Non-invasive road multi-component cover plate fault intelligent detection method and device

Abstract

The invention discloses a non-invasive road multi-component cover plate fault intelligent detection method and device. The method comprises the following steps: collecting vibration data of a cover plate in three directions; performing noise reduction preprocessing on the vibration data in the three directions; respectively carrying out Grubm angle field transformation on the denoised data to obtain grey-scale maps, and fusing the grey-scale maps in three directions according to an RGB form to obtain a GADF color map; building a lightweight twin CNN-BiMGRU-MHAM model, and inputting a color image into a network for training; and performing intelligent fault detection on the road multi-component cover plate by using the trained model, and displaying a fault detection result on a terminal in a data and / or image form. According to the method, information fusion and deep learning technologies are combined, the efficiency and accuracy of fault detection of the coverage board are effectively improved, the detection and maintenance cost is reduced, driving safety is not affected, and a management department can master the health condition of the coverage board in time.

Description

Technical Field

[0001] The present invention relates to the field of smart city and artificial intelligence technology, and in particular to a non-invasive intelligent detection method and device for faults of multi-component road covering plates. Background Art

[0002] In recent years, as cities have gradually expanded, more and more infrastructure, such as power cables, natural gas pipelines, and sewage systems, have been placed underground. In order to facilitate the maintenance of the above equipment, a large number of large-diameter multi-component covering plates are often installed on the road. Due to long-term rolling by vehicles, the covering plate itself and the supporting beams underneath are easily damaged. If they are not repaired and replaced in time, they will pose a threat to road driving safety. Therefore, early diagnosis of the health status of multi-component covering plates is of great significance to resolve potential risks.

[0003] At present, the health status of multi-component cover plates is mainly detected by regular inspections. This method requires frequent opening of the cover plates, which not only increases the burden on maintenance personnel and maintenance costs, but also causes unnecessary interference to normal driving. At the same time, this method is relatively inefficient, the detection accuracy is heavily dependent on manual experience, and it is easy to ignore hidden damage, which is not suitable for large-scale multi-component cover plate fault detection. At present, some studies have attempted to use drones or satellites to take pictures of the cover plates and use visual methods to diagnose their faults. On the one hand, the image quality obtained by this method is easily affected by external factors such as climate, resulting in a high misjudgment rate; on the other hand, this method is powerless to detect hidden faults such as the inside of the cover plate and the supporting beams below. Some other studies have used laser scanners to scan the cover plates, but the cost is high and affects road driving safety.

[0004] Therefore, it is considered to use artificial intelligence technology to develop an intelligent multi-component cover plate health status detection method, so that the health status of the cover plates on the road can be understood quickly and accurately; at the same time, the detection cost is low and the interference to road driving is small. Summary of the invention

[0005] In view of this, in order to solve the problems of low efficiency, poor accuracy, high cost and interference to road driving in the existing diagnostic methods, the present invention provides a non-invasive intelligent detection method and device for multi-component road cover plate faults. The method of the present invention can quickly and accurately understand the health status of the cover plate on the road; at the same time, the detection cost is low and the interference to road driving is small.

[0006] To achieve the above object, the technical solution adopted by the present invention is:

[0007] In a first aspect, an embodiment of the present invention provides a non-invasive intelligent detection method for multi-component road cover faults, the method comprising the following steps:

[0008] S1, collect vibration data of the road multi-component cover plate in the three directions of X / Y / Z;

[0009] S2, performing noise reduction processing on the vibration data in three directions respectively;

[0010] S3, performing GADF transformation on the vibration data after the noise reduction process to obtain GADF grayscale image sets in three directions;

[0011] S4, fusing the GADF grayscale images obtained in the three directions in the form of RGB to obtain a GADF color image set, constructing a sample data set, and dividing the sample data set into a training set, a validation set, and a test set according to a preset ratio;

[0012] S5. Build a lightweight twin CNN-BiMGRU-MHAM model and use sample data to train and test the model;

[0013] S6. Use the trained model to perform intelligent detection of road multi-component cover plate faults, and display the fault detection results on the terminal in the form of data and / or images.

[0014] In an optional embodiment, in step S1, an infrared pulse trigger device is used to collect vibration data of the multi-component road cover: when the switch of the infrared pulse trigger device is triggered, vibration data in three directions of X / Y / Z are captured simultaneously.

[0015] In an optional implementation, in step S2, the vibration data in three directions are subjected to noise reduction processing respectively, and the specific method is as follows:

[0016] Firstly, the vibration data in the X / Y / Z directions are decomposed using a three-layer adaptive wavelet packet decomposition method;

[0017] Secondly, the local variance of each wavelet coefficient is calculated to estimate the noise intensity and select the appropriate threshold;

[0018] Finally, the wavelet coefficients with values ​​greater than the threshold are retained for reconstruction to obtain clean vibration data.

[0019] In an optional implementation, in step S3, GADF transformation is performed on the vibration data after the noise reduction processing, and the specific method is as follows:

[0020] Normalize the vibration data after noise reduction, encode the normalized time series in polar coordinates, encode the newly obtained amplitude as the angle cosine, and encode the time node as the radius r. The formula is:

[0021]

[0022]

[0023] Where: x i is the original time series data; is the normalized time series; is the time series after X normalization; t i is the timestamp; N is the constant factor of the regularized polar coordinate system generating space;

[0024] The trigonometric difference between two points is used to identify the correlation of different time intervals, and the time series is converted into a Gram angle difference field based on a sine function, as follows:

[0025]

[0026] Where GADF represents the Gram angle difference field; I is the unit row vector; for The transpose of n represents the polar angle of vector n.

[0027] In an optional implementation, in step S4, the GADF grayscale images obtained in three directions are fused in the form of RGB to obtain a GADF color image set, and the specific method is:

[0028] The GADF grayscale image with a preset size in the X direction is used as the R channel, the GADF grayscale image with a preset size in the Y direction is used as the G channel, and the GADF grayscale image with a preset size in the Z direction is used as the B channel, so as to construct a GADF color image set.

[0029] In an optional implementation, in step S4, the training set, validation set and test set are divided in a ratio of 7:2:1.

[0030] In an optional embodiment, in step S5, a lightweight twin CNN-BiMGRU-MHAM model is built, and the model is trained and tested using sample data, including:

[0031] A CNN network consisting of 5 convolution layers and 5 pooling layers is used to extract spatial features from GADF color images. The Mogrifier data coupling module is introduced to form a lightweight BiMGRU module together with the GRU lightweight model. Through information context interaction, information features that are temporally correlated are extracted. A multi-head attention mechanism is introduced to extract key core features from the extracted features and remove redundant information, thereby improving the accuracy and speed of network training.

[0032] In the relationship measurement module of the twin network, a comprehensive relationship measurement that integrates distance and angle is used, namely:

[0033] D new (f(X 1 ),f(X 2 ))=ηD(f(X 1 ),f(X 2 ))+λA(f(X 1 ),f(X 2 ))

[0034] In the formula, D(f(X 1 ),f(X 2 )) is the distance relationship measure; A(f(X 1 ),f(X 2 )) is the angle relationship measure; η and λ are weight coefficients;

[0035] Based on the relationship measurement strategy, a similarity loss function with a penalty coefficient is used. The penalty coefficient is increased according to the difficulty of distinguishing between different types of samples, and the value of the misjudgment loss function is increased between easily confused samples, that is:

[0036] S i,j =Sigmoid(D new (f(X i ),f(X j ))

[0037]

[0038] In the formula, S i,j Indicates similarity; Loss new represents the loss function during the training of the twin network; Y i,j Represents the similarity label between two samples; α i,j Indicates the penalty coefficient when samples i and j belong to different types

[0039] In an optional embodiment, in step S6, a user-operated interface platform is built based on Python, and the fault detection results of the multi-component cover plate are displayed in the form of data and / or images through the interface platform.

[0040] In a second aspect, an embodiment of the present invention further provides a non-invasive intelligent detection device for multi-component road cover plate faults, comprising: a data acquisition device, a memory and a processor, wherein:

[0041] The data acquisition device is used to collect vibration data of the road multi-component cover plate in three directions of X / Y / Z;

[0042] The memory stores computer instructions;

[0043] The processor is data-connected to the data acquisition device and the memory. The processor executes the computer instructions to perform the above-mentioned non-invasive intelligent detection method for multi-component road cover plate faults, thereby performing intelligent detection of multi-component road cover plate faults.

[0044] In a third aspect, an embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the above-mentioned non-invasive intelligent detection method for multi-component road cover plate faults is implemented.

[0045] Compared with the prior art, the present invention has at least the following beneficial effects:

[0046] 1. Based on the lightweight twin CNN-BiMGRU-MHAM model, the present invention provides a non-invasive intelligent detection method for multi-component road cover plate faults. The present invention combines information fusion technology and deep learning to perform early detection of the fault status of multi-component road cover plates, thereby improving the detection efficiency and accuracy, saving labor costs, and filling the gap in the field of early fault diagnosis of multi-component cover plates. The method is simple and easy to operate and will not affect road driving safety.

[0047] 2. The present invention provides a non-invasive intelligent detection method for multi-component road cover plate faults, which simultaneously collects vibration data in three directions of X / Y / Z as fault data for feature extraction. Compared with the image vision method, it can better reflect the change law of the internal excitation signal of the multi-component cover plate, which is helpful to detect internal faults of the system.

[0048] 3. The present invention provides a non-invasive intelligent detection method for multi-component road cover plate faults. The designed twin network structure can achieve accurate detection even when the samples are (severely) insufficient. Compared with the traditional single diagnosis model that relies heavily on a large number of samples, it avoids the generation of a large number of samples, saves computing resources, and improves detection efficiency and speed. In addition, it adopts a lightweight network structure and incorporates a unique contextual interaction mechanism and a multi-head attention mechanism, which improves the training speed of the network model while ensuring the detection accuracy.

[0049] 4. The present invention provides a non-invasive intelligent detection method for multi-component road cover plate faults. In the similarity measurement stage, the traditional measurement function is improved, and the differences between two vectors in distance and direction are comprehensively considered to propose a new comprehensive measurement strategy, which can avoid the distance deviation problem caused by the same length but different directions. In terms of network model training, an innovative dynamic weighted loss function is proposed, which can adaptively adjust the size according to the similarity value change trend. When the distinction between different categories is small, the loss function is increased by increasing the penalty factor, thereby improving the training accuracy and speed.

[0050] 5. The non-invasive intelligent detection method for multi-component road cover plate faults of the present invention is designed with a special classifier in the model testing stage to avoid searching for classification results by comparing similarities one by one, thereby greatly improving the detection efficiency.

[0051] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing 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 and the accompanying drawings.

[0052] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0054] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0055] Figure 1 A schematic diagram of the principle of a non-invasive intelligent detection method for multi-component road cover plate faults provided in an embodiment of the present invention.

[0056] Figure 2 A schematic flow chart of a non-invasive intelligent detection method for multi-component road cover plate faults provided in an embodiment of the present invention.

[0057] Figure 3 A schematic diagram of the structure of a lightweight twin CNN-BiMGRU-MHAM model provided in an embodiment of the present invention.

[0058] Figure 4 A schematic diagram of the model training and testing process provided by an embodiment of the present invention.

[0059] Figure 5 A schematic diagram of the changing trends of accuracy and loss function during the model training process provided in an embodiment of the present invention.

[0060] Figure 6 A schematic diagram of a confusion matrix of detection results provided by an embodiment of the present invention.

[0061] Figure 7 A schematic diagram of a user operation interface developed based on Python provided in an embodiment of the present invention.

[0062] Figure 8 A schematic structural diagram of a non-invasive intelligent detection device for multi-component road cover plate faults provided in an embodiment of the present invention.

[0063] Among them, 1-data acquisition device; 2-memory; 3-processor. DETAILED DESCRIPTION

[0064] 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 only part of the embodiments of the present invention, not all of the embodiments.

[0065] In the description of the present invention, it should be noted that: in some processes described in the specification and drawings of this application, multiple operations appearing in a specific order are included, but it should be clearly understood that these operations may not be performed in the order in which they appear in this document or may be performed in parallel. In addition, various serial numbers are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0066] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0067] See also Figure 1 and Figure 2 As shown, the present invention provides a non-invasive intelligent detection method for multi-component road cover faults. By sending the collected vibration data (i.e., vibration signal) into a trained diagnostic network model (a lightweight twin CNN-BiMGRU-MHAM model), the output of the fault status can be realized, and the operation is convenient and fast. In specific implementation, the method flow mainly includes the following steps:

[0068] S1, collecting vibration signals of multi-component cover plates in the X / Y / Z directions to obtain a time domain data set of cover plate vibration;

[0069] In an embodiment of the present invention, an infrared pulse trigger device (data acquisition device) is installed on the multi-component road covering plate. When the front wheel of a vehicle rolls over the multi-component road covering plate (such as a road manhole cover), the switch of the infrared pulse trigger device is triggered to collect vibration data at a frequency of 100KHz until the rear wheel leaves the multi-component covering plate, and then the data collection automatically stops.

[0070] S2. De-noise the data in the X / Y / Z directions respectively. The preferred method is the adaptive wavelet packet decomposition denoising method, which is as follows: first, decompose the time domain signals in the X / Y / Z directions using a three-layer adaptive wavelet packet decomposition method; second, calculate the local variance of each wavelet coefficient to estimate the noise intensity, and then select a suitable threshold; finally, retain the wavelet coefficients with values ​​greater than the threshold for reconstruction to obtain a clean vibration signal.

[0071] S3, perform GADF transformation on the pre-processed data in the X / Y / Z directions respectively to obtain a GADF grayscale image set of 100×100 size. Specifically:

[0072] The vibration data after noise reduction is normalized, and the normalized time series shown in formula (1) is encoded in polar coordinates. The newly obtained amplitude is encoded as the angle cosine, and the time node is encoded as the radius r, as shown in formula (2).

[0073]

[0074]

[0075] Where: x i is the original time series data; is the normalized time series; is the time series after X normalization; t i is the timestamp; N is the constant factor of the regularized polar coordinate system generating space;

[0076] Furthermore, by using the trigonometric difference between two points to identify the correlation of different time intervals, the time series can be transformed into a sine function-based Gram angle difference field (GADF) as shown in formula (3).

[0077]

[0078] Where I is the unit row vector; for The transpose of n represents the polar angle of vector n.

[0079] S4. Use the GADF grayscale image with a size of 100×100 in the X direction as the R channel, the GADF grayscale image with a size of 100×100 in the Y direction as the G channel, and the GADF grayscale image with a size of 100×100 in the Z direction as the B channel to construct a 100×100×3 GADF color image set, and divide it into training set, validation set and test set in a ratio of 7:2:1.

[0080] S5. Build a lightweight twin CNN-BiMGRU-MHAM model as a diagnostic network model for multi-component road cover panels. First, a CNN network consisting of 5 layers of convolution and 5 layers of pooling is used to extract spatial features from GADF color images; then, the Mogrifier data coupling module is introduced to form a lightweight BiMGRU module together with the GRU lightweight model, and the temporally correlated information features are extracted through information context interaction; finally, a multi-head attention mechanism (MHAM) is introduced to extract key core features from the extracted features and remove redundant information, thereby improving the accuracy and speed of network training. In the relationship measurement module of the twin network, the present invention innovatively proposes a comprehensive relationship measurement that integrates distance and angle, as shown in formula (4):

[0081] D new (f(X 1 ),f(X 2 ))=ηD(f(X 1 ),f(X 2 ))+λA(f(X 1 ),f(X 2 )) (4)

[0082] In the formula, D(f(X 1 ),f(X 2 )) is the distance relationship measure; A(f(X 1 ),f(X 2 )) is the angle relationship measure; η and λ are weight coefficients, which are obtained by experience respectively.

[0083] In the embodiment of the present invention, in order to further improve the recognition accuracy of the classifier, the present invention proposes a similarity loss function with a penalty coefficient based on the new relationship measurement strategy, which can increase the penalty coefficient according to the difficulty of distinguishing between different types of samples, and increase the value of the misjudgment loss function between easily confused samples, as shown in formula (5):

[0084] S i,j =Sigmoid(D new (f(X i ),f(X j ))

[0085]

[0086] Where: S i,j Indicates similarity; Loss new represents the loss function of the proposed twin network training; Y i,j Represents the similarity label between two samples; α i,j Indicates the penalty coefficient when sample i and sample j belong to different types. When two samples are not similar, if the network outputs S i,j If it does not approach 0, then assign α i,j Larger values ​​increase the loss value.

[0087] See also Figure 3 As shown, it shows the model structure of the twin CNN-BiMGRU-MHAM model built by the present invention. Each subnetwork includes a CNN network consisting of 5 convolutional layers and 5 pooling layers, a BiMGRU network with contextual information interaction, and a multi-head attention mechanism model. In the relationship measurement module, a new comprehensive measurement function that includes both distance and angle is designed to improve the accuracy of similarity judgment of two samples in a sample pair. The pseudo code of the entire model is shown in Table 1 below.

[0088] Table 1 Pseudo code of the model

[0089]

[0090]

[0091] Figure 4 This is the training and testing flow chart of the twin CNN-BiMGRU-MHAM model of the present invention. The designed twin CNN-BiMGRU-MHAM model is trained with the training set. After the training, the network model is saved. A sample from the training set is selected to form a sample pair with a sample from the test set, and the sample is sent to the trained network model for testing. During the entire training process, Epoch is set to 200, batch_size is set to 10, and the initial learning rate is set to 2×10 -3 , every 20 iterations the learning rate is reduced to half of the previous one, and the loss function threshold is 10 -6 .

[0092] Figure 5 In the case of a small sample of 40 normal state samples and 40 fault state samples, the changing trend of the accuracy curve and loss function curve during the training process of the twin CNN-BiMGRU-MHAM model ( Figure 5 The left part in the middle is the changing trend of the accuracy curve, and the right part is the changing trend of the loss function curve). Figure 5It can be seen that during the training process, when Epoch = 100, the accuracy of the network model tends to stabilize at around 96%, and the loss function gradually shows a convergence trend. In order to reflect the structural complexity of the designed network, the training time of different networks under the same sample condition is shown in Table 2.

[0093] Table 2 Training time of different network models

[0094]

[0095] It can be seen from the training time of different comparison models that the training speed of the twin CNN-BiMGRU-MHAM model of the present invention is higher than that of other network models under different small sample conditions, which once again proves that the network of the present invention is a lightweight network.

[0096] Figure 6 It is the result of testing the twin CNN-BiMGRU-MHAM of the present invention with different input information under 400 training samples and 200 test samples. Figure 6 Part (a) is a diagram of the confusion matrix of the X-direction detection results. Figure 6 Part (b) is a diagram of the confusion matrix of the Y direction detection results. Figure 6 Part (c) is a schematic diagram of the confusion matrix of the Z-direction detection results. Figure 6 Part (d) is a diagram of the confusion matrix of the detection results after using the RGB fusion method. It can be seen that: in terms of accuracy, the network is trained using data in the single direction of X, Y, and Z, and the accuracy is 72.50%, 80.50%, and 90.00% respectively, while the accuracy of the network trained using RGB fusion information is 97.00%, which is an increase of 24.50%, 16.50%, and 7.00% respectively; in terms of F1-score, the network is trained using data in the single direction of X, Y, and Z, and the F1-score values ​​are 72.30%, 80.48%, and 89.99% respectively, while the F1-score value of the network trained using RGB fusion information is 97.00%, which is an increase of 24.70%, 16.52%, and 7.01% respectively. Therefore, whether in terms of accuracy or F1-score, the network model proposed by using RGB training achieves the best performance.

[0097] S6. Develop a user interface platform to display the analysis and test results of the model in the terminal in the form of data or images. The user interface platform here is preferably developed based on the Python language, and the platform has functions such as data uploading, result analysis and display, visual display, and data storage. Through the development of the user interface platform, the fault diagnosis results of the multi-component cover plate are intuitively presented in the form of data and images.

[0098] In a specific embodiment, Figure 7 The user interface developed based on Python is shown in Figure 1. The user interface platform here is developed based on Python language. The platform has functions such as data upload, result analysis and display, visual display, and data storage. Through the development of this user interface platform, the fault diagnosis results of multi-component cover plates are intuitively presented in the form of data and images.

[0099] From the description of the above implementation, those skilled in the art can know that the present invention provides a non-invasive intelligent detection method for multi-component road cover plate faults based on a lightweight twin CNN-BiMGRU-MHAM model, aiming to solve the problems of low efficiency, poor accuracy, high cost and interference with road driving in existing diagnostic methods. The present invention has also made innovations in network structure design and model training methods; in the sub-network structure design, since the GADF image has spatial information and time-related information that reflect the state change of the object, the spatial features are first extracted using the CNN network, and then the time-related features are extracted using the lightweight BiGRU network. In order to improve the network feature extraction capability and avoid information loss, the Mogrifier data coupling module is introduced into the lightweight BiGRU model, and a lightweight BiMGRU model with information context interaction function is designed. In order to improve the feature extraction efficiency, a multi-head attention mechanism (MHAM) is introduced to extract key core features from the BiMGRU output to avoid redundant information, which can improve the training speed of the model. In addition, in terms of network training methods, an innovative dynamic weighted loss function is proposed, which can adaptively adjust the size according to the changing trend of the similarity value. When the distinction between different categories is small, the loss function is increased by increasing the penalty factor, thereby improving the training accuracy and speed.

[0100] Furthermore, the embodiment of the present invention also provides a non-invasive intelligent detection device for multi-component road cover plate faults, which is used for intelligent detection of multi-component road cover plate faults. The device can be implemented by software and / or hardware. Figure 8 As shown, the device includes a data acquisition device 1, a memory 2 and a processor 3. The data acquisition device 1, the memory 2 and the processor 3 can be connected via a bus or other means.

[0101] The data acquisition device 1 is used to collect vibration data of the multi-component cover plate in three directions of X / Y / Z, and send it to the processor 3. The processor 3 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components and other chips, or a combination of the above chips.

[0102] The memory 2, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer executable programs and modules, such as programs or instructions corresponding to the non-invasive road multi-component cover plate fault intelligent detection method.

[0103] The processor 3 executes various functional applications and data processing of the processor by running the non-transient software program or instructions stored in the memory 2, that is, implements the non-invasive road multi-component cover plate fault intelligent detection method in the above method embodiment.

[0104] The memory 2 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created by the processor 2 and the like.

[0105] In addition, the memory 2 may include a high-speed random access memory and may also include a non-transitory memory, and more specific examples of the memory 2 (a non-exhaustive list) include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, and any suitable combination thereof. In some aspects, the memory 2 may also optionally include a memory remotely arranged relative to the processor 3, and these remote memories may be connected to the processor 3 via a network.

[0106] Those skilled in the art will appreciate that the embodiments of the present invention may be provided as methods, devices or computer program products, etc. Therefore, the present invention may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects.

[0107] It should be noted that the word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several distinct components, and by means of a suitably programmed computer.

[0108] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0109] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A non-invasive intelligent detection method for multi-component road cover faults, characterized in that: The method comprises the following steps: S1, collect vibration data of the road multi-component cover plate in the three directions of X / Y / Z; S2, performing noise reduction processing on the vibration data in three directions respectively; S3, performing GADF transformation on the vibration data after the noise reduction process to obtain GADF grayscale image sets in three directions; S4, fusing the GADF grayscale images obtained in the three directions in the form of RGB to obtain a GADF color image set; constructing a sample data set, and dividing the sample data set into a training set, a validation set, and a test set according to a preset ratio; S5. Build a lightweight twin CNN-BiMGRU-MHAM model and use sample data to train and test the model; S6. Use the trained model to perform intelligent detection of road multi-component cover plate faults, and display the fault detection results on the terminal in the form of data and / or images.

2. A non-invasive intelligent detection method for multi-component road cover faults according to claim 1, characterized in that: In step S1, the vibration data of the road multi-component covering plate is collected by using an infrared pulse trigger device: when the switch of the infrared pulse trigger device is triggered, the vibration data in the three directions of X / Y / Z are captured simultaneously.

3. A non-invasive intelligent detection method for multi-component road cover faults according to claim 1, characterized in that: In the step S2, the vibration data in three directions are subjected to noise reduction processing respectively, and the specific method is as follows: Firstly, the vibration data in the X / Y / Z directions are decomposed using a three-layer adaptive wavelet packet decomposition method; Secondly, the local variance of each wavelet coefficient is calculated to estimate the noise intensity and select the appropriate threshold; Finally, the wavelet coefficients with values ​​greater than the threshold are retained for reconstruction to obtain clean vibration data.

4. A non-invasive intelligent detection method for multi-component road cover faults according to claim 1, characterized in that: In the step S3, GADF transformation is performed on the vibration data after the noise reduction processing, and the specific method is as follows: Normalize the vibration data after noise reduction, encode the normalized time series in polar coordinates, encode the newly obtained amplitude as the angle cosine, and encode the time node as the radius r. The formula is: Where: x i is the original time series data; is the normalized time series; is the time series after X normalization; t i is the timestamp; N is the constant factor of the regularized polar coordinate system generating space; The trigonometric difference between two points is used to identify the correlation of different time intervals, and the time series is converted into a Gram angle difference field based on a sine function, as follows: Where GADF represents the Gram angle difference field; I is the unit row vector; for The transpose of n represents the polar angle of vector n.

5. A non-invasive intelligent detection method for multi-component road cover faults according to claim 1, characterized in that: In step S4, the GADF grayscale images obtained in three directions are fused in the form of RGB to obtain a GADF color image set. The specific method is: The GADF grayscale image with a preset size in the X direction is used as the R channel, the GADF grayscale image with a preset size in the Y direction is used as the G channel, and the GADF grayscale image with a preset size in the Z direction is used as the B channel, so as to construct a GADF color image set.

6. A non-invasive intelligent detection method for multi-component road cover faults according to claim 1, characterized in that: In step S4, the training set, the validation set and the test set are divided into a ratio of 7:2:

1.

7. A non-invasive intelligent detection method for multi-component road cover faults according to claim 1, characterized in that: In step S5, a lightweight twin CNN-BiMGRU-MHAM model is built, and the model is trained and tested using sample data, including: A CNN network consisting of 5 convolution layers and 5 pooling layers is used to extract spatial features from GADF color images. The Mogrifier data coupling module is introduced to form a lightweight BiMGRU module together with the GRU lightweight model. Through information context interaction, information features that are temporally correlated are extracted. A multi-head attention mechanism is introduced to extract key core features from the extracted features and remove redundant information, thereby improving the accuracy and speed of network training. In the relationship measurement module of the twin network, a comprehensive relationship measurement that integrates distance and angle is used, namely: D new (f(X1),f(X2))=ηD(f(X1),f(X2))+λA(f(X1),f(X2)) Where D(f(X1), f(X2)) is the distance relationship metric; A(f(X1), f(X2)) is the angle relationship metric; η and λ are weight coefficients; Based on the relationship measurement strategy, a similarity loss function with a penalty coefficient is used. The penalty coefficient is increased according to the difficulty of distinguishing between different types of samples, and the value of the misjudgment loss function is increased between easily confused samples, that is: S i,j =Sigmoid(D new (f(X i ),f(X j )) In the formula, S i,j Indicates similarity; Loss new represents the loss function during the training of the twin network; Y i,j Represents the similarity label between two samples; α i,j Indicates the penalty coefficient when samples i and j belong to different types.

8. A non-invasive intelligent detection method for multi-component road cover faults according to claim 1, characterized in that: In the step S6, a user-operated interface platform is built based on Python, and the fault detection results of the multi-component cover plate are displayed in the form of data and / or images through the interface platform.

9. A non-invasive intelligent detection device for multi-component road cover faults, characterized in that: include: A data acquisition device, a memory and a processor, wherein: The data acquisition device is used to collect vibration data of the road multi-component cover plate in three directions of X / Y / Z; The memory stores computer instructions; The processor is data-connected to the data acquisition device and the memory. The processor executes the computer instructions to perform a non-invasive intelligent detection method for multi-component road cover plate faults as described in any one of claims 1 to 8, and performs intelligent detection of multi-component road cover plate faults.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the non-invasive intelligent detection method for multi-component road cover faults according to any one of claims 1 to 8 is implemented.

Images