Steel wire rope anomaly detection method and system based on multichannel data reconstruction

Abstract

The invention discloses a steel wire rope anomaly detection method and system based on multichannel data reconstruction, and belongs to the technical field of steel wire rope anomaly detection.The method comprises the steps that after the detection position of a steel wire rope is determined, historical state information of the detection position is obtained, the historical state information is preprocessed, and a training data set is obtained; constructing a detection model, and performing model training on the detection model through the training data set; obtaining to-be-detected state information of the detection position, inputting the to-be-detected state information into the trained detection model, and obtaining output data; obtaining a reconstruction scoring function through the to-be-detected state information and the output data, and obtaining an adaptive threshold value through the reconstruction scoring function; and comparing the to-be-detected state information with the adaptive threshold, and determining that the detection position corresponding to the to-be-detected state information outside the adaptive threshold is an abnormal position.

Description

Technical Field

[0001] The present application belongs to the technical field of wire rope anomaly detection, and specifically relates to a wire rope anomaly detection method and system based on multi-channel data reconstruction. Background Art

[0002] Steel wire ropes come in a variety of structures, typically made from high-quality high-carbon steel or alloy steel wires twisted according to a specific helical pattern. This unique twisting process imparts excellent bending properties and a high surface modulus of elasticity. Furthermore, steel wire ropes offer advantages such as light weight, excellent elasticity, and corrosion resistance, making them widely used in mine hoisting, metallurgical hoisting, cable-stayed bridges, cable cars, and aerospace applications.

[0003] As a key load-bearing component, wire rope plays an important role in many fields. However, during use, it is prone to defects such as wire breakage, deformation, fatigue, cracks, and wear. These defects not only reduce the service life of the wire rope but can also cause serious safety accidents. Therefore, it is crucial to promptly detect and address abnormal conditions in wire ropes.

[0004] However, wire rope damage detection faces numerous challenges. For one thing, the diversity of wire rope manufacturing processes, material selection, and structural design complicates damage patterns and increases the difficulty of detection. Furthermore, existing detection equipment suffers from low accuracy and high costs, making it difficult to meet practical application requirements. Currently, many organizations still rely on manual inspections and regular wire rope replacements to ensure safety. However, these methods are not only time-consuming and labor-intensive, but also fail to accurately detect internal damage and can result in financial waste. Summary of the Invention

[0005] Purpose of the invention: This application develops a wire rope anomaly detection method and system based on multi-channel data reconstruction, aiming to solve the above technical problems.

[0006] Technical Solution: In the first aspect, the present application provides a wire rope anomaly detection method based on multi-channel data reconstruction, comprising: Determining a detection position of the wire rope and obtaining historical status information of the detection position; Preprocessing the historical status information to obtain a training data set; Constructing a detection model, and performing model training on the detection model based on the training data set; Acquiring the state information of the detection position to be tested; Inputting the state information to be tested into the trained detection model to obtain output data; Acquire a reconstruction scoring function based on the state information to be measured and the output data; Obtaining an adaptive threshold based on the reconstruction scoring function; The state information to be measured is compared with the adaptive threshold, and a detection position corresponding to the state information to be measured that is outside the adaptive threshold is determined as an abnormal position.

[0007] In some embodiments, the state information of the steel wire rope includes multi-channel state information, and the detection model includes a feature enhancement module, and the feature enhancement module includes: The first submodule is configured to: determining a first feature, where the first feature is a potential feature space vector of the multi-channel state information; Determining state information of a specified channel from the multi-channel state information, and extracting a second feature from the state information of the specified channel; Processing the second feature based on a fully connected layer to obtain a third feature; Performing weighted processing on the first feature and the third feature based on a cross attention mechanism to obtain a fourth feature; Performing adaptive weight feature fusion on the third feature and the fourth feature, and obtaining a fifth feature by combining the residual connection of the output feature of the fully connected layer; A second submodule is connected to the first submodule, and the second submodule is used to: extracting a sixth feature from state information of a next channel adjacent to the designated channel; Processing the sixth feature based on the fully connected layer to obtain a seventh feature; Performing weighted processing on the seventh feature and the fifth feature based on the cross attention mechanism to obtain an eighth feature; Adaptive weight feature fusion is performed on the seventh feature and the eighth feature, and a ninth feature is obtained by combining the residual connection of the output feature of the fully connected layer.

[0008] In some embodiments, the detection model further includes: A multi-channel parallel convolutional encoder module, the multi-channel parallel convolutional encoder module is connected to the feature enhancement module, and the multi-channel parallel convolutional encoder module is used to: Obtain a two-dimensional convolution kernel based on the number of channels, where the number of rows of the two-dimensional convolution kernel is the same as the number of channels; Dividing the data of each row of the two-dimensional convolution kernel into independent row convolution sub-kernels; Match each row convolution sub-kernel to the corresponding channel, and calculate the convolution result of the state information of each channel. Based on the multiple convolution results, obtaining a potential feature space vector of the multi-channel state information; A multi-channel parallel convolution decoder module is connected to the feature enhancement module and is used to increase the dimension of the output data of the feature enhancement module through the fully connected layer and perform decoding reconstruction.

[0009] In some embodiments, the characterization formula for calculating the convolution result of the state information of each of the channels includes: ; in, The output The channel is located at The convolution result at ; For the input The channel is located at The data at is the position offset, is the number of positions; is the weight of the row convolution kernel.

[0010] In some embodiments, the detection model further includes: A channel position encoding module, the channel position encoding module is connected to the multi-channel parallel convolution encoder module, and the channel position encoding module is used to: Obtain input data of the plurality of channels, and match a corresponding position encoding vector for each channel; Concatenate the input data of each channel and the position encoding vector to obtain single-channel state information; Acquire the multi-channel state information based on the plurality of single-channel state information; A channel position decoding module, connected to the multi-channel parallel convolution decoder module, is used to: The output data of the multi-channel parallel convolution decoder module is obtained, and the data with the position coding vector is reconstructed and generated, and the channel to which it belongs is determined based on the position coding vector.

[0011] In some embodiments, when inputting the state information to be measured into the trained detection model, the channel position encoding module is used to: Obtaining input data of a plurality of channels; splicing the input data of each channel and the position coding vector to obtain single-channel state information; wherein the position coding vector of each channel is the same; Concatenate the input data of each channel and the position encoding vector to obtain single-channel state information; The multi-channel state information is acquired based on a plurality of the single-channel state information.

[0012] In some embodiments, the step of obtaining an adaptive threshold based on the reconstructed scoring function includes: Determine a confidence limit, and obtain the adaptive threshold based on the confidence limit, where the characterization formula of the adaptive threshold includes: ; in, is the adaptive threshold; are the confidence limits; is the location parameter, which is used to characterize the central position of the distribution; is the scale parameter, which is used to characterize the degree of dispersion of the distribution; is the shape parameter, which determines the tail shape of the distribution; The extreme value samples are fitted to the generalized extreme value distribution model based on maximum likelihood estimation to obtain the location parameter, the scale parameter, and the shape parameter. The characterization formula of the maximum likelihood estimation includes: ; in, is the maximum likelihood estimate; is the value of the extreme sample. In some embodiments, the step of determining that the detection location corresponding to the state information to be measured is outside the adaptive threshold is an abnormal location includes: In the data with the position coding vector reconstructed and generated by the channel position decoding module, determining that the data outside the adaptive threshold is abnormal data; An abnormal channel is confirmed based on the reconstructed position encoding vector corresponding to the abnormal data.

[0013] In some embodiments, the step of preprocessing the historical state information to obtain a training data set includes: Treat the channel as the first type of row data and the time as the column data, and sort the historical status information of the same channel in the same row based on the time sequence; in multiple column data, sort the sum of the data points in the same column based on the time sequence to generate the second type of row data and obtain the historical data matrix; Performing adaptive double exponential detrending processing on the historical data matrix to obtain detrended data; The detrended data is subjected to maximum pooling processing to obtain normalized data, and the maximum eigenvalue is extracted while maintaining the relative proportion of the data, and then a sliding window is used to perform data interception to obtain the training data set.

[0014] In a second aspect, an embodiment of the present application further provides a wire rope anomaly detection system based on multi-channel data reconstruction, comprising: A historical information acquisition module, the historical information acquisition module is used to determine the detection position of the wire rope and obtain historical status information of the detection position; A historical data processing module, configured to pre-process the historical status information to obtain a training data set; A model training module, wherein the historical data processing module is used to construct a detection model and perform model training on the detection model based on the training data set; A test information acquisition module, the test information acquisition module is used to obtain the test state information of the detection position; A reconstruction prediction module, wherein the reconstruction prediction module is used to input the state information to be measured into the trained detection model to obtain output data; A reconstruction scoring module, wherein the reconstruction scoring module is used to obtain a reconstruction scoring function based on the state information to be measured and the output data; A threshold acquisition module, configured to acquire an adaptive threshold based on the reconstructed scoring function; The abnormality locating module is used to compare the state information to be measured with the adaptive threshold, and determine that the detection position corresponding to the state information to be measured that is outside the adaptive threshold is an abnormal position.

[0015] Beneficial Effects: Compared with the prior art, the embodiment of the present application provides a wire rope anomaly detection method based on multi-channel data reconstruction, which includes obtaining historical state information of the detection position after determining the detection position of the wire rope, preprocessing the historical state information, and obtaining a training data set; constructing a detection model and training the detection model using the training data set; obtaining the state information to be tested at the detection position, and inputting the state information to be tested into the trained detection model to obtain output data; obtaining a reconstruction scoring function based on the state information to be tested and the output data, and obtaining an adaptive threshold based on the reconstruction scoring function; comparing the state information to be tested with the adaptive threshold, and determining that the detection position corresponding to the state information to be tested that is outside the adaptive threshold is an abnormal position. The present application obtains historical state information at the detection position of the wire rope, and uses the historical state information as training data to train the detection model, so that the trained detection model can recognize the state information of the wire rope and output reconstructed data; then, the trained detection model obtains the output result of the test data, obtains the adaptive threshold based on the output result and the reconstruction function of the state information to be tested, and determines that the position corresponding to the state information to be tested that is outside the adaptive threshold is an abnormal position, thereby detecting the abnormal position of the wire rope. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0017] Figure 1 A flowchart of the steps of a wire rope anomaly detection method based on multi-channel data reconstruction provided in an embodiment of the present application; Figure 2 This is a functional flow chart of the first submodule in the wire rope anomaly detection method based on multi-channel data reconstruction provided in an embodiment of the present application; Figure 3 This is a functional flow chart of the second submodule in the wire rope anomaly detection method based on multi-channel data reconstruction provided in an embodiment of the present application; Figure 4 This is a functional flow chart of a multi-channel parallel convolutional encoder module in a wire rope anomaly detection method based on multi-channel data reconstruction provided in an embodiment of the present application; Figure 5 This is a functional flow chart of the channel position encoding module in the wire rope anomaly detection method based on multi-channel data reconstruction provided by an embodiment of the present application when training the detection model; Figure 6 A functional flow chart of the channel position encoding module in the wire rope anomaly detection method based on multi-channel data reconstruction provided in an embodiment of the present application when inputting the state information to be tested into the trained detection model; Figure 7 A flowchart of the steps for preprocessing historical status information in a wire rope anomaly detection method based on multi-channel data reconstruction provided in an embodiment of the present application; Figure 8 A flowchart of the steps for determining an abnormality location in the wire rope abnormality detection method based on multi-channel data reconstruction provided in an embodiment of the present application; Figure 9 A schematic diagram of a wire rope anomaly detection method based on multi-channel data reconstruction provided in an embodiment of the present application; Figure 10 A module connection diagram of a wire rope anomaly detection system based on multi-channel data reconstruction provided in an embodiment of the present application; Figure numerals: 10, historical information acquisition module; 20, historical data processing module; 30, model training module; 40, test information acquisition module; 50, reconstruction prediction module; 60, reconstruction scoring module; 70, threshold acquisition module; 80, anomaly location module. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the embodiments described are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0019] Steel wire ropes come in a variety of structures, typically made from high-quality high-carbon steel or alloy steel wires twisted according to a specific helical pattern. This unique twisting process imparts excellent bending properties and a high surface modulus of elasticity. Furthermore, steel wire ropes offer advantages such as light weight, excellent elasticity, and corrosion resistance, making them widely used in mine hoisting, metallurgical hoisting, cable-stayed bridges, cable cars, and aerospace applications.

[0020] As a key load-bearing component, wire rope plays an important role in many fields. However, during use, it is prone to defects such as wire breakage, deformation, fatigue, cracks, and wear. These defects not only reduce the service life of the wire rope but can also cause serious safety accidents. Therefore, it is crucial to promptly detect and address abnormal conditions in wire ropes.

[0021] However, wire rope damage detection faces numerous challenges. For one thing, the diversity of wire rope manufacturing processes, material selection, and structural design complicates damage patterns and increases the difficulty of detection. Furthermore, existing detection equipment suffers from low accuracy and high costs, making it difficult to meet practical application requirements. Currently, many organizations still rely on manual inspections and regular wire rope replacements to ensure safety. However, these methods are not only time-consuming and labor-intensive, but also fail to accurately detect internal damage and can result in financial waste.

[0022] In view of this, an embodiment of the present application provides a wire rope anomaly detection method based on multi-channel data reconstruction, comprising: obtaining historical state information of the detection position after determining the detection position of the wire rope, preprocessing the historical state information, and obtaining a training data set; constructing a detection model and training the detection model using the training data set; obtaining the state information to be tested at the detection position, and inputting the state information to be tested into the trained detection model to obtain output data; obtaining a reconstruction scoring function based on the state information to be tested and the output data, and obtaining an adaptive threshold based on the reconstruction scoring function; comparing the state information to be tested with the adaptive threshold, and determining that the detection position corresponding to the state information to be tested that is outside the adaptive threshold is an abnormal position. The present application obtains historical state information at the detection position of the wire rope, and uses the historical state information as training data to train the detection model, so that the trained detection model can recognize the state information of the wire rope and output reconstructed data; then, the trained detection model obtains the output result of the test data, obtains the adaptive threshold based on the output result and the reconstruction function of the state information to be tested, and determines that the position corresponding to the state information to be tested that is outside the adaptive threshold is an abnormal position, thereby detecting the abnormal position of the wire rope.

[0023] In some embodiments, see Figure 1 and Figure 9 , Figure 1 A flowchart of the steps of the wire rope anomaly detection method based on multi-channel data reconstruction provided in an embodiment of the present application is provided. Figure 9 This is a schematic diagram of a wire rope anomaly detection method based on multi-channel data reconstruction provided in an embodiment of the present application. The wire rope anomaly detection method based on multi-channel data reconstruction provided in an embodiment of the present application is specifically implemented through steps 100 to 800: Step 100: Determine the detection position of the wire rope and obtain historical status information of the detection position.

[0024] In some embodiments, the data collection method of the embodiments of the present application specifically includes: identifying six locations on the wire rope as detection locations, placing sensors at these locations, and acquiring electromagnetic signals from these locations via the sensors. Each sensor represents a data collection channel, and the collected data is single-channel data. A total of 12 data points numbered 0 to 11 are collected and stored in MAT format. The past data collected using this data collection method is used as historical status information.

[0025] Step 200: Preprocess the historical status information to obtain a training data set.

[0026] In some embodiments, see Figure 7 , Figure 7This is a flowchart of the steps for preprocessing historical status information in the wire rope anomaly detection method based on multi-channel data reconstruction provided by an embodiment of the present application. The method for preprocessing historical status information in the present application is specifically implemented through steps 210 to 230: Step 210: Take the channel as the first type of row data and the time as the column data, and sort the historical status information of the same channel in the same row based on the time sequence; in multiple column data, sort the sum of the data points in the same column based on the time sequence to generate the second type of row data and obtain the historical data matrix.

[0027] In some embodiments, the present application arranges the historical status information of 6 channels in sequence as a type of row data, and sorts the historical status information collected at different times of the same channel in chronological order, sums the data points in the same column to obtain the 7th row data, which is the second type of row data, and finally generates a complete historical data matrix.

[0028] Step 220: Perform adaptive double exponential detrending processing on the historical data matrix to obtain detrended data.

[0029] In some embodiments, when performing adaptive double-exponential detrending processing on the historical data matrix, the short-term moving average and the long-term exponentially weighted moving average of the historical data matrix are calculated respectively, the short-term moving average and the long-term exponentially weighted moving average are linearly combined to obtain a comprehensive trend estimate, and the comprehensive trend estimate is subtracted from the historical data matrix to obtain detrended data, so that short-term fluctuations and abnormal characteristics are more prominent.

[0030] Step 230: Perform maximum pooling on the detrended data to obtain normalized data, extract the maximum eigenvalue while maintaining the relative proportion of the data, and then use a sliding window to intercept the data to obtain a training data set.

[0031] It can be understood that in order to improve the stability and efficiency of model training, this application uses the maximum pooling method to normalize the detrended data and scale the data to the range of [-1, 1]. By extracting the maximum value features in the data while maintaining the relative proportion of the data, the model's ability to capture important features is enhanced while reducing computational complexity. The normalized data is intercepted using a sliding window technique, and a sliding window with a segment length of 20 and a step length of 10 is selected to intercept data segments in turn to construct a test data set, from which some normal data segments are screened as training set data sets for model training and optimization.

[0032] Step 300: Build a detection model and perform model training on the detection model based on the training data set.

[0033] In some embodiments, the state information of the wire rope in the embodiment of the present application includes multi-channel state information, the detection model includes a feature enhancement module, and the feature enhancement module includes a first submodule, see Figure 2 , Figure 2 This is a functional flow chart of the first submodule in the wire rope anomaly detection method based on multi-channel data reconstruction provided in an embodiment of the present application. The first submodule in the present application is specifically used to: Step 301: confirming a first feature, where the first feature is a potential feature space vector of multi-channel state information.

[0034] Step 302: Determine state information of a designated channel in the multi-channel state information, and extract a second feature from the state information of the designated channel.

[0035] Step 303: Process the second feature based on the fully connected layer to obtain the third feature.

[0036] Step 304: Perform weighted processing on the first feature and the third feature based on the cross-attention mechanism to obtain the fourth feature.

[0037] Step 305: Perform adaptive weight feature fusion on the third feature and the fourth feature, and obtain the fifth feature by combining the residual connection of the output feature of the fully connected layer.

[0038] In some embodiments, the feature enhancement module further includes a second submodule, which is connected to the first submodule. Figure 3 , Figure 3 This is a functional flow chart of the second submodule in the wire rope anomaly detection method based on multi-channel data reconstruction provided in an embodiment of the present application. The second submodule in the present application is specifically used to: Step 306: extracting a sixth feature from the state information of the next channel adjacent to the designated channel.

[0039] Step 307: Process the sixth feature based on the fully connected layer to obtain the seventh feature.

[0040] Step 308: Perform weighted processing on the seventh feature and the fifth feature based on the cross-attention mechanism to obtain the eighth feature.

[0041] Step 309: Adaptively weight feature fusion is performed on the seventh feature and the eighth feature, and the ninth feature is obtained by combining the residual connection of the output feature of the fully connected layer.

[0042] It can be understood that the present application performs feature enhancement on multi-channel state information through the first submodule and the second submodule, so that the model can better capture abnormal features in the data.

[0043] In some embodiments, the detection model in the wire rope anomaly detection method based on multi-channel data reconstruction provided by the embodiment of the present application further includes a multi-channel parallel convolution encoder module, which is connected to the feature enhancement module. Figure 4 , Figure 4 This is a functional flow chart of a multi-channel parallel convolutional encoder module in a wire rope anomaly detection method based on multi-channel data reconstruction provided by an embodiment of the present application. The multi-channel parallel convolutional encoder module is specifically used to: Step 310: Obtain a two-dimensional convolution kernel based on the number of channels, where the number of rows of the two-dimensional convolution kernel is the same as the number of channels.

[0044] Step 311: Divide the data of each row of the two-dimensional convolution kernel into independent row convolution sub-kernels.

[0045] Step 312: Match each row convolution sub-kernel to the corresponding channel, and calculate the convolution result of the state information of each channel.

[0046] In some embodiments, a representation formula for calculating the convolution result of the state information of each channel includes: ; in, The output Channels at position The convolution result at ; For the input Channels at position The data at is the position offset, is the number of positions; is the weight of the row convolution kernel.

[0047] Step 313: Based on the multiple convolution results, obtain the potential feature space vector of the multi-channel state information.

[0048] It can be understood that this application designs a multi-channel parallel convolution encoder module for input data with a two-dimensional structure to extract low-dimensional features of the data. Unlike the two-dimensional convolution kernel, this module adopts a multi-row convolution kernel design, that is, a OK The two-dimensional convolution kernel of the column is divided into Independent row convolution sub-kernels, that is, single-channel convolution sub-kernels, each single-channel convolution sub-kernel corresponds to a channel of the input data, and extracts the features of the input data of different channels respectively. During training, the input channel is randomly masked and the random input At the same time, the adaptive kernel clipping strategy is adopted when the number of input channels is Less than the maximum number of rows of the convolution kernel Before activation The weights of the remaining single-channel convolution sub-kernels remain frozen. Taking the 7 rows of data in the example as an example, the 2D convolution kernel can be set to 7 rows. The convolution kernel can still work normally when only inputting data from any channel. Each single-channel convolution sub-kernel operates independently and in parallel to process the data of each channel simultaneously, while maintaining the independence of each channel data and the integrity of local features. It can be seen that the present application extracts the potential features of the input data through a multi-channel parallel convolution encoder, and finally obtains the potential feature space vector representation of its low-dimensional mapping.

[0049] In some embodiments, the detection model in the wire rope anomaly detection method based on multi-channel data reconstruction provided in the embodiments of the present application also includes a multi-channel parallel convolution decoder module, which is connected to the feature enhancement module and is used to increase the dimension of the output data of the feature enhancement module through a fully connected layer and perform decoding and reconstruction.

[0050] The multi-channel parallel convolution decoder module in this application is structurally symmetrical with the multi-channel parallel convolution encoder module. The input of the multi-channel parallel convolution decoder module is the latent feature vector after feature enhancement. The dimension of the feature is increased through the fully connected layer, and the latent feature vector after feature enhancement is decoded into reconstructed data with position encoding.

[0051] The loss function in this application A combination of mean square error (MSE) and weighted absolute error is used to balance the impact of global and local errors. The characterization formulas include:

[0052] in, is the number of data samples; is the sorting number of; is the input data value; To reconstruct data values; is the weighting coefficient; and is the weight coefficient.

[0053] In some embodiments, the detection model in the wire rope anomaly detection method based on multi-channel data reconstruction provided by the embodiment of the present application further includes a channel position encoding module, which is connected to the multi-channel parallel convolution encoder module. Figure 5 , Figure 5 This is a functional flow chart of the channel position encoding module in the wire rope anomaly detection method based on multi-channel data reconstruction provided in an embodiment of the present application when training the detection model. The channel position encoding module is specifically used to: Step 314: Obtain input data of multiple channels and match a corresponding position encoding vector for each channel.

[0054] Step 315: Concatenate the input data and position encoding vector of each channel to obtain single-channel state information.

[0055] Step 316: Obtain multi-channel status information based on the plurality of single-channel status information.

[0056] Specifically, when training the detection model, the channel position encoding module assigns a fixed unique position encoding vector to each channel. When a 7×20 data segment is input, the 20 data points of each channel are concatenated with the position encoding vector of the corresponding channel to obtain a 7×27 coded data with position information. Assume that the data of channel 1 is , the position encoding vector is , concatenate the encoding vector and the data vector to form a feature vector with position information , similarly, the feature vector after channel 2 encoding is , and so on.

[0057] In some embodiments, see Figure 6 , Figure 6 This is a functional flow chart of the channel position encoding module in the wire rope anomaly detection method based on multi-channel data reconstruction provided by an embodiment of the present application when inputting the state information to be measured into the trained detection model. The channel position encoding module is specifically used to: Step 318: Obtain input data of multiple channels.

[0058] Step 319: concatenate the input data and position coding vectors of each channel to obtain single-channel state information; wherein the position coding vectors of each channel are the same.

[0059] Step 320: Concatenate the input data and position encoding vector of each channel to obtain single-channel state information.

[0060] Step 321: Acquire multi-channel status information based on multiple single-channel status information.

[0061] Specifically, when the state information to be tested is input into the trained detection model, the channel position encoding module will encode the channel data of any input position. , the position encoding vectors are , the encoded data is The reconstructed channel position coding information is cut separately to obtain the channel position coding vector. When determining the channel position, if a bit of the channel position coding vector is closest to 1, it is determined to be the channel data.

[0062] In some embodiments, the detection model in the wire rope anomaly detection method based on multi-channel data reconstruction provided in the embodiments of the present application also includes a channel position decoding module, which is connected to the multi-channel parallel convolution decoder module to obtain the output data of the multi-channel parallel convolution decoder module, and reconstruct and generate data with a position coding vector, and determine the channel to which it belongs based on the position coding vector.

[0063] It can be understood that in the detection model under training, the detection model is trained by inputting training data of the position coding vector with channel distinguishing features, so that the detection model can output data of the position coding vector containing the channel distinguishing features after training; then in the trained detection model, the data to be tested in which the position coding vector does not have the channel distinguishing features is reconstructed to generate normal data of the position coding vector with the channel distinguishing features, and the reconstructed position coding vector with the channel distinguishing features is cut separately. When judging the channel position, if a certain bit code of the channel position coding vector is closest to 1, it is judged to be this channel data.

[0064] Step 400: Acquire the state information of the detection position to be tested.

[0065] Step 500: Input the state information to be tested into the trained detection model to obtain output data.

[0066] Specifically, in the trained detection model, data reconstruction is performed on the test data without the position coding vector of the channel distinguishing feature to generate normal data with the position coding vector of the channel distinguishing feature.

[0067] Step 600: Obtain a reconstruction scoring function based on the state information to be measured and the output data.

[0068] Step 700: Obtain an adaptive threshold based on the reconstruction scoring function.

[0069] In some embodiments, the method for obtaining an adaptive threshold based on a reconstructed scoring function is implemented by the following steps: Determine the confidence limit and obtain the adaptive threshold based on the confidence limit. The characterization formula of the adaptive threshold includes: ; in, is the adaptive threshold; is the confidence limit; is the location parameter, which is used to characterize the central position of the distribution; is the scale parameter, which is used to characterize the degree of dispersion of the distribution; is the shape parameter, which determines the tail shape of the distribution; Among them, based on the maximum likelihood estimation, the extreme value samples are fitted to the generalized extreme value distribution model to obtain the location parameter, scale parameter and shape parameter. The characterization formula of the maximum likelihood estimation includes: ; in, is the maximum likelihood estimate; is the value of the extreme sample.

[0070] Step 800: Compare the state information to be measured with the adaptive threshold, and determine that the detection position corresponding to the state information to be measured that is outside the adaptive threshold is an abnormal position.

[0071] In some embodiments, see Figure 8 , Figure 8 This is a flowchart of the steps for determining an abnormal position in the wire rope abnormality detection method based on multi-channel data reconstruction provided in an embodiment of the present application. The method for determining that the detection position corresponding to the state information to be measured is located outside the adaptive threshold is an abnormal position is specifically implemented through steps 810 to 820: Step 810: In the data with the position coding vector reconstructed and generated by the channel position decoding module, determine that the data outside the adaptive threshold is abnormal data.

[0072] Step 820: Identify the abnormal channel based on the reconstructed position encoding vector corresponding to the abnormal data.

[0073] It can be understood that the embodiment of the present application provides a method for detecting anomalies in a wire rope based on multi-channel data reconstruction, which includes obtaining historical state information of the detection position after determining the detection position of the wire rope, preprocessing the historical state information, and obtaining a training data set; constructing a detection model and training the detection model using the training data set; obtaining the state information to be tested at the detection position, and inputting the state information to be tested into the trained detection model to obtain output data; obtaining a reconstruction scoring function based on the state information to be tested and the output data, and obtaining an adaptive threshold based on the reconstruction scoring function; comparing the state information to be tested with the adaptive threshold, and determining that the detection position corresponding to the state information to be tested that is outside the adaptive threshold is an abnormal position. The present application obtains historical state information at the detection position of the wire rope, and uses the historical state information as training data to train the detection model, so that the trained detection model can recognize the state information of the wire rope and output reconstructed data; then, the trained detection model obtains the output result of the test data, obtains the adaptive threshold based on the output result and the reconstruction function of the state information to be tested, and determines that the position corresponding to the state information to be tested that is outside the adaptive threshold is an abnormal position, thereby detecting the abnormal position of the wire rope.

[0074] Accordingly, the embodiment of the present application also provides a wire rope anomaly detection system based on multi-channel data reconstruction, please refer to Figure 10 , Figure 10 This is a module connection diagram of a wire rope anomaly detection system based on multi-channel data reconstruction provided in an embodiment of the present application. The wire rope anomaly detection system based on multi-channel data reconstruction provided in an embodiment of the present application includes: A historical information acquisition module 10 is used to determine the detection position of the wire rope and obtain historical status information of the detection position; The historical data processing module 20 is used to pre-process the historical status information and obtain a training data set; The model training module 30 and the historical data processing module 20 are used to build a detection model and perform model training on the detection model based on the training data set; The test information acquisition module 40 is used to obtain the test state information of the detection position; The reconstruction prediction module 50 is used to input the state information to be tested into the trained detection model to obtain output data; The reconstruction scoring module 60 is used to obtain a reconstruction scoring function based on the state information to be measured and the output data; A threshold acquisition module 70, the threshold acquisition module 70 is used to obtain an adaptive threshold based on the reconstructed scoring function; The abnormality locating module 80 is used to compare the state information to be measured with the adaptive threshold, and determine that the detection position corresponding to the state information to be measured that is outside the adaptive threshold is an abnormal position.

[0075] The present application has provided a detailed introduction to a wire rope anomaly detection method and system based on multi-channel data reconstruction provided by the embodiments of the present application. Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for those skilled in the art, according to the ideas of the present application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A wire rope anomaly detection method based on multi-channel data reconstruction, characterized in that: include: Determine the detection position of the wire rope and obtain historical status information of the detection position; Preprocessing the historical status information to obtain a training data set; Constructing a detection model, and performing model training on the detection model based on the training data set; Acquiring the state information of the detection position to be measured; Inputting the state information to be tested into the trained detection model to obtain output data; Acquire a reconstruction scoring function based on the state information to be measured and the output data; Obtaining an adaptive threshold based on the reconstruction scoring function; The state information to be measured is compared with the adaptive threshold, and a detection position corresponding to the state information to be measured that is outside the adaptive threshold is determined as an abnormal position.

2. The wire rope anomaly detection method based on multi-channel data reconstruction according to claim 1 is characterized in that: The state information of the steel wire rope includes multi-channel state information, and the detection model includes a feature enhancement module, and the feature enhancement module includes: The first submodule is configured to: determining a first feature, where the first feature is a potential feature space vector of the multi-channel state information; Determining state information of a specified channel from the multi-channel state information, and extracting a second feature from the state information of the specified channel; Processing the second feature based on a fully connected layer to obtain a third feature; Performing weighted processing on the first feature and the third feature based on a cross attention mechanism to obtain a fourth feature; Performing adaptive weight feature fusion on the third feature and the fourth feature, and obtaining a fifth feature by combining the residual connection of the output feature of the fully connected layer; A second submodule is connected to the first submodule, and the second submodule is used to: extracting a sixth feature from state information of a next channel adjacent to the designated channel; Processing the sixth feature based on the fully connected layer to obtain a seventh feature; Performing weighted processing on the seventh feature and the fifth feature based on the cross attention mechanism to obtain an eighth feature; Adaptive weight feature fusion is performed on the seventh feature and the eighth feature, and a ninth feature is obtained by combining the residual connection of the output feature of the fully connected layer.

3. The wire rope anomaly detection method based on multi-channel data reconstruction according to claim 2 is characterized in that: The detection model also includes: A multi-channel parallel convolutional encoder module, the multi-channel parallel convolutional encoder module is connected to the feature enhancement module, and the multi-channel parallel convolutional encoder module is used to: Obtain a two-dimensional convolution kernel based on the number of channels, where the number of rows of the two-dimensional convolution kernel is the same as the number of channels; Dividing the data of each row of the two-dimensional convolution kernel into independent row convolution sub-kernels; Match each row convolution sub-kernel to the corresponding channel, and calculate the convolution result of the state information of each channel. Based on the multiple convolution results, obtaining a potential feature space vector of the multi-channel state information; A multi-channel parallel convolution decoder module is connected to the feature enhancement module and is used to increase the dimension of the output data of the feature enhancement module through the fully connected layer and perform decoding reconstruction.

4. The wire rope anomaly detection method based on multi-channel data reconstruction according to claim 3 is characterized in that: The characterization formula for calculating the convolution result of the state information of each channel includes: ； in, The output The channel is located at The convolution result at ; For the input The channel is located at The data at is the position offset, is the number of positions; is the weight of the row convolution kernel.

5. The wire rope anomaly detection method based on multi-channel data reconstruction according to claim 3 is characterized in that: The detection model also includes: A channel position encoding module, the channel position encoding module is connected to the multi-channel parallel convolution encoder module, and the channel position encoding module is used to: Obtain input data of the plurality of channels, and match a corresponding position encoding vector for each channel; Concatenate the input data of each channel and the position encoding vector to obtain single-channel state information; Acquire the multi-channel state information based on the plurality of single-channel state information; A channel position decoding module, connected to the multi-channel parallel convolution decoder module, is used to: The output data of the multi-channel parallel convolution decoder module is obtained, and the data with the position coding vector is reconstructed and generated, and the channel to which it belongs is determined based on the position coding vector.

6. The wire rope anomaly detection method based on multi-channel data reconstruction according to claim 5 is characterized in that: The channel position encoding module is used to input the state information to be measured into the trained detection model: Obtaining input data of a plurality of channels; splicing the input data of each channel and the position coding vector to obtain single-channel state information; wherein the position coding vector of each channel is the same; Concatenate the input data of each channel and the position encoding vector to obtain single-channel state information; The multi-channel state information is acquired based on a plurality of the single-channel state information.

7. The wire rope anomaly detection method based on multi-channel data reconstruction according to claim 5 is characterized in that: The step of obtaining an adaptive threshold based on the reconstructed scoring function comprises: Determine a confidence limit, and obtain the adaptive threshold based on the confidence limit, where the characterization formula of the adaptive threshold includes: ； in, is the adaptive threshold; are the confidence limits; is the location parameter, which is used to characterize the central position of the distribution; is the scale parameter, which is used to characterize the degree of dispersion of the distribution; is the shape parameter, which determines the tail shape of the distribution; The extreme value samples are fitted to the generalized extreme value distribution model based on maximum likelihood estimation to obtain the location parameter, the scale parameter, and the shape parameter. The characterization formula of the maximum likelihood estimation includes: ； in, is the maximum likelihood estimate; is the value of the extreme sample.

8. The method for detecting abnormality of a wire rope based on multi-channel data reconstruction according to claim 5, characterized in that: The step of determining the detection position corresponding to the state information to be measured that is outside the adaptive threshold as an abnormal position includes: In the data with the position coding vector reconstructed and generated by the channel position decoding module, determining that the data outside the adaptive threshold is abnormal data; An abnormal channel is confirmed based on the reconstructed position encoding vector corresponding to the abnormal data.

9. The method for detecting abnormality of a wire rope based on multi-channel data reconstruction according to claim 1, characterized in that: The steps of preprocessing the historical status information to obtain a training data set include: Treat the channel as the first type of row data and the time as the column data, and sort the historical status information of the same channel in the same row based on the time sequence; in multiple column data, sort the sum of the data points in the same column based on the time sequence to generate the second type of row data and obtain the historical data matrix; Performing adaptive double exponential detrending processing on the historical data matrix to obtain detrended data; The detrended data is subjected to maximum pooling processing to obtain normalized data, and the maximum eigenvalue is extracted while maintaining the relative proportion of the data, and then a sliding window is used to perform data interception to obtain the training data set.

10. A wire rope anomaly detection system based on multi-channel data reconstruction, characterized in that: include: A historical information acquisition module (10), the historical information acquisition module (10) being used to determine a detection position of the wire rope and to acquire historical status information of the detection position; A historical data processing module (20), the historical data processing module (20) is used to pre-process the historical state information to obtain a training data set; A model training module (30), wherein the historical data processing module (20) is used to construct a detection model and perform model training on the detection model based on the training data set; A test information acquisition module (40), the test information acquisition module (40) is used to obtain the test state information of the detection position; A reconstruction prediction module (50), the reconstruction prediction module (50) is used to input the state information to be tested into the trained detection model to obtain output data; A reconstruction scoring module (60), the reconstruction scoring module (60) is used to obtain a reconstruction scoring function based on the state information to be measured and the output data; A threshold acquisition module (70), the threshold acquisition module (70) is used to acquire an adaptive threshold based on the reconstruction scoring function; An abnormality locating module (80) is used to compare the state information to be measured with the adaptive threshold, and determine that the detection position corresponding to the state information to be measured that is outside the adaptive threshold is an abnormal position.

Images