Mechanical system structure health monitoring method and device based on hierarchical symbol transfer entropy

Abstract

The invention discloses a mechanical system structure health monitoring method and device based on hierarchical symbol transfer entropy, and relates to the technical field of data processing, and the method comprises the steps: collecting the vibration data of a plurality of channels of an electromechanical system in different health states; calculating a hierarchical symbol transfer entropy of each sub-channel vibration signal in the plurality of channel vibration data; performing superposition calculation on the level symbol transfer entropy of each level of each sub-channel vibration signal to obtain a second-order tensor feature corresponding to each sub-channel vibration signal; superposing the second-order tensor features corresponding to the vibration signals of each sub-channel to obtain second-order tensor feature matrixes corresponding to the vibration data of the multiple channels; training the initial two-dimensional extreme learning machine based on the second-order tensor feature matrix belonging to the training data set to obtain a trained two-dimensional extreme learning machine; the health state of the electromechanical system is monitored based on the two-dimensional extreme learning machine, and the accuracy of monitoring the health state of the electromechanical system is improved.

Description

Technical Field

[0001] The present disclosure relates to the technical field of data processing, and particularly to a method and device for structural health monitoring of a mechanical system based on hierarchical symbol transfer entropy. Background Art

[0002] Mechanical fault diagnosis is of great significance for ensuring the stable operation of electromechanical systems and reducing economic costs. In the past few decades, methods for fault diagnosis of mechanical structure systems based on vibration data have been established, and such methods generally can be divided into the following three steps: data acquisition, fault feature extraction, and feature recognition. Among them, fault feature extraction is the key to accurately identifying mechanical structure faults. Existing feature extraction methods are mainly divided into three categories: fault detection methods based on modern signal processing, automatic feature representation methods based on deep learning, and manual feature extraction methods based on statistical metrics. For the first category, the main purpose is to detect the characteristic fault frequencies of machinery. However, the vibration data of mechanical systems contains complex frequency characteristics, making it impossible for non-professional users to effectively make judgments. For the second category, fault features can be automatically extracted from a large amount of labeled fault data to reduce human arbitrariness. However, collecting a large amount of labeled fault data is both time-consuming and difficult. For the last category, several statistical metrics are usually constructed in the time domain and frequency domain and used as the input of the classifier in the form of vectors. However, the above statistical indicators cannot be guaranteed to be on the same order of magnitude, resulting in some small valuable indicators being difficult to play an important role in the process of mechanical system fault detection.

[0003] When a mechanical system fails, due to the instantaneous changes in friction and damping, the collected vibration data will contain more non-linear and non-stationary characteristics. As an effective tool for revealing the dynamic changes of non-linear data, entropy has received extensive attention in the field of mechanical system health state monitoring and benefited from its excellent performance. More and more, it is used as the characteristic data for the health state monitoring of the mechanical system to solve the problems existing in the existing characteristic data. However, there are still some deficiencies in the practical application of entropy in the health state monitoring of mechanical systems. For example, when entropy measures such as approximate entropy (ApEn), sample entropy (SE), and fuzzy entropy (FE) are applied to the field of mechanical system health state diagnosis as non-linear complexity indicators, when dealing with data of a large length, the calculation efficiency of ApEn, SE, and FE is low. In addition, in order to improve the processing efficiency of data of a large length, the diagnosis of the health state of the mechanical system can also use permutation entropy (PE), conditional entropy of ordered patterns (CEOP), and symbolic dynamic entropy (SDE). However, permutation entropy (PE) and conditional entropy of ordered patterns (CEOP) ignore the differences between the amplitudes of the given signals, resulting in inaccurate diagnosis results of the health state of the mechanical system. Although symbolic dynamic entropy (SDE) can detect the degree of amplitude difference between each data point in the given signal as conditional entropy, its calculation may not conform to the definition of conditional entropy, thus bringing difficulties to the normalization calculation, and further leading to inaccurate monitoring results when monitoring the health state of the mechanical system under complex working conditions and poor robustness of the model for monitoring the health state of the mechanical system. Summary of the Invention

[0004] The present disclosure provides a method and device for structural health monitoring of a mechanical system based on hierarchical symbolic transfer entropy, and the main purpose is to improve the accuracy of the monitoring result of the health state of the mechanical system and improve the robustness of the health state monitoring model of the mechanical system.

[0005] According to the first aspect of the present disclosure, there is provided a method for structural health monitoring of a mechanical system based on hierarchical symbolic transfer entropy, which includes:

[0006] Collect multiple-channel vibration data of the electromechanical system in different health states;

[0007] Calculate the hierarchical symbolic transfer entropy of each sub-channel vibration signal in the multiple-channel vibration data;

[0008] Perform superposition calculation on the hierarchical symbolic transfer entropy of each level of each sub-channel vibration signal to obtain the second-order tensor feature corresponding to each sub-channel vibration signal;

[0009] Perform superposition on the second-order tensor features corresponding to each sub-channel vibration signal to obtain the second-order tensor feature matrix corresponding to the multiple-channel vibration data;

[0010] Training the initial two-dimensional extreme learning machine based on the second-order tensor feature matrix belonging to the training data set to obtain a trained two-dimensional extreme learning machine;

[0011] Monitoring the health state of the electromechanical system based on the two-dimensional extreme learning machine.

[0012] 2. The method according to claim 1, wherein calculating the hierarchical symbolic transfer entropy of each sub-channel vibration signal in the multiple-channel vibration data includes:

[0013] Performing smoothing processing and differential processing on each sub-channel vibration signal at each level in a preset depth level to obtain vibration signals in different frequency bands corresponding to each sub-channel vibration signal;

[0014] Performing symbolic processing on the vibration signals in different frequency bands respectively based on a preset step threshold or a preset sliding window to obtain symbol sequences corresponding to the vibration signals in different frequency bands;

[0015] Calculating the hierarchical entropy between the symbol sequences corresponding to the vibration signals in different frequency bands in one sub-channel vibration signal, and calculating the hierarchical symbolic transfer entropy based on the hierarchical entropy corresponding to each level, and so on, to obtain the hierarchical symbolic transfer entropy of each sub-channel vibration signal.

[0016] 3. The method according to claim 2, wherein the formula for the smoothing processing is expressed as:

[0017] Q 0 =(x n +x n+1 ) / 2 n = 1, 2, …, N - 1

[0018] The formula for the differential processing is expressed as:

[0019] Q 1 =(x n -x n+1 ) / 2 n = 1, 2, …, N - 1

[0020] wherein, x n and x n+1 are consecutive data points of the sub-channel vibration signal, and the continuous sampling values in the sub-channel vibration signal are between the two. The sub-channel vibration signal is time series data, the sub-channel vibration signal is any one of each sub-channel vibration signal, x n is the nth data point in the sub-channel vibration signal, and x n+1 is the one following x nThe subsequent (n + 1)-th data point, Q 0 represents the vibration signal of the low-frequency band part, which includes low-frequency fault information and is obtained by performing smoothing processing on two consecutive data points x n and x n+1 ; Q 1 represents the vibration signal of the high-frequency band part, which includes high-frequency fault information and is obtained by performing difference processing on two consecutive data points x n and x n+1 . The vibration signals of different frequency bands include the vibration signal of the low-frequency band part and the vibration signal of the high-frequency band part. n = 1, 2, …, N - 1 represents the time series index of the vibration signal of the sub-channel, ranging from 1 to N - 1, where N is the total number of data points.

[0021] 4. The method according to claim 3, wherein the formula representation of the fault feature extraction operator included in any level of the preset depth level is:

[0022]

[0023] wherein, is the fault feature extraction operator of the k-th layer, represents a matrix, where the subscript f is a label, the preset depth level is k, and k represents a specific stage in the hierarchical analysis. k is a positive integer, and the symbol represents is a real number matrix, and its dimension depends on the value of k and the total number of data points N.

[0024] 5. The method according to claim 4, wherein calculating the hierarchical entropy between the symbol sequences corresponding to the vibration signals of different frequency bands in a sub-channel vibration signal includes:

[0025] calculating the node components of the nodes where each symbol sequence is located in each level of the preset depth level;

[0026] calculating the hierarchical entropy based on the node components in each level;

[0027] The calculation formula of the node component is:

[0028]

[0029] wherein, X k,e represents the node component at node e of the k-th layer in the preset depth level, r k represents the selected frequency band at the k-th layer, represents at the k-th layer frequency band r kThe operator matrix extracted above is used for feature extraction of vibration signals, where x represents the

[0030] vibration data, and the node component is used to reflect the frequency band characteristics of its corresponding level;

[0031] The calculation formula of node e is:

[0032]

[0033] where r j represents the average operator or difference operator of the j-th layer, 2 k-j is a weighting factor that decreases as j increases and is used to adjust the influence between different levels, and r represents the selected frequency band;

[0034] The calculation formula of the hierarchical entropy is:

[0035]

[0036] where E 层次熵 represents the hierarchical entropy, m is the number of levels of the vibration signal, represents the range of the summation operation, ste represents the calculation symbol of the hierarchical entropy, and X k is the node component of the k-th layer, represents the fault information under different frequency bands, and τ represents the number of steps of transfer.

[0037] 6. The method according to claim 5, wherein the formula for calculating the hierarchical symbol transfer entropy based on the hierarchical entropy corresponding to each level is expressed as:

[0038]

[0039] where E 层次符号转移熵 represents the hierarchical symbol transfer entropy, E 层次熵 represents the hierarchical entropy, m is the number of levels of the vibration signal, and k represents the preset depth level.

[0040] 7. The method according to any one of claims 1-6, wherein the formula of the two-dimensional extreme learning machine is expressed as:

[0041]

[0042] where represents the output weight vector connecting the l-th hidden unit, represents the v-th projection vector connecting the l-th hidden node, where v = 1 or 2, v = 1 represents low-frequency features, v = 2 represents high-frequency features, and n v is the dimension of the projection vector, representing the number of elements contained in the vector of this layer, represents that the vector belongs to n vd-dimensional real space, b l ∈R represents the bias of the l-th hidden unit, g(·) is the sigmoid function, X n is the input signal or data point, t n is the output or target value, representing the output result or predicted value obtained after signal processing, L is the total number of layers, n represents the n-th sample or data point, and C is the number of categories of the output result or predicted value.

[0043] According to a second aspect of the present disclosure, there is provided a mechanical system structural health monitoring device based on hierarchical symbol transfer entropy, including:

[0044] An acquisition unit for acquiring multi-channel vibration data of the electromechanical system in different health states;

[0045] A calculation unit for calculating the hierarchical symbol transfer entropy of each sub-channel vibration signal in the multi-channel vibration data;

[0046] A first superposition unit for performing superposition calculation on the hierarchical symbol transfer entropy of each level of each sub-channel vibration signal to obtain a second-order tensor feature corresponding to each sub-channel vibration signal;

[0047] A second superposition unit for superposing the second-order tensor features corresponding to each sub-channel vibration signal to obtain a second-order tensor feature matrix corresponding to the multi-channel vibration data;

[0048] A training unit for training an initial two-dimensional extreme learning machine based on the second-order tensor feature matrix belonging to the training data set to obtain a trained two-dimensional extreme learning machine;

[0049] A monitoring unit for monitoring the health state of the electromechanical system based on the two-dimensional extreme learning machine.

[0050] According to a third aspect of the present disclosure, there is provided an electronic device, including:

[0051] At least one processor; and

[0052] A memory communicatively connected to the at least one processor; wherein,

[0053] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method described in the foregoing first aspect.

[0054] According to a fourth aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to execute the method described in the foregoing first aspect.

[0055] According to a fifth aspect of the present disclosure, there is provided a computer program product including a computer program which, when executed by a processor, implements the method as described in the foregoing first aspect.

[0056] The method and device for structural health monitoring of a mechanical system based on hierarchical symbolic transfer entropy provided by the present disclosure collect multi-channel vibration data of an electromechanical system in different health states; calculate the hierarchical symbolic transfer entropy of each sub-channel vibration signal in the multi-channel vibration data; perform superposition calculation on the hierarchical symbolic transfer entropy of each level of each sub-channel vibration signal to obtain a second-order tensor feature corresponding to each sub-channel vibration signal; perform superposition on the second-order tensor features corresponding to each sub-channel vibration signal to obtain a second-order tensor feature matrix corresponding to the multi-channel vibration data; train an initial two-dimensional extreme learning machine based on the second-order tensor feature matrix belonging to a training data set to obtain a trained two-dimensional extreme learning machine; monitor the health state of the electromechanical system based on the two-dimensional extreme learning machine. Compared with the related art, by introducing hierarchical symbolic transfer entropy as feature data for monitoring the health state of an electromechanical system, the difference between fault signals and healthy signals in vibration signals can be better distinguished, significantly improving the accuracy of monitoring the health state of the electromechanical system. By adopting a two-dimensional extreme learning machine that can effectively process high-dimensional data, the accuracy of monitoring the health state of the mechanical system is further improved and the robustness of the health state monitoring model is enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The accompanying drawings forming a part of this specification are used to provide a further understanding of the present invention. The schematic embodiments and descriptions thereof of the present invention are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0058] Figure 1 is a schematic flow chart of a method for structural health monitoring of a mechanical system based on hierarchical symbolic transfer entropy provided by an embodiment of the present disclosure;

[0059] Figure 2 is a schematic diagram of a hierarchical analysis tree structure when k = 3 provided by an embodiment of the present disclosure;

[0060] Figure 3 is a schematic flow chart of a method for calculating hierarchical symbolic transfer entropy provided by an embodiment of the present disclosure;

[0061] Figure 4 is a schematic diagram of a comparison result of feature visualization between the proposed hierarchical symbolic transfer entropy and other feature extractors provided by an embodiment of the present disclosure;

[0062] Figure 5A comparison chart of fault accuracies of seven methods under 20 tests provided by an embodiment of the present disclosure;

[0063] Figure 6 A schematic diagram showing the comparison results of accuracy rate, recall rate, and F-Score obtained by different methods provided by an embodiment of the present disclosure;

[0064] Figure 7 A schematic diagram showing the comparison results of feature visualization between the proposed hierarchical symbolic transfer entropy and other collaborative feature extractors provided by an embodiment of the present disclosure;

[0065] Figure 8 A schematic diagram showing the average diagnostic accuracy of 20 tests using six algorithms under different training data percentages provided by an embodiment of the present disclosure;

[0066] Figure 9 A schematic diagram showing the average diagnostic accuracy of 20 tests using six methods under different noise environments provided by an embodiment of the present disclosure;

[0067] Figure 10 A schematic diagram of the structure of a mechanical system structural health monitoring device based on hierarchical symbolic transfer entropy provided by an embodiment of the present disclosure;

[0068] Figure 11 A schematic block diagram of an exemplary electronic device 300 provided by an embodiment of the present disclosure. Detailed implementation manners

[0069] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments. It should be noted that, without conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0070] The following detailed descriptions are all exemplary descriptions, aiming to provide further detailed descriptions of the present invention. Unless otherwise specified, all technical terms adopted by the present invention have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs. The terms used by the present invention are only for describing specific implementation manners and are not intended to limit the exemplary embodiments according to the present invention.

[0071] In addition, the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such used data may be interchanged under appropriate circumstances so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein.

[0072] The following describes the mechanical system structural health monitoring method and device based on hierarchical symbolic transfer entropy according to the embodiments of the present disclosure with reference to the accompanying drawings.

[0073] In order to provide at least one method for structural health monitoring of mechanical systems based on hierarchical symbolic transfer entropy to improve the accuracy of the monitoring results of the health state of mechanical systems. This embodiment provides a method for structural health monitoring of mechanical systems based on hierarchical symbolic transfer entropy.

[0074] Figure 1 Schematic diagram of the process of a method for structural health monitoring of mechanical systems based on hierarchical symbolic transfer entropy provided by an embodiment of the present disclosure. As Figure 1 shown, the method includes the following steps:

[0075] Step 101, collect vibration data of multiple channels of the electromechanical system in different health states;

[0076] Step 102, calculate the hierarchical symbolic transfer entropy of each sub-channel vibration signal in the vibration data of the multiple channels;

[0077] Step 103, perform superposition calculation on the hierarchical symbolic transfer entropy of each level of each sub-channel vibration signal to obtain the second-order tensor feature corresponding to each sub-channel vibration signal;

[0078] Step 104, perform superposition on the second-order tensor features corresponding to each sub-channel vibration signal to obtain the second-order tensor feature matrix corresponding to the vibration data of the multiple channels;

[0079] Step 105, train the initial two-dimensional extreme learning machine based on the second-order tensor feature matrix belonging to the training data set to obtain a trained two-dimensional extreme learning machine;

[0080] Step 106, monitor the health state of the electromechanical system based on the two-dimensional extreme learning machine.

[0081] The method for structural health monitoring of a mechanical system based on hierarchical symbolic transfer entropy provided by the present disclosure collects vibration data of multiple channels of an electromechanical system in different health states; calculates the hierarchical symbolic transfer entropy of each sub-channel vibration signal in the vibration data of the multiple channels; performs superposition calculation on the hierarchical symbolic transfer entropy of each level of each sub-channel vibration signal to obtain a second-order tensor feature corresponding to each sub-channel vibration signal; superimposes the second-order tensor features corresponding to each sub-channel vibration signal to obtain a second-order tensor feature matrix corresponding to the vibration data of the multiple channels; trains an initial two-dimensional extreme learning machine based on the second-order tensor feature matrix belonging to the training data set to obtain a trained two-dimensional extreme learning machine; and monitors the health state of the electromechanical system based on the two-dimensional extreme learning machine. Compared with the related art, by introducing hierarchical symbolic transfer entropy as the feature data for monitoring the health state of an electromechanical system, the difference between the fault signal and the healthy signal in the vibration signal can be better distinguished, and the accuracy of monitoring the health state of the electromechanical system is significantly improved. By adopting a two-dimensional extreme learning machine that can effectively process high-dimensional data, the accuracy of monitoring the health state of the mechanical system is further improved, and the robustness of the health state monitoring model is improved.

[0082] As a refinement of the above embodiment, when performing step 102 of calculating the hierarchical symbolic transfer entropy of each sub-channel vibration signal in the vibration data of the multiple channels, the following implementation manners may also be adopted but are not limited to, for example: performing smoothing processing and differential processing on each sub-channel vibration signal at different levels in each level of a preset depth level to obtain vibration signals of different frequency bands corresponding to each sub-channel vibration signal; performing symbolization processing on the vibration signals of different frequency bands respectively based on a preset step threshold or a preset sliding window to obtain symbol sequences corresponding to the vibration signals of different frequency bands; calculating the hierarchical entropy between the symbol sequences corresponding to the vibration signals of different frequency bands in a sub-channel vibration signal, and calculating the hierarchical symbolic transfer entropy based on the hierarchical entropy corresponding to each level, and so on, to obtain the hierarchical symbolic transfer entropy of each sub-channel vibration signal.

[0083] As a refinement of the above embodiment, in order to extract richer fault information in different frequency bands, the data is first hierarchically decomposed, and an average operator and a differential operator are used to extract fault information in different frequency bands, so as to obtain finer fault information. Assume that the given data is Then the low-frequency and high-frequency fault information can be extracted through the following operations, including: extracting the low-frequency part of the fault information Q through a low-pass filter or a mean operation, that is, smoothing processing 0 ; extracting the high-frequency part of the fault information Q through a high-pass filter or a differential operation, that is, differential processing 1 . The formula for the smoothing processing is expressed as:

[0084] Q 0 =(x n +x n+1 ) / 2n = 1, 2, …, N - 1

[0085] The formula for the differential processing is expressed as:

[0086] Q 1 =(x n -x n+1 ) / 2n = 1, 2, …, N - 1

[0087] Among them, x n and x n+1 are consecutive data points of the sub-channel vibration signal, and the continuous sampling values in the sub-channel vibration signal are between the two. The sub-channel vibration signal is time series data, and the sub-channel vibration signal is any one of the each sub-channel vibration signals. x n is the nth data point in the sub-channel vibration signal, while x n+1 is the (n + 1)th data point following x n immediately. Q 0 represents the vibration signal of the low-frequency band part, which includes low-frequency fault information and is obtained by performing smoothing processing on two consecutive data points x n and x n+1 . Q 1 represents the vibration signal of the high-frequency band part, which includes high-frequency fault information and is obtained by performing differential processing on two consecutive data points x n and x n+1 . The vibration signals of different frequency bands include the vibration signal of the low-frequency band part and the vibration signal of the high-frequency band part. n = 1, 2, …, N - 1 represents the time series index of the sub-channel vibration signal, ranging from 1 to N - 1, where N is the total number of data points.

[0088] As a refinement of the above embodiment, the formula for the fault feature extraction operator included in any level of the preset depth level is expressed as:

[0089]

[0090] Among them, is the fault feature extraction operator of the kth layer, represents a matrix, where the subscript f is a label. The preset depth level is k, and k represents a specific stage in the hierarchical analysis. k is a positive integer. The symbol represents is a real number matrix, and its dimension depends on the value of k and the total number of data points N.

[0091] It is important to understand that through hierarchical analysis, the information capture of different frequency bands can be refined. For example, when k = 3, the signal is decomposed into three different levels, each of which represents the fault characteristics of different frequency bands. Figure 2 A schematic diagram of a hierarchical analysis tree structure when k=3 provided in an embodiment of the present disclosure is shown in FIG. Figure 2 As shown. In the feature extraction method of hierarchical symbol transfer entropy, the basis for hierarchical division is not directly the size of the frequency band, but the information of different frequency bands is captured through hierarchical analysis of the signal. The decomposition of each level is actually to extract the features of different frequency bands (low frequency or high frequency) by smoothing or differencing the signal to different degrees. Therefore, although the division of high-frequency and low-frequency components is the initial step, in the hierarchical analysis, each level will further refine the characteristics of the signal, especially in different frequency ranges. For example, at each level, the low-frequency part may be further decomposed into lower frequency bands or more details, while the high-frequency part may be more finely distinguished. In general, the purpose of hierarchical division is to gradually extract fault information at different scales in the signal, rather than simply segmenting the high-frequency and low-frequency bands again.

[0092] As a refinement of the above embodiment, when calculating the hierarchical entropy between the symbol sequences corresponding to the vibration signals of different frequency bands in the sub-channel vibration signal, the following implementation method may also be adopted but is not limited to, for example: calculating the node components of the nodes where each symbol sequence is located in each level of the preset depth level; calculating the hierarchical entropy based on the node components in each level; the calculation formula of the node components is:

[0093]

[0094] Among them, X k,e represents the node component at the kth node e in the preset depth hierarchy, r k represents the frequency band selected in the kth layer, Indicates that in the k-th layer frequency band r k The operator matrix extracted above is used to extract features from the vibration signal, x represents the vibration data, and the node component is used to reflect the frequency band characteristics of the level to which it belongs, which can be used to further calculate the symbol transfer entropy;

[0095] The calculation formula for node e is:

[0096]

[0097] Among them, r j represents the average operator or difference operator of the jth layer, 2 k-jis a weighting factor that decreases as j increases and is used to adjust the influence between different levels. r represents the selected frequency band;

[0098] The formula for calculating the hierarchical entropy is as follows:

[0099]

[0100] where E 层次熵 represents the hierarchical entropy, m is the number of levels of the vibration signal, represents the range of the summation operation, ste represents the calculation symbol of the hierarchical entropy, and X k is the node component of the k-th layer, representing the fault information under different frequency bands. The transfer entropy measures the degree of information transfer from one node to another and reflects the information flow characteristics of the signal between different levels. τ represents the number of transfer steps.

[0101] As a refinement of the above embodiment, the formula for calculating the hierarchical symbol transfer entropy based on the hierarchical entropy corresponding to each level is expressed as:

[0102]

[0103] where E 层次符号转移熵 represents the hierarchical symbol transfer entropy, E 层次熵 represents the hierarchical entropy, m is the number of levels of the vibration signal, and k represents the preset depth level.

[0104] In some embodiments, the selection of the hierarchical layer k is based on experience. Set k = 3 or k = 4. By adjusting the value of k, the depth of the hierarchy can be flexibly controlled to capture the fault information in different frequency bands. The step size τ is used to control the time step of the signal, thereby further optimizing the calculation of the symbol transfer entropy. Therefore, the flowchart of the proposed hierarchical symbol transfer entropy is as Figure 3 shown Figure 3 is a schematic diagram of the calculation process of the hierarchical symbol transfer entropy provided by the embodiments of the present disclosure.

[0105] As a refinement of the above embodiment, a two-dimensional extreme learning machine is designed to identify the extracted feature matrix representation without the need for vectorization. Given N different training samples where represents the input matrix feature, (C is the number of categories) is the corresponding expected output label. Therefore, the output formula of the two-dimensional extreme learning machine classifier with L hidden units is as follows::

[0106]

[0107] where, represents the output weight vector connecting the l-th hidden unit, denotes the \(v\)-th projection vector connecting the \(l\)-th hidden node, where \(v = 1\) or \(2\), \(v = 1\) represents low-frequency features, \(v = 2\) represents high-frequency features, and \(n\) v is the dimension of the projection vector, indicating the number of elements contained in the vector of this layer, denotes that the vector belongs to \(n\) v dimensional real number space, and \(b\) l \(\in\mathbb{R}\) represents the bias of the \(l\)-th hidden unit, \(g(\cdot)\) is the sigmoid function, and \(X\) n is the input signal or data point, and \(t\) n is the output or target value, representing the output result or predicted value obtained after signal processing. \(L\) is the total number of layers, \(n\) represents the \(n\)-th sample or data point, and \(C\) is the number of categories of the output result or predicted value. The output formula of the two-dimensional extreme learning machine classifier is further expressed as follows:

[0108] \(H\beta=T\)

[0109] where, \(H\) is the output matrix of the hidden layer and can be expressed as:

[0110]

[0111] In the formula denotes the element in matrix \(H\). \(N\) is the number of rows of the matrix, representing the number of samples of the signal or data. \(L\) is the number of columns of the matrix, and \(T\) is the target vector. In addition, \(\beta=[\beta\) 1 ,\(\beta\) 2 ,\(\cdots\),\(\beta\) L T denotes the connection weight vector between the hidden layer and the output layer. \(Y = [y\) 1 ,y\) 2 ,y\) N is the target matrix. In the two-dimensional extreme learning machine classifier, the main objective is to search for the parameter \(\beta\) that can minimize the error between the output matrix and the target matrix. Therefore, the corresponding objective function can be expressed as:

[0112]

[0113] where, is the penalty coefficient that balances the regularizer and the loss, and \(Y\) represents the target value. In addition, the connection weight \(\beta\) is determined by the following formula:

[0114]

[0115] where \(I\) represents the identity matrix of appropriate size, and \(T\) represents the transpose operation of the matrix.

[0116] ​To facilitate the understanding of the above embodiments, this embodiment provides an exemplary illustration to further explain the aforementioned mechanical system structural health monitoring method based on hierarchical symbol transfer entropy as follows:

[0117] (1) Collect multi-channel vibration data of the electromechanical system in different health states: In this step, multi-channel vibration data of the electromechanical system in different health states such as normal, mild fault, and severe fault are collected through devices such as acceleration sensors. Ensure the representativeness and comprehensiveness of the data during collection, set an appropriate sampling frequency (such as above 10 kHz), and perform necessary preprocessing on the collected original signals, such as denoising, standardization, normalization, etc., to ensure the accuracy and reliability of the data.

[0118] (2) Extract fault features from the vibration signals of each sub-channel using the proposed hierarchical symbol transfer entropy: In this step, first, perform symbolization processing on the collected multi-channel vibration signals to extract their hierarchical structure information:

[0119] a Symbolization process: For the vibration signal of each sub-channel, perform symbolization processing on the signal by setting an appropriate threshold or sliding window, and convert it into a symbol sequence. The symbolization process converts the continuous change of the signal into discrete symbols, making it more suitable for entropy calculation.

[0120] b Hierarchical symbol transfer entropy calculation: At each level, calculate the transfer entropy value between symbol sequences. The transfer entropy measures the amount of information transfer from one symbol sequence to another. Each level corresponds to symbol sequences of different scales, and the dynamic characteristics of the signal in different frequency bands are captured through hierarchical decomposition.

[0121] c Fault feature extraction: Use the hierarchical symbol transfer entropy values of each sub-channel signal to extract features that are helpful for fault diagnosis as input features for subsequent fault mode recognition.

[0122] (3) Superimpose the hierarchical symbol transfer entropy values of the calculated training dataset and test dataset to obtain the second-order tensor features of the training dataset and test dataset: The core of this step is to superimpose the extracted hierarchical symbol transfer entropy values to construct a feature matrix suitable for the extreme learning machine:

[0123] a Superimposition method: First, perform weighted superimposition on the symbol transfer entropy values of each level, or add them element by element to obtain the second-order tensor features of each vibration signal. The transfer entropy value of each level reflects different features of the signal, and the superimposition process can integrate information from different levels and improve the discrimination of the features.

[0124] b Second-order tensor construction: The hierarchical symbolic transfer entropy values of each sample in the training dataset and the test dataset are superimposed to form a second-order tensor feature matrix. These second-order tensor feature matrices will be used as the input of the training and test datasets for the subsequent two-dimensional extreme learning machine model to perform training and prediction.

[0125] (4) Train the two-dimensional extreme learning machine designed in this paper using the second-order tensor features of the training dataset: In this step, based on the second-order tensor features of the training dataset obtained in step (3), use the two-dimensional extreme learning machine for training:

[0126] a Two-dimensional extreme learning machine structure: The two-dimensional extreme learning machine is a method that extends the traditional extreme learning machine. Its input layer provides information through the second-order tensor feature matrix, the hidden layer performs feature transformation, and the output layer is responsible for the classification task. The advantage of this model is that it can handle high-dimensional data and can obtain good generalization ability under limited training samples.

[0127] b Training process: Use the second-order tensor features in the training dataset to train the two-dimensional extreme learning machine and optimize its weights and biases. During the training process, use gradient descent or other optimization algorithms to adjust the parameters to improve the accuracy and robustness of the model.

[0128] (5) Input the second-order tensor features of the test dataset into the trained two-dimensional extreme learning machine for the diagnosis of mechanical system fault modes, and verify the effectiveness and superiority of the proposed method through comparative analysis with existing methods: In this step, use the second-order tensor features of the test dataset for fault mode diagnosis:

[0129] a Fault mode diagnosis: Input the second-order tensor features of the test dataset into the trained two-dimensional extreme learning machine to predict the fault modes of the mechanical system. Determine the health status of the mechanical system through the output of the model and classify different fault modes.

[0130] b Comparative analysis: To verify the effectiveness and superiority of the proposed method, select existing technologies (such as traditional entropy algorithms, machine learning methods, etc.) for comparison. Compare the diagnostic performance of different methods through quantitative indicators (such as accuracy, recall rate, F1 score, etc.). The experimental results show that the proposed method is superior to the existing methods in terms of the accuracy and robustness of fault mode recognition, and can effectively improve the accuracy and real-time performance of fault diagnosis.

[0131] To more intuitively demonstrate the technical effects of the method provided by the present disclosure, this embodiment provides an experimental verification and analysis process, specifically including: using a laboratory gearbox fault dataset to verify the feasibility and superiority of the structural health monitoring method proposed by the present disclosure. For parameter settings, in the hierarchical symbolic transfer entropy proposed by the present disclosure, the hierarchical layer k = 3, the embedding dimension m = 3, the number of symbols e = 3, and the time delay d = 2 are set. And in the two-dimensional extreme learning machine designed by the present disclosure, the number of hidden units is set to 90.

[0132] To evaluate the performance of the proposed hierarchical symbolic transfer entropy, the present disclosure first selects existing entropy-based feature extraction schemes, such as hierarchical dispersion entropy (HDE), hierarchical symbolic dynamic entropy (HSDE), improved multi-scale dispersion entropy (IMDE), and multi-scale symbolic diversity entropy (MSDivEn). Then, several state-of-the-art entropy-based collaborative mechanical fault diagnosis methods and hybrid methods (STE + 2-D-ELM) are also selected to verify the superiority of the proposed scheme.

[0133] Collect the laboratory gearbox fault dataset through a planetary gearbox experimental platform (model: HD-CL-012X) to verify the feasibility of the proposed method in gearbox health monitoring. The system mainly includes a three-phase induction motor, a variable frequency drive device, planetary gears, an electromagnetic brake, and an accelerometer position device. Among them, the three-phase induction motor is the motor for providing power, and the planetary gears are the gear system for transmission. The electromagnetic drive is the electromagnetic device for adjusting the motor drive. The variable frequency drive device is the device for adjusting the motor speed. The accelerometer position device is used to measure and control the acceleration of a certain part in the system. In addition, multi-channel data will be collected by a triaxial accelerometer with a load of 0.25 A. In this experiment, considering 7 working condition types of the planetary gearbox, namely normal condition (NC), ring gear crack (RGC), ring gear pitting (RGP), sun gear crack (SGC), sun gear pitting (SGP), planetary gear crack (PGC), and planetary gear pitting (PGP), the rotational speed is set to 1500 rpm, the sampling frequency is set to 25600 hz, and the measurement data under each gearbox working condition is divided into 200 non-overlapping samples with a sample size of 3 × 2048. The planetary gearbox crack faults are divided into the following categories: (a). Ring gear crack (RGC), (b). Planetary gear crack (PGC), (c). Sun gear crack (SGC), (d). Ring gear pitting (RGP), (e). Planetary gear pitting (PGP), (f). Sun gear pitting (SGP)

[0134] To verify the effectiveness and superiority of the proposed hierarchical symbolic transfer entropy, one-dimensional vibration data was collected from the X direction of a triaxial accelerometer. The hierarchical symbolic transfer entropy was used to represent the fault characteristics, and the t-SNE technique was used to visualize the extracted hierarchical symbolic transfer entropy values in a two-dimensional space. The feature visualization results are as shown in Figure 4 (a). The results in Figure 4 (a) show that the proposed hierarchical symbolic transfer entropy has good clustering and separability. At the same time, HDE, HSDE, IMDE, and MSDivEn were used to analyze the same vibration data, and the corresponding feature visualization results are as shown in Figure 4 (b)-(e). As can be seen from Figure 4 , compared with the other four single-channel extractors, the hierarchical symbolic transfer entropy proposed in this disclosure has better clustering within the same category and better separability between different categories. Figure 4 Feature visualization comparison results of the proposed hierarchical symbolic transfer entropy and other feature extractors: (a). The proposed method, (b). MSDivEn, (c). IMDE, (d). HDE, (e). HSDE.

[0135] In addition, to verify the performance of the proposed fusion of hierarchical symbolic transfer entropy and two-dimensional extreme learning machine in mechanical system structural health monitoring, multivariate data measured from a triaxial accelerometer was collected. First, the structural health monitoring method proposed in this disclosure and the existing entropy-based collaborative fault diagnosis strategy were used to analyze the collected multi-channel data. At the same time, the fusion hierarchical symbolic transfer entropy was constructed through average information, and the two-dimensional extreme learning machine was used as the classifier. To suppress the influence of randomness on the classification rate, 20 tests were conducted, and the corresponding diagnostic comparison analysis results are as shown in Figure 5 . As shown by Figure 5 , the diagnostic accuracy of the method proposed in this disclosure is above 99%, and its standard deviation is the lowest among all methods, which can prove that the performance of the proposed method is superior to other entropy-based collaborative health monitoring technologies. In addition, several indicators (i.e., accuracy, recall, and F-Score) were introduced to further evaluate the performance of the proposed method and other comparison methods, and the results are as shown in Figure 6 . As can be seen from the results shown in Figure 6 , the accuracy, recall, and F-Score values of the proposed method are the highest among all comparison methods under all conditions. At the same time, the fault characteristics extracted by the hierarchical symbolic transfer entropy proposed in this disclosure and the symbolic transfer entropy (STE), RCmvDE, RCmvSDE, mvHPE, and veMDE were visualized in a two-dimensional space, and the results are as shown in Figure 7 . According to the visualization analysis results shown in Figure 7 , it is further proved that the proposed hierarchical symbolic transfer entropy has the best clustering and separability performance in the process of fault feature extraction.Figure 7 . Feature visualization comparison results of the proposed hierarchical symbolic transfer entropy and other collaborative feature extractors: (a). The proposed method, (b). STE, (c). RCmvMDE, (d). RCmvMSDE, (e). mvHPE, (f). veMDE

[0136] Subsequently, the present disclosure set five different percentages of training samples: 10%, 20%, 30%, 40% and 50% to evaluate the impact of different percentages of training data on the proposed method. For each case, the present disclosure conducted 20 trials to suppress the influence of randomness on the diagnostic accuracy, and the results are as Figure 8 shown. The Figure 8 results show that the proposed method has the highest diagnostic accuracy for various cases and the lowest standard deviation. Even when the training data is insufficient and the proportion of training data is only 10%, the average diagnostic accuracy of the method proposed in the present disclosure is above 95%.

[0137] Finally, to evaluate the anti-noise ability of the proposed method in fault diagnosis, the present disclosure simulated different noise environments, set the signal-to-noise ratio range to [0, 40], and the step size to 5 dB. Further, the proposed method and other fault diagnosis methods were used to process multi-channel data in different noise environments. Each method was randomly tested 20 times, and the final classification results in different noise environments are as Figure 9 shown. The Figure 9 results show that the fault diagnosis method proposed in the present disclosure has the highest diagnostic accuracy and the smallest standard deviation. In particular, when the signal-to-noise ratio is 5 dB, the diagnostic accuracy of the proposed method is close to 100%. In summary, through the above analysis, it is verified that the proposed method has better robustness than the existing entropy-based collaborative fault diagnosis methods.

[0138] In summary, the embodiments of the present disclosure can achieve the following effects:

[0139] 1. By introducing hierarchical symbolic transfer entropy as the feature data for the health state monitoring of electromechanical systems, the difference between fault signals and healthy signals in vibration signals can be better distinguished, significantly improving the accuracy of the health state monitoring of electromechanical systems. By adopting a two-dimensional extreme learning machine that can effectively process high-dimensional data, the accuracy of the health state monitoring of mechanical systems is further improved, and the robustness of the health state monitoring model is enhanced.

[0140] 2. A method for structural health monitoring of mechanical systems based on hierarchical symbolic transfer entropy is proposed, aiming to improve the health monitoring and fault diagnosis capabilities of mechanical systems under different working conditions. Through the feature extraction strategy - hierarchical symbolic transfer entropy, the problem of poor resolvability of existing entropy-based feature extraction methods in the face of complex mechanical fault signals is solved, thereby realizing a more accurate and efficient health state assessment.

[0141] 3. A mechanical system fault diagnosis framework integrating hierarchical symbolic transfer entropy and two-dimensional extreme learning machine is proposed. This framework can construct a second-order tensor representation by stacking the hierarchical symbolic transfer entropy values extracted from the sub-signal channel data through a two-dimensional extreme learning machine, so as to obtain the multi-channel hierarchical symbolic transfer entropy values. The two-dimensional extreme learning machine designed in this disclosure has low requirements for the random weights of the hidden layer, reduces the space and computational complexity, and enhances the classification ability.

[0142] Corresponding to the above-mentioned mechanical system structural health monitoring method based on hierarchical symbolic transfer entropy, the present invention also proposes a mechanical system structural health monitoring device based on hierarchical symbolic transfer entropy. Since the device embodiment of the present invention corresponds to the above-mentioned method embodiment, for the details not disclosed in the device embodiment, reference can be made to the above-mentioned method embodiment, and no further description will be given in the present invention.

[0143] Figure 10 The structural schematic diagram of a mechanical system structural health monitoring device provided by an embodiment of this disclosure is shown in Figure 10 as follows, including:

[0144] An acquisition unit 21, configured to acquire multi-channel vibration data of the electromechanical system in different health states;

[0145] A calculation unit 22, configured to calculate the hierarchical symbolic transfer entropy of each sub-channel vibration signal in the multi-channel vibration data;

[0146] A first stacking unit 23, configured to perform stacking calculation on the hierarchical symbolic transfer entropy of each level of each sub-channel vibration signal to obtain the second-order tensor features corresponding to each sub-channel vibration signal;

[0147] A second stacking unit 24, configured to stack the second-order tensor features corresponding to each sub-channel vibration signal to obtain a second-order tensor feature matrix corresponding to the multi-channel vibration data;

[0148] A training unit 25, configured to train an initial two-dimensional extreme learning machine based on the second-order tensor feature matrix belonging to the training data set to obtain a trained two-dimensional extreme learning machine;

[0149] A monitoring unit 26, configured to monitor the health state of the electromechanical system based on the two-dimensional extreme learning machine.

[0150] The mechanical system structural health monitoring device based on hierarchical symbol transfer entropy provided by the present disclosure collects vibration data of multiple channels of an electromechanical system in different health states; calculates the hierarchical symbol transfer entropy of each sub-channel vibration signal in the multiple-channel vibration data; performs superposition calculation on the hierarchical symbol transfer entropy of each level of each sub-channel vibration signal to obtain a second-order tensor feature corresponding to each sub-channel vibration signal; superimposes the second-order tensor features corresponding to each sub-channel vibration signal to obtain a second-order tensor feature matrix corresponding to the multiple-channel vibration data; trains an initial two-dimensional extreme learning machine based on the second-order tensor feature matrix belonging to the training data set to obtain a trained two-dimensional extreme learning machine; monitors the health state of the electromechanical system based on the two-dimensional extreme learning machine. Compared with the related technology, by introducing hierarchical symbol transfer entropy as the characteristic data for monitoring the health state of the electromechanical system, the difference between the fault signal and the healthy signal in the vibration signal can be better distinguished, and the accuracy of monitoring the health state of the electromechanical system is significantly improved. By using a two-dimensional extreme learning machine that can effectively process high-dimensional data, the accuracy of monitoring the health state of the mechanical system is further improved and the robustness of the health state monitoring model is enhanced.

[0151] It should be noted that the foregoing explanation of the method embodiments also applies to the device of this embodiment, with the same principle, and will not be limited in this embodiment.

[0152] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0153] Figure 11 FIG. shows a schematic block diagram of an exemplary electronic device 300 that can be used to implement embodiments of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as, for example, a laptop computer, a desktop computer, a workbench, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as, for example, a personal digital processor, a cellular phone, a smart phone, a wearable device, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely exemplary and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0154] As Figure 11As shown, device 300 includes a computing unit 301, which can perform various appropriate actions and processes according to computer programs stored in a ROM (Read-Only Memory) 302 or computer programs loaded from a storage unit 308 into a RAM (Random Access Memory) 303. In the RAM 303, various programs and data required for the operation of device 300 can also be stored. The computing unit 301, ROM 302, and RAM 303 are connected to each other via a bus 304. An I / O (Input / Output) interface 305 is also connected to the bus 304.

[0155] Multiple components in device 300 are connected to the I / O interface 305, including: an input unit 306, such as a keyboard, mouse, etc.; an output unit 307, such as various types of displays, speakers, etc.; a storage unit 308, such as a disk, optical disc, etc.; and a communication unit 309, such as a network card, modem, wireless communication transceiver, etc. The communication unit 309 allows device 300 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0156] The computing unit 301 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 301 include but are not limited to a CPU (Central Processing Unit), a GPU (Graphic Processing Units), various dedicated AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, a DSP (Digital Signal Processor), and any appropriate processor, controller, microcontroller, etc. The computing unit 301 executes the various methods and processes described above, such as the mechanical system structural health monitoring method based on hierarchical symbolic transfer entropy. For example, in some embodiments, the mechanical system structural health monitoring method based on hierarchical symbolic transfer entropy can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as the storage unit 308. In some embodiments, part or all of the computer program can be loaded and / or installed onto device 300 via the ROM 302 and / or the communication unit 309. When the computer program is loaded into the RAM 303 and executed by the computing unit 301, one or more steps of the method described above can be executed. Alternatively, in other embodiments, the computing unit 301 can be configured to execute the aforementioned mechanical system structural health monitoring method based on hierarchical symbolic transfer entropy in any other appropriate manner (e.g., by means of firmware).

[0157] The various embodiments of the systems and techniques described above in this specification can be implemented in digital electronic circuitry, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application Specific Standard Products), SOCs (System On Chip), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be a special-purpose or general-purpose programmable processor that receives data and instructions from a storage system, at least one input device, and at least one output device, and transmits the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0158] The program code for implementing the methods of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the program codes are executed by the processor or controller, the functions / operations specified in the flowchart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0159] In the context of this disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer disk, a hard disk, a RAM, a ROM, an EPROM (Electrically Programmable Read-Only Memory), or a flash memory, an optical fiber, a CD-ROM (Compact Disc Read-Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0160] To provide for interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can also be used to provide for interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, speech input, or tactile input).

[0161] The systems and techniques described herein can be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a LAN (Local Area Network), a WAN (Wide Area Network), the Internet, and a blockchain network.

[0162] A computer system may include a client and a server. The client and the server are generally far from each other and usually interact via a communication network. The client-server relationship is generated by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system, solving the defects of difficult management and weak business scalability existing in traditional physical hosts and VPS services ("Virtual Private Server", or simply "VPS" for short). The server can also be a server of a distributed system or a server combined with blockchain.

[0163] Among them, it should be noted that artificial intelligence is a discipline that studies how to make a computer simulate certain thinking processes and intelligent behaviors of humans (such as learning, reasoning, thinking, planning, etc.), and it has both hardware-level technologies and software-level technologies. Artificial intelligence hardware technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, and big data processing; artificial intelligence software technologies mainly include several major directions such as computer vision technology, speech recognition technology, natural language processing technology, and machine learning / deep learning, big data processing technology, and knowledge graph technology.

Claims

1. A method for structural health monitoring of mechanical systems based on hierarchical symbol transfer entropy, characterized in that: include: Collect multi-channel vibration data of electromechanical systems in different health states; Calculating the hierarchical sign transfer entropy of each sub-channel vibration signal in the plurality of channel vibration data; Performing superposition calculation on the hierarchical symbol transfer entropy of each level of each sub-channel vibration signal to obtain the second-order tensor feature corresponding to each sub-channel vibration signal; Superimposing the second-order tensor features corresponding to each of the sub-channel vibration signals to obtain the second-order tensor feature matrices corresponding to the plurality of channel vibration data; Training an initial two-dimensional extreme learning machine based on the second-order tensor feature matrix belonging to the training data set to obtain a trained two-dimensional extreme learning machine; The health status of the electromechanical system is monitored based on the two-dimensional extreme learning machine.

2. The method according to claim 1, characterized in that The calculating of the hierarchical symbol transfer entropy of each sub-channel vibration signal in the plurality of channel vibration data comprises: At each level in the preset depth level, each sub-channel vibration signal is subjected to different levels of smoothing and differential processing to obtain vibration signals of different frequency bands corresponding to each sub-channel vibration signal; Based on a preset step size threshold or a preset sliding window, the vibration signals of the different frequency bands are respectively symbolized to obtain symbol sequences corresponding to the vibration signals of the different frequency bands; Calculate the hierarchical entropy between the symbol sequences corresponding to vibration signals of different frequency bands in a sub-channel vibration signal, calculate the hierarchical symbol transfer entropy based on the hierarchical entropy corresponding to each level, and so on to obtain the hierarchical symbol transfer entropy of each sub-channel vibration signal.

3. The method according to claim 2, characterized in that The formula for the smoothing process is expressed as: Q0=(x n +x n+1 ) / 2n=1,2,…,N-1 The differential processing formula is expressed as: Q1=(x n -x n+1 ) / 2n=1,2,…,N-1 Among them, x n and x n+1 is a continuous data point of the sub-channel vibration signal, between which are continuous sampling values ​​in the sub-channel vibration signal, the sub-channel vibration signal is time series data, the sub-channel vibration signal is any one of the sub-channel vibration signals, x n is the nth data point of the vibration signal in the subchannel, and x n+1 is immediately followed by x n The subsequent (n+1)th data point, Q0, represents the vibration signal of the low-frequency band, which includes low-frequency fault information, and is obtained by comparing two consecutive data points x n and x n+1 Smoothing is performed to obtain Q1, which represents the vibration signal of the high-frequency band part, which includes high-frequency fault information, and is obtained by performing smoothing on two consecutive data points x n and x n+1 Differential processing is performed to obtain the vibration signals of different frequency bands including the vibration signal of the low-frequency band part and the vibration signal of the high-frequency band part, n=1, 2, ..., N-1 represents the time series index of the sub-channel vibration signal, ranging from 1 to N-1, where N is the total number of data points.

4. The method according to claim 3, characterized in that The formula of the fault feature extraction operator included in any level of the preset depth level is expressed as: in, f={0,1} is the fault feature extraction operator of the kth layer, represents a matrix, where the subscript f is a label, the preset depth level is k, and k represents a specific stage in the hierarchical analysis, k is a positive integer, and the symbol express is a real matrix whose dimensions depend on the value of k and the total number of data points N.

5. The method according to claim 4, characterized in that The step of calculating the hierarchical entropy between the symbol sequences corresponding to the vibration signals of different frequency bands in the sub-channel vibration signal comprises: Calculate the node components of the nodes where each symbol sequence is located in each level of the preset depth level; Calculating the hierarchical entropy based on the node components in each hierarchical level; The calculation formula of the node component is: Among them, X k,e represents the node component at the kth node e in the preset depth hierarchy, r k represents the frequency band selected in the kth layer, Indicates that in the k-th layer frequency band r k The operator matrix extracted from the above is used to extract the features of the vibration signal. Vibration data, the node component is used to reflect the frequency band characteristics of the level to which it belongs; The calculation formula for node e is: Among them, r j represents the average operator or difference operator of the jth layer, 2 k-j is a weighting factor that decreases as j increases and is used to adjust the influence between different levels. r represents the selected frequency band; The calculation formula of the hierarchical entropy is: Among them, E 层次熵 represents the hierarchical entropy, m is the number of levels of the vibration signal, represents the range of the summation operation, ste represents the calculation symbol of the hierarchical entropy, X k is the node component of the kth layer, representing the fault information in different frequency bands, and τ represents the number of transfer steps.

6. The method according to claim 5, characterized in that The formula for calculating the hierarchical symbol transfer entropy based on the hierarchical entropy corresponding to each hierarchical level is expressed as: Among them, E 层次符号转移熵 represents the hierarchical symbol transfer entropy, E 层次熵 represents the hierarchical entropy, m is the number of levels of the vibration signal, and k represents the preset depth level.

7. The method according to any one of claims 1 to 6, characterized in that: The formula of the two-dimensional extreme learning machine is expressed as: in, represents the output weight vector connecting the lth hidden unit, represents the vth projection vector connecting the lth hidden node, where v = 1 or 2, v = 1 represents low-frequency features, v = 2 represents high-frequency features, and n v is the dimension of the projection vector, indicating the number of elements contained in the layer vector. Indicates that the vector belongs to n v dimensional real space, b l ∈R represents the bias of the lth hidden unit, g(·) is the sigmoid function, X n is the input signal or data point, t n is the output or target value, which represents the output result or predicted value after signal processing, L is the total number of layers, n represents the nth sample or data point, and C is the number of categories of the output result or predicted value.

8. A mechanical system structural health monitoring device based on hierarchical symbol transfer entropy, characterized in that: include: A collection unit, used to collect vibration data of multiple channels of the electromechanical system in different health states; A calculation unit, used for calculating the hierarchical symbol transfer entropy of each sub-channel vibration signal in the plurality of channel vibration data; A first superposition unit is used to perform superposition calculation on the hierarchical symbol transfer entropy of each level of each sub-channel vibration signal to obtain a second-order tensor feature corresponding to each sub-channel vibration signal; A second superposition unit is used to superimpose the second-order tensor features corresponding to each of the sub-channel vibration signals to obtain a second-order tensor feature matrix corresponding to each of the plurality of channel vibration data; A training unit, used for training an initial two-dimensional extreme learning machine based on the second-order tensor feature matrix belonging to the training data set to obtain a trained two-dimensional extreme learning machine; A monitoring unit is used to monitor the health status of the electromechanical system based on the two-dimensional extreme learning machine.

9. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-7.

Images