A smart data acquisition and analysis monitoring method and system for urban manhole safety

Abstract

This application discloses an intelligent data acquisition and analysis monitoring method and system for urban manhole safety, relating to the field of urban manhole monitoring and control technology. The method includes acquiring vibration data of the manhole cover; inputting the vibration data into an MSCNN model to obtain abnormal vibration data; the MSCNN model is a pre-trained model that takes vibration data as input and abnormal vibration data as output to identify abnormal vibration data within the vibration data; inputting the abnormal vibration data into a VGG-CNN model for damage identification and classification to obtain a damage category prediction result for the manhole cover; and the VGG-CNN model is a pre-trained model that takes abnormal vibration data as input and the damage category prediction result as output to predict the health status of the manhole cover based on the abnormal vibration data. This application can achieve accurate damage identification and classification and health status assessment of manhole covers, ensuring the safety of pedestrians and vehicles around urban manhole covers.

Description

Technical Field

[0001] This application relates to the field of urban well monitoring and control technology, and in particular to an intelligent data acquisition and analysis monitoring method and system for urban well safety. Background Technology

[0002] Currently, traditional urban manhole monitoring methods typically rely on regular manual inspections and simple sensor monitoring tools, such as basic vibration sensors, triaxial accelerometers, and temperature and humidity sensors. These monitoring methods can provide some degree of monitoring and protection, such as identifying whether manhole covers have been illegally moved or detecting obvious physical damage. However, these methods usually only detect problems after they have become obvious or have caused significant consequences, failing to predict the extent of damage in advance or accurately assess the health status of the manhole covers. For example, minor cracks, slight deformations, or minor loosening—if these subtle changes are not quickly identified and addressed, they may develop into more serious problems, even posing significant safety hazards to pedestrians and vehicles around the manhole cover. Summary of the Invention

[0003] The purpose of this application is to provide an intelligent data acquisition and analysis monitoring method and system for urban manhole safety, which can achieve accurate damage identification and classification and health status assessment of manhole covers, ensuring the safety of pedestrians and vehicles around urban manhole covers.

[0004] To achieve the above objectives, this application provides the following solution.

[0005] In a first aspect, this application provides an intelligent data acquisition and analysis monitoring method for urban well safety, which includes the following steps.

[0006] Obtain vibration data of the manhole cover.

[0007] The vibration data is input into the MSCNN model to obtain abnormal vibration data; the MSCNN model is a pre-trained model that takes the vibration data as input and the abnormal vibration data as output to identify the abnormal vibration data in the vibration data; the abnormal vibration data is the data corresponding to the abnormal vibration state in the vibration data.

[0008] The abnormal vibration data is input into the VGG-CNN model for damage identification and classification to obtain the damage category prediction result of the manhole cover; the VGG-CNN model is a pre-trained model that takes the abnormal vibration data as input and the damage category prediction result as output, and is used to predict the health status of the manhole cover based on the abnormal vibration data; the damage category prediction result is the probability value corresponding to each damage category of the manhole cover.

[0009] Optionally, vibration data of the manhole cover can be obtained, specifically including the following steps.

[0010] An accelerometer is installed at the center of the back of the manhole cover. The accelerometer is used to measure the vertical vibration data of the manhole cover at a predetermined frequency.

[0011] The vibration data is collected by monitoring the manhole cover using the acceleration sensor.

[0012] Optionally, after the step of acquiring vibration data of the manhole cover, the intelligent acquisition and analysis monitoring method for urban manhole safety further includes the following steps.

[0013] The vibration data is preprocessed by data cleaning, data normalization and data noise reduction to obtain preprocessed vibration data; the preprocessed vibration data is used as input to the MSCNN model to obtain the abnormal vibration data.

[0014] Optionally, the vibration data can be preprocessed by data cleaning, data normalization and data noise reduction to obtain preprocessed vibration data, specifically including the following steps.

[0015] Remove outliers from the vibration data and fill in any missing values ​​to obtain cleaned vibration data.

[0016] The cleaned vibration data is then normalized to obtain normalized vibration data.

[0017] A wavelet denoising algorithm is used to denoise the normalized vibration data to obtain the preprocessed vibration data.

[0018] Optionally, after inputting the abnormal vibration data into the VGG-CNN model for damage identification and classification to obtain the damage category prediction result of the manhole cover, the intelligent acquisition and analysis monitoring method for urban manhole safety further includes the following steps.

[0019] Based on the damage category prediction results, the health status level of the manhole cover is determined.

[0020] Optionally, the health status level of the manhole cover is divided into minor damage level, moderate damage level and severe damage level in order of increasing degree of damage.

[0021] Optionally, the damage categories corresponding to the minor damage level include minor cracks, minor deformation, and minor loosening; the damage categories corresponding to the moderate damage level include moderate cracks, moderate deformation, and moderate loosening; and the damage categories corresponding to the severe damage level include severe cracks, severe deformation, and severe loosening.

[0022] Optionally, after determining the health status level of the manhole cover based on the damage category prediction results, the intelligent data acquisition and analysis monitoring method for urban manhole safety further includes the following steps.

[0023] For manhole covers with severe cracks, severe deformation, and / or severe loosening, the location of the manhole cover is obtained, and the health status level and location of the manhole cover are reported so that inspection personnel can repair the manhole cover in a timely manner.

[0024] Optionally, the sampling frequency of the accelerometer is 2500Hz.

[0025] Secondly, this application proposes an intelligent data acquisition and analysis monitoring system for urban manhole safety, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the intelligent data acquisition and analysis monitoring method for urban manhole safety described in the first aspect.

[0026] Based on the specific embodiments provided in this application, the following technical effects are disclosed.

[0027] This application provides an intelligent data acquisition and analysis monitoring method and system for urban manhole safety. It combines MSCNN and VGG-CNN models and applies them to the monitoring of manhole covers in urban areas. First, the MSCNN model accurately identifies abnormal vibration data based on vibration data. The abnormal vibration data output by the MSCNN model is then used as input to the VGG-CNN model. By inputting the abnormal vibration data into the VGG-CNN model, damage identification, classification, and health status assessment of the manhole cover are performed, resulting in accurate and reliable damage category predictions. This enables accurate damage identification and classification of manhole covers, facilitating effective management of urban manhole covers and ensuring the safety of pedestrians and vehicles around urban manhole covers. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 This is a flowchart illustrating an intelligent data acquisition and analysis monitoring method for urban manhole safety, provided as an embodiment of this application.

[0030] Figure 2 This is a schematic diagram of the abnormal vibration data identification process provided in an embodiment of this application.

[0031] Figure 3 This is a schematic diagram of the preprocessing flow provided in an embodiment of this application.

[0032] Figure 4 This is a schematic diagram of the structure of an MSCNN model provided in an embodiment of this application.

[0033] Figure 5 This is a schematic diagram of the convolution process of an MSCNN model provided in an embodiment of this application.

[0034] Figure 6 This is a schematic diagram of the parameter optimization process of a VGG-CNN model provided in an embodiment of this application.

[0035] Figure 7 This is a schematic diagram of the structure of a VGG-CNN model provided in an embodiment of this application. Detailed Implementation

[0036] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0037] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0038] like Figure 1 As shown in this embodiment, an intelligent data acquisition and analysis monitoring method for urban manhole safety is proposed. This method mainly combines the MSCNN (Multi-Scale Convolutional Neural Networks) model and the VGG-CNN (Visual Geometry Group-Convolutional Neural Network) model and applies them to the monitoring of manhole covers in urban areas. The aim is to achieve accurate damage identification and classification and health status assessment of manhole covers, thereby ensuring the safety of pedestrians and vehicles around urban manhole covers. The method includes the following steps.

[0039] Step S1: Obtain vibration data of the manhole cover.

[0040] Step S2: Input the vibration data into the MSCNN model to obtain abnormal vibration data.

[0041] The MSCNN model is a pre-trained model that takes the vibration data as input and the abnormal vibration data as output, and is used to identify the abnormal vibration data in the vibration data; the abnormal vibration data is the data corresponding to the abnormal vibration state in the vibration data.

[0042] Step S3: Input the abnormal vibration data into the VGG-CNN model for damage identification and classification to obtain the damage category prediction result of the manhole cover.

[0043] The VGG-CNN model is a pre-trained model that takes the abnormal vibration data as input and the damage category prediction result as output, and is used to predict the health status of the manhole cover based on the abnormal vibration data; the damage category prediction result is the probability value corresponding to each damage category of the manhole cover.

[0044] In this embodiment, step S1, which involves acquiring vibration data of the manhole cover, specifically includes the following steps.

[0045] Step S11: Install an acceleration sensor at the center of the back of the manhole cover. The acceleration sensor is used to measure the vertical acceleration signal of the manhole cover at a predetermined frequency as vibration data.

[0046] In this embodiment, an ADXL1001 accelerometer is used. This accelerometer is installed at the center of the back of the manhole cover and is used to measure the vertical acceleration signal of the manhole cover at a predetermined frequency as vibration data, thereby achieving real-time acquisition of the manhole cover's vibration data. The sampling frequency of the accelerometer in this embodiment is 2500Hz.

[0047] Step S12: Use the acceleration sensor to monitor the manhole cover and collect the vibration data.

[0048] In this embodiment, after the step of obtaining vibration data of the manhole cover in step S1, the intelligent acquisition and analysis monitoring method for urban manhole safety further includes the following steps.

[0049] The vibration data undergoes preprocessing including data cleaning, data normalization, and data noise reduction to obtain preprocessed vibration data. This preprocessed vibration data is then used as input to the MSCNN model to obtain the abnormal vibration data.

[0050] In this embodiment, the vibration data is preprocessed by data cleaning, data normalization and data noise reduction to obtain preprocessed vibration data, specifically including the following steps.

[0051] (1) Remove outliers from the vibration data and fill in the missing values ​​in the vibration data to obtain cleaned vibration data.

[0052] (2) Normalize the vibration data after data cleaning to obtain normalized vibration data.

[0053] (3) The wavelet denoising algorithm is used to denoise the vibration data after the data normalization to obtain the preprocessed vibration data.

[0054] This embodiment utilizes an accelerometer to collect vibration data of the manhole cover in real time at a predetermined frequency, and then performs preprocessing such as data cleaning, data normalization, and data noise reduction on the vibration data. Figure 2 As shown, the preprocessed vibration data is input into the microcontroller of the manhole cover terminal. The microcontroller uses a pre-built MSCNN model to analyze the vibration data, determining if it is abnormal. If so, the abnormal vibration data is written to an SD card and then fed back to the microcontroller. The abnormal vibration data is then transmitted to the server via a 4G module; otherwise, it is discarded. The abnormal vibration data uploaded to the server is input into a pre-built VGG-CNN model for identification, classification, and health status assessment of the manhole cover, determining its health status level. When the manhole cover's health status is determined to be severely damaged (i.e., at a dangerous level), the server reports the health status level and location of the manhole cover to inspection personnel in real time. This allows inspection personnel to promptly repair severely damaged manhole covers and notify the manhole cover terminal to issue an alarm, reminding pedestrians and vehicles to pass with caution. Furthermore, in unmanned inspection scenarios, this achieves intelligent digital management of urban manholes through "sensing-diagnosis-alarm."

[0055] In this embodiment, after acquiring the vibration data, the data is first preprocessed, including data cleaning, data normalization, and data noise reduction. Since the accelerometer outputs an analog signal, ADC sampling inevitably results in outliers and missing values. Based on the fundamental parameters of the accelerometer, it can be determined that when the voltage exceeds the power supply voltage V... DD In this case, the outlier should be removed and filled with the average of its five neighboring samples. Missing values ​​are filled with the average of their five neighboring samples, following the same principle. The expression for data normalization is as follows.

[0056] .

[0057] Among them, X norm This represents the normalized result of the vibration data, where X represents the vibration data. max X represents the maximum value of the vibration data. min This represents the minimum value of the vibration data.

[0058] Since construction and underground facilities often occur around urban manholes, low-frequency or high-frequency vibration noise may be generated. Therefore, it is necessary to denoise the collected vibration signal sequence. In this embodiment, wavelet denoising algorithm is used for denoising. The steps of wavelet denoising algorithm include: (1) Determining the number of decomposition levels. The preprocessed vibration signal is decomposed into multiple resolution levels through wavelet transform at multiple scales to obtain a series of wavelet coefficients of different frequency bands. Low-frequency components usually represent the overall trend of the signal, while high-frequency components contain detailed information and noise. (2) Determining the threshold standard. A soft threshold denoising algorithm is adopted, a filtering threshold is set, and based on the filtering threshold, the wavelet coefficients of different frequency bands are quantized using a threshold function. Wavelet coefficients smaller than the filtering threshold are removed, and wavelet coefficients greater than or equal to the filtering threshold are kept unchanged or reduced to achieve multi-scale wavelet coefficient denoising. (3) Reconstructing the signal by inverse wavelet transform. Through inverse wavelet transform, the wavelet coefficients processed above are reconstructed into a signal to obtain the denoised vibration signal, thus achieving denoising of the vibration signal. The preprocessing process is as follows: Figure 3 As shown in the figure. Investigation revealed that abnormal vibration states such as minor cracks, slight deformation, and slight loosening of the manhole cover correspond to low-frequency signals of 0-1000Hz. In this embodiment, the sampling frequency is set to 2500Hz, and the Mallet algorithm is used to implement discrete wavelet transform. Components with frequencies less than 20Hz do not require further differentiation; therefore, the resolution level of the wavelet transform multi-scale decomposition in this embodiment is 5 levels, resulting in a series of wavelet coefficients in different frequency bands. Then, a soft thresholding denoising algorithm is used to process the wavelet coefficients, and the threshold estimation method employs unbiased risk estimation of the filtering threshold.

[0059] In this embodiment, after step S3, in which the abnormal vibration data is input into the VGG-CNN model for damage identification and classification to obtain the damage category prediction result of the manhole cover, the intelligent acquisition and analysis monitoring method for urban manhole safety further includes the following steps.

[0060] Step S4: Determine the health status level of the manhole cover based on the damage category prediction results.

[0061] In this embodiment, the health status levels of the manhole cover, ranked from least to most severe damage, include minor damage, moderate damage, and severe damage. Minor damage includes minor cracks, minor deformation, and minor loosening. Moderate damage includes moderate cracks, moderate deformation, and moderate loosening. Severe damage includes severe cracks, severe deformation, and severe loosening.

[0062] In this embodiment, after step S4, which determines the health status level of the manhole cover based on the damage category prediction result, the intelligent data acquisition and analysis monitoring method for urban manhole safety further includes the following steps.

[0063] Step S5: For manhole covers with severe cracks, severe deformation and / or severe loosening, obtain the location of the manhole cover and report the health status level and location of the manhole cover so that inspection personnel can repair the manhole cover in a timely manner.

[0064] Figure 4 The structure of the MSCNN model is shown. This model includes an input layer, convolutional layers, concatenation layers, maxpooling layers, and an output layer. This structure allows for parallel extraction of input layer data and in-depth feature extraction layer by layer. Furthermore, with only 8 layers, the computational complexity is reduced, improving real-time performance. The input to the MSCNN model is vibration data, specifically preprocessed vibration data, which is in the form of a vibration signal sequence. Therefore, the input layer is the preprocessed vibration signal sequence with a size of 2500. The main feature extraction part has four convolutional layers, each with a length of 1... 1 convolutional layer, 1 2 convolutional layers, 1 3 convolutional layers and 1 The MSCNN model has 5 convolutional layers with ReLU activation. Feature extraction from the input data is performed in parallel to preserve as many features as possible. The convolution process of the MSCNN model is as follows: Figure 5As shown, features are extracted sequentially between convolutional layers to ensure the depth of feature information extracted by the kernels of each layer. The Concat layer concatenates the features of the abnormal vibration signals (cracks, deformation, and loosening) extracted by each convolutional layer. The bias is an additional input to the neuron, used to adjust the neuron's activation level. In the Concat layer, the bias provides a fixed offset, helping the neuron better adapt to different input data, thereby improving the generalization ability of the MSCNN model. Since the features extracted by each convolutional layer from the input layer contain a large amount of repetitive information, and the abnormal vibration signals to be identified have high energy, the max pooling layer effectively identifies this characteristic, avoiding unnecessary computational waste. The MSCNN model is a binary classification model; the output layer uses the sigmoid function, and the loss function is the binary cross-entropy loss function. The output is the probability value that the current vibration data is abnormal vibration data.

[0065] Before using the MSCNN model, it needs to be trained and its parameters optimized. First, 2400 sets of vibration signal sequences were collected from healthy and abnormal manhole covers using a manhole cover monitor. These included 1800 sets of abnormal vibration signal sequences and 600 sets of normal vibration signal sequences. It's important to note that the 1800 sets of abnormal vibration signal sequences were collected from manhole covers categorized into nine types: minor cracks, slight deformation, slight loosening, moderate cracks, moderate deformation, moderate loosening, severe cracks, severe deformation, and severe loosening, with 200 sets from each type. This facilitates more refined diagnosis using the VGG-CNN model later. Then, the vibration signal sequences were labeled to form a dataset, with normal vibration signal sequences labeled as... The abnormal vibration signal sequence was labeled as 1, and then the dataset was divided into three parts in a 7:2:1 ratio: 1680 training sets, 480 test sets, and 240 validation sets. The hyperparameters of the MSCNN model were set as follows: learning rate η = 0.01, batch size = 32, and training epochs = 300. After 300 training epochs, the model accuracy was 92.99%. A series of abnormal vibration candidate probability thresholds t = 0.1, 0.2, ..., 0.9 were determined. 100 validation set samples were input into the MSCNN model to obtain the predicted probability of each sample. p For each threshold t The predicted probability is calculated using the following formula. p Convert to category tags It includes two classification labels, 0 and 1, where 1 is a positive class, representing abnormal vibration data; and 0 is a negative class, representing normal vibration data.

[0066] .

[0067] Then, the evaluation index is calculated. In this embodiment, the F1 score is used as the evaluation index, and the calculation formula is as follows.

[0068] .

[0069] Where F1 is the F1 score, Precision is the proportion of samples predicted as positive that are actually positive, and Recall is the proportion of all positive samples successfully predicted as positive. This embodiment calculates and determines that when... t The optimal classification threshold is 0.7.

[0070] In this embodiment, the model parameters of the VGG-CNN model need to be trained before using it. First, the original vibration data consists of 1800 sets of abnormal vibration signal sequences. These sequences are then re-labeled; for example, a vibration signal sequence with a small crack is labeled as 1, a slightly deformed sequence as 2, a slightly loosened sequence as 3, a moderately cracked sequence as 4, a moderately deformed sequence as 5, a moderately loosened sequence as 6, a severely cracked sequence as 7, a severely deformed sequence as 8, and a severely loosened sequence as 9. Next, the VGG-CNN model is built, and the hyperparameters for model training are set: batch size 32, epoch size 400, learning rate η 0.01, and a dropout rate of 0.1 to prevent overfitting. Finally, the VGG-CNN model is trained, and after 400 epochs, its accuracy reaches 94%. The parameter optimization process for the VGG-CNN model is as follows: Figure 6 As shown.

[0071] In this embodiment, when building the VGG-CNN model, firstly, the abnormal vibration data is preprocessed, converting a one-dimensional vector into a two-dimensional matrix; then, features are extracted through three convolutional layers (C1, C2, and C3), followed by downsampling through two max-pooling layers (P1 and P2); next, features are integrated using two fully connected layers (F1 and F2), and finally, classification is performed through a single softmax layer. The structure of the VGG-CNN model is as follows: Figure 7 As shown. The input data for the input layer is the abnormal vibration data received from the manhole cover terminal by the server, and its size is 2500. Before the abnormal vibration data is input into the convolutional layer C1, the abnormal vibration data (1 The one-dimensional vector transformation of (2500) is (50) 50) Two-dimensional vectors; Convolutional layer C1 uses 16 (5 5) convolutional kernels are applied to the input layer data to form (16) 50 50) Tensor; the output tensor of convolutional layer C1 is then passed through a tensor of size (2 2) The maximum pooling layer P1 is obtained as (16 25 25) Tensors; Convolutional layer C2 then calculates the output tensor of max-pooling layer P1, using 32 (3) tensors. 3) Convolution is performed using convolution kernels, and the output tensor size is (32... 25 25). Convolutional layer C3 then computes the output tensor, using 64 (3) tensors. 3) Convolution is performed using convolution kernels, and the output tensor size is (64... 25 25). Max pooling layer P2 performs max pooling calculation on the output tensor of convolutional layer C3. The size of max pooling layer P2 is (64). 5 5), the output tensor size is (64 5 5). The fully connected layer F1 contains 256 neurons. The input tensor, after flattening, results in a vector of length 1600. Therefore, the size of its weight matrix is ​​(256...). 1600), the output tensor size is (1 256); The principle of the fully connected layer F2 is the same as that of the fully connected layer F1, and the final output tensor is (1 9). Finally, the Softmax layer is used for classification to obtain the probability values ​​of 9 cases: micro-cracks, slight deformation, slight loosening, moderate cracks, moderate deformation, moderate loosening, severe cracks, severe deformation, and severe loosening.

[0072] In this embodiment, the health status levels of manhole covers are categorized in ascending order of damage severity: minor damage, moderate damage, and severe damage. Minor damage includes minor cracks, minor deformation, and minor loosening; moderate damage includes moderate cracks, moderate deformation, and moderate loosening; and severe damage includes severe cracks, severe deformation, and severe loosening. In this embodiment, the damage category corresponding to the highest probability value is taken as the health status of the manhole cover corresponding to the vibration signal sequence. For example, assuming the VGG-CNN model predicts that a certain manhole cover has the highest probability value of 97% for moderate cracks, then the health status of the manhole cover can be determined as moderate cracks, and the health status level of the manhole cover is moderate damage.

[0073] In this embodiment, when the VGG-CNN model infers that the manhole cover corresponding to the vibration signal sequence is at a severe damage level (including severe cracks, severe deformation, and / or severe loosening), it reports the information. In other cases, the information is recorded and written into the FLASH memory inside the STM32 chip, and reported periodically when the 4G module is idle.

[0074] Current traditional urban manhole monitoring methods suffer from a heavy burden due to the massive amount of vibration signal data collected and the need for long-distance transmission to cloud platforms. This increases the information transmission burden and storage pressure on cloud platforms. Furthermore, the vast majority of this data is useless noise, resulting in a significant waste of cloud platform storage resources. This embodiment, based on neural network technology, constructs and trains MSCNN and VGG-CNN models to significantly improve the accuracy of identifying abnormal vibration data. It effectively filters out invalid vibration data, allowing only valid abnormal vibration data to be sent to the server, reducing storage pressure, saving storage space, and solving the current problems of high storage pressure and significant waste of storage resources.

[0075] Currently, due to the limited computing power of manhole cover terminals, it is necessary to analyze the collected abnormal vibration data on the server side to identify minute cracks, slight deformations, or loose displacements, and to assess the health status of the manhole cover. This embodiment combines the MSCNN and VGG-CNN models. First, the MSCNN model is used to identify abnormal vibration data in the vibration data, and then the VGG-CNN model is used to identify, classify, and assess the health status of the abnormal vibration data. This enables accurate identification of abnormal vibration data and accurate assessment of the health status of the manhole cover, solving the problem that the current manhole cover terminal controller has limited computing power and cannot identify, classify, and assess the health status of abnormal vibration data.

[0076] In one exemplary embodiment, an intelligent data acquisition and analysis monitoring system for urban manhole safety is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the intelligent data acquisition and analysis monitoring method for urban manhole safety.

[0077] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0078] This specification uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for intelligent data acquisition, analysis, and monitoring of urban manhole safety, characterized in that, The intelligent data acquisition and analysis monitoring method for urban manhole safety combines the MSCNN model and the VGG-CNN model and applies it to the monitoring scenario of manhole covers in urban manholes. The intelligent data acquisition and analysis monitoring method for urban well safety includes: Obtain vibration data of the manhole cover; The vibration data is input into the MSCNN model to obtain abnormal vibration data; the MSCNN model is a pre-trained model that takes the vibration data as input and the abnormal vibration data as output to identify the abnormal vibration data in the vibration data; the abnormal vibration data is the data corresponding to the abnormal vibration state in the vibration data. The abnormal vibration data is input into the VGG-CNN model for damage identification and classification to obtain the damage category prediction result of the manhole cover; the VGG-CNN model is a pre-trained model that takes the abnormal vibration data as input and the damage category prediction result as output, and is used to predict the health status of the manhole cover based on the abnormal vibration data; the damage category prediction result is the probability value corresponding to each damage category of the manhole cover. The health status level of the manhole cover is determined based on the damage category prediction results; The MSCNN model accurately identifies abnormal vibration data based on vibration data. The abnormal vibration data output by the MSCNN model is used as the input of the VGG-CNN model. Then, by inputting the abnormal vibration data into the VGG-CNN model, the VGG-CNN model is used to identify and classify damage to the manhole cover and assess its health status. The preprocessed vibration data is input into the microcontroller of the manhole cover terminal. The microcontroller uses a pre-built MSCNN model to analyze the vibration data and determine whether it is abnormal. If it is, the abnormal vibration data is written to the SD card and then fed back to the microcontroller. The abnormal vibration data is also transmitted to the server via the 4G module. Otherwise, it is discarded. The abnormal vibration data uploaded to the server is input into a pre-built VGG-CNN model to identify, classify, and assess the health status of the manhole cover, thus determining the health status level. When the VGG-CNN model infers that the manhole cover corresponding to the vibration signal sequence is severely damaged, it is reported. Other cases are recorded, written to the FLASH memory inside the STM32 chip, and reported periodically when the 4G module is idle. The MSCNN model structure includes an input layer, convolutional layers, a concatenation layer, a max pooling layer, and an output layer. The input data of the input layer is preprocessed vibration data. The convolutional layers include 1×1, 1×2, 1×3, and 1×5 convolutional layers. Feature extraction of the preprocessed vibration data adopts a parallel mode. The convolution process of the 1×1, 1×2, 1×3, and 1×5 convolutional layers is that features are extracted sequentially between each convolutional layer. The VGG-CNN model structure includes an input layer, a convolutional layer C1, a max-pooling layer P1, a convolutional layer C2, a convolutional layer C3, a max-pooling layer P2, a fully connected layer F1, and a fully connected layer F2. The input data for the input layer is abnormal vibration data. The convolutional layer C1 uses 16 5×5 convolutional kernels to convolve the input data, forming a tensor of size 16×50×50. This 16×50×50 tensor is then passed through a 2×2 max-pooling layer P1 to obtain a larger value. The size of the tensor is 16×25×25. The convolutional layer C2 uses 32 3×3 convolutional kernels for convolution, and the output tensor size is 32×25×25. The convolutional layer C3 uses 64 3×3 convolutional kernels for convolution, and the output tensor size is 64×25×25. The max pooling layer P2 performs max pooling calculation on the output tensor of the convolutional layer C3, which has a size of 64×25×25. The size of the max pooling layer P2 is 64×5×5, and the output tensor size is 64×5×5.

2. The intelligent data acquisition and analysis monitoring method for urban manhole safety according to claim 1, characterized in that, Obtain vibration data from manhole covers, specifically including: An accelerometer is installed at the center of the back of the manhole cover. The accelerometer is used to measure the vertical vibration data of the manhole cover at a predetermined frequency. The vibration data is collected by monitoring the manhole cover using the acceleration sensor.

3. The intelligent data acquisition and analysis monitoring method for urban manhole safety according to claim 1, characterized in that, Following the step of acquiring vibration data from manhole covers, the intelligent data acquisition and analysis monitoring method for urban manhole safety further includes: The vibration data is preprocessed by data cleaning, data normalization and data noise reduction to obtain preprocessed vibration data; the preprocessed vibration data is used as input to the MSCNN model to obtain the abnormal vibration data.

4. The intelligent data acquisition and analysis monitoring method for urban manhole safety according to claim 3, characterized in that, The vibration data undergoes preprocessing including data cleaning, data normalization, and data noise reduction to obtain preprocessed vibration data, specifically including: Remove outliers from the vibration data and fill in missing values ​​to obtain cleaned vibration data. The cleaned vibration data is normalized to obtain normalized vibration data. A wavelet denoising algorithm is used to denoise the normalized vibration data to obtain the preprocessed vibration data.

5. The intelligent data acquisition and analysis monitoring method for urban manhole safety according to claim 1, characterized in that, After inputting the abnormal vibration data into the VGG-CNN model for damage identification and classification to obtain the damage category prediction result of the manhole cover, the intelligent data acquisition and analysis monitoring method for urban manhole safety further includes: Based on the damage category prediction results, the health status level of the manhole cover is determined.

6. The intelligent data acquisition and analysis monitoring method for urban manhole safety according to claim 5, characterized in that, The health status levels of the manhole covers, in ascending order of damage severity, include minor damage, moderate damage, and severe damage.

7. The intelligent data acquisition and analysis monitoring method for urban manhole safety according to claim 6, characterized in that, The damage categories corresponding to the minor damage level include minor cracks, minor deformation, and minor loosening; the damage categories corresponding to the moderate damage level include moderate cracks, moderate deformation, and moderate loosening; and the damage categories corresponding to the severe damage level include severe cracks, severe deformation, and severe loosening.

8. The intelligent data acquisition and analysis monitoring method for urban manhole safety according to claim 7, characterized in that, After determining the health status level of the manhole cover based on the damage category prediction results, the intelligent data acquisition and analysis monitoring method for urban manhole safety further includes: For manhole covers with severe cracks, severe deformation, and / or severe loosening, the location of the manhole cover is obtained, and the health status level and location of the manhole cover are reported so that inspection personnel can repair the manhole cover in a timely manner.

9. The intelligent data acquisition and analysis monitoring method for urban manhole safety according to claim 2, characterized in that, The accelerometer has a sampling frequency of 2500Hz.

10. An intelligent data acquisition and analysis monitoring system for urban manhole safety, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the intelligent acquisition and analysis monitoring method for urban manhole safety as described in any one of claims 1-9.

Images