Gearbox fault detection method, device, equipment, medium and program
By constructing the training sample set and sparse representation classification method, the limitations of sensor monitoring and vibration analysis in vehicle gearbox fault detection are solved, timely and accurate identification of fault categories is achieved, and the flexibility and accuracy of detection are enhanced.
Patent Information
- Application Number
- CN202410859673.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-28
- Publication Date
- 2024-11-12
AI Technical Summary
The existing technology cannot determine the fault category of automobile gearboxes in a timely, accurate and flexible manner. Sensor monitoring is limited to the sensor coverage, vibration analysis requires professional equipment and is susceptible to environmental noise interference, and fault mode recognition is not flexible enough for new fault modes.
Based on gearbox vibration signals of different fault categories, a training sample set is constructed, a sparse representation classification is used for the initial dictionary matrix, a discriminant reconstruction constraint term and discriminant constraint, an objective function is updated to update the dictionary matrix and coefficient matrix, and feature expansion is performed to improve the dictionary. The feature representation ability of the current fault category is determined.
It realizes timely and accurate detection of vehicle gearbox faults, enhances the discriminant nature of the dictionary and signal representation classification performance, adapts to the translational unchanging characteristics of the signal, and improves the flexibility and accuracy of fault detection.
Smart Images

Figure CN118940082A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of fault detection, and in particular to a gearbox fault detection method, device, equipment, medium and program. Background Art
[0002] The automotive gearbox is one of the core components of the automotive transmission system, and its normal operation is very important to vehicle performance and safety. Gearbox failure may lead to problems such as reduced vehicle performance, unstable speed, and increased noise, so timely and accurate fault diagnosis of the gearbox is crucial.
[0003] Several commonly used methods for detecting automotive gearbox failures include sensor monitoring, vibration analysis, and fault pattern recognition. Sensor monitoring has the advantages of strong real-time performance and direct access to key parameter data of gearboxes, which facilitates the rapid detection of abnormalities. However, it can only monitor the parameters covered by the sensor, and some hidden faults may not be discovered in time. In addition, it is necessary to accurately set the warning value, otherwise false alarms or missed alarms may occur. Vibration analysis can detect changes in gearbox vibration signals and is sensitive to faults of components such as gears and bearings, but it requires professional equipment and technical support, and has high requirements for the processing and analysis of vibration signals, and is easily disturbed by environmental noise. Fault pattern recognition methods are data-driven and can identify different types of fault modes. They have a certain degree of intelligence and accuracy. However, most methods require a large amount of historical data for training and may not be flexible enough for new fault modes. Summary of the invention
[0004] In view of this, the present invention provides a gearbox fault detection method, device, equipment, medium and program to solve the problem that traditional technology cannot determine the gearbox fault category in a timely, accurate and flexible manner.
[0005] In a first aspect, the present invention provides a gearbox fault detection method, the method comprising:
[0006] Based on the gearbox vibration signals of different fault categories, a training sample set is constructed;
[0007] Using the initial dictionary matrix to perform sparse representation classification on the training sample set, an initial coefficient matrix corresponding to the initial dictionary matrix is obtained;
[0008] Based on the initial dictionary matrix, the initial coefficient matrix and the training sample set, a discriminant reconstruction constraint item of the initial dictionary matrix and a discriminant constraint of the initial coefficient matrix are constructed, and an objective function is constructed based on the discriminant reconstruction constraint item and the discriminant constraint;
[0009] Based on the objective function and the training sample set, the initial dictionary matrix and the initial coefficient matrix are updated to obtain the ideal dictionary matrix and the ideal coefficient matrix;
[0010] The target dictionary matrix is obtained by feature expansion of the ideal dictionary matrix. The current fault category corresponding to the current vibration signal of the gearbox is determined using the target dictionary matrix and the ideal coefficient matrix.
[0011] The present invention constructs a rich training sample set based on the vibration signals of the gearbox of different fault categories, uses the initial dictionary matrix to perform sparse representation classification on the training sample set, obtains the initial coefficient matrix, and constructs the discriminant reconstruction constraint items of the initial dictionary matrix, the discriminant constraints of the initial coefficient matrix, and the objective function, so as to facilitate training based on the objective function and the training sample set to obtain the ideal dictionary matrix and the ideal coefficient matrix. Finally, the feature expansion of the ideal dictionary matrix is performed to obtain the target dictionary matrix, the feature representation capability of the dictionary is improved, and the current fault category corresponding to the current vibration signal of the gearbox is determined using the target dictionary matrix and the ideal coefficient matrix. The present invention takes into account the translation invariant characteristics of the signal and can detect faults in the gearbox in a timely and accurate manner.
[0012] In an optional implementation, based on the initial dictionary matrix, the initial coefficient matrix and the training sample set, the discriminant reconstruction constraint item of the initial dictionary matrix and the discriminant constraint of the initial coefficient matrix are constructed, including:
[0013] Based on the preset constraint conditions, the discriminant reconstruction constraint items corresponding to the initial dictionary matrix are obtained; the preset constraint conditions are used to perform sparse representation classification on the training sample set through the initial dictionary matrix;
[0014] The transposed matrix of the initial coefficient matrix is determined, and the discriminant constraints of the initial coefficient matrix are obtained according to the transposed matrix and the initial coefficient matrix.
[0015] Therefore, by constructing discriminative reconstruction constraint terms, the discriminability of the initial dictionary matrix is enhanced, and by imposing discriminative constraints on the initial coefficient matrix, it is beneficial to enhance the discriminability of dictionary atoms, thereby improving the signal representation classification performance of the dictionary.
[0016] In an optional implementation, the discriminant constraint of the initial coefficient matrix is obtained according to the transposed matrix and the initial coefficient matrix, including:
[0017] A first discriminant constraint is obtained based on a first-class inter-dispersion matrix and a first-class intra-dispersion matrix corresponding to the initial coefficient matrix;
[0018] Based on the second type of between-dispersion matrix and the second type of within-dispersion matrix corresponding to the transposed matrix, a second discriminant constraint is obtained;
[0019] Based on the first discriminant constraint and the second discriminant constraint, the discriminant constraint of the initial coefficient matrix is obtained.
[0020] Therefore, by imposing discriminative constraints on the initial coefficient matrix, it is beneficial to enhance the discriminativeness of dictionary atoms, thereby improving the signal representation classification performance of the dictionary.
[0021] In an optional implementation, the parameters of the objective function include an initial dictionary matrix and an initial coefficient matrix; training is performed based on the objective function and the training sample set, and the initial dictionary matrix and the initial coefficient matrix are updated to obtain an ideal dictionary matrix and an ideal coefficient matrix, including:
[0022] The value of the initial coefficient matrix in the objective function is fixed, and the initial dictionary matrix is iteratively updated using the training sample set to obtain an ideal dictionary matrix corresponding to the minimum value of the objective function;
[0023] The value of the initial dictionary matrix in the objective function is fixed, and the initial coefficient matrix is iteratively updated using the training sample set to obtain the ideal coefficient matrix corresponding to the minimum value of the objective function.
[0024] Thus, by fixing the value of the initial coefficient matrix or the initial dictionary matrix in the objective function and iterating in a loop, the model training is completed, and an ideal dictionary matrix and an ideal coefficient matrix that meet the objective function are obtained, so as to facilitate the subsequent fault detection of the vibration signal of the gearbox.
[0025] In an optional implementation, feature expansion is performed on the ideal dictionary matrix to obtain a target dictionary matrix, including:
[0026] Shift and intercept multiple dictionary column vectors in the ideal dictionary matrix respectively to obtain multiple dictionary shift features corresponding to each dictionary column vector;
[0027] For each dictionary column vector, multiple dictionary shift features corresponding to the dictionary column vector are horizontally merged to obtain a target sub-dictionary corresponding to the dictionary column vector;
[0028] Based on the target sub-dictionaries corresponding to the multiple dictionary column vectors, a target dictionary matrix is obtained.
[0029] Thus, atomic feature expansion is performed through shifting and intercepting, and the feature expansion of the ideal dictionary matrix is performed to obtain the target dictionary matrix, so that the same features are located in multiple phases, thereby improving the feature representation ability of the dictionary.
[0030] In an optional implementation, each target sub-dictionary in the target dictionary matrix corresponds to a different fault category, and the ideal coefficient matrix includes sub-coefficient matrices corresponding to different fault categories; using the target dictionary matrix and the ideal coefficient matrix, determining the current fault category corresponding to the current vibration signal of the gearbox includes:
[0031] Based on the ideal coefficient matrix, the average vector of each sub-coefficient matrix is calculated;
[0032] Based on the target sub-dictionary and the average vector, the reconstruction error of the current vibration signal of the gearbox on different target sub-dictionaries is calculated, and the minimum reconstruction error is determined;
[0033] According to the fault category of the target sub-dictionary corresponding to the minimum value of the reconstruction error, the current fault category corresponding to the current vibration signal is obtained.
[0034] Therefore, the classification accuracy is improved by using the average vector of each sub-coefficient matrix in the ideal coefficient matrix and the corresponding reconstruction error as classification information during fault detection.
[0035] In an optional implementation, a training sample set is constructed based on gearbox vibration signals of different fault categories, including:
[0036] Segment the gearbox vibration signals of different fault categories to obtain the gearbox vibration short signal corresponding to each fault category;
[0037] For each fault category, feature expansion is performed on the gearbox vibration short signal corresponding to the fault category to obtain a training sample set.
[0038] Therefore, by segmenting and expanding the features of the gearbox vibration signal at the data level, a training sample set is obtained, which is beneficial to enhancing the dictionary feature representation capability and improving the signal reconstruction performance.
[0039] In a second aspect, the present invention provides a gearbox fault detection device, the device comprising:
[0040] An acquisition module is used to construct a training sample set based on the vibration signals of the gearbox of different fault categories;
[0041] The first processing module is used to perform sparse representation classification on the training sample set using the initial dictionary matrix to obtain an initial coefficient matrix corresponding to the initial dictionary matrix;
[0042] The second processing module is used to construct the discriminant reconstruction constraint item of the initial dictionary matrix and the discriminant constraint of the initial coefficient matrix based on the initial dictionary matrix, the initial coefficient matrix and the training sample set, and construct the objective function based on the discriminant reconstruction constraint item and the discriminant constraint;
[0043] A training module is used to perform training based on an objective function and a training sample set, update an initial dictionary matrix and an initial coefficient matrix, and obtain an ideal dictionary matrix and an ideal coefficient matrix;
[0044] The determination module is used to perform feature expansion on the ideal dictionary matrix to obtain a target dictionary matrix, and to determine the current fault category corresponding to the current vibration signal of the gearbox using the target dictionary matrix and the ideal coefficient matrix.
[0045] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the gearbox fault detection method of the first aspect or any corresponding embodiment thereof by executing the computer instructions.
[0046] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the gearbox fault detection method of the first aspect or any corresponding embodiment thereof.
[0047] In a fifth aspect, the present invention provides a computer program product, comprising computer instructions for causing a computer to execute the gearbox fault detection method of the first aspect or any corresponding embodiment thereof.
[0048] The beneficial effects of the present invention are:
[0049] Based on the vibration signals of the gearbox of different fault categories, a rich training sample set is constructed, and the training sample set is sparsely represented and classified using the initial dictionary matrix to obtain the initial coefficient matrix, and the discriminant reconstruction constraint items of the initial dictionary matrix, the discriminant constraints of the initial coefficient matrix, and the objective function are constructed to facilitate training based on the objective function and the training sample set to obtain the ideal dictionary matrix and the ideal coefficient matrix. Finally, the ideal dictionary matrix is feature expanded to obtain the target dictionary matrix to improve the feature representation capability of the dictionary, and the target dictionary matrix and the ideal coefficient matrix are used to determine the current fault category corresponding to the current vibration signal of the gearbox. The present invention takes into account the translation invariant characteristics of the signal and can detect gearbox faults in a timely and accurate manner. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0051] Figure 1 is a schematic flow chart of a gearbox fault detection method according to an embodiment of the present invention;
[0052] Figure 2 is a schematic diagram of a data-level feature extension according to an embodiment of the present invention;
[0053] Figure 3 is a schematic diagram of a signal sparse representation according to an embodiment of the present invention;
[0054] Figure 4 is a basic principle diagram of coefficient representation classification according to an embodiment of the present invention;
[0055] Figure 5 is a flow chart of another gearbox fault detection method according to an embodiment of the present invention;
[0056] Fig. 6A is a schematic diagram of a principle of a preset constraint condition according to an embodiment of the present invention;
[0057] Figure 6B is a schematic diagram of another principle of a preset constraint condition according to an embodiment of the present invention;
[0058] Figure 6C is a schematic diagram of another principle of a preset constraint condition according to an embodiment of the present invention;
[0059] Figure 7 is a schematic diagram of atomic level feature expansion according to an embodiment of the present invention;
[0060] Figure 8 is a schematic diagram of a classification effect according to an embodiment of the present invention;
[0061] Fig. 9 is a structural block diagram of a gearbox fault detection device according to an embodiment of the present invention;
[0062] Fig.10 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION
[0063] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.
[0064] The automotive gearbox is one of the core components of the automotive transmission system, and its normal operation is very important to vehicle performance and safety. Gearbox failure may lead to problems such as reduced vehicle performance, unstable speed, and increased noise, so timely and accurate fault diagnosis of the gearbox is crucial.
[0065] Several commonly used methods for detecting automotive gearbox failures include sensor monitoring, vibration analysis, and fault pattern recognition. Sensor monitoring has the advantages of strong real-time performance and direct access to key parameter data of gearboxes, which facilitates the rapid detection of abnormalities. However, it can only monitor the parameters covered by the sensor, and some hidden faults may not be discovered in time. In addition, it is necessary to accurately set the warning value, otherwise false alarms or missed alarms may occur. Vibration analysis can detect changes in gearbox vibration signals and is sensitive to faults of components such as gears and bearings, but it requires professional equipment and technical support, and has high requirements for the processing and analysis of vibration signals, and is easily disturbed by environmental noise. Fault pattern recognition methods are data-driven and can identify different types of fault modes. They have a certain degree of intelligence and accuracy. However, most methods require a large amount of historical data for training and may not be flexible enough for new fault modes.
[0066] The embodiment of the present invention proposes a gearbox fault detection scheme, which realizes gearbox fault detection and confirmation based on sparse representation classification method. The sparse representation classification method has excellent feature representation and extraction capabilities, which makes the representation of signals more concise. It has shown excellent results in many fields such as mechanical fault diagnosis and is one of the research hotspots in recent years. Sparse representation uses sparse information of feature components to construct an atomic dictionary, and reconstructs feature information with as few atoms as possible. It has good interpretability and fewer model parameters, and at the same time, the demand for training sample size is also smaller. Considering that the transient impact components generated by local defects of gears and bearings in the gearbox are often sparse, and randomly distributed environmental noise does not have sparse characteristics, sparse representation technology can be used to achieve effective diagnosis of automotive gearbox faults.
[0067] According to an embodiment of the present invention, a gearbox fault detection method embodiment is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.
[0068] In this embodiment, a gearbox fault detection method is provided, which can be used for a computer device or electronic device such as a computer, a mobile phone, a tablet computer, etc. for gearbox fault detection. Figure 1 is a flow chart of a gearbox fault detection method according to an embodiment of the present invention. Figure 1 As shown, the process includes the following steps:
[0069] Step S101: construct a training sample set based on gearbox vibration signals of different fault categories.
[0070] In some optional implementations, the above step S101 includes:
[0071] Step a1, segmenting the gearbox vibration signals of different fault categories to obtain the gearbox vibration short signal corresponding to each fault category.
[0072] Specifically, the failure of the automobile gearbox mainly occurs in the bearing and gear parts, and the bearings and gears are located inside the gearbox. It is difficult to monitor the vibration signals directly emitted by the rotation of the bearings and gears. On the basis of ensuring the integrity of the gearbox, it is only possible to monitor the vibration signals of the bearings and gears transmitted to the gearbox housing. In order to ensure that the collected signals contain more useful information, the sensors should be arranged at the position where the transmission path of the bearings and gears is the shortest. After the vibration sensor is arranged, start the engine to make the gearbox work. After the sensor collects the data, it saves the data to the host computer.
[0073] Generally speaking, the gearbox is allowed to work for a continuous period of time under different fault conditions, and the time series "long" signal corresponding to the gearbox vibration signal is collected. Next, the data needs to be segmented to obtain multiple "short" signals, that is, the short gearbox vibration signal corresponding to each fault category, to construct the initial data set.
[0074] Step a2: for each fault category, feature expansion is performed on the gearbox vibration short signal corresponding to the fault category to obtain a training sample set.
[0075] Specifically, Figure 2 As shown, by means of shifting and intercepting, the initial data set obtained in step a1 is feature expanded at the data level to construct a training sample set.
[0076] In common discriminative dictionary learning models, there are few methods that use inter-class and intra-class information as discriminative constraints. At the same time, most methods ignore the translation invariance of the signal. The embodiment of the present invention performs feature expansion based on translation invariance before dictionary learning to enhance the dictionary feature representation capability and improve signal reconstruction performance. Feature expansion at the data level can, on the one hand, achieve data enhancement to expand the amount of training data and provide data support for fault diagnosis with fewer data samples; on the other hand, it can construct a training sample set with a certain translation invariance, thereby promoting the dictionary to learn translation invariant atoms.
[0077] Therefore, by segmenting and expanding the features of the gearbox vibration signal at the data level, a training sample set is obtained, which is beneficial to enhancing the dictionary feature representation capability and improving the signal reconstruction performance.
[0078] Step S102, using the initial dictionary matrix to perform sparse representation classification on the training sample set to obtain an initial coefficient matrix corresponding to the initial dictionary matrix.
[0079] It should be understood that in the sparse representation model, first, a dictionary matrix needs to be constructed. Each column of this matrix represents a dictionary atom. The number of rows of the matrix is the length of the dictionary atom, and the atom length is generally equal to the length of the signal to be represented. If the number of rows of the dictionary is less than the number of columns, it can be called an overcomplete dictionary. Then, the signal to be represented can be represented or sufficiently approximated by a linear combination of multiple dictionary atoms, and the corresponding coefficients are the sparse representation or sparse coding coefficients. If there are fewer non-zero values in the sparse representation coefficients, it means that fewer atoms can be used to reconstruct the signal, achieving a more concise representation of the signal.
[0080] Given the signal y ∈ R s×1 as an s-dimensional column vector, and the overcomplete dictionary D = [d1, d2,..., d k ∈ R s×k (s < k), where the s-dimensional column vector d i ∈ R s×1 is a dictionary atom, and the sparse representation coefficient vector x = [x1, x2,..., x k ∈ R k×1 , then the signal y can be represented as:
[0081]
[0082] Exemplarily, the schematic diagram of the sparse representation of the signal y is as Figure 3 shown.
[0083] Due to the redundancy of the dictionary atoms, the same signal can be represented or sufficiently approximated by a linear combination of multiple groups of atoms. The sparse representation aims to find the most concise representation method, requiring the combination coefficients to be as sparse as possible, that is, the norm of the coefficient vector x is as small as possible. Therefore, the sparse representation model of the signal can be expressed as:
[0084]
[0085] In the formula, ε is the upper limit of the reconstruction error of the signal, will ensure that the obtained representation coefficients are as sparse as possible, while ensures that the signal reconstruction error is within the limited range.
[0086] In addition, taking the sparsity of the representation coefficients as a constraint term to solve the minimum reconstruction error can also obtain the sparse representation coefficients. At this time, the sparse representation model is converted to:
[0087]
[0088] In the formula, T0 is the given sparsity. According to the Lagrange multiplier method, the sparse representation model shown in formula (3) can be rewritten as the following unconstrained optimization problem:
[0089]
[0090] Where λ>0 is a regularization parameter used to control the balance between reconstruction error and sparsity.
[0091] The main idea of the sparse representation classification method based on reconstruction error is that when the number of training samples corresponding to a specific class sample is sufficient, the sample can be linearly represented or sufficiently approximated by these training samples, and the contribution of training samples of other categories to the reconstruction of the specific class sample is very small or 0, that is, the linear relationship between samples of the same category is relatively strong, and the linear relationship between samples of different categories is relatively weak. Therefore, the result of the sparse representation of samples is often a linear combination of atoms of the same category in the dictionary, thereby converting a classification task into a sparse representation solution problem. The principle is as follows Figure 4 shown.
[0092] Specifically, a dictionary matrix can be initialized to obtain an initial dictionary matrix, and based on the above method, the initial dictionary matrix is used to perform sparse representation classification on the training sample set to obtain an initial coefficient matrix (essentially a sparse representation coefficient matrix) corresponding to the initial dictionary matrix.
[0093] Step S103, based on the initial dictionary matrix, the initial coefficient matrix and the training sample set, construct the discriminant reconstruction constraint items of the initial dictionary matrix and the discriminant constraints of the initial coefficient matrix, and construct the objective function based on the discriminant reconstruction constraint items and the discriminant constraints.
[0094] Specifically, in order to obtain a sparser solution, the classification algorithm based on dictionary learning needs to enhance the discriminability between dictionary atoms of each category, that is, it is expected that samples of a specific class are reconstructed by dictionary atoms of the corresponding class, the sparse representation coefficients of samples of the same class are similar, and the sparse representation coefficients of samples of different classes are quite different. This goal is consistent with the nature of the Fisher discriminant criterion. Therefore, applying Fisher discriminant constraints on sparse representation coefficients in dictionary learning will help enhance the discriminability of dictionary atoms, thereby improving the signal representation classification performance of the dictionary.
[0095] Step S104 , training is performed based on the objective function and the training sample set, and the initial dictionary matrix and the initial coefficient matrix are updated to obtain an ideal dictionary matrix and an ideal coefficient matrix.
[0096] Specifically, the training goal is to minimize the objective function, and the initial dictionary matrix and the initial coefficient matrix are iteratively updated using the training sample set to obtain an ideal dictionary matrix and an ideal coefficient matrix that meet the requirements of the objective function.
[0097] Step S105 , feature expansion is performed on the ideal dictionary matrix to obtain a target dictionary matrix, and the target dictionary matrix and the ideal coefficient matrix are used to determine the current fault category corresponding to the current vibration signal of the gearbox.
[0098] Specifically, the result of sparse representation classification should not be sensitive to the translation of the input signal in the time domain, and if the phase of the learned atom is consistent with the training sample, it cannot represent samples with different phases well. In order to achieve accurate classification, the dictionary atoms need to represent the corresponding class samples with multiple possible phases well. The embodiment of the present invention implements feature expansion through the method of shift interception at the sample data level and the dictionary atom level, and uses the signal translation invariance to improve the feature representation ability of the dictionary. After the dictionary learning is completed, each category of atoms completes the feature learning of the corresponding category of faults. At this time, atomic feature expansion is performed through shift interception, and the feature expansion of the ideal dictionary matrix is performed to obtain the target dictionary matrix, so that the same features are located in multiple phases, thereby improving the feature representation ability of the dictionary.
[0099] The gearbox fault detection method provided in this embodiment constructs a rich training sample set based on the gearbox vibration signal of different fault categories, uses the initial dictionary matrix to perform sparse representation classification on the training sample set to obtain the initial coefficient matrix, and constructs the discriminant reconstruction constraint items of the initial dictionary matrix, the discriminant constraints of the initial coefficient matrix, and the objective function, so as to facilitate training based on the objective function and the training sample set to obtain the ideal dictionary matrix and the ideal coefficient matrix. Finally, the ideal dictionary matrix is feature expanded to obtain the target dictionary matrix to improve the feature representation ability of the dictionary, and the target dictionary matrix and the ideal coefficient matrix are used to determine the current fault category corresponding to the current vibration signal of the gearbox. The present invention takes into account the translation invariant characteristics of the signal and can detect gearbox faults in a timely and accurate manner.
[0100] In this embodiment, a gearbox fault detection method is provided, which can be used for a computer device or electronic device such as a computer, a mobile phone, a tablet computer, etc. for gearbox fault detection. Figure 5 is a flow chart of a gearbox fault detection method according to an embodiment of the present invention. Figure 5 As shown, the process includes the following steps:
[0101] Step S501: construct a training sample set based on the gearbox vibration signals of different fault categories. Figure 1 Step S101 of the illustrated embodiment will not be described in detail here.
[0102] Step S502: Use the initial dictionary matrix to perform sparse representation classification on the training sample set to obtain an initial coefficient matrix corresponding to the initial dictionary matrix. Figure 1 Step S102 of the illustrated embodiment will not be described in detail here.
[0103] Step S503, based on the initial dictionary matrix, the initial coefficient matrix and the training sample set, construct the discriminant reconstruction constraint items of the initial dictionary matrix and the discriminant constraints of the initial coefficient matrix, and construct the objective function based on the discriminant reconstruction constraint items and the discriminant constraints.
[0104] Specifically, the above step S503 includes:
[0105] Step S5031, based on preset constraints, obtain the discriminant reconstruction constraint items corresponding to the initial dictionary matrix; the preset constraints are used to perform sparse representation classification on the training sample set through the initial dictionary matrix.
[0106] Specifically, in the traditional dictionary learning model, minimizing the reconstruction error of D and X for the entire sample set Y is often used as a constraint, that is, Although a smaller reconstruction error can sometimes be obtained, the dictionary atoms selected to reconstruct the signal often come from multiple categories, which indicates that the discriminability between the dictionary atoms of different categories is weak, which is not conducive to obtaining a sparser solution. In order to make the different atoms more discriminative, for the i-th sub-training sample set Y i , we expect to meet the following preset constraints:
[0107] 1) The entire dictionary D can represent Y well i ;
[0108] 2) The i-th sub-dictionary D i Can well represent Y i ;
[0109] 3) The j-th sub-dictionary D j Can't express Y well i .
[0110] Therefore, we can get the following discriminant reconstruction constraint r(Y i ,D,X i ):
[0111]
[0112] It is necessary to explain the relevant symbols of the present invention that there are C types of samples, and the number of training samples of the i-th type is n. i , sample y∈R s×1 is a column vector of length s, then the corresponding training sample set of the i-th category is The entire training sample set is Y = [Y1, Y2, ..., Y C ]∈R s×n ,in, is the total number of samples. Corresponding to the i-th sub-dictionary Among them, the dictionary atom d∈R s×1is a column vector of length s, k i is the number of atoms in the ith sub-dictionary, and the entire dictionary set D = [D1, D2, ..., D C ]∈R s×k ,in, is the total number of dictionary atoms. X=[X1,X2,…,X C ]∈R k×n is the encoding coefficient of the training sample set Y corresponding to the dictionary set D, is the i-th sub-training sample set Y i The encoding coefficients corresponding to the dictionary set D, is the i-th sub-training sample set Y i Corresponding sub-dictionary D j The coding coefficient of .
[0113] For example, it can be achieved by FIG. 6A to FIG. 6C Better explanation of the discriminant reconstruction constraint r(Y i ,D,X i ). Fig. 6A correspond Indicates that dictionary D can well represent Y i , but R i Deviation Y i Farther, indicating D i Cannot express Y well i Therefore, adding the constraint Make D i Can better represent Y i ,like Figure 6B As shown, other sub-dictionaries may also better represent Y i ,like Figure 6B R i-1 , which will reduce the discriminability of the dictionary. To this end, add constraints Make other dictionaries unable to represent Y well i ,like Figure 6C Therefore, compared with the unconstrained reconstruction error term in the traditional dictionary learning model, the discriminant reconstruction constraint term r(Y i ,D,X i ) can better enhance the discriminability of the dictionary.
[0114] Step S5032, determine the transposed matrix of the initial coefficient matrix, and obtain the discriminant constraints of the initial coefficient matrix based on the transposed matrix and the initial coefficient matrix.
[0115] In some optional implementations, the above step S5032 includes:
[0116] Step b1, obtaining a first discriminant constraint based on a first inter-class dispersion matrix and a first intra-class dispersion matrix corresponding to the initial coefficient matrix.
[0117] For sample y, the dictionary D can be expressed as Among them, d ij Represents the sub-dictionary D i Ideally, the non-zero values of the sparse coefficient vector x only appear at the positions corresponding to atoms of the same category as the test sample y, that is, if y is the i-th sample, then
[0118] Specifically, the inter-class dispersion matrix S of the initial coefficient matrix (sparse representation coefficient matrix) X is B (X) and the intra-class dispersion matrix S W (X) are as follows:
[0119]
[0120] Among them, m i and m are X i and the mean column vector of X, then the Fisher discriminant constraint of X can be defined as:
[0121]
[0122] Step b2: obtaining a second discriminant constraint based on a second type of between-dispersion matrix and a second type of within-dispersion matrix corresponding to the transposed matrix.
[0123] Similarly, the transposed matrix X of the initial coefficient matrix T The Fisher discriminant constraint is:
[0124]
[0125] Among them, λ1 and λ2 are constants, and adding and The role is to solve tr(S B (X) and tr(S B (X T ) is a non-convex problem.
[0126] Step b3, obtaining the discriminant constraint of the initial coefficient matrix based on the first discriminant constraint and the second discriminant constraint.
[0127] Specifically, in order to obtain a sparser solution, the classification algorithm based on dictionary learning needs to enhance the discriminability between dictionary atoms of each category, that is, it is expected that samples of a specific class are reconstructed by dictionary atoms of the corresponding class, the sparse representation coefficients of samples of the same class are similar, and the sparse representation coefficients of samples of different classes are quite different. This goal is consistent with the nature of the Fisher discriminant criterion. Therefore, applying Fisher discriminant constraints on sparse representation coefficients in dictionary learning will help enhance the discriminability of dictionary atoms, thereby improving the signal representation classification performance of the dictionary.
[0128] In common discriminant dictionary learning models, few use inter-class and intra-class information as discriminant constraints. At the same time, most methods ignore the translation invariance of signals. The present invention proposes a Fisher discriminant constraint translation invariant dictionary learning method, which performs feature expansion based on translation invariance in the two stages before and after dictionary learning, enhances the dictionary feature representation ability, and improves signal reconstruction performance. In the dictionary learning stage, by applying Fisher discriminant constraints to the coding coefficient matrix and its transposed matrix, the discriminability of the dictionary is enhanced, and the classification diagnosis performance is improved.
[0129] Therefore, by constructing discriminative reconstruction constraint terms, the discriminability of the initial dictionary matrix is enhanced, and by imposing discriminative constraints on the initial coefficient matrix, it is beneficial to enhance the discriminability of dictionary atoms, thereby improving the signal representation classification performance of the dictionary.
[0130] Step S5033: construct an objective function based on the discriminant reconstruction constraint item and the discriminant constraint.
[0131] Specifically, in combination with the discriminant reconstruction constraint term and the Fisher discriminant constraint term shown in formulas (4), (8) and (9), the present invention proposes the following objective function:
[0132]
[0133] In the formula,
[0134]
[0135] Where X1 is the sparsity constraint, and α, β, and λ are all constants. To simplify the calculation, f(X) and f(X T ) where λ1 and λ2 are set to 1.
[0136] In the objective function of the present invention, r(Y, D, X) can make the i-th class sample be simultaneously included in the dictionary D and D i It can be well represented by other sub-dictionaries, which not only ensures the signal reconstruction ability of the dictionary, but also enhances the discriminability of the dictionary. In addition, f(X) and f(X T) makes the representation coefficients of samples of the same type more similar, while the differences between different types are greater, further enhancing the discriminability of the dictionary.
[0137] Step S504 , training is performed based on the objective function and the training sample set, and the initial dictionary matrix and the initial coefficient matrix are updated to obtain an ideal dictionary matrix and an ideal coefficient matrix.
[0138] Specifically, the parameters of the objective function mainly include an initial dictionary matrix and an initial coefficient matrix. The above step S504 includes:
[0139] Step S5041, fix the value of the initial coefficient matrix in the objective function, and iteratively update the initial dictionary matrix using the training sample set to obtain an ideal dictionary matrix corresponding to the minimum value of the objective function.
[0140] Specifically, the purpose of model training is to minimize the objective function. However, since the objective function is relatively complex and difficult to solve directly, it is necessary to fix one of D and X and perform model training through alternating iterative updates.
[0141] When X in the objective function is fixed, the problem of solving D can be transformed into:
[0142]
[0143] And r(Y,D,X) can be transformed into the following form:
[0144] r(Y,D,X)=-2tr(ED T )+tr(FD T D) (13)
[0145] In the formula, function Indicates the following transformation of the block matrix A:
[0146]
[0147] Therefore, D can be solved as follows:
[0148]
[0149] Furthermore, for the problem shown in formula (15), an online dictionary learning method can be directly used to solve it. For details, please refer to the description of related technologies, which will not be repeated here.
[0150] Step S5042, fix the value of the initial dictionary matrix in the objective function, and iteratively update the initial coefficient matrix using the training sample set to obtain an ideal coefficient matrix corresponding to the minimum value of the objective function.
[0151] Similarly, when D in the objective function is fixed, the problem of solving X can be transformed into:
[0152]
[0153] in,
[0154]
[0155] Furthermore, the problem shown in formula (16) can be solved by a fast iterative threshold shrinkage algorithm. For details, please refer to the description of the relevant technology, which will not be repeated here.
[0156] Thus, by fixing the value of the initial coefficient matrix or the initial dictionary matrix in the objective function and iterating in a loop, the model training is completed, and an ideal dictionary matrix and an ideal coefficient matrix that meet the objective function are obtained, so as to facilitate the subsequent fault detection of the vibration signal of the gearbox.
[0157] Step S505 , feature expansion is performed on the ideal dictionary matrix to obtain a target dictionary matrix, and the target dictionary matrix and the ideal coefficient matrix are used to determine the current fault category corresponding to the current vibration signal of the gearbox.
[0158] It should be noted that each target sub-dictionary in the target dictionary matrix corresponds to a different fault category, and the ideal coefficient matrix includes sub-coefficient matrices corresponding to different fault categories.
[0159] In some optional implementations, the above step S505 includes:
[0160] Step c1, shift and intercept multiple dictionary column vectors in the ideal dictionary matrix respectively, and obtain multiple dictionary shift features corresponding to each dictionary column vector.
[0161] For example, Figure 7 As shown, by shifting and truncating the original dictionary (here, the ideal dictionary matrix), multiple dictionary shift features corresponding to each dictionary column vector are obtained.
[0162] Step c2: for each dictionary column vector, multiple dictionary shift features corresponding to the dictionary column vector are horizontally merged to obtain a target sub-dictionary corresponding to the dictionary column vector.
[0163] For example, see again Figure 7 , for each dictionary column vector, a horizontal dictionary merge is performed on multiple dictionary shift features corresponding to the dictionary column vector to obtain the target sub-dictionary in the extended dictionary.
[0164] Step c3, obtaining a target dictionary matrix based on the target sub-dictionaries corresponding to the multiple dictionary column vectors.
[0165] For example, see again Figure 7 , based on the target sub-dictionaries corresponding to multiple dictionary column vectors, an extended dictionary (here referred to as the target dictionary matrix) is obtained.
[0166] Specifically, the result of sparse representation classification should not be sensitive to the translation of the input signal in the time domain. If the phase of the learned atom is consistent with the training sample, it cannot represent samples with different phases well. In order to achieve accurate classification, the dictionary atoms need to represent the corresponding class samples with multiple possible phases well. Therefore, the present invention realizes feature expansion by shifting and intercepting the method from the sample data level and the dictionary atom level, and uses the signal translation invariance to improve the feature representation ability of the dictionary. The principles are as follows: Figure 2 and Figure 7 shown.
[0167] See again Figure 2 ,Feature expansion at the data level, on the one hand, can achieve data enhancement to expand the amount of training data and provide data support for fault diagnosis with fewer data samples; on the other hand, it can construct a training data set with certain translation invariance, thereby promoting the dictionary to learn translation invariant atoms.
[0168] like Figure 7 As shown in the figure, after the dictionary learning is completed, each category of atoms completes the feature learning of the corresponding category of faults. At this time, the atomic features are expanded through shifting and intercepting, and the feature expansion of the ideal dictionary matrix is performed to obtain the target dictionary matrix, so that the same features are located in multiple phases, thereby improving the feature representation ability of the dictionary.
[0169] Step c4, based on the ideal coefficient matrix, calculate and obtain the average vector of each sub-coefficient matrix.
[0170] Specifically, based on the trained dictionary D, the test samples can be classified and identified according to different information, such as judging the corresponding category according to the reconstruction error size of the test signal on each type of sub-dictionary.
[0171] In the embodiment of the present invention, since Fisher discriminant constraints are applied to the coding coefficient matrix (initial coefficient matrix) and its transposed matrix at the same time, the various coding coefficient matrices X obtained by dictionary learning are i There will be stronger discrimination between them. i Represents X i The average vector of the test sample y t belongs to the i-th category, then y t In the sub-dictionary D i The reconstruction error on is small, and the encoding coefficient on dictionary D will also be close to m i Therefore, the reconstruction error is i All of them will be used as classification information of the method of the present invention to improve the classification accuracy.
[0172] Step c5, based on the target sub-dictionary and the average vector, calculate the reconstruction error of the current vibration signal of the gearbox on different target sub-dictionaries, and determine the minimum reconstruction error.
[0173] Specifically, the current vibration signal y of the gearbox t When performing sparse representation, in order to eliminate the interference between different sub-dictionaries and reduce the complexity of sparse representation, a local classification method can be used, namely:
[0174]
[0175] In the formula, ζ1 and ζ2 are constants. Then the classification index, i.e. the reconstruction error e i It can be defined as:
[0176]
[0177] Step c6, obtaining the current fault category corresponding to the current vibration signal according to the fault category of the target sub-dictionary corresponding to the minimum reconstruction error value.
[0178] Finally, the current vibration signal y t The current fault category is the minimum reconstruction error e i The corresponding categories are:
[0179]
[0180] Therefore, the classification accuracy is improved by using the average vector of each sub-coefficient matrix in the ideal coefficient matrix and the corresponding reconstruction error as classification information during fault detection.
[0181] The gearbox fault detection method provided in this embodiment is based on the segmentation and feature expansion of the gearbox vibration signal of different fault categories, constructs a rich training sample set, uses the initial dictionary matrix to perform sparse representation classification on the training sample set, obtains the initial coefficient matrix, and constructs the discriminant reconstruction constraint items of the initial dictionary matrix, the discriminant constraints of the initial coefficient matrix, and the objective function, which is conducive to enhancing the discriminability of dictionary atoms, thereby improving the signal representation classification performance of the dictionary, so as to facilitate training based on the objective function and the training sample set to obtain the ideal dictionary matrix and the ideal coefficient matrix. Finally, the ideal dictionary matrix is feature expanded to obtain the target dictionary matrix, improve the feature representation ability of the dictionary, and use the target dictionary matrix and the ideal coefficient matrix to determine the current fault category corresponding to the current vibration signal of the gearbox. The present invention takes into account the translation invariant characteristics of the signal and can detect gearbox faults in a timely and accurate manner.
[0182] The gearbox fault detection method of the present invention is described in detail below with reference to a specific application example.
[0183] This application example uses the data from the Case Western Reserve University bearing data set, which contains bearing vibration data under multiple working conditions (different speeds, different loads), multiple sampling positions, multiple fault levels, and multiple fault types. It is similar to the bearing faults generated in actual operation, so it is used by many scholars for bearing research. In this application example, part of the data from the drive end at 12kHz is selected to construct the data set, and the data samples of different speeds and different loads are merged. Only three fault types (inner ring fault, outer ring fault, rolling element fault) and three fault levels (0.007", 0.014", 0.021") are considered, and they are divided into 9 different bearing fault data types, plus normal bearing data for a total of 10 categories.
[0184] like Figure 8 As shown, Figure 8 The horizontal axis represents the predicted category, and the vertical axis represents the true category. For the Case Western Reserve University bearing data set, the diagnostic accuracy of the method of the present invention can reach 95.3%, and this effect can be applied to practical applications.
[0185] Compared with the traditional method, the present invention applies Fisher discriminant constraints to both the coding coefficient matrix and the transposed matrix, making the dictionary more discriminative and improving the classification performance of the method. All the information of the original data is retained, and the fault feature information of bearings, gears, etc. is adaptively mined through dictionary learning, avoiding the loss of feature information caused by autocorrelation constraints. It not only has high accuracy and generalization ability, but also can maintain good stability and reliability under different working conditions. In addition, the translation invariance of the signal is taken into account, and the dictionary is made translation invariant through feature expansion, which effectively improves the feature representation ability of the dictionary. This is also one of the main reasons why the method of the present invention is superior to some traditional methods.
[0186] In this embodiment, a gearbox fault detection device is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made are not repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.
[0187] This embodiment provides a gear box fault detection device, such as Fig. 9 As shown, including:
[0188] An acquisition module 901 is used to construct a training sample set based on gearbox vibration signals of different fault categories;
[0189] The first processing module 902 is used to perform sparse representation classification on the training sample set using the initial dictionary matrix to obtain an initial coefficient matrix corresponding to the initial dictionary matrix;
[0190] The second processing module 903 is used to construct the discriminant reconstruction constraint item of the initial dictionary matrix and the discriminant constraint of the initial coefficient matrix based on the initial dictionary matrix, the initial coefficient matrix and the training sample set, and construct the objective function based on the discriminant reconstruction constraint item and the discriminant constraint;
[0191] A training module 904 is used to perform training based on the objective function and the training sample set, update the initial dictionary matrix and the initial coefficient matrix, and obtain an ideal dictionary matrix and an ideal coefficient matrix;
[0192] The determination module 905 is used to perform feature expansion on the ideal dictionary matrix to obtain a target dictionary matrix, and to determine the current fault category corresponding to the current vibration signal of the gearbox using the target dictionary matrix and the ideal coefficient matrix.
[0193] In some optional implementations, the acquisition module 901 is further used to:
[0194] Segment the gearbox vibration signals of different fault categories to obtain the gearbox vibration short signal corresponding to each fault category;
[0195] For each fault category, feature expansion is performed on the gearbox vibration short signal corresponding to the fault category to obtain a training sample set.
[0196] In some optional implementations, the second processing module 903 is further configured to:
[0197] Based on the preset constraint conditions, the discriminant reconstruction constraint items corresponding to the initial dictionary matrix are obtained; the preset constraint conditions are used to perform sparse representation classification on the training sample set through the initial dictionary matrix;
[0198] The transposed matrix of the initial coefficient matrix is determined, and the discriminant constraints of the initial coefficient matrix are obtained according to the transposed matrix and the initial coefficient matrix.
[0199] In some optional implementations, the second processing module 903 is further configured to:
[0200] A first discriminant constraint is obtained based on a first-class inter-dispersion matrix and a first-class intra-dispersion matrix corresponding to the initial coefficient matrix;
[0201] Based on the second type of between-dispersion matrix and the second type of within-dispersion matrix corresponding to the transposed matrix, a second discriminant constraint is obtained;
[0202] Based on the first discriminant constraint and the second discriminant constraint, the discriminant constraint of the initial coefficient matrix is obtained.
[0203] In some optional implementations, the parameters of the objective function include an initial dictionary matrix and an initial coefficient matrix; the training module 904 is further used to:
[0204] The value of the initial coefficient matrix in the objective function is fixed, and the initial dictionary matrix is iteratively updated using the training sample set to obtain an ideal dictionary matrix corresponding to the minimum value of the objective function;
[0205] The value of the initial dictionary matrix in the objective function is fixed, and the initial coefficient matrix is iteratively updated using the training sample set to obtain the ideal coefficient matrix corresponding to the minimum value of the objective function.
[0206] In some optional implementations, the determination module 905 is further configured to:
[0207] Shift and intercept multiple dictionary column vectors in the ideal dictionary matrix respectively to obtain multiple dictionary shift features corresponding to each dictionary column vector;
[0208] For each dictionary column vector, multiple dictionary shift features corresponding to the dictionary column vector are horizontally merged to obtain a target sub-dictionary corresponding to the dictionary column vector;
[0209] Based on the target sub-dictionaries corresponding to the multiple dictionary column vectors, a target dictionary matrix is obtained.
[0210] In some optional implementations, each target sub-dictionary in the target dictionary matrix corresponds to a different fault category, and the ideal coefficient matrix includes sub-coefficient matrices corresponding to different fault categories; the determination module 905 is further used to:
[0211] Based on the ideal coefficient matrix, the average vector of each sub-coefficient matrix is calculated;
[0212] Based on the target sub-dictionary and the average vector, the reconstruction error of the current vibration signal of the gearbox on different target sub-dictionaries is calculated, and the minimum reconstruction error is determined;
[0213] According to the fault category of the target sub-dictionary corresponding to the minimum value of the reconstruction error, the current fault category corresponding to the current vibration signal is obtained.
[0214] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.
[0215] The gearbox fault detection device in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.
[0216] The embodiment of the present invention also provides a computer device having the above Fig. 9 The gearbox fault detection device shown.
[0217] See also Fig.10 , Fig.10 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Fig.10 As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Fig.10 A processor 10 is taken as an example.
[0218] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.
[0219] The memory 20 stores instructions executable by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.
[0220] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0221] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.
[0222] The computer device also includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 may be connected via a bus or other means. Fig.10 The example of connecting through bus is taken in the following.
[0223] The input device 30 can receive input digital or character information, and generate key signal input related to the user settings and function control of the computer device, such as a touch screen, a keypad, a mouse, a track pad, a touch pad, an indicator bar, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 may include a display device, an auxiliary lighting device (e.g., an LED) and a tactile feedback device (e.g., a vibration motor), etc. The above-mentioned display device includes but is not limited to a liquid crystal display, a light emitting diode, a display and a plasma display. In some optional embodiments, the display device can be a touch screen.
[0224] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.
[0225] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.
[0226] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.
Claims
1. A gearbox fault detection method, characterized in that: The method comprises: Based on the gearbox vibration signals of different fault categories, a training sample set is constructed; Using an initial dictionary matrix to perform sparse representation classification on the training sample set, to obtain an initial coefficient matrix corresponding to the initial dictionary matrix; Based on the initial dictionary matrix, the initial coefficient matrix and the training sample set, constructing a discriminant reconstruction constraint item of the initial dictionary matrix and a discriminant constraint of the initial coefficient matrix, and constructing an objective function based on the discriminant reconstruction constraint item and the discriminant constraint; Perform training based on the objective function and the training sample set, update the initial dictionary matrix and the initial coefficient matrix, and obtain an ideal dictionary matrix and an ideal coefficient matrix; The target dictionary matrix is obtained by feature expansion of the ideal dictionary matrix. The current fault category corresponding to the current vibration signal of the gearbox is determined using the target dictionary matrix and the ideal coefficient matrix.
2. The method according to claim 1, characterized in that The method of constructing the discriminant reconstruction constraint items of the initial dictionary matrix and the discriminant constraints of the initial coefficient matrix based on the initial dictionary matrix, the initial coefficient matrix and the training sample set includes: Based on preset constraints, a discriminant reconstruction constraint item corresponding to the initial dictionary matrix is obtained; the preset constraints are used to perform sparse representation classification on the training sample set through the initial dictionary matrix; A transposed matrix of the initial coefficient matrix is determined, and a discriminant constraint of the initial coefficient matrix is obtained based on the transposed matrix and the initial coefficient matrix.
3. The method according to claim 2, characterized in that The step of obtaining the discriminant constraint of the initial coefficient matrix according to the transposed matrix and the initial coefficient matrix comprises: Obtaining a first discriminant constraint based on a first inter-class dispersion matrix and a first intra-class dispersion matrix corresponding to the initial coefficient matrix; Obtain a second discriminant constraint based on a second type of between-dispersion matrix and a second type of within-dispersion matrix corresponding to the transposed matrix; Based on the first discriminant constraint and the second discriminant constraint, a discriminant constraint of the initial coefficient matrix is obtained.
4. The method according to claim 1, characterized in that: The parameters of the objective function include an initial dictionary matrix and an initial coefficient matrix; the training based on the objective function and the training sample set, updating the initial dictionary matrix and the initial coefficient matrix, and obtaining an ideal dictionary matrix and an ideal coefficient matrix, includes: Fixing the value of the initial coefficient matrix in the objective function, and iteratively updating the initial dictionary matrix using the training sample set to obtain an ideal dictionary matrix corresponding to the minimum value of the objective function; The value of the initial dictionary matrix in the objective function is fixed, and the initial coefficient matrix is iteratively updated using the training sample set to obtain an ideal coefficient matrix corresponding to the minimum value of the objective function.
5. The method according to claim 4, characterized in that The step of performing feature expansion on the ideal dictionary matrix to obtain a target dictionary matrix includes: Shifting and intercepting a plurality of dictionary column vectors in the ideal dictionary matrix respectively to obtain a plurality of dictionary shift features corresponding to each dictionary column vector; For each dictionary column vector, multiple dictionary shift features corresponding to the dictionary column vector are horizontally merged to obtain a target sub-dictionary corresponding to the dictionary column vector; Based on the target sub-dictionaries corresponding to the multiple dictionary column vectors, a target dictionary matrix is obtained.
6. The method according to claim 5, characterized in that Each target sub-dictionary in the target dictionary matrix corresponds to a different fault category, and the ideal coefficient matrix includes sub-coefficient matrices corresponding to different fault categories; The method of using the target dictionary matrix and the ideal coefficient matrix to determine the current fault category corresponding to the current vibration signal of the gearbox includes: Based on the ideal coefficient matrix, an average vector of each sub-coefficient matrix is calculated; Based on the target sub-dictionary and the average vector, calculating the reconstruction error of the current vibration signal of the gearbox on different target sub-dictionaries, and determining the minimum reconstruction error; According to the fault category of the target sub-dictionary corresponding to the minimum reconstruction error value, a current fault category corresponding to the current vibration signal is obtained.
7. The method according to any one of claims 1 to 6, characterized in that The training sample set is constructed based on the gearbox vibration signals of different fault categories, including: Segment the gearbox vibration signals of different fault categories to obtain the gearbox vibration short signal corresponding to each fault category; For each fault category, feature expansion is performed on the gearbox vibration short signal corresponding to the fault category to obtain a training sample set.
8. A gearbox fault detection device, characterized in that: The device comprises: An acquisition module is used to construct a training sample set based on gearbox vibration signals of different fault categories; A first processing module is used to perform sparse representation classification on the training sample set using an initial dictionary matrix to obtain an initial coefficient matrix corresponding to the initial dictionary matrix; A second processing module is used to construct a discriminant reconstruction constraint item of the initial dictionary matrix and a discriminant constraint of the initial coefficient matrix based on the initial dictionary matrix, the initial coefficient matrix and the training sample set, and to construct an objective function based on the discriminant reconstruction constraint item and the discriminant constraint; A training module, used for performing training based on the objective function and the training sample set, updating the initial dictionary matrix and the initial coefficient matrix, and obtaining an ideal dictionary matrix and an ideal coefficient matrix; The determination module is used to perform feature expansion on the ideal dictionary matrix to obtain a target dictionary matrix, and to determine the current fault category corresponding to the current vibration signal of the gearbox using the target dictionary matrix and the ideal coefficient matrix.
9. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the gearbox fault detection method according to any one of claims 1 to 7 by executing the computer instructions.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the gearbox fault detection method according to any one of claims 1 to 7.
11. A computer program product, characterized in that The method comprises computer instructions for causing a computer to execute the gearbox fault detection method according to any one of claims 1 to 7.
Citation Information
Patent Citations
Rolling bearing fault identification method based on learning dictionary and singular value decomposition
CN110348491A
Fabric image defect detection method and system based on category constraint dictionary learning model
CN113793319A
Intelligent monitoring method and apparatus for abnormal working conditions in heavy metal wastewater treatment process based on transfer learning and storage medium
US20230089156A1
