Anchor strength detection method and system

Abstract

This invention provides a method and system for anchor bolt strength detection, relating to the field of anchor bolt strength detection technology. The method includes: first, preprocessing and denoising the original signal using fractional Fourier transform and local projection matrix methods to obtain a denoised signal; then, inputting the denoised signal into a feature extraction model to extract multidimensional features, including time complexity features, instantaneous frequency features, and recursive features; next, obtaining sparse representation features through feature concatenation and dimensionality reduction; inputting the sparse representation features into a prediction model, combining them with a Gaussian process regression model for strength prediction, and calculating uncertainty information; finally, filtering and deleting abnormal prediction values ​​using an anomaly detection model to obtain accurate anchor bolt strength detection values. This invention's detection method can improve the accuracy and reliability of anchor bolt strength prediction and is suitable for long-term monitoring and safety assessment in complex environments.

Description

Technical Field

[0001] The present invention relates to the technical field of anchor rod strength detection, and in particular to an anchor rod strength detection method and system. Background Art

[0002] At present, traditional anchor strength detection methods mostly rely on mechanical or manual mechanical testing instruments, mainly including pull-out tests, pressure sensor monitoring and other methods. These methods require a large amount of field data collection and field operations, and often ignore the dynamic response characteristics of anchors in actual projects. At the same time, due to the complex interaction between the stress state of the anchor and the surrounding environment (such as soil, rock, etc.), the traditional static detection method cannot timely reflect the strength changes of the anchor under actual load. In recent years, basalt fiber anchors, as a new type of material, have been widely used in underground projects due to their excellent mechanical properties and corrosion resistance. Compared with traditional steel anchors, basalt fiber anchors have higher strength-to-weight ratios, better corrosion resistance and lower maintenance costs, and are particularly suitable for special environments such as groundwater wetness and saline-alkali land. However, basalt fiber anchors face more complex technical challenges in strength monitoring and detection. Due to the differences in its material properties from traditional steel anchors, traditional detection methods are difficult to adapt to its dynamic response and complex stress behavior, especially in different construction environments and long-term service. The strength decay and performance changes of anchors are difficult to accurately evaluate.

[0003] Therefore, there is an urgent need for an anchor strength detection method and system to solve the technical problems of strength detection of traditional steel anchors and basalt fiber anchors. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for detecting the strength of an anchor rod to improve the above-mentioned problem. In order to achieve the above-mentioned purpose, the technical solution adopted by the present invention is as follows:

[0005] In a first aspect, the present application provides a method for detecting the strength of an anchor rod, comprising:

[0006] Obtaining the original anchor bolt dynamic response signal collected by the sensor;

[0007] Preprocessing and denoising the original anchor bolt dynamic response signal, wherein the original anchor bolt dynamic response signal is preprocessed by fractional Fourier transform, and the preprocessed signal is denoised by using a preset local projection matrix to obtain a denoised anchor bolt dynamic response signal;

[0008] The denoised anchor bolt dynamic response signal is sent to a preset feature extraction model for feature extraction to obtain multi-dimensional features, wherein the multi-dimensional features include a time complexity feature vector, an instantaneous frequency feature vector, and a recursive feature vector;

[0009] Performing feature concatenation on the multi-dimensional features, and performing dimensionality reduction and sparse representation classification on the concatenated high-dimensional feature vector to obtain sparse representation features;

[0010] The sparse representation features are sent to a preset prediction model for training and prediction, and the predicted anchor strength prediction value and its uncertainty information are processed by a preset Gaussian process regression model to obtain a predicted anchor strength value;

[0011] The sparse representation features and all the predicted strength values ​​of the anchor rods are sent to an anomaly detection model for anomaly detection, and the abnormal predicted strength value data is deleted to obtain the strength detection value of the anchor rod.

[0012] In a second aspect, the present application also provides an anchor strength detection system, comprising:

[0013] An acquisition unit, used for acquiring an original anchor bolt dynamic response signal collected by a sensor;

[0014] A first processing unit is used to preprocess and reduce noise on the original anchor bolt dynamic response signal, wherein the original anchor bolt dynamic response signal is preprocessed by fractional Fourier transform, and the preprocessed signal is reduced noise by using a preset local projection matrix to obtain a denoised anchor bolt dynamic response signal;

[0015] A second processing unit is used to send the denoised anchor bolt dynamic response signal to a preset feature extraction model for feature extraction to obtain multi-dimensional features, wherein the multi-dimensional features include a time complexity feature vector, an instantaneous frequency feature vector, and a recursive feature vector;

[0016] A third processing unit is used to perform feature concatenation on the multi-dimensional features, and perform dimensionality reduction and sparse representation classification on the high-dimensional feature vector obtained by concatenation to obtain sparse representation features;

[0017] A fourth processing unit is used to send the sparse representation features to a preset prediction model for training and prediction, and to process the predicted anchor strength prediction value and its uncertainty information through a preset Gaussian process regression model to obtain a predicted anchor strength value;

[0018] The fifth processing unit is used to send the sparse representation features and all the predicted strength values ​​of the anchor rods to the anomaly detection model for anomaly detection, and delete the abnormal predicted strength value data to obtain the strength detection value of the anchor rod.

[0019] The beneficial effects of the present invention are:

[0020] This method combines advanced signal processing and machine learning technologies such as fractional Fourier transform, adaptive filtering, feature extraction algorithm, sparse representation classification and Gaussian process regression to extract key information from the dynamic response signal of the anchor in real time and accurately, and combine the anomaly detection model to efficiently predict the anchor strength. Compared with traditional detection methods, the anchor strength detection method of this application not only has higher detection accuracy and real-time performance, but also can effectively eliminate the interference of external environmental factors on the measurement results, providing more reliable protection for engineering safety, especially in the dynamic response and performance evaluation of basalt fiber anchors, it has significant advantages.

[0021] Since the application environment of basalt fiber anchors is often complex and their service life is long, abnormal signals caused by sensor failure or environmental interference may appear during long-term monitoring. Through the abnormal detection model in the present invention, these inaccurate prediction data can be automatically identified and deleted, ensuring the stability and reliability of the anchor strength detection value, avoiding misjudgment caused by individual abnormal data, and improving the robustness of the entire system under complex working conditions.

[0022] Basalt fiber anchors have different mechanical properties from traditional steel anchors, and their dynamic response signals may present more complex nonlinear characteristics. By using fractional Fourier transform based on the minimum entropy criterion and combining recursive feature analysis, the present invention can identify and extract the unique dynamic response patterns of basalt fiber anchors, and transform these features into high-dimensional feature vectors through feature extraction models, thereby improving the ability to recognize its complex response patterns.

[0023] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or be understood by implementing the embodiments of the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0025] Figure 1 A schematic diagram of the anchor strength detection method according to an embodiment of the present invention;

[0026] Figure 2 It is a schematic diagram of the structure of the anchor strength detection system described in an embodiment of the present invention.

[0027] In the figure: 701, acquisition unit; 702, first processing unit; 703, second processing unit; 704, third processing unit; 705, fourth processing unit; 706, fifth processing unit. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0029] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0030] Embodiment 1:

[0031] This embodiment provides a method for detecting the strength of an anchor rod.

[0032] See also Figure 1 , the figure shows that the method includes step S1, step S2, step S3, step S4, step S5 and step S6.

[0033] Step S1, obtaining the original anchor bolt dynamic response signal collected by the sensor;

[0034] It is understood that in this step, "anchor strength" refers to the maximum ability of the anchor to maintain its structural integrity and function under external loads or other forces. Specifically, anchor strength reflects the ability of the anchor to support rock mass, tunnels or other supporting structures, and is an important indicator for evaluating whether the anchor can work normally and maintain structural safety in engineering applications.

[0035] It is understandable that the dynamic response signal of the anchor is generally collected in real time by sensors installed on the anchor (such as strain gauges, accelerometers or displacement sensors, etc.), which can detect the physical response of the anchor under different loads, such as stress, displacement and vibration. These signals reflect the stress state and deformation of the anchor under actual working conditions, and can provide valuable raw data for anchor health monitoring. In order to ensure the accuracy and effectiveness of the data, the selection and layout of sensors are crucial. Generally speaking, sensors should be arranged in key positions, such as the joints of the anchor, places where stress is concentrated, or areas where damage may occur. The original dynamic response signal often contains noise, interference and unstable factors. Therefore, when collecting signals, the influence of environmental conditions, such as temperature changes, electromagnetic interference and other factors, should be considered. The collected original signal provides basic data for subsequent signal preprocessing and feature extraction. The technical effect of this step is to obtain original, comprehensive and potential information-rich dynamic response data, which lays a solid foundation for subsequent signal processing and strength prediction.

[0036] The sensor collects the dynamic response of basalt fiber anchors and traditional steel anchors in real time after being affected by external loads, geological conditions or environmental fluctuations during operation. The sensor transmits the collected signals to the central processing unit through the data acquisition system. The signals are usually analog signals, which are converted and transmitted to the computer for analysis and storage.

[0037] Among them, the installation of sensors can avoid stress concentration by selecting appropriate bonding methods and materials, such as using flexible adhesives or mechanical clamping methods to ensure that the impact on the anchor structure is minimized. When installing sensors, non-penetrating installation methods can be adopted to ensure that holes do not appear in areas with greater stress. During the wiring process, flexible cables and non-destructive fixing methods can be used to avoid tight contact between the cable and the anchor surface or improper fixing, so as to avoid additional stress on the anchor.

[0038] Step S2, preprocessing and denoising the original anchor bolt dynamic response signal, wherein the original anchor bolt dynamic response signal is preprocessed by fractional Fourier transform, and the preprocessed signal is denoised by using a preset local projection matrix to obtain a denoised anchor bolt dynamic response signal;

[0039] It can be understood that the fractional Fourier transform improves the signal capture capability in a time-varying environment by optimizing the time-frequency representation of the signal, so that subsequent processing can better reflect the dynamic characteristics of the anchor, especially the performance in nonlinear or complex vibration modes. Local projection matrix denoising provides a denoising method for local features, so that the signal can retain its important features while denoising, avoiding the loss of useful information that may be caused by global denoising. In this step, step S2 includes step S21, step S22, step S23 and step S24.

[0040] Step S21, determining the optimal fractional angle of the original anchor bolt dynamic response signal based on the minimum entropy criterion, wherein the original anchor bolt dynamic response signal is subjected to a short-time Fourier transform, and the optimal fractional angle is determined by a preset minimum entropy angle determination formula;

[0041] It can be understood that the short-time Fourier transform converts the original anchor dynamic response signal into a time-frequency domain representation. In order to make up for the limitations of the short-time Fourier transform, the fractional-order Fourier transform is introduced, which provides a more flexible conversion method between the time-frequency domain and can adjust the representation of the signal according to different fractional-order angles. By introducing a fractional-order angle, the fractional-order Fourier transform can show the characteristics of the signal at different time-frequency resolutions, thereby improving the ability to capture non-stationary and transient signals when processing anchor signals. The minimum entropy criterion is used as a signal optimization method to determine the fractional-order angle that is most suitable for signal representation.

[0042] Among them, the minimum entropy angle determination formula is as follows:

[0043] ;

[0044] ;

[0045] in, is the entropy value, is the candidate fractional angle, is the amplitude of the fractional-order signal, is the number of signal sample points, For the A fractional-order signal, is the fractional angle that minimizes the entropy value.

[0046] Step S22, performing a fractional-order Fourier transform on the optimal fractional-order angle to obtain a fractional-order domain signal corresponding to the original anchor bolt dynamic response signal;

[0047] Among them, the formula for calculating the fractional Fourier transform is as follows:

[0048] ;

[0049] in, represents the fractional-order signal, represents the original signal, represents the kernel function, Indicates time.

[0050] It can be understood that by applying the optimal fractional angle to the original signal, the fractional Fourier transform maps the signal from the traditional time domain to the fractional domain. In this domain, the time-frequency characteristics of the signal are described more flexibly and meticulously. According to the selected fractional angle, the different frequency components of the signal will be more reasonably distributed to the time axis and frequency axis, thereby revealing details such as the instantaneous characteristics, mutation points and local changes of the signal. After the fractional Fourier transform, the fractional domain representation of the original signal can more clearly show the dynamic behavior of the signal at different frequencies and time scales. This representation is particularly important for the analysis of the dynamic response signal of the anchor bolt, because such signals usually have non-stationary, time-varying and nonlinear characteristics, and conventional time domain or frequency domain analysis methods may not be able to fully capture these characteristics.

[0051] Step S23, sending the fractional-order domain signal corresponding to the original anchor bolt dynamic response signal to a preset adaptive filter for noise reduction, to obtain the fractional-order domain signal after initial noise reduction;

[0052] It can be understood that the adaptive filter is different from the traditional linear filter in that its filter coefficients are not pre-set, but are automatically adjusted according to the characteristics of the input signal. By minimizing the mean square error, the adaptive filter can adjust its parameters in real time to optimize the signal processing effect. In the previous step, the fractional Fourier transform has converted the original anchor dynamic response signal into a fractional domain signal. At this stage, the time-frequency characteristics of the signal are optimized, which can provide a more accurate frequency domain representation. However, since different types of noise often exist in actual signals, such as sensor errors, environmental interference, etc., these noises may affect the subsequent analysis results. Therefore, it is necessary to eliminate these noise components through adaptive filters.

[0053] Step S24, performing fuzzy clustering segmentation and local projection denoising on the fractional-order domain signal after the initial denoising in sequence, and merging the obtained projection denoising signals of all regions to obtain the denoised dynamic response signal of the anchor bolt.

[0054] It can be understood that fuzzy clustering is a technique for dividing data points into different clusters. It is different from the traditional hard clustering method. The fuzzy clustering method allows a data point to belong to multiple clusters, and each cluster has a membership value. This method can effectively handle fuzzy boundaries and uncertainties in signals. Specifically, the process of fuzzy clustering segmentation includes: determining multiple cluster centers of the signal through the optimization algorithm fuzzy C-means algorithm, calculating the membership of each signal sample to each cluster center, and then forming the distribution of clusters, and segmenting the signal according to the membership value to form multiple sub-signal areas. After the signal is divided into multiple regions, each region is further denoised by local projection denoising. Local projection denoising is a spatial filtering technology that removes local noise components by projecting and filtering the signal in a local area. The core of this process is to project the signal into a low-dimensional space, retain the main components of the signal, and suppress noise that is not related to the signal. The projection method is performed by the least squares method to ensure the accurate restoration of the local signal.

[0055] Step S3, sending the denoised anchor bolt dynamic response signal to a preset feature extraction model for feature extraction to obtain multidimensional features, wherein the multidimensional features include a time complexity feature vector, an instantaneous frequency feature vector, and a recursive feature vector;

[0056] It can be understood that by extracting multi-scale permutation entropy, instantaneous frequency and recursive features, the time domain and frequency domain information of the signal can be fully captured, providing more abundant and accurate input features for subsequent intensity prediction and anomaly detection. These features can not only effectively reflect the time-varying characteristics, frequency changes and recursive relationships of the signal, but also have high robustness and can maintain good performance in various complex environments. In this step, step S3 includes step S31, step S32 and step S34.

[0057] Step S31, performing time scale coarse-graining on the denoised anchor dynamic response signal, wherein the denoised signal is coarse-grained to generate subsequences of different time scales, and then the time complexity feature vector is calculated by multi-scale permutation entropy to obtain the time complexity feature vector of the denoised anchor dynamic response signal;

[0058] It can be understood that for each subsequence at each time scale, it is first converted into a symbol sequence or a permutation sequence, that is, the signal values ​​are sorted by size to generate a corresponding permutation pattern. Then the complexity of the signal is quantified by calculating the probability distribution of the permutation pattern. The permutation entropy is calculated at different time scales to obtain the complexity measure at that scale. Finally, the multi-scale permutation entropy is the weighted or averaged result of the permutation entropy values ​​at different scales, which represents the complexity characteristics of the signal at multiple levels of scale.

[0059] The calculation formula of permutation entropy is as follows:

[0060] ;

[0061] in, is the permutation entropy of the denoised signal, is the embedding dimension, is the similarity tolerance, is the signal length, is the vector logarithm of the embedding dimension, is the vector logarithm of the next embedding dimension.

[0062] Step S32, extracting instantaneous frequency feature vectors of the subsequences of different time scales by Hilbert-Huang transform, and obtaining instantaneous frequency feature vectors of the dynamic response signal of the anchor bolt after denoising;

[0063] It can be understood that the mathematical expression of the Hilbert transform is as follows:

[0064] ;

[0065] in, is the original signal The Hilbert transform of represents the imaginary unit, represents the signal after Hilbert transform.

[0066] Among them, the mathematical expression of the instantaneous frequency eigenvector is as follows:

[0067] ;

[0068] in, represents the instantaneous frequency eigenvector of the anchor bolt dynamic response signal, Indicates time.

[0069] Step S33: performing dimensional recursive analysis based on the instantaneous frequency eigenvector, wherein the instantaneous frequency eigenvector is subjected to recursive feature calculation to obtain a recursive eigenvector of the denoised anchor bolt dynamic response signal.

[0070] It can be understood that first, the instantaneous frequency eigenvector is mapped to the high-dimensional phase space. The goal of this step is to make the time series of the signal show richer structural features. In the high-dimensional space, the temporal characteristics of the signal are more finely displayed, so that more subtle dynamic behaviors can be revealed. By recursively calculating the instantaneous frequency eigenvector in the high-dimensional embedding space, the delayed coordinate embedding method is used to further evaluate the dynamic behavior of the signal in the nonlinear system. Then the distance between the points in the phase space is calculated, and the recursive graph is constructed. Then, the diagonal length distribution is extracted from the recursive graph to reflect the predictability of the signal, and the vertical line length distribution is extracted to reflect the non-stationarity of the signal. The entropy value is extracted to measure the complexity of the structure in the recursive graph.

[0071] Step S4, performing feature splicing on the multi-dimensional features, and performing dimensionality reduction and sparse representation classification on the high-dimensional feature vector obtained by splicing to obtain sparse representation features;

[0072] It is understandable that different types of features are integrated into a high-dimensional vector to form a vector that comprehensively describes the signal features. Through the combination of feature concatenation, dimensionality reduction and sparse representation, the expression of features is optimized, so that subsequent model training can more efficiently extract key information for anchor strength prediction. The processing of dimensionality reduction and sparse representation can greatly reduce computational complexity, reduce memory consumption, and improve the accuracy and robustness of anchor strength prediction. It is understandable that in this step, step S4 includes step S41, step S42 and step S43.

[0073] Step S41, concatenating the time complexity feature vector, the instantaneous frequency feature vector and the recursive feature vector to obtain a high-dimensional feature vector;

[0074] It can be understood that this step directly connects vectors of different features to form a new, longer feature vector. The concatenation process can be regarded as a fusion of signal information, which integrates information of multiple dimensions and provides a more comprehensive signal description.

[0075] Step S42, mapping the high-dimensional feature vector to a low-dimensional space by using a t-distributed random neighbor embedding method to obtain a reduced-dimensional feature vector;

[0076] Since the dimension of the concatenated feature vector is high, directly using these high-dimensional features will face the problem of dimensionality disaster, which will affect the efficiency of subsequent calculations and the performance of the model. Therefore, dimensionality reduction is performed through t-distributed random neighbor embedding. Here, the main purpose of dimensionality reduction is to map high-dimensional features to a lower-dimensional space, while retaining the main information of the data and removing redundant information to reduce computational complexity and improve data interpretability. Through this dimensionality reduction process, the most representative features can be extracted to ensure that subsequent processing is more efficient. Among them, this step achieves high-dimensional to low-dimensional mapping by optimizing the following loss function:

[0077] ;

[0078] in, represents the value of the loss function, Representing data points in high-dimensional space and data points The conditional probability of Represents data points in low-dimensional space and data points The conditional probability of The combinations represented are summed.

[0079] Step S43: According to the sparse representation classification algorithm, the reduced-dimensional feature vector is represented as a sparse linear combination of the dictionary matrix, and the sparse coefficient vector is solved based on a preset minimization reconstruction error method and a preset sparsity constraint to obtain a sparse representation feature.

[0080] It is understandable that after dimensionality reduction, the obtained feature vectors will be classified by sparse representation method. The key to sparse representation is to represent the feature vector as a linear combination of dictionary matrices, and the goal is to reconstruct the signal by finding the least non-zero coefficients. Specifically, the process of sparse representation involves solving the sparse coefficient vector by minimizing the reconstruction error and sparsity constraints. The formula for solving the sparse coefficient vector is as follows:

[0081] ;

[0082] in, is a sparse coefficient vector, express of norm, To adjust the parameters, is the pre-learned dictionary matrix, Indicates signal characteristics.

[0083] Step S5, sending the sparse representation features to a preset prediction model for training and prediction, and processing the predicted anchor strength prediction value and its uncertainty information through a preset Gaussian process regression model to obtain a predicted anchor strength value;

[0084] It is understandable that this step, by combining the Gaussian process regression model, can not only obtain accurate anchor strength prediction values, but also quantify the credibility of the prediction results. In this way, a strength prediction with high credibility can be provided, while providing important support for risk assessment and decision-making in engineering applications. Especially in complex or unknown environments, the model can reveal the potential risks of the prediction results through uncertainty information, thereby enhancing the robustness and practicality of the model. In this step, step S5 includes step S51, step S52 and step S53.

[0085] Step S51, training a preset prediction model based on a preset correlation vector machine learning method and the sparse representation feature, and predicting the anchor strength value and uncertainty information through the trained prediction model, wherein the uncertainty information is a variance value of the anchor strength value;

[0086] It can be understood that the relevance vector machine is an extension of the support vector machine, which provides uncertainty estimation in the prediction process through the Bayesian method. Compared with the traditional support vector machine, the relevance vector machine can not only give the prediction results, but also calculate the uncertainty (i.e. variance) of the prediction, where the posterior distribution of each training sample is calculated through Bayesian inference, so that the prediction function of the model is obtained as follows:

[0087] ;

[0088] in, Represents the prediction result, is the weight vector, represents the feature mapping function, Represents the bias term.

[0089] The calculation formula for uncertain information is as follows:

[0090] ;

[0091] in, represents the variance, represents the feature matrix, The inverse of the precision matrix representing the parameters.

[0092] Step S52: constructing a Gaussian process regression model based on the sparse representation features and the anchor strength value, sending the uncertain information to the Gaussian regression model, and calculating the predicted mean and confidence interval of the anchor strength by the formula in the Gaussian regression model;

[0093] It is understandable that Gaussian process regression can not only provide accurate prediction values ​​of anchor strength, but also quantify the uncertainty of the prediction, helping decision makers to judge the reliability of anchor health status. This provides more comprehensive information for risk assessment and maintenance decision-making in actual projects. Among them, the calculation formula for the predicted mean value of anchor strength is as follows:

[0094] ;

[0095] in, Indicates the predicted point The predicted mean of Indicates the predicted point and training dataset The covariance vector between Represents the training data set The covariance matrix between represents the variance of the noise, Represents the training data values ​​of the target variable.

[0096] Among them, the confidence interval (i.e. variance) calculation formula of anchor strength is as follows:

[0097] ;

[0098] in, Indicates the predicted point The predicted mean of Indicates the predicted point and training dataset The covariance vector between Represents the training data set The covariance matrix between represents the variance of the noise, Represents the training data set and prediction points The covariance vector between .

[0099] Step S53: multiply the lower limit of the preset confidence interval weight by the preset weight, and obtain the predicted strength value of the anchor rod by subtracting the weighted confidence interval value from the predicted mean.

[0100] It can be understood that this step significantly improves the accuracy and reliability of anchor strength prediction, especially when facing complex environmental factors or data uncertainties. It can effectively avoid erroneous decisions caused by excessively high or low prediction errors, thereby providing more robust results in practical applications.

[0101] Step S6: Send the sparse representation features and all the predicted strength values ​​of the anchor rods to an anomaly detection model for anomaly detection, and delete the abnormal predicted strength value data to obtain the strength detection value of the anchor rod.

[0102] It can be understood that this step identifies and eliminates abnormal data points that are significantly deviated from normal data patterns, thereby improving the accuracy and reliability of anchor strength detection. In this step, step S6 includes step S61, step S62 and step S63.

[0103] Step S61, concatenating the sparse representation features and all the predicted strength values ​​of the anchors into a joint feature vector, and constructing a preset number of random trees, randomly selecting features and dividing samples each time, calculating the isolated path length of each sample and the average neighboring distance of the sample within the k-nearest neighbors, and obtaining samples whose isolated path length is less than a preset threshold and whose neighboring distance is greater than a preset threshold;

[0104] It can be understood that this step learns the structure of the data by constructing a certain number of random trees (each tree randomly selects features and divides samples). The construction process of each tree is carried out independently, and the diversity of the data is simulated by randomly selecting a subset of the data. Calculate the isolated path length: The isolated path length refers to the path length required for a sample to be isolated in a tree. Simply put, if a sample is easily divided into a separate branch by the tree, its isolated path length will be shorter; conversely, if the sample is located in a dense area of ​​the data set, its isolated path length will be longer. Therefore, the isolated path length can be used to measure whether a sample is an outlier, where abnormal samples usually have shorter isolated paths.

[0105] The k-nearest neighbor algorithm analyzes the local density of a sample by calculating the distance between the sample and its nearest neighbor. The smaller the average neighbor distance of a sample, the denser the area it is in; on the contrary, a larger average neighbor distance may mean that the sample is located in a sparse area and is a potential outlier.

[0106] It can be understood that by considering both the isolated path length and the proximity distance, the algorithm can comprehensively evaluate the "isolation" and "density" of the sample in the data distribution, thereby more accurately identifying anomalies. Compared with traditional methods that rely on only a single criterion, this dual standard improves the accuracy of anomaly detection.

[0107] Step S62, constructing an abnormal candidate set from samples whose isolated path lengths are less than a preset threshold and whose neighboring distances are greater than a preset threshold, training a sparse autoencoder with the abnormal candidate set, and filtering abnormal prediction strength value data with the trained sparse autoencoder;

[0108] It can be understood that the sparse autoencoder imposes sparsity constraints during the training process, usually through L1 regularization, forcing the network to generate sparse hidden layer representations during the learning process. This sparsity helps to extract the most representative information in the data, so that the network can better capture the characteristics of abnormal samples. During the training process, the sparse autoencoder learns the model by minimizing the error between the input data and the reconstructed data. Samples with large reconstruction errors often indicate that they have a large deviation from most normal samples in the feature space, so they are usually considered to be abnormal data. During the training process of the abnormal candidate set, outliers can be further screened out by calculating the reconstruction error of each sample. Once the sparse autoencoder is trained, abnormal samples can be identified by its reconstruction ability. Specifically, samples with large reconstruction errors will be regarded as abnormal samples. They are located in a relatively separated area from other samples in the data space, so they may be abnormal signals caused by external factors (such as damage, breakage, etc.). The training goal of the sparse autoencoder in this step is to minimize the reconstruction error. If the reconstruction error exceeds the set threshold, it is determined to be a true abnormal point.

[0109] Step S63, delete the abnormal predicted strength value data to obtain the strength detection value of the anchor rod.

[0110] It can be understood that by deleting abnormal data, erroneous predictions caused by external interference, equipment failure or other uncontrollable factors can be effectively removed, system errors can be reduced, and the accuracy and reliability of anchor strength detection can be improved.

[0111] Embodiment 2:

[0112] like Figure 2 As shown, this embodiment provides an anchor strength detection system, see Figure 2 The system includes an acquisition unit 701 , a first processing unit 702 , a second processing unit 703 , a third processing unit 704 , a fourth processing unit 705 and a fifth processing unit 706 .

[0113] An acquisition unit 701 is used to acquire an original anchor bolt dynamic response signal collected by a sensor;

[0114] The first processing unit 702 is used to preprocess and reduce noise on the original anchor bolt dynamic response signal, wherein the original anchor bolt dynamic response signal is preprocessed by fractional Fourier transform, and the preprocessed signal is reduced noise by using a preset local projection matrix to obtain a denoised anchor bolt dynamic response signal;

[0115] The second processing unit 703 is used to send the denoised anchor bolt dynamic response signal to a preset feature extraction model for feature extraction to obtain multi-dimensional features, wherein the multi-dimensional features include a time complexity feature vector, an instantaneous frequency feature vector, and a recursive feature vector;

[0116] The third processing unit 704 is used to perform feature concatenation on the multi-dimensional features, and perform dimension reduction and sparse representation classification on the high-dimensional feature vector obtained by concatenation to obtain sparse representation features;

[0117] The fourth processing unit 705 is used to send the sparse representation features to a preset prediction model for training and prediction, and process the predicted anchor strength prediction value and its uncertainty information through a preset Gaussian process regression model to obtain a predicted anchor strength value;

[0118] The fifth processing unit 706 is used to send the sparse representation features and all the predicted strength values ​​of the anchor rods to the anomaly detection model for anomaly detection, and delete the abnormal predicted strength value data to obtain the strength detection value of the anchor rod.

[0119] It should be noted that, regarding the system in the embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0120] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0121] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for detecting anchor strength, characterized in that: include: Obtaining the original anchor bolt dynamic response signal collected by the sensor; Preprocessing and denoising the original anchor bolt dynamic response signal, wherein the original anchor bolt dynamic response signal is preprocessed by fractional Fourier transform, and the preprocessed signal is denoised by using a preset local projection matrix to obtain a denoised anchor bolt dynamic response signal; The denoised anchor bolt dynamic response signal is sent to a preset feature extraction model for feature extraction to obtain multi-dimensional features, wherein the multi-dimensional features include a time complexity feature vector, an instantaneous frequency feature vector, and a recursive feature vector; Performing feature concatenation on the multi-dimensional features, and performing dimensionality reduction and sparse representation classification on the concatenated high-dimensional feature vector to obtain sparse representation features; The sparse representation features are sent to a preset prediction model for training and prediction, and the predicted anchor strength prediction value and its uncertainty information are processed by a preset Gaussian process regression model to obtain a predicted anchor strength value; The sparse representation features and all the predicted strength values ​​of the anchor rods are sent to an anomaly detection model for anomaly detection, and the abnormal predicted strength value data is deleted to obtain the strength detection value of the anchor rod; The sparse representation features are sent to a preset prediction model for training and prediction, and the predicted anchor strength prediction value and its uncertainty information are processed by a preset Gaussian process regression model to obtain the predicted anchor strength value, including: Based on the preset correlation vector machine learning method and the sparse representation feature, a preset prediction model is trained, and the anchor strength value and uncertainty information are predicted by the trained prediction model, wherein the uncertainty information is the variance value of the anchor strength value; A Gaussian process regression model is constructed based on the sparse representation features and the anchor strength value, and the uncertain information is sent to the Gaussian regression model, and the predicted mean and confidence interval of the anchor strength are calculated by the formula in the Gaussian regression model; The lower limit of the preset confidence interval weight and the preset weight are multiplied to calculate the product, and the predicted strength value of the anchor rod is obtained by subtracting the weighted confidence interval value from the predicted mean.

2. The anchor strength detection method according to claim 1 is characterized in that , preprocessing and denoising the original anchor bolt dynamic response signal, wherein the original anchor bolt dynamic response signal is preprocessed by fractional Fourier transform, and the preprocessed signal is denoised by using a preset local projection matrix to obtain a denoised anchor bolt dynamic response signal, including: Determine the optimal fractional angle of the original anchor bolt dynamic response signal based on the minimum entropy criterion, wherein the original anchor bolt dynamic response signal is subjected to a short-time Fourier transform, and the optimal fractional angle is determined by a preset minimum entropy angle determination formula; Performing fractional Fourier transform on the optimal fractional angle to obtain the fractional domain signal corresponding to the original anchor bolt dynamic response signal; Sending the fractional-order domain signal corresponding to the original anchor bolt dynamic response signal to a preset adaptive filter for noise reduction to obtain the fractional-order domain signal after initial noise reduction; The fractional-order domain signal after the initial denoising is subjected to fuzzy clustering segmentation and local projection denoising in sequence, and the obtained projection denoising signals of all regions are merged to obtain the denoised dynamic response signal of the anchor bolt.

3. The anchor strength detection method according to claim 1, characterized in that ,The denoised anchor dynamic response signal is sent to a preset feature extraction model for feature extraction to obtain multidimensional features, which include time complexity feature vectors, instantaneous frequency feature vectors and recursive feature vectors, including: The denoised anchor bolt dynamic response signal is subjected to time scale coarse-graining, wherein the denoised signal is subjected to coarse-graining processing to generate subsequences of different time scales, and then the time complexity feature vector is calculated by multi-scale permutation entropy to obtain the time complexity feature vector of the denoised anchor bolt dynamic response signal; Extracting instantaneous frequency feature vectors of the subsequences of different time scales by Hilbert-Huang transform to obtain instantaneous frequency feature vectors of the dynamic response signal of the anchor bolt after denoising; A dimensional recursive analysis is performed based on the instantaneous frequency eigenvector, wherein the recursive eigenvector of the denoised dynamic response signal of the anchor bolt is obtained by performing recursive feature calculation on the instantaneous frequency eigenvector.

4. The anchor strength detection method according to claim 1, characterized in that , the multi-dimensional features are concatenated, and the concatenated high-dimensional feature vectors are subjected to dimensionality reduction and sparse representation classification to obtain sparse representation features, including: Concatenating the time complexity feature vector, the instantaneous frequency feature vector and the recursive feature vector to obtain a high-dimensional feature vector; The high-dimensional feature vector is mapped to a low-dimensional space through the t-distributed random neighbor embedding method to obtain a reduced-dimensional feature vector; According to the sparse representation classification algorithm, the reduced-dimensional feature vector is represented as a sparse linear combination of a dictionary matrix, and the sparse coefficient vector is solved based on a preset minimization reconstruction error method and a preset sparsity constraint to obtain a sparse representation feature.

5. An anchor strength detection system, characterized in that: include: An acquisition unit, used for acquiring an original anchor bolt dynamic response signal collected by a sensor; A first processing unit is used to preprocess and reduce noise on the original anchor bolt dynamic response signal, wherein the original anchor bolt dynamic response signal is preprocessed by fractional Fourier transform, and the preprocessed signal is reduced noise by using a preset local projection matrix to obtain a denoised anchor bolt dynamic response signal; A second processing unit is used to send the denoised anchor bolt dynamic response signal to a preset feature extraction model for feature extraction to obtain multi-dimensional features, wherein the multi-dimensional features include a time complexity feature vector, an instantaneous frequency feature vector, and a recursive feature vector; A third processing unit is used to perform feature concatenation on the multi-dimensional features, and perform dimensionality reduction and sparse representation classification on the high-dimensional feature vector obtained by concatenation to obtain sparse representation features; A fourth processing unit is used to send the sparse representation features to a preset prediction model for training and prediction, and to process the predicted anchor strength prediction value and its uncertainty information through a preset Gaussian process regression model to obtain a predicted anchor strength value; A fifth processing unit, configured to send the sparse representation features and all the predicted strength values ​​of the anchor rods to an anomaly detection model for anomaly detection, and delete the abnormal predicted strength value data to obtain the strength detection value of the anchor rod; Wherein, the fourth processing unit includes: A first prediction subunit is used to train a preset prediction model based on a preset correlation vector machine learning method and the sparse representation feature, and predict the anchor strength value and uncertainty information through the trained prediction model, wherein the uncertainty information is a variance value of the anchor strength value; The second prediction subunit is used to construct a Gaussian process regression model based on the sparse representation features and the anchor strength value, and send the uncertainty information to the Gaussian regression model to calculate the predicted mean and confidence interval of the anchor strength through the formula in the Gaussian regression model; The third prediction subunit is used to calculate the product of the lower limit of the preset confidence interval weight and the preset weight, and obtain the predicted strength value of the anchor rod by subtracting the weighted confidence interval value from the predicted mean.

6. The anchor strength detection system according to claim 5, characterized in that: The first processing unit comprises: The first processing subunit is used to determine the optimal fractional angle of the original anchor bolt dynamic response signal based on the minimum entropy criterion, wherein the original anchor bolt dynamic response signal is subjected to a short-time Fourier transform, and the optimal fractional angle is determined by a preset minimum entropy angle determination formula; The second processing subunit is used to perform a fractional-order Fourier transform on the optimal fractional-order angle to obtain a fractional-order domain signal corresponding to the original anchor bolt dynamic response signal; The third processing sub-unit is used to send the fractional-order domain signal corresponding to the original anchor bolt dynamic response signal to a preset adaptive filter for noise reduction to obtain the fractional-order domain signal after initial noise reduction; The fourth processing subunit is used to perform fuzzy clustering segmentation and local projection denoising on the fractional-order domain signal after the initial denoising in sequence, and merge the obtained projection denoising signals of all regions to obtain the denoised dynamic response signal of the anchor bolt.

7. The anchor strength detection system according to claim 5, characterized in that: The second processing unit comprises: The first calculation subunit is used to perform time scale coarse-graining on the denoised anchor bolt dynamic response signal, wherein the denoised signal is coarse-grained to generate subsequences of different time scales, and then the time complexity feature vector is calculated by multi-scale permutation entropy to obtain the time complexity feature vector of the denoised anchor bolt dynamic response signal; The second calculation subunit is used to extract the instantaneous frequency feature vector of the subsequences of different time scales by Hilbert-Huang transform to obtain the instantaneous frequency feature vector of the dynamic response signal of the anchor bolt after denoising; The third calculation subunit is used to perform dimensional recursive analysis based on the instantaneous frequency feature vector, wherein the recursive feature vector of the anchor bolt dynamic response signal after denoising is obtained by performing recursive feature calculation on the instantaneous frequency feature vector.

8. The anchor strength detection system according to claim 5, characterized in that: The third processing unit comprises: A fifth processing subunit, configured to concatenate the time complexity feature vector, the instantaneous frequency feature vector and the recursive feature vector to obtain a high-dimensional feature vector; A sixth processing subunit, used for mapping the high-dimensional feature vector to a low-dimensional space by using a t-distributed random neighbor embedding method to obtain a reduced-dimensional feature vector; The seventh processing subunit is used to represent the reduced-dimensional feature vector as a sparse linear combination of a dictionary matrix according to a sparse representation classification algorithm, and solve the sparse coefficient vector based on a preset minimization reconstruction error method and a preset sparsity constraint to obtain a sparse representation feature.

Images