Early fault detection method for armored vehicle under multiple working conditions based on LNS-QKECA

Abstract

The application discloses a kind of early fault detection methods of armored vehicle under multiple working conditions based on LNS-QKECA.The specific steps are as follows: step 1 collects the operating data of each component of armored vehicle as characteristic parameter, and is stored in database;Step 2. The characteristic parameter obtained in step 1 is pretreated and a sample feature matrix is constructed;Step 3. Obtain training sample data to construct and train model, and use new data to iteratively train the model to obtain a higher precision model;Step 4. Early fault detection is carried out on each component of armored vehicle, and the data is recorded and stored in database, which provides basis for subsequent analysis.A kind of early fault detection system of armored vehicle under multiple working conditions based on local neighborhood standardization and secondary kernel entropy component analysis includes data acquisition module, model construction and training module and early fault detection module.The application can convert multiple working conditions of armored vehicle into a single working condition, improve the accuracy of early fault detection and has high practicality.

Description

Technical Field

[0001] This invention relates to the field of fault detection, specifically to a method for early fault detection of armored vehicles under multiple operating conditions based on Local Nearest Neighbor Normalization and Quadratic Kernel Entropy Component Analysis (LNS-QKECA). Background Technology

[0002] With the maturity of intelligent manufacturing technology and the development of emerging technologies, modern armored vehicles possess increasingly rich functions and undertake more complex tasks. This also means armored vehicles face more complex operating environments, posing a significant challenge to fault detection. Existing multi-condition fault detection methods for armored vehicles involve pre-setting multiple operating conditions, conducting test runs, and establishing corresponding fault detection models based on the characteristics of each condition. During use, after identifying the operating condition, the armored vehicle uses the corresponding fault detection model for fault detection. However, in existing test runs, the pre-set scenarios cannot fully cover all operating environments, making it difficult to accurately match the current operating condition during use and affecting fault detection results.

[0003] Therefore, it is necessary to design an early fault detection model that can improve the accuracy of fault detection for armored vehicles under various operating conditions. Summary of the Invention

[0004] To address the aforementioned problems, the present invention aims to provide an early fault detection method for armored vehicles under multiple operating conditions based on local nearest neighbor standardization and quadratic kernel entropy component analysis. This method can transform the multi-condition problem into a single-condition problem through the local nearest neighbor standardization algorithm, and obtain the early fault detection results of armored vehicles using quadratic kernel entropy component analysis based on information theory and multivariate statistics.

[0005] The technical solution adopted by the present invention to achieve the above objectives is as follows:

[0006] The LNS-QKECA-based method for early fault detection of armored vehicles under multiple operating conditions includes the following steps:

[0007] 1) Collect operational data of various components of the armored vehicle as characteristic parameters and store them in the database;

[0008] 2) Preprocess the feature parameters and construct the sample feature matrix;

[0009] 3) Construct a fault diagnosis model using the sample feature matrix, and iteratively train the model using the feature parameters collected in real time;

[0010] 4) Early fault detection is performed on various components of armored vehicles using a trained fault diagnosis model, and the data is recorded and stored in a database.

[0011] Step 2) includes the following steps:

[0012] 2.1) Calculate the time-domain and frequency-domain features of the feature parameters, and summarize the time-domain and frequency-domain features of each feature parameter to construct a feature parameter matrix;

[0013] 2.2) Using the local nearest neighbor normalization algorithm, local information and nearest neighbor information are used to replace global information to preprocess the feature parameter matrices under multiple working conditions, obtain the sample feature matrix under a single working condition, and store it in the historical database as a modeling sample.

[0014] The local nearest neighbor normalization algorithm is as follows:

[0015]

[0016] Where, n(X) i ) represents the i-th sample X i Let m[n(X) be the k-nearest neighbor set, n and m be the data sample size and the number of feature parameters, respectively. i )] and s[n(X i )] are the mean and standard deviation of the nearest neighbor set, respectively.

[0017]

[0018]

[0019] Step 3) includes the following steps:

[0020] 3.1) The kernel entropy component analysis algorithm is used to perform data dimensionality reduction and first feature extraction on the modeling samples to obtain the feature matrix after the first kernel entropy analysis;

[0021] 3.2) Use multi-scale sample entropy to perform a second feature extraction on the feature matrix after the first analysis to obtain the feature matrix after the second kernel entropy analysis;

[0022] 3.3) Calculate the SPE control limit and T after the second kernel entropy analysis. 2 Control limits, and set the SPE control limits to T 2 The control limits are fused together to form a comprehensive control limit for the fault detection model, and the comprehensive control limit value is stored in the database.

[0023] The kernel entropy component analysis algorithm projects data into a high-dimensional space using a kernel function and extracts features from an information theory perspective using Renyi entropy, thereby completing dimensionality reduction and initial feature extraction of the data. Specifically:

[0024] 3.1.1) For a dataset D generated by the probability density function p(x), (x1,x2,…,x…) n), calculate Renyi entropy H(p):

[0025] H(p)=-logV(p)=-log∫p 2 (x)dx

[0026] Where V(p) is an intermediate variable;

[0027] 3.1.2) Estimate p(x) using the Parzen window method:

[0028]

[0029] Where N is the sample size. Let x be an estimate of p(x). i ∈D, k σ (x,x i ) is the Mercer kernel function;

[0030] 3.1.3) Estimate V(p):

[0031]

[0032] in, Let V(p) be the estimated value, representing the Renyi entropy contribution; I is an n-dimensional column vector with all elements equal to 1; K is an N×N kernel function matrix.

[0033] 3.1.4) Perform eigenvalue decomposition and simplification on K:

[0034] K = EΛE T

[0035] Among them, Λ=diag(λ1,λ2,..,λ n Let E be a diagonal matrix of eigenvalues, and let E = diag(e1, e2, ..., e...). n ) is the eigenvector matrix, and thus we obtain:

[0036]

[0037] 3.1.5) Select the eigenvalues ​​Λ of the top A features that contribute the most to Renyi entropy. A =diag(λ1,λ2,..,λ) a ) and the corresponding first A feature vectors E A =diag(e1,e2,..,e a After standardization, the projection matrix Ψ = {ψ1,ψ2,…,ψ} is obtained. A If}, then the dimensionality-reduced sample matrix is ​​Y = KΨ.

[0038] The multi-scale sample entropy measures the probability of generation of new sample patterns from multiple time dimensions by artificially adding multiple granularities, so as to measure the probability of data deviating from the original state, and perform secondary feature extraction. The specific steps are:

[0039] 3.2.1) For the sequence Y i = {y i (1), y i (2), …, y i (N)}, 0 < i < A, the j-th coarse-grained sequence T (τ) is:

[0040]

[0041] wherein, τ is a scale factor, and the length of the coarse-grained sequence relative to the original time sequence is

[0042] 3.2.2) When τ takes different values, calculate the SE value of T (τ) , which is MSE:

[0043]

[0044] wherein, r is a similarity tolerance, m is an embedding dimension, E SE (·) is a sample entropy value, and are respectively the number of m-dimensional and m+1-dimensional spatial vectors of the coarse-grained sequence;

[0045] 3.2.3) Perform weighting processing on MSE so that the weight value w i is distributed around 1:

[0046]

[0047] wherein, m(E MSE (y,τ,m,r)) is the mean value of E MSE (y,τ,m,r), and the principal component matrix Y after multi-scale sample entropy transformation is:

[0048] Q = WY wherein W = diag(w1,w2,…,w n ) is the transformed multi-scale sample entropy matrix, and Q is the transformed principal component matrix, that is, the feature matrix after secondary kernel entropy analysis.

[0049] The comprehensive control limit is:

[0050]

[0051] wherein, δ 2 is a delta distribution, χ 2The distribution follows a chi-square distribution, where Φ is an intermediate variable.

[0052]

[0053] Where P is the load matrix, P = KW, This is the residual load matrix.

[0054] Step 4) specifically involves:

[0055] The system acquires the diagnostic data of armored vehicles in real time and processes it using a trained fault diagnosis model to obtain the comprehensive control limit as a comprehensive statistic. If the comprehensive statistic is greater than the comprehensive control limit, the armored vehicle is in an abnormal state; if the comprehensive statistic is less than or equal to the comprehensive control limit, the armored vehicle is in a normal state.

[0056] The LNS-QKECA-based multi-condition early fault detection system for armored vehicles includes:

[0057] The data acquisition module is used to collect operational data of various components of the armored vehicle as characteristic parameters and store them in the database;

[0058] The data processing module is used to preprocess the feature parameters and construct the sample feature matrix;

[0059] The fault diagnosis model building module is used to build a fault diagnosis model using the sample feature matrix and to iteratively train the model using the feature parameters collected in real time.

[0060] The fault diagnosis module is used to perform early fault detection on various components of armored vehicles using a trained fault diagnosis model, record the data, and store it in a database.

[0061] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the LNS-QKECA-based method for early fault detection of armored vehicles under multiple operating conditions.

[0062] The present invention has the following beneficial effects and advantages:

[0063] 1. This invention uses a local nearest neighbor normalization algorithm to transform multi-condition problems into single-condition problems, thus avoiding the problem of decreased detection accuracy of fault detection models when complex and variable conditions occur in traditional methods.

[0064] 2. By using a secondary kernel entropy component analysis algorithm, this invention extracts the state characteristics of armored vehicles more deeply, making the state information representation of armored vehicles more accurate.

[0065] 3. This invention uses a comprehensive statistical measure that integrates SPE and T2 statistics to characterize the principal component space and residual space of the armored vehicle's state space, thus providing a more comprehensive and intuitive representation of the armored vehicle's state. Attached Figure Description

[0066] Figure 1 This is a flowchart illustrating the present invention.

[0067] Figure 2 The flowchart shows the algorithm for local nearest neighbor standardization and secondary kernel entropy component analysis. Detailed Implementation

[0068] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0069] like Figure 1 As shown, a multi-condition early fault detection method for armored vehicles based on local nearest neighbor normalization and secondary kernel entropy component analysis is presented. The specific steps are as follows:

[0070] Step 1. Collect operational data of various components of the armored vehicle as feature parameters and store them in the database;

[0071] Step 2. Preprocess the feature parameters obtained in Step 1 and construct the sample feature matrix;

[0072] Step 3. Obtain training sample data to build and train the model, and use new data to iteratively train the model to obtain a more accurate model;

[0073] Step 4. Conduct early fault detection on various components of the armored vehicle, record the data and store it in the database to provide a basis for subsequent analysis.

[0074] The collection of operational data from various components of the armored vehicle includes:

[0075] Establish a sample library and a historical database, collect operational data of various components of armored vehicles under various working conditions as samples, conduct initial screening of the samples, analyze and identify abnormal data in the samples, and correct, supplement or delete the data.

[0076] The data collected in step 1 includes characteristics that characterize the armored vehicle's condition, such as engine speed, load, torque, engine water temperature, coolant temperature, and oil pressure.

[0077] The feature parameters are preprocessed and a feature parameter matrix is ​​constructed, including:

[0078] The time-domain and frequency-domain features of the feature parameters are calculated, and the time-domain and frequency-domain features of each feature parameter are summarized to construct a feature parameter matrix. Using the local nearest neighbor normalization algorithm, the feature parameter matrices under multiple operating conditions are preprocessed to obtain the sample feature matrix for a single operating condition, which is then stored in the historical database as modeling samples.

[0079] Step 2 describes using local nearest neighbor standardization to handle multi-condition problems, replacing traditional global information with local and nearest neighbor information to remove the multi-condition attributes of the data.

[0080] like Figure 2 As shown, the step of acquiring training sample data to construct and train the model includes:

[0081] Model building and training are performed using modeling samples from a historical database. Kernel entropy component analysis is used to perform dimensionality reduction and first feature extraction on the modeling samples, obtaining a feature matrix after the first kernel entropy analysis. Multi-scale sample entropy is then used to perform a second feature extraction on the feature matrix after the first analysis, obtaining a feature matrix after the second kernel entropy analysis. The SPE control limits and T are calculated after the second kernel entropy analysis. 2 Control limits, and set the SPE control limits to T 2 The control limits are fused to obtain the comprehensive control limits of the early fault detection model. The comprehensive control limit values ​​are then stored in a database.

[0082] The kernel entropy component analysis described in step 3 projects the data into a high-dimensional space using a kernel function, and then uses Renyi entropy to extract features from an information theory perspective, thereby completing the dimensionality reduction and initial feature extraction of the data.

[0083] The multi-scale sample entropy mentioned in step 3 measures the probability of new sample patterns from multiple time dimensions by artificially adding multiple granularities, thereby measuring the probability of data deviating from the original state, and thus performing a second feature extraction.

[0084] Step 3 includes constructing an early fault detection model using historical normal operation samples of armored vehicles through a secondary kernel entropy component analysis algorithm.

[0085] The two entropy types used in the aforementioned quadratic kernel entropy component analysis algorithm are Renyi entropy and multi-scale sample entropy, respectively. A hybrid kernel is used, comprising both Gaussian and sigmoid kernel functions. The control limits are a comprehensive control limit, integrating the SPE statistic and the T-statistic. 2 Statistics.

[0086] The aforementioned early fault detection of various components of armored vehicles includes:

[0087] The time-domain and frequency-domain features of samples at a given time point in the sample database are calculated, and the time-domain and frequency-domain features of all samples are summarized to construct a feature parameter matrix. The feature parameter matrix is ​​preprocessed using the local nearest neighbor normalization algorithm, and the resulting sample feature matrix is ​​used as the detection sample. Kernel entropy component analysis is used to perform dimensionality reduction and first feature extraction on the detection samples, obtaining the feature matrix after the first kernel entropy analysis. Multi-scale sample entropy is used to perform a second feature extraction on the feature matrix after the first analysis, obtaining the feature matrix after the second kernel entropy analysis. The SPE statistic and T0 of the feature matrix after the second kernel entropy component analysis are calculated. 2 The SPE statistic is used to compare the SPE statistic with the T statistic. 2 The statistical measures are fused to obtain a comprehensive statistical measure. The comprehensive statistical measure of the detected samples is compared with the comprehensive control limit of the early fault detection model. If the comprehensive statistical measure is greater than the comprehensive control limit, the armored vehicle is in an abnormal state; if the comprehensive statistical measure is less than the comprehensive control limit, the armored vehicle is in a normal state.

[0088] Record data and store it in a database to provide a basis for subsequent analysis, including:

[0089] The results of early fault detection are stored in a database. After completing early fault detection for samples at all time points, an early abnormal state monitoring map for a certain period is obtained. This provides a basis for subsequent fault diagnosis, remaining life prediction, and maintenance.

[0090] Example

[0091] This invention provides a method for early fault detection of armored vehicles under multiple operating conditions based on local nearest neighbor normalization and secondary kernel entropy component analysis, comprising the following steps:

[0092] Step 1: Collect operational data of each component of the armored vehicle as feature parameters and store them in the database;

[0093] Characteristic variables that can characterize the state of an armored vehicle include engine speed, load, torque, engine coolant temperature, coolant temperature, oil pressure, and oil temperature. These parameters can be collected by sensors on the armored vehicle.

[0094] Step 2: Preprocess the acquired feature parameters and construct the sample feature matrix;

[0095] This step calculates the time-domain and frequency-domain features of the armored vehicle state characteristics collected from the database, and summarizes them to construct a feature parameter matrix. The calculation formulas for the selected time-domain and frequency-domain features are shown in Table 1.

[0096] Table 1 Selection of Time-Domain and Frequency-Domain Features

[0097]

[0098]

[0099] The local nearest neighbor normalization algorithm is used to eliminate the influence of multiple operating conditions on the data, transforming the multi-condition problem into a single-condition problem, thus completing the data preprocessing. The modeling data is represented by a two-dimensional matrix X (n×m), where n and m are the data sample size and the number of feature parameters, respectively. The local nearest neighbor normalization algorithm first determines each sample X... i The k-nearest neighbor set n(X) i And calculate the average value m[n(X) of the nearest neighbor set. i )] and standard deviation s[n(X i If the nearest neighbor standardization process is then performed, then the process is as follows:

[0100]

[0101] in

[0102]

[0103]

[0104] Step 3: Obtain training sample data to build and train the model, and use new data to iteratively train the model to obtain a more accurate model;

[0105] An early fault detection model was constructed using a secondary kernel entropy component analysis algorithm and historical normal operation samples of armored vehicles.

[0106] The second-order kernel entropy component analysis algorithm calculates the following:

[0107] Renyi entropy calculation formula:

[0108] H(p)=-logV(p)=-log∫p 2 (x)dx

[0109] Since p(x) is often difficult to obtain from prior knowledge, it is generally estimated using the Parzen window method. The estimation method is as follows:

[0110]

[0111] Where, x i ∈d;k σ (x,x i ) is the Mercer kernel function; by estimating V(p), we can obtain

[0112]

[0113] wherein, I is an n-dimensional column vector with all elements being 1; K is an N×N kernel function matrix. After eigen decomposition and simplification of K, the following formula can be obtained

[0114] K=EΛE T

[0115] wherein, Λ=diag(λ1,λ2,..,λ n ) is a diagonal matrix of eigenvalues, E=diag(e1,e2,..,e n ) is an eigenvector matrix. Further derivation can obtain

[0116]

[0117] select the first A eigenvalues Λ with the largest contribution to Renyi entropy A =diag(λ1,λ2,..,λ a ) and the corresponding first A eigenvectors E A =diag(e1,e2,..,e a ), after standardization, the projection matrix Ψ={ψ1,ψ2,…,ψ A} is obtained, and then the dimension-reduced sample matrix is Y=KΨ.

[0118] Calculation method of multi-scale sample entropy:

[0119] For sequence Y i ={y i (1),y i (2),…,y i (N)},0<i<A, the coarse-grained sequence T (τ) is

[0120]

[0121] wherein τ is a scale factor, which is a positive integer, and the length of the coarse-grained sequence is that of the original time series

[0122] When τ takes different values, calculate the SE value of T (τ) , which is MSE:

[0123]

[0124] wherein r is a similarity tolerance; m is an embedding dimension; E SE (·) is a sample entropy value; and are respectively the number of m-dimensional and m+1-dimensional spatial vectors of the coarse-grained sequence. Weighting processing is performed on MSE and the weighted values are distributed around 1

[0125]

[0126] Where m(E) MSE (y,τ,m,r)) is E MSE The mean of (y,τ,m,r).

[0127] The calculation method for the comprehensive statistic is as follows:

[0128]

[0129] in

[0130]

[0131] Where δ 2 For delta distribution, χ 2 It follows a chi-square distribution. The residual load matrix is ​​calculated using the same method from the remaining eigenvalues ​​and eigenvectors of the first A units that were not selected.

[0132] Step 4: Conduct early fault detection on various components of the armored vehicle, record the data and store it in the database to provide a basis for subsequent analysis.

[0133] The time-domain and frequency-domain features of samples at a given time point in the sample database are calculated, and the time-domain and frequency-domain features of all samples are summarized to construct a feature parameter matrix. The feature parameter matrix is ​​preprocessed using the local nearest neighbor normalization algorithm, and the resulting sample feature matrix is ​​used as the detection sample. Kernel entropy component analysis is used to perform dimensionality reduction and first feature extraction on the detection samples, obtaining the feature matrix after the first kernel entropy analysis. Multi-scale sample entropy is used to perform a second feature extraction on the feature matrix after the first analysis, obtaining the feature matrix after the second kernel entropy analysis. The SPE statistic and T0 of the feature matrix after the second kernel entropy component analysis are calculated. 2 The SPE statistic is used to compare the SPE statistic with the T statistic. 2 Statistical fusion yields a comprehensive statistic. The comprehensive statistic of the detected samples is compared with the comprehensive control limit of the early fault detection model. If the comprehensive statistic is greater than the comprehensive control limit, the armored vehicle is in an abnormal state; if the comprehensive statistic is less than the comprehensive control limit, the armored vehicle is in a normal state. The early fault detection results of the armored vehicle are stored in a database to provide data support for subsequent analysis.

[0134] The implementation methods described in this specification are interconnected and cannot be implemented independently of any single step. For details on the differences and similarities between each implementation method, please refer to the specification section. It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0135] The above description is merely an embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, extensions, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. A method for early fault detection of armored vehicles under multiple operating conditions based on LNS-QKECA, characterized in that, Includes the following steps: 1) Collect operational data of various components of the armored vehicle as characteristic parameters and store them in the database; 2) Preprocess the feature parameters and construct the sample feature matrix; 3) Construct a fault diagnosis model using the sample feature matrix, and iteratively train the model using the feature parameters collected in real time; 4) Early fault detection is performed on various components of armored vehicles using a trained fault diagnosis model, and the data is recorded and stored in a database; Step 3) includes the following steps: 3.1) The kernel entropy component analysis algorithm is used to perform data dimensionality reduction and first feature extraction on the modeling samples to obtain the feature matrix after the first kernel entropy analysis; 3.2) Use multi-scale sample entropy to perform a second feature extraction on the feature matrix after the first analysis to obtain the feature matrix after the second kernel entropy analysis; 3.3) Calculate the SPE control limit and T after the second kernel entropy analysis. 2 Control limits, and set the SPE control limits to T 2 The control limits are fused together to form a comprehensive control limit for the fault detection model, and the comprehensive control limit value is stored in the database.

2. The method for early fault detection of armored vehicles under multiple operating conditions based on LNS-QKECA according to claim 1, characterized in that, Step 2) includes the following steps: 2.1) Calculate the time-domain and frequency-domain features of the feature parameters, and summarize the time-domain and frequency-domain features of each feature parameter to construct a feature parameter matrix; 2.2) Using the local nearest neighbor normalization algorithm, local information and nearest neighbor information are used to replace global information to preprocess the feature parameter matrices under multiple working conditions, obtain the sample feature matrix under a single working condition, and store it in the historical database as a modeling sample.

3. The method for early fault detection of armored vehicles under multiple operating conditions based on LNS-QKECA according to claim 2, characterized in that, The local nearest neighbor normalization algorithm is as follows: in, For the i-th sample The k-nearest neighbor set, and These are the data sample size and the number of feature parameters, respectively. and These are the mean and standard deviation of the nearest neighbor set, respectively. 。 4. The method for early fault detection of armored vehicles under multiple operating conditions based on LNS-QKECA according to claim 1, characterized in that, The kernel entropy component analysis algorithm projects data into a high-dimensional space using a kernel function and extracts features from an information theory perspective using Renyi entropy, thereby completing dimensionality reduction and initial feature extraction of the data. Specifically: 3.1.1) For a probability density function The generated dataset Calculate Renyi entropy : in, As an intermediate variable; 3.1.2) Using the Parzen window method Make an estimate: Where N is the sample size. for The estimated value, For Mercer kernel functions; 3.1.3) Regarding Make an estimate: in, for The estimated value is used as the Renyi entropy contribution. For all elements are 1 3D column vector; for The kernel function matrix; 3.1.4) Perform eigenvalue decomposition and simplification on K: in, A diagonal matrix of eigenvalues. Given the eigenvector matrix, we can then obtain: ； 3.1.5) Select the eigenvalues ​​of the top A features that contribute the most to Renyi entropy. With the corresponding first A feature vectors After standardization, the projection matrix is ​​obtained. Then the dimensionality-reduced sample matrix is .

5. The method for early fault detection of armored vehicles under multiple operating conditions based on LNS-QKECA according to claim 1, characterized in that, The multi-scale sample entropy measures the probability of new sample patterns arising from multiple time dimensions by artificially adding various granularities, thereby measuring the probability of data deviating from the original state and performing a second feature extraction, specifically: 3.2.1) For sequences Its j-th coarse-grained sequence for: in, The scale factor is used to coarse-grained sequence length, which is the length of the original time series. ; 3.2.2) When Calculate when taking different values The SE value is also known as MSE: in, For similarity tolerance, For the embedding dimension, It is the sample entropy value. and These are coarse-grained sequences. dimension, Number of vectors in the dimensional space; 3.2.3) Weight the MSE and adjust the weighted values. Distributed around 1: in, for The mean, principal component matrix After multi-scale sample entropy changes, the result is: in, This is the transformed multi-scale sample entropy matrix. This is the transformed principal component matrix, i.e., the characteristic matrix after the second kernel entropy analysis.

6. The method for early fault detection of armored vehicles under multiple operating conditions based on LNS-QKECA according to claim 1, characterized in that, The comprehensive control limit for: in, It is a delta distribution. It follows a chi-square distribution. As an intermediate variable, Where P is the load matrix. , This is the residual load matrix.

7. The method for early fault detection of armored vehicles under multiple operating conditions based on LNS-QKECA according to claim 1, characterized in that, Step 4) specifically involves: The system acquires the diagnostic data of armored vehicles in real time and processes it using a trained fault diagnosis model to obtain the comprehensive control limit as a comprehensive statistic. If the comprehensive statistic is greater than the comprehensive control limit, the armored vehicle is in an abnormal state; if the comprehensive statistic is less than or equal to the comprehensive control limit, the armored vehicle is in a normal state.

8. A multi-condition early fault detection system for armored vehicles based on LNS-QKECA, used to implement the multi-condition early fault detection method for armored vehicles based on LNS-QKECA as described in any one of claims 1-7, characterized in that, include: The data acquisition module is used to collect operational data of various components of the armored vehicle as characteristic parameters and store them in the database; The data processing module is used to preprocess the feature parameters and construct the sample feature matrix; The fault diagnosis model building module is used to build a fault diagnosis model using the sample feature matrix and to iteratively train the model using the feature parameters collected in real time. The fault diagnosis module is used to perform early fault detection on various components of armored vehicles using a trained fault diagnosis model, record the data, and store it in a database.

9. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the LNS-QKECA-based method for early fault detection of armored vehicles under multiple operating conditions as described in any one of claims 1-7.

Images