Bearing fault diagnosis method and device based on improved neural chaotic learning

Abstract

The application relates to a bearing fault diagnosis method and device based on improved neural chaotic learning. The method comprises the following steps: obtaining a use data matrix; based on a GLS neuron neural activation algorithm, obtaining a use neural activity trajectory matrix according to the use data matrix and preset hyperparameters; based on a topological transitivity-symbol sequence algorithm, performing feature extraction on the use neural activity trajectory matrix according to the hyperparameters to obtain a use normalized feature matrix; calculating a use mean representation vector according to the use normalized feature matrix; and based on an enhanced classifier, inputting the use mean representation vector to output a use fault classification result vector. The method solves the problem that the prior art is sensitive to partial loss of a data matrix, and has poor diagnosis effect on small samples and unbalanced data matrices.

Description

Technical Field

[0001] This application relates to the field of neural networks, specifically to a bearing fault diagnosis method and apparatus based on improved neural chaotic learning. Background Technology

[0002] Rolling bearings, as crucial precision components in machine tool manufacturing processes, frequently experience equipment failures or downtime due to design flaws and harsh operating conditions. Early fault diagnosis and treatment before bearing failure reaches their final stage can prevent safety accidents and property losses caused by escalating faults, thus possessing significant socio-economic value. Therefore, research into intelligent bearing fault diagnosis methods is of paramount importance.

[0003] In existing technologies, the use of dual training methods can accurately classify faults under small-scale training sample conditions; the selection method based on the modal component of the maximum kurtosis value of each IMF; and the combination of the AdaBoost algorithm and the decision tree classifier have achieved better performance.

[0004] However, existing technologies are highly sensitive to partial missing data sets and are less effective at diagnosing small or imbalanced datasets. Summary of the Invention

[0005] To address the aforementioned problems, the purpose of this application is to provide a bearing fault diagnosis method and apparatus based on improved neural chaotic learning.

[0006] According to a first aspect of the embodiments of this application, a bearing fault diagnosis method based on improved neural chaotic learning is provided, comprising: acquiring a data matrix; obtaining a neural activity trajectory matrix based on the data matrix and preset hyperparameters using the GLS neuron activation algorithm; extracting features from the neural activity trajectory matrix based on the hyperparameters using the topological transitivity-symbolic sequence algorithm to obtain a normalized feature matrix; calculating a mean representation vector based on the mean representation vector; and outputting a fault classification result vector based on an augmented classifier, using the mean representation vector as input.

[0007] Furthermore, the GLS-based neuron activation algorithm, which obtains the neural activity trajectory matrix based on the data matrix and preset hyperparameters, includes: wherein the hyperparameters include initial neural activity values ​​and epsilon neighborhood values ​​of neural stimuli; setting the parameters of the GLS neuron activation algorithm based on the initial neural activity values; and obtaining the neural activity trajectory matrix based on the GLS neuron activation algorithm, according to the data matrix and the epsilon neighborhood values ​​of neural stimuli.

[0008] Furthermore, the step of obtaining the neural activity trajectory matrix based on the GLS neuron activation algorithm according to the usage data matrix and the epsilon neighborhood value of the neural stimulus includes: obtaining each usage data element in the usage data matrix;

[0009] The neighborhood set corresponding to the usage data element is obtained based on the epsilon neighborhood value of the neural stimulus and the first element of each column of the usage data matrix; the columns of the usage data matrix correspond one-to-one with the epsilon neighborhood values ​​in the neighborhood set; based on the GLS neuron activation algorithm, the usage data matrix is ​​input to obtain an initial usage neural activity trajectory matrix, the columns of the initial usage neural activity trajectory matrix correspond one-to-one with the columns of the usage data matrix; it is detected whether there is an element in each column of the initial usage neural activity trajectory matrix that reaches the corresponding neighborhood value. If not, the current column of the usage neural activity trajectory matrix is ​​stored, and the detection of the next column continues. If there is, according to the element that reaches the corresponding neighborhood value, the remaining elements of the corresponding column of the usage neural activity trajectory matrix are assigned values, and the current column is stored, and the detection of the next column continues; the usage neural activity trajectory matrix is ​​obtained based on the stored columns of each initial usage neural activity trajectory matrix. Furthermore, the topology-transitivity-symbol sequence algorithm, based on the hyperparameters, extracts features from the neural activity trajectory matrix to obtain a normalized feature matrix, including: wherein the hyperparameters include a discrimination threshold; based on the neural activity trajectory matrix; based on the neural activity trajectory matrix, obtaining the total firing time of each corresponding neuron in the matrix; traversing each neural activity trajectory element in the neural activity trajectory matrix; for each neural activity trajectory element, determining whether the current neural activity trajectory element is higher than the discrimination threshold; if so, accumulating the firing time of the neuron corresponding to the neural activity trajectory element; if not, determining the next neural activity trajectory element; obtaining topology-transitivity-symbol sequence features based on the total neuron firing time and the firing time of each neuron higher than the discrimination threshold; obtaining a normalized feature matrix based on each topology-transitivity-symbol sequence feature; wherein the discrimination threshold determines the skewness of the neural activity trajectory values.

[0010] According to a second aspect of the embodiments of this application, a bearing fault diagnosis device based on improved neural chaotic learning is provided, characterized in that it includes: a data acquisition module for acquiring a data matrix; a data processing module for obtaining a neural activity trajectory matrix based on the GLS neuron activation algorithm, according to the data matrix and preset hyperparameters; a feature extraction module for extracting features from the neural activity trajectory set based on the topological transitivity-symmetric sequence algorithm, according to the hyperparameters, to obtain a normalized feature matrix; a feature processing module for calculating a mean representation vector based on the normalized feature matrix; and a fault diagnosis module for outputting a fault classification result vector based on an augmented classifier, using the mean representation vector as input.

[0011] Furthermore, the data processing module is specifically used for: wherein the hyperparameters include initial neural activity values ​​and epsilon neighborhood values ​​of neural stimulation; setting parameters of the GLS neuron activation algorithm based on the initial neural activity values; and obtaining a neural activity trajectory matrix based on the GLS neuron activation algorithm, according to the data matrix used and the epsilon neighborhood values ​​of the neural stimulation.

[0012] Furthermore, the data processing module is further configured to: acquire each usage data element in the usage data matrix; obtain a neighborhood set corresponding to each usage data element based on the epsilon neighborhood value of the neural stimulus and the first element of each column of the usage data matrix; the columns of the usage data matrix correspond one-to-one with the epsilon neighborhood values ​​in the neighborhood set; based on the GLS neuron activation algorithm, input the usage data matrix to obtain an initial usage neural activity trajectory matrix, the columns of the initial usage neural activity trajectory matrix correspond one-to-one with the columns of the usage data matrix; detect whether there is an element in each column of the initial usage neural activity trajectory matrix that reaches the corresponding neighborhood value; if not, store the current column of the usage neural activity trajectory matrix and continue to detect the next column; if so, assign values ​​to the remaining elements of the corresponding column of the usage neural activity trajectory matrix based on the element that reaches the corresponding neighborhood value, store the current column, and continue to detect the next column; obtain the usage neural activity trajectory matrix based on the stored columns of each initial usage neural activity trajectory matrix.

[0013] Further, the feature extraction module is specifically used for: wherein the hyperparameters include a discrimination threshold; based on the neural activity trajectory matrix; based on the neural activity trajectory matrix, obtaining the total firing time of neurons corresponding to each matrix; traversing each neural activity trajectory element in the neural activity trajectory matrix; for each neural activity trajectory element, determining whether the current neural activity trajectory element is higher than the discrimination threshold; if so, accumulating the firing time of the neurons corresponding to the neural activity trajectory element; if not, determining the next neural activity trajectory element; obtaining topological transitivity-symbol sequence features based on the total neuron firing time and the firing time of each neuron higher than the discrimination threshold; obtaining a normalized feature matrix based on each topological transitivity-symbol sequence feature; wherein the discrimination threshold determines the skewness of the neural activity trajectory values.

[0014] According to a third aspect of the embodiments of this application, an electronic device is provided, comprising: a processor; a memory for storing processor-executable instructions; and a processor for reading executable instructions from the memory and executing the instructions to implement a bearing fault diagnosis method based on improved neural chaotic learning provided in the first aspect of this application.

[0015] According to a fourth aspect of the embodiments of this application, a computer-readable storage medium is provided, on which computer program instructions are stored, which, when executed by a processor, implement the steps of the bearing fault diagnosis method based on improved neural chaotic learning provided in the first aspect of this application.

[0016] The technical solutions provided by the embodiments of this application may include the following beneficial effects:

[0017] This application obtains a data matrix and, based on the GLS neuron activation algorithm, generates a neural activity trajectory matrix using the data matrix and hyperparameters. The data, after periodic processing with low feature density, is processed by GLS neurons to generate chaotic neural activity, producing a non-periodic neural activity trajectory matrix with higher feature density. The higher density features have a greater proportion of specific features, which is beneficial for subsequent feature extraction. Based on the topological transitivity-symbolic sequence algorithm, features are extracted from the neural activity trajectory matrix to obtain a normalized feature matrix. This utilizes the principles and characteristics of chaos to improve feature extraction, thus enhancing the classifier and improving recognition efficiency. Finally, the mean of the normalized feature matrix is ​​taken to obtain a mean representation vector, making the sample features more objective and reducing the impact of outliers on the classification results. Based on the enhanced classifier, the mean representation vector is used as input to output a classification result vector. Therefore, this application can solve the problems of difficult feature extraction for bearing faults and poor diagnostic performance on small samples and imbalanced datasets under conditions of insufficient training samples. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in one or more embodiments of this specification or in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the use of a bearing fault diagnosis method based on improved neural chaotic learning, according to an exemplary embodiment.

[0020] Figure 2 This is a flowchart illustrating the training and verification process of a bearing fault diagnosis method based on improved neural chaotic learning, according to an exemplary embodiment.

[0021] Figure 3 This is a macroscopic F1 score variation graph illustrating a bearing fault diagnosis method based on improved neural chaotic learning, according to an exemplary embodiment.

[0022] Figure 4 This is a confusion matrix diagram of diagnostic results on a test set of a bearing fault diagnosis method based on improved neural chaotic learning, according to an exemplary embodiment.

[0023] Figure 5 This is an average macroscopic F1 score graph of five diagnostic models for a bearing fault diagnosis method based on improved neural chaotic learning, according to an exemplary embodiment.

[0024] Figure 6 This is an apparatus diagram illustrating a bearing fault diagnosis method based on improved neural chaotic learning, according to an exemplary embodiment. Detailed Implementation

[0025] To enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this document.

[0026] The network model framework (NL-AdaBoost) in this application is a single-layer chaotic neuron structure, consisting of an input layer and an output layer. The input layer consists of GLS neurons, used to extract feature models from input samples. The output layer nodes store the representation vectors of the fault categories.

[0027] Exemplary Method 1

[0028] like Figure 1 As shown in the embodiment of this application, a bearing fault diagnosis method based on improved neural chaotic learning is provided, including the following steps:

[0029] In step S101, the data matrix is ​​obtained.

[0030] In this application, the time-domain signal is converted into a frequency-domain signal using a fast Fourier transform (FFT), and the time-frequency and frequency-domain signals are concatenated to obtain a single sample. This processing method not only preserves the original feature information of the signal but also retains its one-dimensional characteristics, preparing it for neural activation and feature extraction.

[0031] In step S102, based on the GLS neuron activation algorithm, the neural activity trajectory matrix is ​​obtained according to the data matrix and preset hyperparameters.

[0032] Hyperparameters include initial neural activity values ​​and epsilon neighborhood values ​​of neural stimulation.

[0033] Based on the initial neural activity value, the parameters of the GLS neuron neural activation algorithm are set. In this application, the initial neural activity value is q, that is, the GLS neuron has q units of initial neural activity in the initial state.

[0034] The entire input data sample matrix, also known as the data matrix, is represented as an m×n dimensional matrix, where m represents the number of samples and n represents the number of sampling points in each sample. After inputting the data, the GLS neuron has n units of neural activity.

[0035] The input data samples are represented as X1, X2, ..., X s Based on the GLS neuron activation algorithm, the initial neural activity trajectory matrix A1(t), A2(t), ..., A is obtained by inputting the data matrix. s (t).

[0036] The neighborhood value corresponding to the data element is calculated based on the epsilon neighborhood value of the neural stimulus and the data matrix used.

[0037] The formula using the neighborhood value corresponding to the data element is expressed as follows:

[0038]

[0039] Where i represents a sample, i = 1, 2, ..., s, k represents a different feature in each sample, the feature is the first value of the column of the data matrix, the columns of the data matrix correspond one-to-one with the epsilon neighborhood values ​​in the neighborhood set, and ε is the epsilon neighborhood value of the neural stimulus and ε > 0.

[0040] The neighborhood set corresponding to each data element corresponds to the number of columns in the neural activity trajectory matrix, with each column sharing a common neighborhood value.

[0041] Input the data matrix to obtain the initial neural activity trajectory matrix. Check if there are any elements in each column of the initial neural activity trajectory matrix that have reached the corresponding neighborhood value. If not, store the current column of the neural activity trajectory matrix and continue to check the next column. If there are, assign values ​​to the remaining elements in the corresponding column of the neural activity trajectory matrix according to the elements that have reached the corresponding neighborhood value, store the current column, and continue to check the next column. Based on the stored columns of each initial neural activity trajectory matrix, obtain the neural activity trajectory matrix.

[0042] For example, input matrix X1, based on the GLS neuron activation algorithm, yields the initial neural activity trajectory matrix A1(t), where the first element of each column of input matrix X1 is... The epsilon neighborhood value of the neural stimulus is calculated to obtain This is used to determine whether the corresponding column of the initial neural activity trajectory matrix A1(t) needs to stop calculation. The columns of the initial neural activity trajectory matrix correspond one-to-one with the columns of the data matrix used. Determine the elements in the first column of A1(t) If the first column element has not been reached The range is determined by storing the first column, and then the other columns are evaluated. If the third row of the first column... arrive Scope, will The value is assigned to zero and stored in the neural activity trajectory matrix. Other columns are then evaluated. Once all matrix elements have been evaluated, the neural activity trajectory matrix is ​​obtained based on the stored columns of the initial neural activity trajectory matrix.

[0043] After processing periodic data with low feature density, the data is used to generate chaotic neural activity through GLS neurons to produce a set of non-periodic neural activity trajectories with higher feature density. This provides data with higher feature density for subsequent feature extraction calculations. Feature density refers to the proportion of a finite number of feature segments in a sample within the entire sample.

[0044] Broadly speaking Series (Generalized) A GLS (Geosystems Series) is a piecewise linear one-dimensional chaotic map. Tent graphs and binary graphs are commonly used GLS graphs; we use the former as the chaotic neurons in the architecture. The skewness of the GLS graph is controlled by the discrimination threshold 'a'. By changing 'a', chaotic neurons can exhibit weak chaos and strong chaos respectively according to the Lyapunov exponent. Here, "chaos" is a phenomenon where the motion of a system tends to be uncertain, and "strong" and "weak" are used to measure the degree of this uncertainty. Chaotic features can provide linear separability between classes.

[0045] In step S103, based on the topological transitivity-symbolic sequence algorithm, features are extracted from the set of neural activity trajectories according to the hyperparameters to obtain a normalized feature set.

[0046] Hyperparameters include the discrimination threshold α.

[0047] Obtain the neural activity trajectory matrices from the set of neural activity trajectories used; based on the neural activity trajectory values ​​in the neural activity trajectory matrices, obtain the total firing time of each corresponding neuron. Where u represents each element i = 1, 2, ..., s in the matrix.

[0048] Iterate through each neural activity trajectory value in the neural activity trajectory matrix;

[0049] For each neural activity trajectory value, determine whether the current neural activity trajectory value is higher than the discrimination threshold a. If so, assign the neural activity trajectory value to the corresponding neuron firing time. Accumulate the values; if not, determine the next neural activity trajectory value.

[0050] Based on the total firing time of the neurons and the firing time of each of the neurons above the discrimination threshold Obtain topological transitivity-symbol sequence features The calculation formula is as follows:

[0051]

[0052] Based on the topological transitivity-symbol sequence features described above, a normalized feature matrix U is obtained. The matrix size is m×n, and there are s matrices in total, denoted as U. 1 U 2 ,…,U S .

[0053] By utilizing the chaotic properties and calculating the discrimination threshold 'a', a normalized feature matrix U with higher feature density is obtained, which ensures improved classification accuracy in this application.

[0054] In step S104, the mean-represented vector is calculated based on the normalized feature matrix.

[0055] The normalized feature matrix U 1 U 2 ,…,U S Let V be the eigenvector. 1 V 2 ,…,V S The average representation vector is calculated using the following formula:

[0056]

[0057] The average representation vector M is obtained from the calculation. 1 M 2 M s .

[0058] A normalized feature matrix constitutes a feature. Taking the mean of the normalized feature matrix of the samples yields a mean representation vector, which is more representative of the sample features and more objective. At the same time, it can reduce the impact of outlier samples on the classification results.

[0059] In step S105, based on the augmented classifier, the vector representing the result of the fault classification is output as the input.

[0060] The output is calculated using the fault classification result vector through an enhanced classification expression, as shown in the following formula:

[0061] ψ (z) =argmaxf(z i ,y i )

[0062] in d is the number of features, G v (i) represents the weight parameters, i = 1, 2, ..., d.

[0063] Exemplary Method Two

[0064] like Figure 2 As shown in the embodiments of this application, a training and verification method for bearing fault diagnosis based on improved neural chaotic learning is provided, including the following steps:

[0065] In this exemplary method, hyperparameters are calculated, the model is trained, and the method of this application is verified using Data Set 1.

[0066] Data in Data Set 1 comes from a bearing of model NICE, with a ball diameter of 0.235 cm and 8 balls. The dataset includes both inner and outer ring faults. This dataset contains actual fault data and fault data from a laboratory bearing test bench. Detailed descriptions of each type of fault are shown in Table 1.

[0067] Table 1 Fault Description

[0068]

[0069]

[0070] The experimental dataset was composed of three faults each from the inner and outer ring faults under different loads, as well as three normal operating conditions, for a total of 10 fault types. Eighty experimental samples were selected from each fault type, for a total of 800 samples. Each sample had 2048 sampling points, with the training set accounting for 80% and the test set accounting for 20%.

[0071] In step S201, hyperparameters are calculated.

[0072] In the embodiments of this application, there are three hyperparameters: initial neural activity q, discrimination threshold a, and epsilon neighborhood ε of neural stimulation.

[0073] This application uses K-fold cross-validation to determine the optimal parameters of the model, which can comprehensively evaluate the overall performance of the model and hyperparameters. Compared with a fixed single training set, each sample in cross-validation will be used as both the training set and the validation set, resulting in more reliable final results.

[0074] A fixed set of hyperparameters (initial neural activity q = 0.31, discrimination threshold a = 0.92) was used, and ε was increased from 0.15 to 0.16 in increments of 0.0001, resulting in 100 training iterations on the training set with different ε parameters. Finally, the macroscopic F1 score for each sample set was calculated.

[0075] The macro F1 score, which balances the precision and recall of the classification model, was chosen as the classification metric. The macro F1 score can be viewed as a weighted average of the model's precision and recall, and is suitable for measuring the accuracy of imbalanced data.

[0076] The value range is [0,1]. The F1 score for each class is calculated using the formula. i :

[0077]

[0078] Calculate the macro F1 score for each class. i :

[0079]

[0080] After 100 training iterations, the changes in the macroscopic F1 score corresponding to hyperparameter adjustments are as follows: Figure 2 As shown, the F1 score reaches 1.00 in the range of 0.153-0.157. Therefore, the final optimal hyperparameters obtained in this application are q = 0.31, a = 0.92, and ε = 0.155.

[0081] In step S202, the classifier is trained.

[0082] Using hyperparameters, based on the GLS neuron activation algorithm, topological transitivity-symbolic sequence algorithm, and feature processing, we computed data set one and obtained the feature training classifier for data set one.

[0083] The initial weight parameters for the boost classifier are:

[0084] G1=(ω 11 ,ω 12 ,…,ω 1i ,…,ω 1d )

[0085] in,

[0086] Each base classifier partitions the sample space into Z1, Z2, ..., Zn. r r refers to the number of samples in the classifier.

[0087] Calculate weight G v , where v = 1, 2, ..., d, and d refers to the number of features.

[0088] Set the classifier output Where j = 1, 2, ..., r; y i =1,2,…,s, For all H v The sum of (i). The enhanced classification expression is calculated using the following formula:

[0089] ψ (z) =argmaxf(z i ,y i )

[0090] in

[0091] In step S203, the bearing fault diagnosis method based on improved neural chaotic learning is verified.

[0092] The test set validates the classification ability of the optimal hyperparameters of NL-AdaBoost. Ten repeated experiments were conducted, and the average of the precision, recall, and F1 score for each fault type was recorded. The test set includes 10 fault types, with 16 groups per type, for a total of 160 data sets.

[0093] The average precision, recall, and F1 score for each fault type were calculated using the formula on the test set, as shown in Table 2.

[0094] Table 2 Diagnostic results of the test set

[0095]

[0096] The F1 score for the entire test set was 0.975, demonstrating that the method described in this application has a good classification effect on bearing fault data.

[0097] To visually demonstrate the fault classification, a fault diagnosis result confusion matrix was plotted on the test set, based on one set of test results. Figure 3 As shown, it can be seen that except for Type 4, Type 7 and Type 9 which have 1 and 2 judgment errors respectively, the fault types of the rest are accurately identified.

[0098] Exemplary Method 3

[0099] In this exemplary method, the calculation results of this application are verified by using data set 2 and compared with the effects of common classification algorithms, proving that the model proposed in this application can effectively improve the fault diagnosis results.

[0100] Data set two includes data on German FLURO ER12K wide inner ring ball bearings. Three-channel accelerometers (top, left, and right) were used to collect bearing vibration signals at a sampling frequency of 5000Hz for 10 seconds at a rotational speed of 500 r / min. The vibration signal samples included five states: normal operation, inner ring fault, outer ring fault, cage fault, and rolling element fault. Eighty sets of experimental samples were selected from each type of fault, for a total of 800 sets. In the training set, the number of training samples for each type of fault increased from 1, 2, ..., 16 sets, while the number of samples in the test set remained consistent with data set one.

[0101] Four deep learning algorithms were used, including the native ChaosNet (ChaosFEX+SVM), K-Nearest Neighbors (KNN) classification algorithm, SVM, and Naive Bayes classifier. A comparative experiment was conducted between Bayes and this application. During training, to analyze the relationship between the number of training set samples and the model's diagnostic accuracy, the number of training samples for each type of fault was set to 1, 2, ..., 16. Then, the model's generalization ability was verified on the test set, and the average macroscopic F1 score for each case was calculated. Figure 5The curves show the average macroscopic F1 score changes for each type of deep learning algorithm under 16 different conditions.

[0102] Depend on Figure 5 It can be seen that the fault diagnosis accuracy of the five algorithms gradually increases with the increase of the number of training samples. Among them, the Naive Bayes classifier has the lowest fault diagnosis accuracy, with a macro F1 score of only 0.39 when the training samples reach 9 groups. When the model diagnosis accuracy tends to stabilize, the macro F1 scores of the other four algorithms are all greater than 0.8. When the training samples are insufficient, the native ChaosNet (ChaosFEX+SVM) and the proposed method NL-AdaBoost (ChaosFEX+Adaboost) show good classification performance, with macro F1 scores greater than 0.9.

[0103] In other words, the bearing fault diagnosis model based on improved neural chaotic learning proposed in this application has good stability, accuracy and generalization ability.

[0104] Exemplary device

[0105] In the embodiments of this application, such as Figure 6 It includes a data acquisition module 601, a data processing module 602, a feature extraction module 603, a feature processing module 604, and a fault diagnosis module 605.

[0106] Data acquisition module 601 is used to acquire the data matrix.

[0107] Data processing module 602 is used to obtain a neural activity trajectory matrix based on the GLS neuron activation algorithm, according to the usage data matrix and preset hyperparameters; wherein, the hyperparameters include initial neural activity values ​​and epsilon neighborhood values ​​of neural stimuli; set the parameters of the GLS neuron activation algorithm according to the initial neural activity values; obtain each usage data element in the usage data matrix; obtain the neighborhood set corresponding to the usage data element according to the epsilon neighborhood values ​​of the neural stimuli and the first element of each column of the usage data matrix; the columns of the usage data matrix correspond one-to-one with the epsilon neighborhood values ​​in the neighborhood sets; based on the GLS... The neuron activation algorithm takes the data matrix as input to obtain an initial neural activity trajectory matrix, where the columns of the initial neural activity trajectory matrix correspond one-to-one with the columns of the data matrix. It then checks whether any element in each column of the initial neural activity trajectory matrix reaches its corresponding neighborhood value. If not, it stores the current column of the neural activity trajectory matrix and continues checking the next column. If so, it assigns values ​​to the remaining elements in the corresponding column of the neural activity trajectory matrix based on the elements that have reached their corresponding neighborhood values, stores the current column, and continues checking the next column. Based on the stored columns of each initial neural activity trajectory matrix, it obtains the neural activity trajectory matrix.

[0108] The feature extraction module 603 is used to extract features from the neural activity trajectory matrix based on the topological transitivity-symbol sequence algorithm and according to the hyperparameters, to obtain a normalized feature matrix; wherein the hyperparameters include a discrimination threshold; based on the neural activity trajectory matrix; based on the neural activity trajectory matrix, the total firing time of neurons corresponding to each matrix is ​​obtained; each neural activity trajectory element in the neural activity trajectory matrix is ​​traversed; for each neural activity trajectory element, it is determined whether the current neural activity trajectory element is higher than the discrimination threshold. If so, the firing time of the neuron corresponding to the neural activity trajectory element is accumulated; if not, the next neural activity trajectory element is determined; based on the total firing time of neurons and the firing time of each neuron higher than the discrimination threshold, topological transitivity-symbol sequence features are obtained; based on each topological transitivity-symbol sequence feature, a normalized feature matrix is ​​obtained; wherein the discrimination threshold determines the skewness of the neural activity trajectory values.

[0109] Feature processing module 604 is used to calculate a mean-represented vector based on the normalized feature matrix.

[0110] The fault diagnosis module 605 is used to output a fault classification result vector based on an augmented classifier, taking the mean representation vector as input.

[0111] Exemplary electronic devices

[0112] This embodiment proposes an electronic device, including: one or more processors, an internal memory, and an external memory. The internal memory stores instructions, which, when executed by the one or more processors, cause the one or more processors to perform a bearing fault diagnosis method based on improved neural chaotic learning as described in any of the preceding embodiments.

[0113] The processor is used to execute all or part of the steps in the bearing fault diagnosis method based on improved neural chaotic learning as described in the embodiments. The memory is used to store various types of data, which may include, for example, instructions for any application or method in the electronic device, as well as application-related data.

[0114] The processor can be implemented as an Application Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), controller, microcontroller, microprocessor, or other electronic components, and is used to execute the bearing fault diagnosis method based on improved neural chaotic learning described in the embodiment.

[0115] Computer storage media

[0116] A computer program is stored on the computer-readable storage medium, which, when executed by one or more processors, implements a bearing fault diagnosis method based on improved neural chaotic learning as described in any of the foregoing embodiments.

[0117] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0118] In the 1930s, improvements to a technology could be clearly distinguished as either hardware improvements (e.g., improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the methodology). However, with technological advancements, many improvements to the methodology today can be considered direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved methodology into the hardware circuit. Therefore, it cannot be said that an improvement to the methodology cannot be implemented using a hardware physical module. For example, a Programmable Logic Device (PLD) (e.g., a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logic function is determined by the user programming the device. Designers can program a digital system themselves to "integrate" it onto a PLD, without needing chip manufacturers to design and manufacture dedicated integrated circuit chips. Furthermore, nowadays, instead of manually manufacturing integrated circuit chips, this programming is mostly implemented using "logic compiler" software. Similar to the software compiler used in program development, the original code before compilation must be written in a specific programming language, called a Hardware Description Language (HDL). There are many HDLs, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, and RHDL (Ruby Hardware Description Language). Currently, the most commonly used are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should understand that by simply performing some logic programming on the method flow using one of these hardware description languages ​​and programming it into an integrated circuit, the hardware circuit implementing the logical method flow can be easily obtained.

[0119] The controller can be implemented in any suitable manner. For example, it can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicon Labs C8051F320. A memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also recognize that, in addition to implementing the controller in purely computer-readable program code form, the same functionality can be achieved by logically programming the method steps to make the controller take the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the means included therein for implementing various functions can also be considered as structures within the hardware component. Alternatively, the means for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0120] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0121] For ease of description, the above apparatus is described by dividing it into various functional units. Of course, when implementing the embodiments of this specification, the functions of each unit can be implemented in one or more software and / or hardware.

[0122] Those skilled in the art will understand that one or more embodiments of this specification can be provided as a method, system, or computer program product. Therefore, one or more embodiments of this specification may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this specification may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0123] This specification is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this specification. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0124] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0125] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0126] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0127] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0128] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0129] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0130] One or more embodiments of this specification can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a particular task or implement a particular abstract data type. One or more embodiments of this specification can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0131] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0132] The above description is merely an embodiment of this document and is not intended to limit the scope of this document. Various modifications and variations can be made to this document by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this document should be included within the scope of the claims of this document.

Claims

1. A bearing fault diagnosis method based on improved neural chaotic learning, characterized in that, The method comprises the following steps: obtaining a use data matrix; obtaining a use neural activity trajectory matrix based on a GLS neuron neural activation algorithm and the use data matrix and preset hyperparameters; extracting features from the use neural activity trajectory matrix based on a topological transitivity-symbolic sequence algorithm and the hyperparameters to obtain a use normalized feature matrix; calculating a use mean representation vector based on the use normalized feature matrix; inputting the use mean representation vector into an enhanced classifier to output a use fault classification result vector; the GLS neuron neural activation algorithm is used to obtain a use neural activity trajectory matrix based on the use data matrix and preset hyperparameters, The method comprises the following steps: wherein the hyperparameters include an initial neural activity value and an epsilon neighborhood value of neural stimulation; parameters of the GLS neuron neural activation algorithm are set based on the initial neural activity value; a use neural activity trajectory matrix is obtained based on the GLS neuron neural activation algorithm, the use data matrix and the epsilon neighborhood value of neural stimulation; the topological transitivity-symbolic sequence algorithm is used to extract features from the use neural activity trajectory matrix based on the hyperparameters to obtain a use normalized feature matrix, which comprises the following steps: wherein the hyperparameters include a discrimination threshold value; based on the use neural activity trajectory matrix; based on the use neural activity trajectory matrix, a total neural firing time corresponding to each matrix is obtained; each use neural activity trajectory element in the use neural activity trajectory matrix is traversed, and for each use neural activity trajectory element, it is determined whether the current use neural activity trajectory element is higher than the discrimination threshold value; if yes, the neural firing time corresponding to the neural activity trajectory element is accumulated, and if not, the next neural activity trajectory element is determined; topological transitivity-symbolic sequence features are obtained based on the total neural firing time and the neural firing time of each neural activity trajectory element that is higher than the discrimination threshold value; a normalized feature matrix is obtained based on each topological transitivity-symbolic sequence feature; wherein the discrimination threshold value determines the skewness of the neural activity trajectory value.

2. The method of claim 1, wherein, the GLS neuron neural activation algorithm is used to obtain a use neural activity trajectory matrix based on the use data matrix and the epsilon neighborhood value of neural stimulation, which comprises the following steps: each use data element in the use data matrix is obtained; a use data element corresponding neighborhood set is obtained based on the epsilon neighborhood value of neural stimulation and the first element of each column of the use data matrix; the columns of the use data matrix correspond one by one to the epsilon neighborhood values in the neighborhood set; based on the GLS neuron neural activation algorithm, the use data matrix is inputted to obtain an initial use neural activity trajectory matrix, and the columns of the initial use neural activity trajectory matrix correspond one by one to the columns of the use data matrix; detecting whether there is an element reaching a corresponding epsilon neighborhood value in each column of the initial usage neural activity trajectory matrix, if not, storing the current column of the usage neural activity trajectory matrix, and continuing to detect the next column, if yes, assigning values to the remaining elements of the corresponding column of the usage neural activity trajectory matrix according to the element reaching the corresponding epsilon neighborhood value, storing the current column, and continuing to detect the next column; obtaining a usage neural activity trajectory matrix according to the stored columns of each initial usage neural activity trajectory matrix.

3. A bearing fault diagnosis device based on improved neural chaotic learning, characterized by, comprising: a data acquisition module, configured to acquire a usage data matrix; a data processing module, configured to obtain a usage neural activity trajectory matrix according to the usage data matrix and a preset hyperparameter based on a GLS neuron neural activation algorithm; a feature extraction module, configured to perform feature extraction on the usage neural activity trajectory matrix according to the hyperparameter based on a topological transitivity-symbolic sequence algorithm, and obtain a usage normalized feature matrix; a feature processing module, configured to calculate a usage mean representation vector according to the usage normalized feature matrix; a fault diagnosis module, configured to output a usage fault classification result vector based on an enhanced classifier with the usage mean representation vector as input; the data processing module is specifically configured to: wherein the hyperparameter comprises an initial neural activity value and an epsilon neighborhood value of neural stimulation; setting parameters of the GLS neuron neural activation algorithm according to the initial neural activity value; obtaining a usage neural activity trajectory matrix according to the usage data matrix and the epsilon neighborhood value of neural stimulation based on the GLS neuron neural activation algorithm; the feature extraction module is specifically configured to: wherein the hyperparameter comprises a discrimination threshold value; according to the usage neural activity trajectory matrix; obtaining total neural firing time of each matrix corresponding neuron according to the usage neural activity trajectory matrix; traversing each usage neural activity trajectory element in the usage neural activity trajectory matrix; judging whether the current usage neural activity trajectory element is higher than the discrimination threshold value for each usage neural activity trajectory element, if yes, accumulating the neural firing time of the neural activity trajectory element, if not, judging the next neural activity trajectory element; obtaining topological transitivity-symbolic sequence features according to the total neural firing time and the firing time of each neuron higher than the discrimination threshold value; obtaining a normalized feature matrix according to each topological transitivity-symbolic sequence feature; wherein the discrimination threshold value determines the skewness of the neural activity trajectory value.

4. The apparatus of claim 3, wherein, the data processing module is further configured to: acquiring each usage data element in the usage data matrix; obtaining a usage data element corresponding neighborhood set according to the epsilon neighborhood value of neural stimulation and the first element of each column of the usage data matrix; the column of the usage data matrix corresponds to the epsilon neighborhood value in the neighborhood set one by one; inputting the usage data matrix based on the GLS neuron neural activation algorithm to obtain an initial usage neural activity trajectory matrix, the column of the initial usage neural activity trajectory matrix corresponds to the column of the usage data matrix one by one; detecting whether there is an element reaching the corresponding neighborhood value in each column of the initial use neural activity trajectory matrix, if not, storing the current column of the use neural activity trajectory matrix, and continuing to detect the next column, if yes, assigning values to the remaining elements of the corresponding column of the use neural activity trajectory matrix according to the element reaching the corresponding neighborhood value, storing the current column, and continuing to detect the next column; obtaining a use neural activity trajectory matrix according to the stored columns of the initial use neural activity trajectory matrix.

5. An electronic device, comprising: comprise: a memory for storing a computer program; a processor for implementing the steps of the bearing fault diagnosis method based on improved neural chaotic learning according to any one of claims 1 to 2 when executing the computer program.

6. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the bearing fault diagnosis method based on improved neural chaotic learning according to any one of claims 1 to 2.

Images