, a set of parallel convolutional auto-encoding structure is constructed, and the sample features are decomposed into uncertainty fault features and category difference features; an adaptive threshold strategy is used to update the threshold value, and fault diagnosis is carried out; finally, the unsupervised analysis method is used to predict the number of unknown fault categories, and the diagnosis result is output, so as to realize fault diagnosis. The present application method considers the feature decomposition idea, the sample features are decomposed into uncertainty fault features and difference features through the convolutional auto-encoding structure, the hidden fault information and clustering geometric information are fully considered, and the model feature extraction capability is improved from the perspective of enhancing the feature explainability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault diagnosis, in particular to an unknown fault diagnosis method, device and medium for a landing gear system. BACKGROUND

[0002] The landing gear system is one of the important components of the aircraft electromechanical system, mainly composed of a retracting system and a front wheel turning system. Key components such as torque motors, speed sensors and angle sensors are prone to failure, which affects the control of the aircraft and may cause stoppage events or even casualties. As a complex nonlinear system, the rule-based maintenance and fault diagnosis method for the landing gear system is affected by the maturity of the rule base and expert experience, resulting in long time consumption and low efficiency. Therefore, it is necessary to introduce intelligent fault diagnosis algorithms to improve the diagnosis accuracy and efficiency.

[0003] With the rapid development of science and technology, the amount of aircraft onboard sensor data available has shown explosive growth. The landing gear system has complex cross-linking relationships, and the signal types and data volume are huge. It is necessary to clarify the aircraft fault characterization information from a large amount of high-dimensional and heterogeneous data. Data-driven deep learning methods can perform deep information mining and feature dimensionality reduction on high-dimensional data, and can be used for research on civil aircraft fault feature characterization methods. Due to the high safety, high reliability and multi-task characteristics of aircraft, it is difficult to collect fault data, and the available abnormal data also has the problems of false positives and fault homogenization, which leads to the problem of unknown faults in actual monitoring. Common deep learning models belong to supervised learning and rely on a large number of labeled samples for training, thereby realizing the classification and diagnosis of known faults. However, they have limitations in identifying unknown and invisible faults.

[0004] In summary, how to effectively solve the problem that supervised models cannot intelligently diagnose unknown faults is a problem that needs to be solved by technical personnel in the field. SUMMARY

[0005] Therefore, the present application provides an unknown fault diagnosis method, device and medium for a landing gear system, which introduces a new discrimination mechanism to learn unknown fault information representation, thereby solving the problem that supervised models cannot intelligently diagnose unknown faults.

[0006] To achieve the above purpose, the present application provides the following technical solutions:

[0007] An unknown fault diagnosis method for a landing gear system, comprising the following steps:

[0008] S100: Data acquisition, at least one group of sensing signals is collected by a sensor and a sample data set is established;

[0009] S200: data preprocessing, dividing the sample data set into a training set and a test set, wherein the number of fault types in the training set is less than that in the test set to ensure that the unknown faults defined in the test set do not participate in the training process;

[0010] S300: constructing an unknown fault diagnosis model, establishing a group of parallel convolutional auto-encoding structures to obtain uncertainty fault features and difference features, and simultaneously adopting a correlation loss function to constrain the feature reconstruction process;

[0011] Each convolutional auto-encoding structure includes an encoding layer and a decoding layer, the encoding layer includes at least two convolutional layers and a pooling layer, the decoding layer includes three fully connected layers, the last layer of the encoding layer is taken as a feature output, and a classification layer is connected to constitute a semi-supervised model structure, and an unknown fault diagnosis model is obtained;

[0012] S400: parameter updating, inputting the training set into the unknown fault diagnosis model, randomly extracting a fixed number of data in the training set to participate in the training process, iteratively optimizing the model parameters, and stopping training and saving the model when a set iteration step is met or a precision threshold condition is met;

[0013] S500: threshold updating, in the model test process, an adaptive threshold strategy based on distance difference ratio is used to adaptively determine the unknown fault threshold, realize threshold updating, and complete the model training of the unknown fault diagnosis model;

[0014] S600: fault diagnosis, inputting the to-be-tested data into the trained unknown fault diagnosis model, judging whether the to-be-tested data belongs to a known fault or an unknown fault, and judging the fault type, outputting the known fault, unknown fault and unknown fault type number as the diagnosis result, and completing the fault diagnosis.

[0015] Preferably, in S100, the sensors are arranged at key measurement points of the landing gear system, and one or more groups of sensing signals reflecting the health status of the landing gear system are collected, the sensing signals are classified according to different working conditions and different health statuses, and a sample data set is established.

[0016] Preferably, in S200, the sliding data sampling is specifically: using a fixed window size for the original sample, and obtaining multiple samples with a window size from left to right.

[0017] Preferably, in S300, the uncertainty fault features and difference features satisfy algebra and attributes, and the algebraic sum of the two represents the complete features of the sample, and the calculation steps are as follows:

[0018] S301: constructing a group of parallel and asymmetric convolutional auto-encoding structures as a feature extractor, and calling the feature extractor to extract features from the sample data set;

[0019] S302: The reconstruction loss function is used as the correlation loss function, the reconstruction loss function is called to train the feature extractor, and the uncertainty feature extractor and the difference feature extractor are obtained; during the training process, the uncertainty feature extractor adds an additional uncertainty loss function, and the difference feature extractor adds an additional difference loss function;

[0020] S303: The uncertainty feature extractor and the difference feature extractor are called to decompose the features extracted in S301, and the uncertainty features and the difference features are obtained.

[0021] Preferably, in S400, the total loss function of the parameter updating process is as follows:

[0022] The convolutional auto-encoding structure constructed uses a cross-entropy loss function to introduce label information;

[0023] An additional L2 norm regularization term is added to the convolutional auto-encoding structure as a constraint term for constraining the parameter range;

[0024] The total loss function of the uncertainty feature extractor is composed of the reconstruction loss function, the cross-entropy loss function, the uncertainty loss function, and the constraint term;

[0025] The total loss function of the difference feature extractor is composed of the reconstruction loss function, the cross-entropy loss function, the difference loss function, and the constraint term.

[0026] Preferably, in S500, the adaptive threshold strategy based on the distance difference ratio is as follows:

[0027] S501: Calculate the difference ratio, use the Euclidean distance to calculate the difference value of each sample in the test set and the samples in the training set, and the difference values of all samples form a difference value set;

[0028] S502: Sort the difference value set from small to large to obtain a sorted difference value set;

[0029] S503: Calculate the difference change extreme value, slice the sorted difference value set to calculate the change rate of the continuous two difference values, obtain the difference value change rate set, and the maximum value of the middle section in the difference value change rate set is the difference change extreme value;

[0030] S504: Calculate the unknown fault threshold, take the difference value corresponding to the sequence number of the difference change extreme value as the unknown fault division threshold, and samples with a difference value greater than the unknown fault threshold are divided into unknown faults.

[0031] Preferably, in S600, it is judged whether the to-be-tested data belongs to a known fault or an unknown fault, and the fault type is judged, which is specifically:

[0032] When the distance value of the to-be-tested data is lower than the threshold value, the to-be-tested data belongs to a known fault, and a classifier is called to determine the fault type; when the distance value of the to-be-tested data is greater than or equal to the threshold value, it is determined that the to-be-tested data belongs to an unknown fault; finally, unsupervised analysis is used to realize unknown fault class number prediction.

[0033] Preferably, the unsupervised analysis is specifically k-means++ unsupervised clustering, and a silhouette coefficient and a multivariate probability density function are used to evaluate the k-means++ unsupervised clustering result to realize unknown fault class number prediction.

[0034] Compared with the prior art, the method has the beneficial effects that:

[0035] 1) The method considers the feature decomposition idea, constructs a parallel stack convolutional auto-encoding structure to decompose sample features into uncertain fault features and difference features, fully considers the hidden fault information and clustering geometric information, and improves the feature extraction capability of the model from the perspective of enhancing the interpretability of the features.

[0036] 2) The method proposes an unknown fault adaptive threshold strategy based on a distance difference ratio, adaptively determines the unknown sample threshold according to a difference value set, and combines the Euclidean distance to calculate the distance between the sample and the known sample to realize unknown fault diagnosis.

[0037] 3) The method combines a silhouette coefficient and a multivariate normal distribution probability density function to form a clustering distribution factor, and combines a k-means++ clustering method to efficiently evaluate the clustering effect and predict the unknown fault class number.

[0038] In addition, the present application also includes a computer device comprising a memory and a processor;

[0039] The memory is used to store a computer program that can run on the processor;

[0040] The processor is used to execute the computer program to realize the steps of the unknown fault diagnosis method of the landing gear system as described above.

[0041] In addition, the present application also includes a computer readable storage medium, the computer readable storage medium stores a computer program, and the computer program is executed by the processor to realize the steps of the unknown fault diagnosis method of the landing gear system as described above. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0043] Figure 1 The overall flowchart of the landing gear system unknown fault diagnosis method provided by the embodiment of the present application is shown in the figure.

[0044] Figure 2 The network model structure diagram of the convolutional auto-encoding structure in the embodiment of the present application is shown in the figure.

[0045] Figure 3 The overall structure schematic diagram of the embodiment of the present application is shown in the figure.

[0046] Figure 4 The accuracy comparison diagram of the experimental results of the method and different methods of the present application is shown in the figure.

[0047] Figure 5 The confusion matrix diagram of the method of the present application in a certain experiment on the B group experiment is shown in the figure. DETAILED DESCRIPTION

[0048] The embodiment of the present application discloses a landing gear system unknown fault diagnosis method, device and medium, and the method effectively solves the problem that a supervised model cannot intelligently diagnose unknown faults.

[0049] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments only constitute some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0050] Please refer to Figure 1 , Figure 2 and Figure 3 , Figure 1 The overall flowchart of the landing gear system unknown fault diagnosis method provided by the embodiment of the present application is shown in the figure. Figure 2 The network model structure diagram of the convolutional auto-encoding structure in the embodiment of the present application is shown in the figure. Figure 3 The overall structure schematic diagram of the embodiment of the present application is shown in the figure.

[0051] In a specific embodiment, the embodiment provides a landing gear system unknown fault diagnosis method, comprising the following steps:

[0052] S100: Data collection, install sensors for collecting state information on the landing gear system, collect one or more sets of sensor signals and sample data sets. In a specific embodiment, the sensors are arranged at key measurement points of the landing gear system, and one or more sets of sensor signals reflecting the health status of the landing gear system are collected, which are classified according to different working conditions and different health statuses, and sample data sets are established.

[0053] S200: Data preprocessing, when the number of sample data sets is insufficient, sliding data sampling is used to enhance the sample; the sample data sets are divided into training set and test set, wherein the number of fault types in the training set is less than that in the test set, and the unknown faults defined in the test set are not involved in the training process. In a specific embodiment, the sliding data sampling is as follows: a fixed window size is used for the original sample, and multiple samples with window size are obtained by shifting from left to right.

[0054] S300: Constructing an unknown fault diagnosis model, a group of parallel convolutional auto-encoding structures are established to obtain uncertainty fault features and difference features, and a correlation loss function is used to constrain the feature reconstruction process of the parallel auto-encoding;

[0055] Each convolutional auto-encoding structure includes an encoding layer and a decoding layer, the encoding layer includes at least two convolutional layers and a pooling layer, the decoding layer includes three fully connected layers, the last layer of the encoding layer is used as a feature output, and a Softmax classification layer is connected to form a semi-supervised model structure, and an unknown fault diagnosis model is obtained;

[0056] S400: Parameter update, input the training set into the unknown fault diagnosis model, randomly extract a fixed number of data in the training set for training process, and iteratively optimize the model parameters. When the set iteration step is met or the accuracy threshold condition is met, stop training and save the model;

[0057] S500: Threshold update, in the model test process, an adaptive threshold strategy based on distance difference ratio is used to adaptively determine the unknown fault threshold, realize threshold update, and complete the model training of the unknown fault diagnosis model;

[0058] S600: Fault diagnosis, input the to-be-tested data into the trained unknown fault diagnosis model, judge whether the to-be-tested data belongs to known fault or unknown fault, and judge the fault type, output the known fault, unknown fault and unknown fault type number as the diagnosis result, and complete the fault diagnosis.

[0059] In a specific embodiment, in S300, the uncertainty fault features and difference features satisfy algebraic sum and attribute, and the algebraic sum of the two represents the complete features of the sample, and the calculation steps are as follows:

[0060] S301: Construct a set of parallel and asymmetric convolutional auto-encoding structures as a feature extractor g(·), call the feature extractor to extract features from a sample data set; let the sample data set be Sample x i The features f extracted by the feature extractor can be expressed as:

[0061] f = g(x i ) (1)

[0062] S302: Use a reconstruction loss function as a correlation loss function, call the reconstruction loss function to train the feature extractor, and obtain an uncertainty feature extractor and a difference feature extractor; during training, the uncertainty feature extractor adds an additional uncertainty loss function, and the difference feature extractor adds an additional difference loss function.

[0063] In a specific embodiment, the sum of the reconstruction samples of the uncertainty features and the reconstruction samples of the difference features approximates the original sample, and the expression of the reconstruction loss function L FD-AE is as follows:

[0064]

[0065] Where n represents the number of samples; G u (·) represents the decoding function of the uncertainty feature extractor; g u (·) represents the uncertainty feature extractor; G d (·) represents the decoding function of the difference feature extractor; g d (·) represents the difference feature extractor.

[0066] In a specific embodiment, the expression of the uncertainty loss function L u is as follows:

[0067]

[0068] Where n nor represents the number of normal state samples; represents the i-th normal sample.

[0069] In a specific embodiment, the difference loss function L d is calculated by the intra-class average radius R c and the inter-class average distance D c , and the expression is as follows:

[0070]

[0071] Specifically, assuming that the total number of classes in the sample data set is C, the i-th class contains n i samples xi Then the fitting center and the class radius of the first class can be calculated by the following formula:

[0072]

[0073]

[0074]

[0075]

[0076] wherein, represents the jth sample in the ith class data.

[0077] S303: calling the uncertainty feature extractor and the difference feature extractor to decompose the feature f extracted in S301, to obtain the uncertainty feature f u and the difference feature f d . The expression is as follows:

[0078] f = g u (x i ) + g d (x i ) = f u + f d (9)

[0079] Preferably, in S400, the total loss function of the parameter updating process is as follows:

[0080] The constructed convolutional auto-encoding structure uses a cross-entropy loss function to introduce label information; the expression of the cross-entropy loss function L CE is as follows:

[0081]

[0082] wherein, y i represents the true label of the ith sample; represents the predicted label of the ith sample.

[0083] An additional L2 norm regularization term is added in the convolutional auto-encoding structure as a constraint term, which is used to constrain the parameter range; the expression of the constraint term L2 is as follows:

[0084]

[0085] wherein, n w represents the total number of optimization parameters; w j represents the jth optimization parameter.

[0086] The total loss function of the uncertainty feature extractor is composed of a reconstruction loss function, a cross-entropy loss function, an uncertainty loss function and a constraint term; the expression of the total loss function L of the uncertainty feature extractor is as follows: AE-U

[0087] L AE-U =α1L FD-AE +β1L CE +λ1L u +γ1L2(w AE-U ) (12)

[0088] wherein α1 represents a reconstruction loss weight factor of the uncertainty feature extractor; β1 represents a cross-entropy loss weight factor of the uncertainty feature extractor; λ1 represents a feature constraint factor of the uncertainty feature extractor; γ1 represents a regularization weight of the uncertainty feature extractor; and w AE-U represents an optimized parameter set of the uncertainty feature extractor.

[0089] The total loss function of the difference feature extractor is composed of a reconstruction loss function, a cross-entropy loss function, a difference loss function and a constraint term; the expression of the total loss function L of the difference feature extractor is as follows: AE-D

[0090] L AE-D =α2L FD-AE +β2L CE +λ2L d +γ2L2(w AE-D ) (13)

[0091] wherein α2 represents a reconstruction loss weight factor of the difference feature extractor; β2 represents a cross-entropy loss weight factor of the difference feature extractor; λ2 represents a feature constraint factor of the difference feature extractor; γ2 represents a regularization weight of the difference feature extractor; and w AE-D represents an optimized parameter set of the difference feature extractor.

[0092] Preferably, in S500, the adaptive threshold strategy based on the distance difference ratio is as follows:

[0093] S501: Calculate the difference ratio, and calculate the difference value of each sample in the test set from the sample in the training set using the Euclidean distance ο(x,y). All the difference values of the samples form a difference value set φ, and the calculation expression is as follows:

[0094]

[0095] ​​

[0096]

[0097] wherein x and y are two sets of samples with equal dimensions respectively; m represents the dimension of the sample; n tr represents the number of training set samples; n te represents the number of test set samples.

[0098] S502: Sort the difference value set φ from small to large to obtain the ordered difference value set The relationship is

[0099] S503: Calculate the difference change extreme value, slice the ordered difference value set φ s to obtain the change rate δ of the two continuous difference values, obtain the difference value change rate set P, and the maximum value of the interval in the difference value change rate set is the difference change extreme value. The calculation expression is as follows:

[0100]

[0101] P = { δ j | 1 ≤ j ≤ (n te - 1)} (18)

[0102] S504: Calculate the unknown fault threshold, take the difference value p corresponding to the sequence number t of the difference change extreme value as the unknown fault threshold, and divide the samples with the difference value greater than the unknown fault threshold into unknown faults; the sequence number t of the difference change extreme value is expressed as follows:

[0103] t = argmax { δ j | 0.1 × (n te - 1) ≤ j ≤ 0.9 × (n te - 1)} (19)

[0104] Preferably, in S600, it is judged whether the to-be-tested data belongs to a known fault or an unknown fault, and the fault type is judged, specifically:

[0105] When the distance value of the to-be-tested data is lower than the threshold value, it belongs to a known fault, and a Softmax classifier is called to judge the fault type; when the distance value of the to-be-tested data is greater than or equal to the threshold value, it is judged to belong to an unknown fault; finally, unsupervised analysis is adopted to realize unknown fault category number prediction.

[0106] Preferably, the unsupervised analysis is specifically k-means++ unsupervised clustering, and the k-means++ unsupervised clustering result is evaluated by using a silhouette coefficient and a multivariate probability density function to realize unknown fault category number prediction.

[0107] The specific steps of the unsupervised analysis are as follows:

[0108] S601: using the k-means++ method to the difference feature set g of the test set d (D te ) to perform unsupervised clustering;

[0109] S602: calculate the profile coefficient S(x), and the calculation expression of the profile coefficient S(x) is as follows

[0110]

[0111] Wherein, a(x) represents the average distance between sample x and the cluster center to which it belongs; b(x) represents the average distance between sample x and its nearest cluster center.

[0112] S603: calculate the multivariate probability density evaluation factor Q(x) using the multivariate probability density function, and the multivariate probability density evaluation factor is the minimum value of the density evaluation factors corresponding to all categories, and the calculation expression is as follows:

[0113]

[0114]

[0115] Q(x) = min Q i (x) (23)

[0116] Wherein, ρ i (x) represents the multivariate probability density function value of the i-th category; μ i and ∑ i respectively represent the mean vector and the covariance matrix of the i-th category data.

[0117] S604: get the clustering distribution factor ω c When the value of ω c is maximum, the corresponding clustering category is the predicted optimal category number Calculate the difference between the predicted optimal category number and the known category number C k , to get the unknown fault category number C u , and the calculation expression is as follows:

[0118]

[0119]

[0120] Implementation case:

[0121] To verify the performance of the method, a certain type of aircraft landing gear system is simulated and modeled, and fault injection is implemented. The typical fault modes are shown in Table 1. Due to the complex coupling relationship of the landing gear system components, multiple parameter measurement points are selected to reflect the system state. The 16 characteristic parameters are shown in Table 2. In order to expand the number of samples, the original data is sampled by using the sliding data sampling method, each characteristic dimension is 500, 16 characteristics are spliced to form a sample, the sample dimension is 8000, including normal state, 7 kinds of fault types, after sample enhancement by using the sliding data sampling method, 198 samples of each fault type are obtained, a total of 1386 samples. The sample set is divided into data sets, and the ratio of the training set and the test set is 7:3. In this embodiment, the model structure parameters and the training parameters are shown in Tables 3 and 4, respectively.

[0122] Table 1 Typical fault model table of landing gear system

[0123]

[0124]

[0125] Table 2 Characteristic parameter table of aircraft landing gear system

[0126] Serial number Data type Unit Range Parameter Chinese name 1 Discrete ° |70| Control angle 2 Continuous ° |4e-3| Armature rotation angle 3 Continuous m |3.4e-5| Armature displacement 4 Continuous m |0.002| Spool position 5 Continuous m 3 / s·sec |1.33e-4| A chamber flow 6 Continuous m 3 / s·sec |1.33e-4| B chamber flow 7 Continuous m 3 / s·sec |2.2e-8| Total flow 8 Continuous m 3 ]]> |4.1e-4| A chamber volume 9 Continuous m 3 ]]> |4.1e-4| B chamber volume 10 Continuous / 0-3.204e11 Effective volume modulus A 11 Continuous / 0-3.204e11 Effective volume modulus B 12 Continuous Pa 10M A chamber pressure 13 Continuous Pa 10M B chamber pressure 14 Continuous N |1500| Piston friction 15 Continuous m / s 0-0.0722 Piston speed 16 Continuous ° -70-70 Airplane rotation angle

[0127] Table 3 Model structure parameter table

[0128]

[0129] Table 4 Model training parameter setting table

[0130] Parameter name Step number Learning rate Batch size [Alpha]1 [Alpha]1 ​ [gamma]1 [alpha]2 <![CDATA[β2]]> ​ [gamma]2 Value 150 0.005 100 2 1.5 6 0.0001 1 1 3 0.0001

[0131] In order to explore the diagnostic performance of the method, a set of experimental schemes are designed, and the specific schemes are shown in Table 5. Randomly select 2 kinds of fault types as unknown faults, and do not participate in the training process. The known fault types participate in the model training process. In order to reduce the influence of randomness of the experiment, each group of experiments is repeated 10 times, and the average value is taken.

[0132] Table 5 Experimental scheme design table

[0133]

[0134] The experimental results are shown in Table 6. The method of the present application achieves good diagnosis results in most cases, with the lowest unknown fault diagnosis accuracy rate reaching 84.46%, and the total diagnosis accuracy rate being maintained above 83.29%. From the accuracy rate analysis, in the A, B, C, and D data sets, the known fault accuracy rate is dominant, and the distance difference ratio criterion can better adaptively determine the optimal detection threshold; in the E and F data sets, the known fault accuracy rate is relatively low, which can more easily affect the detection accuracy of the distance difference ratio criterion. Therefore, from the experimental results, it can be seen that the performance of the feature extractor has an important influence on the detection accuracy of the method of the present application. When the feature extraction effect is weak, the threshold tightness may change, and the known or unknown faults may be misclassified; when the feature extraction effect is good, a more accurate unknown fault diagnosis threshold can be obtained, and the diagnosis accuracy of both can be improved.

[0135] Table 6: Multiple experimental results of the method of the present application

[0136] Dataset Total accuracy Known fault accuracy Unknown fault accuracy A 94.98％ 96.98％ 90.19％ B 98.89％ 98.56％ 99.69％ C 91.80％ 94.57％ 84.46％ D 98.46％ 98.06％ 99.42％ E 88.90％ 84.91％ 99.17％ F 83.29％ 82.61％ 86.49％

[0137] Further, the distance difference ratio unknown fault detection mechanism of the present application is combined with different methods, and the experimental results on different data sets are shown in Figure 4 The experimental results are the average of 10 experiments. From the figure, it can be seen that the method of the present application shows the best stability and achieves better diagnosis accuracy for known faults and unknown faults in multiple data sets. It can be seen that the HSDAE method has relatively weak feature extraction capability for high-dimensional data. Although it can separate unknown faults in most cases, the recognition accuracy of known faults is relatively low, making the unknown fault detection threshold too loose, and not achieving ideal diagnosis results. CAE and CNN show similar performance in most cases. In a few data sets, the threshold determination method based on distance ratio can show better unknown fault detection effect, but for different data sets, the stability of these two methods is slightly lower than that of the method of the present application. In summary, the feature decomposition and constraint of the method of the present application can improve the feature extraction capability of the model, thereby improving the stability of the model and achieving unknown fault diagnosis of the aircraft landing gear system.

[0138] In order to more intuitively analyze the diagnosis results, the confusion matrix is used to visualize the classification results of a certain experiment of group B, as shown in Figure 5As shown in the figure, the known fault classifier is accurate and reliable, and can correctly classify most samples, with only two misclassified samples, which are identified as hydraulic cylinder leakage instead of nozzle blockage. Based on the accuracy of the known fault classifier, the experimental results show that the identification threshold of the unknown fault identifier is loose, and 2 normal samples and 2 slide valve sticking samples are identified as unknown faults; but all unknown faults are identified, and the unknown fault diagnosis accuracy is 100%, and no fault sample is identified as a normal sample, which is acceptable in actual working conditions.

[0139] In addition, the embodiment also discloses a computer device, comprising a memory and a processor;

[0140] The memory is used for storing a computer program which can run on the processor;

[0141] The processor is used for implementing the steps of the unknown fault diagnosis method of the landing gear system when the computer program is executed.

[0142] For example, the computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present application. The one or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the computer device.

[0143] The computer device can be a mobile phone, a desktop computer, a notebook computer, a palm computer and a cloud server, etc. The computer device can include, but is not limited to, a processor, a memory. For example, the computer device can also include an input / output device, a network access device, a bus, etc.

[0144] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. The processor is the control center of the computer device, and connects all parts of the computer device through various interfaces and lines.

[0145] The memory can be used to store the computer program and / or modules, and the processor realizes the computer program by running or executing the computer program and / or modules stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application program required by a function (such as a sound playing function, an image playing function, etc.), etc.; and the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory devices.

[0146] The modules / units integrated in the computer device, if realized in the form of software function units and sold or used as independent products, can be stored in a computer readable storage medium. Based on such understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. The computer program can realize the steps of the above-mentioned various method embodiments when executed by a processor. The computer program includes computer program code, which can be in the form of source code, object code, an executable file, or some intermediate form. The computer readable medium can include any entity or device capable of carrying the computer program code, a recording medium, a U disk, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc. It should be noted that the computer readable medium can include or exclude contents according to the requirements of legislation and patent practice in a jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.

[0147] In addition, the embodiment of the present application also includes a computer readable storage medium, and the computer readable storage medium stores a computer program, and the computer program is executed by a processor to realize the steps of the unknown fault diagnosis method of the landing gear system.

[0148] Compared with the prior art, the present application optimizes the feature extraction process of the extractor from two explanation angles of uncertainty features and difference features by constructing a semi-supervised parallel stacked convolutional auto-encoding model; wherein the convolutional layer is used as an encoder structure, the fully connected layer is used as a decoder structure, and the nonlinear feature extraction capability of the model is enhanced. An unknown fault detection method based on distance difference ratio is proposed, the Euclidean distance between the test sample and the training set sample is calculated, the optimal threshold is obtained according to the change extreme value, and the complexity of selecting the threshold is reduced. Finally, K-means++ is used for unsupervised clustering evaluation of the model, a clustering factor composed of a contour coefficient and a multivariate probability density function is constructed to evaluate the clustering effect, and unknown fault class number prediction is realized. Through the implementation case verification, the method of the present application can realize the unknown fault diagnosis of the landing gear system.

[0149] It should be noted that the above-described device embodiments are only schematic and that each unit shown as a separate unit can or can not be physically separate, and each unit shown as a unit can or can not be a physical unit, i.e., can be located in one place or can be distributed on a plurality of network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the present embodiment. In addition, the device embodiments provided in the present application have a connection relationship between the modules, which indicates that there is a communication connection between them, which can be realized as one or more communication buses or signal lines. Those skilled in the art can understand and implement it without creative labor.

[0150] The embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be mutually referred to.

[0151] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to the embodiments will be apparent 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 application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A landing gear system unknown fault diagnosis method characterized by, Comprising the following steps: S100: data acquisition, at least one set of sensor signals is collected by a sensor and a sample data set is established; S200: data preprocessing, the sample data set is divided into a training set and a test set, wherein the number of fault types in the training set is less than that in the test set, to ensure that the unknown faults defined in the test set do not participate in the training process; S300: constructing an unknown fault diagnosis model, a set of parallel convolutional auto-encoding structures are established to obtain uncertainty fault features and difference features, and a correlation loss function is used to constrain the feature reconstruction process; Each convolutional auto-encoding structure includes an encoding layer and a decoding layer, the encoding layer includes at least two convolutional layers and a pooling layer, the decoding layer includes three fully connected layers, the last layer of the encoding layer is used as the feature output, and a classification layer is connected to form a semi-supervised model structure, and an unknown fault diagnosis model is obtained; In S300, the uncertainty fault features and difference features satisfy the algebraic sum and the attribute, and the algebraic sum of the two represents the complete features of the sample, and the calculation steps are as follows: S301: constructing a set of parallel and asymmetric convolutional auto-encoding structures as a feature extractor, and calling the feature extractor to extract features from the sample data set; S302: using a reconstruction loss function as a correlation loss function, calling the reconstruction loss function to train the feature extractor, and obtaining an uncertainty feature extractor and a difference feature extractor; during the training process, the uncertainty feature extractor adds an additional uncertainty loss function, and the difference feature extractor adds an additional difference loss function; The reconstruction loss function is: where n denotes the number of samples; G u (·) denotes the decoding function of the uncertainty feature extractor; g u (·) denotes the uncertainty feature extractor; G d (·) denotes the decoding function of the difference feature extractor; g d (·) denotes the difference feature extractor; The difference function is obtained by the ratio of the average intra-class radius to the average inter-class distance; S303: calling the uncertainty feature extractor and the difference feature extractor to decompose the features extracted in S301, to obtain uncertainty features and difference features; S400: parameter updating, inputting the training set into the unknown fault diagnosis model, randomly selecting a fixed number of data in the training set to participate in the training process, and iteratively optimizing the model parameters; when the set iteration step is met or the accuracy threshold condition is met, the training is stopped and the model is saved; S500: threshold updating, in the model test process, an adaptive threshold strategy based on distance difference ratio is used to adaptively determine the unknown fault threshold, realize threshold updating, and complete the model training of the unknown fault diagnosis model; S600: fault diagnosis, inputting the to-be-tested data into the trained unknown fault diagnosis model, judging whether the to-be-tested data belongs to a known fault or an unknown fault, and judging the fault type, outputting the known fault, unknown fault and unknown fault type number as the diagnosis result, and completing the fault diagnosis.

2. The landing gear system unknown fault diagnosis method according to claim 1, characterized by, In S100, the sensor is arranged at a key measurement point of the landing gear system, and one or more sets of sensor signals reflecting the health state of the landing gear system are collected, the sensor signals are classified according to different working conditions and different health states, and a sample data set is established.

3. The landing gear system unknown fault diagnosis method according to claim 1, characterized by, In S200, when the sample data set is insufficient in sample quantity, the sample is enhanced by using sliding data sampling, and the sliding data sampling is specifically as follows: a fixed window size is used for the original sample, and a plurality of samples with the window size are obtained by shifting from left to right.

4. The landing gear system unknown fault diagnosis method according to claim 1, characterized by, In S400, the total loss function of the parameter updating process is as follows: The constructed convolutional auto-encoding structure uses a cross-entropy loss function to introduce label information. An additional L2 norm regularization term is added to the convolutional auto-encoding structure as a constraint term, which is used to constrain the parameter range. The total loss function of the uncertainty feature extractor is composed of a reconstruction loss function, a cross-entropy loss function, an uncertainty loss function and a constraint term. The total loss function of the difference feature extractor is composed of a reconstruction loss function, a cross-entropy loss function, a difference loss function and a constraint term.

5. The landing gear system unknown fault diagnosis method according to claim 1, characterized by, In S500, the adaptive threshold strategy based on the distance difference ratio is as follows: S501: Calculate the difference ratio. The Euclidean distance is used to calculate the difference value of each sample in the test set and the sample in the training set, and the difference value set of all samples is obtained. S502: Sort the difference value set from small to large to obtain the ordered difference value set. S503: Calculate the difference change extreme value. The change rate of the continuous two difference values is calculated by slicing the ordered difference value set, and the difference value change rate set is obtained. The maximum value of the middle section in the difference value change rate set is the difference change extreme value. S504: Calculate the unknown fault threshold. Take the difference value corresponding to the sequence number of the difference change extreme value as the unknown fault threshold. For the sample with a difference value greater than the unknown fault threshold, it is divided into unknown fault.

6. The landing gear system unknown fault diagnosis method according to claim 1, characterized by, In S600, it is judged whether the to-be-tested data belongs to known fault or unknown fault, and the fault type is judged, which is specifically as follows: When the distance value of the to-be-tested data is lower than the threshold, it belongs to known fault, and the classifier is called to judge the fault type. When the distance value of the to-be-tested data is greater than or equal to the threshold, it is judged to belong to unknown fault. Finally, the unsupervised analysis is used to realize the unknown fault class number prediction.

7. The landing gear system unknown fault diagnosis method according to claim 6, characterized in that, The unsupervised analysis is specifically as follows: the k-means++ unsupervised clustering is performed, the k-means++ unsupervised clustering result is evaluated by using the silhouette coefficient and the multivariate probability density function, and the unknown fault class number prediction is realized.

8. A computer device, comprising: The memory and the processor are included. The memory is used to store a computer program which can run on the processor. The processor is used to execute the computer program to realize the steps of the unknown fault diagnosis method of the landing gear system according to any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer program is stored on the computer readable storage medium, and the computer program is executed by the processor to realize the steps of the unknown fault diagnosis method of the landing gear system according to any one of claims 1-7.

Images