A condition assessment method for vehicle gearbox

Abstract

The invention relates to the technical field of gearbox state assessment and discloses a vehicle gearbox state assessment method, comprising the following steps: S1, collecting physical and chemical performance data of oil in the gearbox; S2, preprocessing the collected data by an independent component analysis (ICA) method to screen out a data set that can be used as a model input; dividing the model input data set into a test data set and a training data set; S3, improving a red-billed blue magpie optimization algorithm (RBMO), comprising initializing the RBMO population by a Latin hypercube sampling method to improve population initialization uniformity; introducing a spiral search strategy in the RBMO foraging phase to improve the red-billed blue magpie search efficiency; introducing a weight factor in the RBMO attack phase to optimize the red-billed blue magpie position update to obtain an improved red-billed blue magpie optimization algorithm (IRBMO), etc. The invention overcomes the defect of blind parameter selection in the training process and improves the prediction accuracy of the regression prediction model.

Description

Technical Field

[0001] The present invention relates to the technical field of gearbox condition assessment, and in particular to a method for assessing the condition of a vehicle gearbox. Background Art

[0002] Gearboxes are core components of vehicle power transmission systems and are often placed in complex and harsh operating environments, subject to enormous impact loads, high temperatures, vibrations, and wear and tear from long-term operation. During high-intensity combat missions, common types of gearbox failures include gear wear, bearing fatigue failure, lubricant degradation, and abnormal meshing of transmission components. If these failures are not detected promptly, they can lead to power outages, vehicle downtime, or even complete equipment failure, posing a serious threat to mission execution and operational safety. Condition assessments can provide real-time insights into the operating status of gearboxes, enabling the timely identification of potential failure risks and preventing vehicle downtime or mission failure due to gearbox failure at critical moments. This ensures the continued combat capability of armored vehicles in complex battlefield environments. Therefore, performing condition assessments on gearboxes is particularly important.

[0003] In the field of mechanical system condition assessment and fault diagnosis, anomaly detection technology has been widely used for condition monitoring of industrial equipment due to its sensitivity to early-stage faults and real-time performance. Support vector machines (SVMs) are mature machine learning algorithms with strong generalization and nonlinear processing capabilities. However, when the amount of fault data is large (such as long-term monitoring of high-frequency vibration signals), training speed slows significantly, making it difficult to meet real-time requirements. Gearbox faults often involve multi-state classification (such as normal, tooth root cracks, broken teeth, and wear). SVMs are native binary classifiers and need to be expanded through "one-to-one" or "one-to-many" strategies, which may introduce class conflicts and computational redundancy.

[0004] The single-class support vector machine (OCSVM), developed on this basis, is specifically designed for anomaly detection. By training the model using only normal data, OCSVM effectively identifies unknown abnormal conditions and is particularly suitable for condition assessment of complex mechanical systems such as gearboxes. Compared to other algorithms, the single-class support vector machine exhibits strong adaptability to the scarcity of abnormal samples, excellent handling of nonlinear relationships, and good generalization capabilities, resulting in highly accurate condition assessment results. Summary of the Invention

[0005] In response to the above-mentioned shortcomings and deficiencies in the prior art, the present invention provides a vehicle gearbox condition assessment method. The method improves the Red-billed Blue Magpie Optimization (RBMO) algorithm and initializes the RBMO population using the Latin Hypercube sampling method to improve the uniformity of population initialization. A spiral search strategy is introduced during the RBMO foraging phase to improve the search efficiency of the Red-billed Blue Magpie. A weight factor is introduced during the RBMO attack phase to optimize the position update of the Red-billed Blue Magpie. The key parameters of the one-class support vector machine (OCSVM) are optimized to construct an IRBMO-OCSVM fault diagnosis model. This method overcomes the defect of blind parameter selection during the training process. The method comprises the following steps:

[0006] S1. Collect the physical and chemical properties data of the oil in the gearbox;

[0007] S2. Preprocess the collected data using independent component analysis (ICA) to select a dataset that can be used as model input; divide the model input dataset into a test dataset and a training dataset;

[0008] S3. Improvements to the Red-billed Blue Magpie Optimization Algorithm (RBMO), including initializing the RBMO population through Latin hypercube sampling to improve population initialization uniformity; introducing a spiral search strategy during the RBMO foraging phase to improve the search efficiency of the Red-billed Blue Magpie; and introducing a weight factor during the RBMO attack phase to optimize the position update of the Red-billed Blue Magpie, resulting in an improved Red-billed Blue Magpie Optimization Algorithm (IRBMO).

[0009] S4, using the improved red-billed blue magpie optimization algorithm IRBMO to optimize the key parameters of the single-class support vector machine OCSVM, constructing the IRBMO-OCSVM fault diagnosis model, and training the fault diagnosis model using the training data set in S2;

[0010] S5. Input the test data set in S2 into the trained IRBMO–OCSVM model, perform fault diagnosis on the gearbox, and output the prediction results.

[0011] In step S3, the RBMO population is initialized by the Latin hypercube sampling method, and the formula is:

[0012] ; ;

[0013] Where: a i and b i For an interval [a i , b i ] at both ends; is to divide the interval [a i , b i ] is divided into the length of the i-th subinterval in n subintervals; k is a random number from 1 to n; xij is a point selected from the kth subinterval; rand(0,1) is a random number uniformly distributed between [0,1];

[0014] The spiral search strategy is introduced in the RBMO foraging stage to optimize the position update of the red-billed blue magpie:

[0015] 0≤Rand1<0.5;

[0016] 0.5≤Rand1≤1;

[0017] Where, t represents the current number of iterations; X i (t+1) represents the position of the i-th individual in the t+1 iteration, p is a random number between [2,5], which represents the number of red-billed blue magpies in a small group, q is a random number between [10,n], which represents the number of red-billed blue magpies in a group, X m represents the randomly selected mth individual, X i represents the i-th individual, X rs (t) represents the randomly selected individual in the current iteration; Rand1 is a random number; It is an exponential function used to adjust the expansion or contraction of the spiral, and c and l control the tightness and periodicity of the spiral respectively;

[0018] In the RBMO attack phase, the weight factor is introduced to optimize the position update of the red-billed blue magpie:

[0019] 0≤Rand n<0.5;

[0020] 0.5≤Rand n≤1;

[0021] ;

[0022] ;

[0023] Where, X food(t) Indicates the location of the food; Rand n represents the random number used to generate the standard normal distribution; T is the maximum number of iterations; CF is a dynamic variable that changes with the number of iterations; is the weight factor;

[0024] Finally, the global optimal value is determined as follows:

[0025] ;

[0026] in and They represent the fitness values of the i-th red-billed blue magpie before and after the position update.

[0027] The improved red-billed blue magpie optimization algorithm IRBMO in step S4 optimizes key parameters of the single-class support vector machine OCSVM, and the optimized parameters include kernel parameters, regularization parameters, and slack variables.

[0028] Compared with the prior art, the present invention has the following beneficial technical effects and advantages:

[0029] Using independent component analysis to pre-process the data can extract independent key signal features, separate the nonlinear characteristics between the physical and chemical information of the oil, remove the noise in the data, and improve the data quality;

[0030] An improved Red-billed Blue Magpie optimization algorithm (IRBMO) was used to optimize the key parameters of the one-class support vector machine (OCSVM). An IRBMO-OCSVM fault diagnosis model was constructed, which made up for the blindness of parameter selection during the training process and improved the prediction accuracy of the regression prediction model. The Latin hypercube sampling method was used to initialize the RBMO population to improve the uniformity of population initialization. A spiral search strategy was introduced in the RBMO foraging phase to improve the search efficiency of the Red-billed Blue Magpie. A weight factor was introduced in the RBMO attack phase to optimize the Red-billed Blue Magpie's position update to avoid falling into local optimality. DETAILED DESCRIPTION

[0031] The present invention is described in detail below: The present invention provides a method for evaluating the condition of a vehicle gearbox, comprising the following steps:

[0032] S1. Collect the physical and chemical properties data of the oil in the gearbox.

[0033] S2. Preprocess the collected data through independent component analysis (ICA) to separate the nonlinear characteristics between the physical and chemical information of the oil, remove noise in the data, improve data quality, and screen out data sets that can be used as model input; divide the model input data set into a test data set and a training data set.

[0034] Specifically, the method includes the following steps:

[0035] S201. Centering and whitening data: This step aims to simplify the separation process of independent components. The data is preprocessed by centering and whitening so that it has zero mean and unit variance, and the components are independent of each other.

[0036] Centering data: The mean of each variable (data dimension) is subtracted so that the mean of the processed data is zero. The formula is:

[0037] (1),

[0038] in is the data matrix after centralization; μ is the mean of the original data X.

[0039] Whitening data: The data is transformed into a new space so that its components are independent of each other and have a variance of 1.

[0040] The formula is: (2),

[0041] where X white is the whitened data matrix; E and D are the eigenvector and eigenvalue matrices of the data covariance matrix, respectively.

[0042] S202, Optimizing the Objective and Applying the ICA Algorithm: Recovering the statistically independent source signals from the whitened data by optimizing an objective function that quantifies the statistical independence of the components.

[0043] The most common approach to optimizing the objective is to maximize non-Gaussianity, because according to the central limit theorem, the stronger the non-Gaussianity, the higher the independence; commonly used non-Gaussianity measures are fourth-order cumulants (kurtosis) or negentropy.

[0044] Choose a nonlinear function to approximate the non-Gaussianity of the independent components;

[0045] Iteratively update the weight matrix W:

[0046] ,

[0047] Where W is the weight matrix; g is a nonlinear function; g′ is the derivative of g; {} indicates expectation; W (next) is the unmixing matrix updated during the ICA iteration process.

[0048] S203, normalization and rotation:

[0049] The weight matrix W is normalized and possibly orthogonally rotated to further improve source independence.

[0050] Normalization formula:

[0051] ,

[0052] Ensure the stability and convergence of the weight matrix;

[0053] Finally, the output of the ICA model is:

[0054] ,

[0055] S represents the independent components in the original data, where W T is the unmixing matrix learned from the data.

[0056] Through the above steps, ICA is able to extract meaningful signals from complex multivariate data, which are as statistically independent as possible.

[0057] S3. Improvements to the Red-billed Blue Magpie Optimization Algorithm (RBMO) are made, including initializing the RBMO population through the Latin hypercube sampling method to improve the uniformity of population initialization; introducing a spiral search strategy in the foraging phase of RBMO to improve the search efficiency of the Red-billed Blue Magpie; and introducing a weight factor in the attack phase of RBMO to optimize the position update of the Red-billed Blue Magpie, thereby obtaining an improved Red-billed Blue Magpie Optimization Algorithm (IRBMO).

[0058] Specifically, in step S3, the RBMO population is initialized using the Latin hypercube sampling method. Since Latin hypercube sampling can generate sample points in a multidimensional space, it is an improved stratified sampling technique that ensures uniform and representative coverage of the entire range of each parameter. This helps improve the global search capability of the Red-billed Blue Magpie algorithm. The formula is:

[0059] (6);

[0060] (7),

[0061] Where: a i and b i For an interval [a i , b i ] at both ends; is to divide the interval [a i , b i ] is divided into the length of the i-th subinterval in n subintervals; k is a random number from 1 to n; x ij is a point selected from the kth subinterval; rand(0,1) is a random number uniformly distributed between [0,1].

[0062] During the RBMO foraging phase, that is, when looking for food, red-billed blue magpies usually act in small groups (2 to 5) or flocks (more than 10) to improve search efficiency.

[0063] Set the random number Rand1. When 0≤Rand1<0.5, the red-billed blue magpie population forms multiple small groups to search. The formula is: , 0≤Rand1<0.5; (8),

[0064] When 0≤Rand1<0.5, the red-billed blue magpie population searches in groups, and the formula is:

[0065] , 0.5≤Rand1≤1; (9),

[0066] Where, t represents the current number of iterations; X i (t+1) represents the position of the i-th individual in the t+1 iteration, p is a random number between [2,5], which represents the number of red-billed blue magpies in a small group, q is a random number between [10,n], which represents the number of red-billed blue magpies in a group, X m represents the randomly selected mth individual, X i represents the i-th individual, X rs (t) represents a randomly selected individual in the current iteration.

[0067] The spiral search strategy can effectively expand the search space, especially in complex or multi-peak optimization problems, helping the algorithm explore multiple potential good areas, thereby increasing the possibility of finding the global optimal solution. The spiral search strategy is introduced during the RBMO foraging phase to optimize the position update of the red-billed blue magpie. The update formulas for the red-billed blue magpie searching in small groups and in large groups are:

[0068] 0≤Rand1<0.5;(10)

[0069] 0.5≤Rand1≤1; (11),

[0070] Where, It is an exponential function used to adjust the expansion or contraction of the spiral, and c and l control the tightness and periodicity of the spiral respectively.

[0071] In the RBMO attack phase, the random number Rand n is set. When 0≤Rand n<0.5, the red-billed blue magpie acts in small groups, and its main target is usually small prey or plants. The corresponding mathematical model is as follows:

[0072] , 0≤Rand n<0.5; (12).

[0073] When 0.5<Rand n≤1, red-billed blue magpies act in groups and can jointly target large prey such as large insects or small vertebrates. The corresponding mathematical model is as follows:

[0074] 0.5≤Rand n≤1; (13),

[0075] ; (14),

[0076] Where, X food(t) Represents the location of the food; Rand n represents the random number used to generate the standard normal distribution; T is the maximum number of iterations; CF is a dynamic variable that changes with the number of iterations.

[0077] At this time, a weight factor is introduced to optimize the position update of the red-billed blue magpie, so that each individual position can communicate with other individuals in the local search space when moving towards the global optimal position, making the local search ability of the algorithm stronger. This embodiment uses a sinusoidal weight factor to control the influence of the prey target on the position update of the red-billed blue magpie. The formula is:

[0078] 0≤Rand n<0.5; (15),

[0079] 0.5<Rand n≤1; (16),

[0080] ; (17),

[0081] in, is the weight factor.

[0082] Finally, the global optimal value is determined: In addition to foraging and attacking food, the red-billed blue magpie also stores excess food. This process retains the solution information, making it easier for individuals to find the global optimal value. The formula is:

[0083] ; (18),

[0084] in and They represent the fitness values of the i-th red-billed blue magpie before and after the position update.

[0085] S4. The improved red-billed blue magpie optimization algorithm IRBMO is used to optimize the key parameters of the single-class support vector machine OCSVM, including kernel parameters, regularization parameters, and slack variables, to construct the IRBMO-OCSVM fault diagnosis model, and the fault diagnosis model is trained using the training data set in S2.

[0086] The parameter optimization process is as follows:

[0087] (1) Initialization parameters: kernel parameters, regularization parameters, and slack variables.

[0088] (2) Initialize the population according to formula (6), calculate the fitness value of the individuals in the population, and determine the optimal position.

[0089] (3) Perform foraging behavior. If 0≤Rand1<0.5, update the individual's location information according to formula (10); if 0.5≤Rand1≤1, update the individual's location information according to formula (11).

[0090] (4) Conduct aggressive behavior, if 0≤Rand n<0.5 , then update the individual position information according to formula (15) and conduct small group attack behavior; if 0.5≤Rand n<1 , then update the individual information according to formula (16) and perform the attack behavior in a group manner.

[0091] (5) Compare the fitness values before and after the update. If Then update the optimal individual position; otherwise, do not update the individual position.

[0092] (6) Determine whether the algorithm meets the termination condition, that is, whether the maximum number of iterations has been reached. If the maximum number of iterations has been reached, output the optimal parameters; if the maximum number of iterations has not been reached, repeat (3)-(5).

[0093] (7) Input the output optimal parameter combination into OCSVM.

[0094] S5. Input the test data set in S2 into the trained IRBMO–OCSVM model, perform fault diagnosis on the gearbox, and output the prediction results.

[0095] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limiting the present invention. Any changes, modifications, substitutions and variations of the above embodiments by a person skilled in the art fall within the scope of the present invention.

[0096] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for evaluating the condition of a vehicle gearbox, characterized in that: The following steps are involved: S1. Collect the physical and chemical properties data of the oil in the gearbox; S2. Preprocess the collected data using independent component analysis (ICA) to select a dataset that can be used as model input; divide the model input dataset into a test dataset and a training dataset; S3. Improvements to the Red-billed Blue Magpie Optimization Algorithm (RBMO), including initializing the RBMO population through Latin hypercube sampling to improve population initialization uniformity; introducing a spiral search strategy during the RBMO foraging phase to improve the search efficiency of the Red-billed Blue Magpie; and introducing a weight factor during the RBMO attack phase to optimize the position update of the Red-billed Blue Magpie, resulting in an improved Red-billed Blue Magpie Optimization Algorithm (IRBMO). Specifically, the RBMO population is initialized by the Latin hypercube sampling method, and the formula is: ； ； Where: a i and b i For an interval [a i , b i ] at both ends; is to divide the interval [a i , b i ] is divided into the length of the i-th subinterval in n subintervals; k is a random number from 1 to n; x ij is a point selected from the kth subinterval; rand(0,1) is a random number uniformly distributed between [0,1]; The spiral search strategy is introduced in the RBMO foraging stage to optimize the position update of the red-billed blue magpie: 0≤Rand1<0.5; 0.5≤Rand1≤1; Where, t is the current iteration number; X i (t+1) represents the position of the i-th individual in the t+1 iteration, p is a random number between [2,5], which represents the number of red-billed blue magpies in a small group, q is a random number between [10,n], which represents the number of red-billed blue magpies in a group, X m represents the randomly selected mth individual, X i represents the i-th individual, X rs (t) represents the randomly selected individual in the current iteration; Rand1 is a random number; It is an exponential function used to adjust the expansion or contraction of the spiral, and c and l control the tightness and periodicity of the spiral respectively; In the RBMO attack phase, the weight factor is introduced to optimize the position update of the red-billed blue magpie: , 0≤Rand n<0.5; , 0.5≤Rand n≤1; ； ； Where, X food(t) Indicates the location of the food; Rand n represents the random number used to generate the standard normal distribution; T is the maximum number of iterations; CF is a dynamic variable that changes with the number of iterations; is the weight factor; Finally, the global optimal value is determined as follows: ； in and Respectively represent the fitness values of the i-th red-billed blue magpie before and after the position update; S4, using the improved red-billed blue magpie optimization algorithm IRBMO to optimize the key parameters of the single-class support vector machine OCSVM, constructing the IRBMO-OCSVM fault diagnosis model, and training the fault diagnosis model using the training data set in S2; S5. Input the test data set in S2 into the trained IRBMO–OCSVM model, perform fault diagnosis on the gearbox, and output the prediction results.

2. The method for evaluating the condition of a vehicle gearbox according to claim 1, wherein: The improved red-billed blue magpie optimization algorithm IRBMO in step S4 optimizes key parameters of the single-class support vector machine OCSVM, and the optimized parameters include kernel parameters, regularization parameters, and slack variables.

Images