Efficient crack detection and intelligent risk assessment system and method

Abstract

The invention provides an efficient crack detection and intelligent risk assessment system, belongs to the technical field of structural health monitoring, and is particularly suitable for crack detection in buildings, bridges, tunnels and industrial equipment. According to the system, comprehensive detection on the surface and the interior of a crack is realized by combining an optical imaging sensor, a laser scanning sensor and an ultrasonic sensor through a multi-mode sensor fusion technology. The system comprises a data acquisition unit, a data processing unit, a multi-modal data fusion module, a three-dimensional modeling module and an intelligent decision support module. Through an adaptive learning algorithm and a deep learning technology, the system can dynamically optimize a detection model and provide a high-precision detection result. Meanwhile, the system supports three-dimensional reconstruction and visual display of the crack, carries out real-time assessment on the crack risk through an intelligent decision support module, automatically generates a health assessment report and gives out an early warning. Compared with traditional manual detection, the system has the advantages that the comprehensiveness, the automation degree and the efficiency of detection are greatly improved, and the system is suitable for structural health monitoring in various complex scenes.

Description

Technical Field

[0001] The present invention relates to the technical field of crack detection, and in particular to a high-efficiency crack detection and intelligent risk assessment system and method. Background Art

[0002] Crack detection and risk assessment are key components of structural health monitoring, particularly in infrastructure such as buildings, bridges, and tunnels. Traditional manual inspections are time-consuming and subjective, and single-sensor systems (such as optical imaging or laser scanning) often have blind spots when capturing internal and external crack characteristics, making them inadequate for the safety needs of modern, complex infrastructure.

[0003] Chinese patent CN117741808A discloses a geological fracture detection method based on geological exploration. By combining multi-wavelength ground penetrating radar (GPR) and a microseismic sensing network, it significantly improves the accuracy and depth resolution of geological fracture detection. This integrated approach enables detection to cover a wider depth range and identify fractures with greater accuracy, thereby significantly reducing the risk of misjudgment and providing a more reliable basis for geological safety assessment. However, the patent's method for obtaining fracture information is relatively simple and cannot obtain comprehensive fracture characteristic information. Furthermore, the patent's adaptability is relatively limited, performing well in specific scenarios but having poor adaptability to environmental changes.

[0004] In order to solve the above problems, there is an urgent need for an intelligent crack detection system that can automatically adapt to complex scenarios, provide accurate detection results and perform health assessments. Summary of the Invention

[0005] The present invention aims to provide an efficient crack detection and intelligent risk assessment system and method to address the issues raised in the background art. By integrating multimodal fusion, adaptive learning, and intelligent decision support, this invention utilizes data fusion, online learning, 3D reconstruction, and risk assessment technologies to significantly enhance the comprehensiveness and automation of crack detection.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] An efficient crack detection and intelligent risk assessment system, comprising:

[0008] A multimodal sensor module for detecting the surface and internal features of the crack. The multimodal sensor includes: an optical imaging sensor for acquiring high-resolution images of the crack surface; a laser scanning sensor for acquiring the three-dimensional geometry of the crack; and an ultrasonic sensor for detecting the depth and internal structure of the crack.

[0009] a data acquisition unit connected to the multimodal sensor module and configured to synchronously acquire optical, laser, and ultrasonic data;

[0010] A data processing unit is used to pre-process the collected data, including image denoising, data smoothing and time alignment processing;

[0011] Multimodal data fusion module, which fuses the data from multimodal sensors based on deep learning algorithms to extract comprehensive characteristics of cracks;

[0012] 3D modeling module, which generates a 3D model of the crack based on the fused data;

[0013] The intelligent decision support module is used to analyze and assess risks based on crack characteristics and generate structural health assessment reports and early warning information for cracks.

[0014] Preferably, the optical imaging sensor has a resolution of 640×480 pixels and a thermal sensitivity of less than 30 mK, and is installed directly in front of the crack surface to obtain a clear image of the crack surface;

[0015] The laser scanning sensor has a scanning angle of 190° and a maximum measuring distance of 80 meters. It is installed on the side of the crack to measure the horizontal and vertical changes of the crack;

[0016] The ultrasonic sensor has a frequency range of 0.5MHz to 10MHz and a measurement accuracy of 0.01mm. It is installed above the crack and moved by a robotic arm to detect the crack depth and changes in its internal structure.

[0017] Preferably, the data processing unit uses a median filter algorithm to denoise the optical image, uses a Gaussian filter to smooth the laser scanning data, and uses linear interpolation to time-align the ultrasonic data to ensure synchronization and consistency of multimodal data.

[0018] Preferably, the multimodal data fusion module uses a convolutional neural network (CNN) to extract features from various sensor data. The convolutional neural network (CNN) includes 4 convolutional layers and 2 fully connected layers, which are used to extract features such as the edge, depth and morphology of the cracks.

[0019] Preferably, the three-dimensional modeling module generates a three-dimensional model of the crack using a voxelization algorithm based on the fused data, with the voxel size set to 1mm×1mm×1mm. The three-dimensional model is used to accurately describe the spatial structure and expansion trend of the crack.

[0020] Preferably, the intelligent decision support module performs real-time analysis of the depth, width, and expansion speed of the crack based on a deep reinforcement learning algorithm, and automatically generates maintenance recommendations and warning signals based on the analysis results. When the crack depth exceeds 2 mm or the width exceeds 0.5 mm, an early warning is automatically triggered.

[0021] Preferably, the intelligent decision support module assesses crack risk and optimizes maintenance strategies using a Q-learning reinforcement learning algorithm. The module generates a structural health assessment report based on real-time data. The report includes crack risk levels, recommended maintenance measures, and early warning prompts.

[0022] Preferably, the data acquisition unit is connected to the multimodal sensor via an optical fiber or an Ethernet cable, and the data acquisition frequency is 10 Hz to ensure real-time synchronous transmission and processing of data from each sensor.

[0023] Preferably, the system further comprises an auxiliary sensor module for detecting environmental parameters, wherein the environmental parameters include temperature, humidity and vibration, and the auxiliary sensor module is used to optimize the data processing algorithm in the crack detection and assessment process.

[0024] The present invention further protects a method for intelligent crack measurement, comprising the following contents:

[0025] S1. Collecting crack surface and internal feature data through a multimodal sensor module, wherein the multimodal sensor module includes an optical imaging sensor, a laser scanning sensor, and an ultrasonic sensor;

[0026] S2. Preprocessing the collected data by the data acquisition unit, wherein the preprocessing includes image denoising, data smoothing, and time alignment to ensure data quality and consistency;

[0027] S3. Use the multimodal data fusion module to fuse optical, laser, and ultrasonic data through deep learning algorithms to extract comprehensive crack characteristics;

[0028] S4. Produce a three-dimensional model of the crack through a three-dimensional modeling module to show the spatial structure, position and morphology of the crack;

[0029] S5. Analyze crack feature data, assess crack severity, and generate maintenance recommendations through the intelligent decision support module.

[0030] S6. Generate a structural health assessment report, providing detailed crack characteristics, 3D models, and maintenance plans.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] (1) Application of multimodal fusion technology: The system combines optical, laser, and ultrasonic sensors to comprehensively detect the surface and internal characteristics of cracks. Traditional systems rely on a single sensor and can usually only obtain information on the crack surface, but cannot effectively identify the depth and internal changes of the crack.

[0033] (2) Adaptive learning and online optimization: This system introduces an adaptive learning algorithm that can dynamically optimize the crack detection model based on real-time detection data. This allows the system to maintain high accuracy and reliability under different detection environments and crack characteristics. Traditional methods usually perform well in specific scenarios but have poor adaptability to environmental changes.

[0034] (3) Automated Decision Support and Early Warning Mechanism: This invention features an intelligent decision support system that automatically performs risk assessments after detection and issues early warnings and maintenance recommendations based on the severity of cracks. Compared to traditional manual detection and analysis, the intelligent decision support system significantly improves detection efficiency and decision accuracy.

[0035] (4) 3D reconstruction and visualization: The system supports 3D modeling of cracks based on fused multimodal data and provides real-time visualization. This function not only helps inspectors intuitively understand the spatial distribution and morphology of cracks, but also provides a reference for subsequent structural maintenance.

[0036] (5) Automation and efficiency: This system improves detection efficiency and reduces the need for human intervention through the collaborative work of multiple sensors and automated data processing. In complex infrastructure such as bridges or tunnels, large-scale, long-term continuous monitoring can be achieved, thereby significantly improving work efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, a brief introduction to the drawings involved in the embodiments is now provided. It is obvious that the drawings described below are only schematic illustrations of some embodiments of the present invention. Those skilled in the art can construct other forms of drawings based on these drawings without inventive effort.

[0038] Figure 1 This is a flow chart of a crack intelligent measurement method proposed by the present invention;

[0039] Figure 2 This is a flow chart of the intelligent crack measurement method mentioned in Example 1 of the present invention;

[0040] Figure 3 This is a workflow diagram of the intelligent decision support module mentioned in Example 1 of the present invention;

[0041] Figure 4This is the multimodal data fusion flow chart mentioned in Example 1 of the present invention;

[0042] Figure 5 This is a simplified block diagram of the sensor and data processing mentioned in Example 1 of the present invention. DETAILED DESCRIPTION

[0043] The following is a clear and comprehensive description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. It should be noted that the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. 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.

[0044] The purpose of this invention is to provide a novel intelligent crack measurement system that combines optical, laser, and ultrasonic sensors to comprehensively detect crack surfaces and their interiors. Through adaptive learning and intelligent decision-making support technology, it addresses the shortcomings of existing detection methods in terms of accuracy, efficiency, and adaptability.

[0045] In order to make the above-mentioned objectives, features and advantages of the present invention more clear, the technical solutions of the present invention are described in detail below with reference to the accompanying drawings and specific implementation methods.

[0046] Example 1:

[0047] See also Figure 2-5 This embodiment proposes an efficient crack detection and intelligent risk assessment method. The method is implemented based on an efficient crack detection and intelligent risk assessment system. The system includes:

[0048] A multimodal sensor module for detecting the surface and internal features of the crack. The multimodal sensor includes: an optical imaging sensor for acquiring high-resolution images of the crack surface; a laser scanning sensor for acquiring the three-dimensional geometry of the crack; and an ultrasonic sensor for detecting the depth and internal structure of the crack.

[0049] a data acquisition unit connected to the multimodal sensor module and configured to synchronously acquire optical, laser, and ultrasonic data;

[0050] A data processing unit is used to pre-process the collected data, including image denoising, data smoothing and time alignment processing;

[0051] Multimodal data fusion module, which fuses the data from multimodal sensors based on deep learning algorithms to extract comprehensive characteristics of cracks;

[0052] 3D modeling module, which generates a 3D model of the crack based on the fused data;

[0053] Intelligent decision support module, used to analyze and assess risks based on crack characteristics, and generate structural health assessment reports and early warning information for cracks;

[0054] Auxiliary sensor modules are used to optimize data processing algorithms during crack detection and assessment; where environmental parameters include temperature, humidity and vibration.

[0055] The above method specifically includes the following steps:

[0056] Step 1: Build the system architecture: Construct the hardware and software architecture of the crack detection system, including the following:

[0057] Step 1.1, hardware setup:

[0058] The hardware includes optical imaging sensors, laser scanning sensors, and ultrasonic sensors, which are used to obtain multi-dimensional information about cracks. The specific configuration and installation details are as follows:

[0059] Optical sensor: Use an optical imaging sensor with a resolution of 640 x 480 pixels and a thermal sensitivity of less than 30 mK. Mount the sensor 1 m in front of the crack surface, ensuring its optical axis is perpendicular to the crack surface to prevent image distortion. Mount the sensor on a tripod to ensure stable and reliable image acquisition, even if the house moves or vibrates during measurement.

[0060] Laser scanning sensor: A laser scanning sensor with a 190° scanning angle range and a resolution of 0.25° is used, with a maximum measurement distance of 80m. The laser scanning sensor is installed to the side of the crack, 2m from the crack, and at a height of 0.5m to 1.5m to ensure accurate scanning of both lateral and longitudinal variations in the crack. The sensor scanning plane should be perpendicular to the crack direction to obtain comprehensive information about the crack geometry.

[0061] Ultrasonic sensor: A frequency range of 0.5MHz to 10MHz with a measurement accuracy of 0.01mm was selected. The ultrasonic sensor was installed above the crack, approximately 50cm away. The probe was moved smoothly by a robotic arm, scanning along the crack at a rate of one second per second to detect the crack depth and internal structure.

[0062] Step 1.2, connect the data acquisition unit to the control unit:

[0063] The sensors are connected to a high-speed data acquisition unit via optical fiber or Ethernet cables. The data acquisition unit should be capable of processing high-speed data from multiple sensors. This unit collects the sensor input signals in real time and transmits them to the central control unit via a high-speed data channel.

[0064] The control unit uses an industrial computer with a high-performance processor to run the appropriate data acquisition and processing software. The control unit centrally manages the sensors through a graphical interface and processes the collected data in real time.

[0065] Step 1.3, installation and debugging:

[0066] Sensor fixation: Use a stable bracket system to fix the optical imaging sensor and laser scanning sensor in the predetermined position to ensure that there is no displacement or angle change during the detection process;

[0067] Laser scanning sensor debugging: Set the laser scanning sensor's scanning angle to 120° in the software to cover the left and right areas of the crack. Use standard calibration tools to calibrate the laser scanning sensor's distance measurement function to ensure that its measurement error is controlled within ±1mm;

[0068] Ultrasonic sensor adjustment: The ultrasonic sensor is flexibly mounted using a robotic arm to ensure perpendicular contact between the probe and the crack surface. The ultrasonic sensor is calibrated using a standard thickness test block and its measurement frequency is adjusted to 2 MHz, ensuring a crack depth accuracy of 0.01 mm.

[0069] Data acquisition unit configuration: In the data acquisition unit, the input channels of the multimodal sensor are configured, and the acquisition frequency is set to 10 Hz to ensure the synchronous acquisition and real-time display of sensor data.

[0070] System electrical connection: All sensors are connected to the data acquisition unit through dedicated cables, and a regulated power supply is used to ensure stable power supply to the equipment to avoid interference from power fluctuations during the measurement process.

[0071] Step 1.4, Software System Construction: The software consists of multiple modules, including data preprocessing, multimodal data fusion, 3D modeling, adaptive learning, and intelligent decision support. The software system is developed using a graphical programming language and Python.

[0072] In step 1.4, the software system is built as follows:

[0073] Step 1.4.1, data processing module:

[0074] Optical imaging processing: A real-time median filter algorithm is used to denoise the optical image, using a 3×3 pixel window size to ensure crack image clarity and remove noise. The optical sensor is geometrically corrected using calibration tools within the software to eliminate optical distortion that may occur during the imaging process and ensure the geometric accuracy of the image data.

[0075] Laser Scanning Data Processing: Gaussian filtering is applied to the laser scanning data with a standard deviation of 1.0 to ensure smoothness and continuity. This filtered data provides an accurate description of the fracture geometry. The coordinate transformation module converts the laser scanning data into a world coordinate system compatible with other sensors to ensure data consistency.

[0076] Ultrasonic data processing: Receives real-time ultrasonic thickness measurement data and uses a linear interpolation algorithm to time-align the data. The interpolation step size is 0.01s to ensure that the ultrasonic data is aligned with other sensor data on the same time axis.

[0077] Step 1.4.2, Multimodal Data Fusion Module: This module spatially aligns the optical image, laser scan, and ultrasonic data to achieve data fusion. First, the pixels in the optical image are converted to physical coordinates using an extrinsic matrix generated by a calibration tool. The fusion algorithm then performs a weighted average of each sensor data type, with initial weights of 0.4 for optical, 0.3 for laser, and 0.3 for ultrasonic. During the actual detection process, the weights are adjusted in real time by an adaptive algorithm based on the sensor's detection accuracy.

[0078] Step 1.4.3, 3D Modeling Module: Use the 3D graphics library to construct a 3D model of the crack. Based on the fused multimodal data, a voxelization algorithm is used to generate a 3D mesh model. The voxel size is set to 1×1×1 mm to ensure model precision. The system also provides real-time visualization of the 3D model, allowing users to observe the 3D morphology and location distribution of the cracks through the software interface.

[0079] Step 1.4.4, Adaptive Learning Module: Use a convolutional neural network (CNN) to automatically extract and learn crack features. This convolutional network consists of four convolutional layers and two fully connected layers, with a 3×3 kernel size. It extracts features such as crack edges and depth. The software uses an API to call the convolutional neural network for real-time feature extraction and classification. The model is continuously optimized based on new inspection data through an online learning mechanism to ensure the accuracy of inspection results.

[0080] Step 1.4.5, Intelligent Decision Support System: The decision support system generates a crack health assessment based on the inspection data. A built-in decision tree model analyzes the inspection results. When a crack exceeds 2mm in depth or 0.5mm in width, the system automatically triggers an early warning signal and generates specific maintenance recommendations. This early warning mechanism displays real-time alarm information on the software interface and generates a complete inspection report.

[0081] Step 2, data collection and processing;

[0082] Multi-dimensional crack data, including surface and internal characteristics, is collected from each sensor. Pre-processing techniques such as image denoising, data smoothing, and time alignment are applied to ensure data accuracy and consistency. Specifically, the following are performed:

[0083] Step 2.1, data acquisition: Obtain multi-dimensional information of the cracks through optical imaging sensors, laser scanning sensors, and ultrasonic sensors;

[0084] Step 2.2, data preprocessing: Preprocessing includes optical image denoising, laser scanning data smoothing, and ultrasonic data time alignment;

[0085] Furthermore, in step 2.2, the optical image is denoised as follows:

[0086] Step 2.2.1, input data preparation: the input is a noisy optical image I(x,y), and the output is a denoised image

[0087] Step 2.2.2, encoder part: use 3 convolutional layers for feature extraction, and the calculation formula of each convolutional layer is:

[0088]

[0089] in, is the output feature map of the lth convolutional layer at position (x, y); is the convolution kernel weight matrix of the lth layer; is the bias vector of the lth layer; * is the convolution operator, which is used to convolve the input feature map with the convolution kernel; ReLU is the activation function, which is used to introduce nonlinear features;

[0090] Step 2.2.3, latent feature representation: The last layer feature map output by the encoder is represented as the latent feature h encoded , its size is m×n×c, where m and n are the height and width of the feature map, and c is the number of channels;

[0091] Step 2.2.4, decoder part: The decoder transforms the latent feature h encoded Return the denoised image to its original size:

[0092]

[0093] in, The denoised image output by the decoder; is the weight matrix of the deconvolution kernel of the lth layer; Input feature map to the l-1 layer decoder; Sigmoid is used to limit the output to [0,1];

[0094] Step 2.2.5, loss function calculation: Use mean square error (MSE) as the loss function to measure the denoising effect:

[0095]

[0096] Where I(x,y) is the pixel value of the original noisy image; is the pixel value of the denoised image;

[0097] Step 2.2.6, model training: Use the Adam optimizer to train the convolutional autoencoder, set the learning rate to 0.001, the batch size to 32, and the number of training iterations to 50 rounds;

[0098] In step 2.3, a one-dimensional convolutional neural network (1D CNN) is used to smooth the laser scanning data and extract the depth features of the cracks. Specifically, the following are performed:

[0099] Step 2.3.1, input data; the input is the laser scanning data sequence L(x), and the output is the smoothed data sequence

[0100] Step 2.3.2, Convolutional Layer: Use multiple convolutional layers to extract high-level features of the aurora data;

[0101]

[0102] in, is the output value of the lth convolutional layer at position x; is the weight matrix of the convolution kernel of the lth layer; is the input value of the l-1th convolutional layer at position x+i; is the bias term of the lth layer; k is the radius of the convolution kernel;

[0103] Step 2.3.3, Pooling layer: Use the pooling layer to downsample the output of the convolutional layer to reduce the dimension of the data;

[0104]

[0105] in, is the output value of the lth pooling layer at position x; p is the pooling window size;

[0106] Step 2.3.4, Fully Connected Layer: Flatten the output of the pooling layer into a one-dimensional vector and input it to the fully connected layer;

[0107]

[0108] Where y is the smoothed laser depth data; W fc is the weight matrix of the fully connected layer; b fcis the bias term of the fully connected layer; To flatten the output of the pooling layer into a one-dimensional vector;

[0109] Step 2.3.5, loss function calculation: Use mean square error (MSE) as the loss function to measure the smoothing effect;

[0110]

[0111] Where M is the total length of the laser data sequence; L true (x) is the manually labeled smoothed real data; y(x) is the smoothed data predicted by the model;

[0112] Step 2.3.6, model training: Use the noisy laser data and the real smooth data to train the model, with a learning rate of 0.0001, a batch size of 64, and 100 training rounds;

[0113] Step 2.4, Ultrasonic Data Time Alignment and Automatic Calibration: Use a long short-term memory network (LSTM) to time-align the ultrasonic data and implement automatic calibration, specifically including the following:

[0114] Step 2.4.1, input data: input is unaligned ultrasonic time series data U(t), output is aligned ultrasonic data U aligned (t);

[0115] Step 2.4.2, LSTM unit calculation:

[0116] f t =σ(W f ·[h t-1 ,U t ]+b f )

[0117] i t =σ(W i ·[h t-1 ,U t ]+b i )

[0118]

[0119] o t =σ(W o ·[h t-1 ,U t ]+b o )

[0120] h t =o t *tanh(C t )

[0121] Among them, f t is the forget gate, which is used to determine the importance of current information; i t is the input gate, which is used to determine the impact of the current input on the memory; is the candidate memory cell state; C t is the update status of the memory unit; o t is the output gate, which is used to determine the impact of the current state on the output; h t is the hidden state of the current time step; σ is the activation function; W f , W i , W C , W o is the weight matrix of each gate; b f , b i , b C , b o is the bias term of each gate;

[0122] Step 2.4.3, output calculation: convert the hidden state h t Pass it to the fully connected layer to calculate the aligned ultrasonic data:

[0123] U aligned (t) = W h ·h t +b h

[0124] Among them, W h is the weight matrix of the fully connected layer; b h is the bias term of the fully connected layer;

[0125] Step 2.4.4, loss function calculation: Use the time series prediction loss function to measure the alignment effect:

[0126]

[0127] Where T is the total length of the time series; U true (t) is the real ultrasonic data after alignment; U aligned (t) is the alignment data predicted by the model;

[0128] Step 2.4.5, model training: Use the real aligned data and unaligned data to train the LSTM model, with a learning rate of 0.001, a batch size of 32, and 150 training rounds.

[0129] Step 3: multimodal data fusion and model training;

[0130] Data from different sensors is integrated through a deep learning framework to extract comprehensive crack features. Adaptive learning algorithms are used to continuously optimize the detection model to accommodate different detection scenarios and changes in crack features. This includes the following:

[0131] Step 3.1, multimodal data fusion:

[0132] The multimodal data fusion module adaptively fuses optical images, laser scanning data, and ultrasonic data through a deep learning framework to generate comprehensive crack feature information, including the following:

[0133] Step 3.1.1, Data Fusion Step: Use Convolutional Neural Network (CNN) to extract features from each sensor data:

[0134] Optical image characteristics: F optical =CNN optical (I filtered (x,y))

[0135] Laser data characteristics: F laser =CNN laser (I filtered (x,y))

[0136] Ultrasonic data characteristics: F ultrasonic =CNN ultrasonic (U aligned (t))

[0137] Among them, F optical 、F laser 、F ultrasonic represent the extracted features of optical, laser, and ultrasonic data, respectively;

[0138] Step 3.1.2, feature fusion: combine the features extracted by each sensor:

[0139] F fused =concat(F optical ,F laser ,F ultrasonic )

[0140] Among them, F fused is the fused feature vector;

[0141] Step 3.1.3, Deep Fusion Model: Further process the fusion features through the fully connected layer:

[0142] F output =ReLU(W f ·F fused +b f )

[0143] Among them, F output is the final feature output after fusion; W f is the weight matrix of the fully connected layer; b f is the bias term of the fully connected layer;

[0144] Step 3.1.4: The fused data is used for 3D modeling: a 3D model of the crack is generated based on the fused features, and a 3D network is constructed using a voxelization method:

[0145]

[0146] Where V(x, y, z) represents the voxel value in the three-dimensional model; threshold is the threshold used to determine the presence of cracks;

[0147] Step 3.2, Adaptive Learning and Online Optimization: Introducing a deep neural network (DNN) to automatically learn and optimize crack features to meet the needs of crack detection in different scenarios. Specifically, it includes the following:

[0148] Step 3.2.1, deep learning model structure:

[0149] Model architecture: The deep neural network consists of multiple hidden layers, each using the ReLU activation function:

[0150] h (l) =ReLU(W (l) ·h (l-1) +b (l) )

[0151] Among them, h (l) is the output of layer l; W (l) is the weight matrix of the lth layer; b (l) is the bias of layer l;

[0152] Output layer: Use the softmax function to output the probability distribution of crack types:

[0153]

[0154] Among them, z is the linear combination result of the output layer; K is the total number of categories;

[0155] Step 3.2.2, online learning mechanism:

[0156] Model update: Use cross entropy loss function for model training:

[0157]

[0158] Where N is the total number of samples, y i is the true label;

[0159] Gradient descent method: Update model parameters:

[0160]

[0161] Among them, θ is the model parameter and η is the learning rate;

[0162] Online adaptability: The model is dynamically updated based on new data collected in real time, ensuring it adapts to different fracture characteristics.

[0163] Step 4: Intelligent decision-making and early warning;

[0164] After the detection is completed, the system automatically conducts a risk assessment of the cracks and generates a health assessment report; specifically, it includes the following:

[0165] Step 4.1, Intelligent Decision Support System: The intelligent decision support system automatically generates a structural health assessment report and provides maintenance recommendations based on deep reinforcement learning (DRL).

[0166] State definition: The state s represents the multimodal characteristics of the crack, such as depth, width, and expansion rate:

[0167] s=[d depth ,d width ,v expansion ]

[0168] Among them, d depth Indicates the crack depth, d width represents the crack width, v expansion Indicates the crack expansion rate;

[0169] Step 4.2, Action Space: Action a is a maintenance operation that the system can perform. The selection of each action depends on the current state s and is decided by a deep reinforcement learning model. It is defined as follows:

[0170] Action a1: When the crack risk reaches the warning threshold, the system automatically issues an alarm:

[0171]

[0172] Among them, r(a1) represents the reward value of executing action a1, threshold d Indicates the warning threshold of crack depth, threshold v Indicates the crack expansion speed warning threshold;

[0173] Action a2: Generate a maintenance recommendation report based on the crack detection results:

[0174] r(a2)=f(d depth ,d width ,vexpansion )

[0175] Where r(a2) represents the reward value for executing action a2, f represents the function used to evaluate the effectiveness of the maintenance suggestion, and generates the corresponding maintenance strategy based on the current crack characteristics;

[0176] Action a3: Automatically arrange on-site inspections in high-risk areas:

[0177]

[0178] Among them, r(a3) represents the reward value for executing action a3, and risk level represents the risk level determined based on the crack detection results;

[0179] Step 4.3, Reinforcement Learning Algorithm: Update the policy using Q-learning:

[0180]

[0181] Where Q(s,a) represents the value of performing action a in the current state s; r represents the current reward; s′ represents the new state after performing the action; a represents the learning rate; γ represents the discount factor;

[0182] Step 4.4, strategy selection: Use the -greedy strategy to select actions:

[0183]

[0184] Step 4.5, assess the risk level: Based on the depth and expansion rate of the crack, use a threshold to determine the risk level of the crack (low, medium, high);

[0185] Step 4.6, generate report: Combine status information and action selection to generate a structural health report, outputting specific maintenance recommendations and emergency treatment measures.

[0186] The remaining technical features of this embodiment can be flexibly selected by those skilled in the art based on actual circumstances to meet different specific practical needs. However, it is obvious to those skilled in the art that these specific details are not required to practice the present invention. In other examples, in order to avoid obscuring the present invention, the composition, structure, or components of the formula are not specifically described, and are all within the technical protection scope defined by the technical solution claimed in the claims of the present invention.

[0187] Modifications and variations made by those skilled in the art without departing from the spirit and scope of the present invention are intended to be within the scope of the claims appended hereto. In the foregoing description, numerous specific details are set forth to provide a thorough understanding of the present invention. However, it will be apparent to those skilled in the art that these specific details are not necessarily required to practice the present invention. In other instances, well-known techniques, such as specific construction details, operating conditions, and other technical requirements, are not described in detail to avoid obscuring the present invention.

[0188] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. An efficient crack detection and intelligent risk assessment system, characterized in that: The system comprises: A multimodal sensor module for detecting the surface and internal features of the crack. The multimodal sensor includes: an optical imaging sensor for acquiring high-resolution images of the crack surface; a laser scanning sensor for acquiring the three-dimensional geometry of the crack; and an ultrasonic sensor for detecting the depth and internal structure of the crack. a data acquisition unit connected to the multimodal sensor module and configured to synchronously acquire optical, laser, and ultrasonic data; A data processing unit is used to pre-process the collected data, including image denoising, data smoothing and time alignment processing; Multimodal data fusion module, which fuses the data from multimodal sensors based on deep learning algorithms to extract comprehensive characteristics of cracks; 3D modeling module, which generates a 3D model of the crack based on the fused data; The intelligent decision support module is used to analyze and assess risks based on crack characteristics and generate structural health assessment reports and early warning information for cracks.

2. The efficient crack detection and intelligent risk assessment system according to claim 1 is characterized in that: The optical imaging sensor has a resolution of 640×480 pixels and a thermal sensitivity of less than 30 mK. The optical imaging sensor is installed directly in front of the crack surface to obtain a clear image of the crack surface; The laser scanning sensor has a scanning angle of 190° and a maximum measuring distance of 80 meters. It is installed on the side of the crack to measure the horizontal and vertical changes of the crack; The ultrasonic sensor has a frequency range of 0.5MHz to 10MHz and a measurement accuracy of 0.01mm. It is installed above the crack and moved by a robotic arm to detect the crack depth and changes in its internal structure.

3. The efficient crack detection and intelligent risk assessment system according to claim 2 is characterized in that: The data processing unit uses a median filter algorithm to denoise the optical image, uses a Gaussian filter to smooth the laser scanning data, and performs time alignment on the ultrasonic data through linear interpolation to ensure the synchronization and consistency of multimodal data.

4. The efficient crack detection and intelligent risk assessment system according to claim 3 is characterized in that: The multimodal data fusion module uses a convolutional neural network to extract features from various sensor data. The convolutional neural network includes four convolutional layers and two fully connected layers to extract the edge, depth and morphological features of the cracks.

5. The efficient crack detection and intelligent risk assessment system according to claim 4 is characterized in that: The three-dimensional modeling module generates a three-dimensional model of the crack using a voxelization algorithm based on the fused data, with the voxel size set to 1 mm×1 mm×1 mm. The three-dimensional model is used to accurately describe the spatial structure and expansion trend of the crack.

6. The efficient crack detection and intelligent risk assessment system according to claim 5, characterized in that: The intelligent decision support module uses a deep reinforcement learning algorithm to perform real-time analysis of the depth, width, and expansion speed of cracks, and automatically generates maintenance recommendations and warning signals based on the analysis results. When the crack depth exceeds 2mm or the width exceeds 0.5mm, an early warning is automatically triggered.

7. The efficient crack detection and intelligent risk assessment system according to claim 6, characterized in that: The intelligent decision support module uses a Q-learning reinforcement learning algorithm to assess crack risks and optimize maintenance strategies. It combines real-time data to generate a structural health assessment report that includes crack risk levels, recommended maintenance measures, and early warning prompts.

8. The efficient crack detection and intelligent risk assessment system according to claim 7, characterized in that: The data acquisition unit is connected to the multimodal sensor via an optical fiber or Ethernet cable, and the data acquisition frequency is 10 Hz to ensure real-time synchronous transmission and processing of the data from each sensor.

9. The efficient crack detection and intelligent risk assessment system according to claim 8, characterized in that: The system also includes an auxiliary sensor module for detecting environmental parameters, including temperature, humidity, and vibration, and the auxiliary sensor module is used to optimize the data processing algorithm in the crack detection and assessment process.

10. An efficient crack detection and intelligent risk assessment method implemented by the system according to any one of claims 1 to 9, characterized in that: Includes the following: S1. Collecting crack surface and internal feature data through a multimodal sensor module, wherein the multimodal sensor module includes an optical imaging sensor, a laser scanning sensor, and an ultrasonic sensor; S2. Preprocessing the collected data by the data acquisition unit, wherein the preprocessing includes image denoising, data smoothing, and time alignment to ensure data quality and consistency; S3. Use the multimodal data fusion module to fuse optical, laser, and ultrasonic data through deep learning algorithms to extract comprehensive crack characteristics; S4. Produce a three-dimensional model of the crack through a three-dimensional modeling module to show the spatial structure, position and morphology of the crack; S5. Analyze crack feature data, assess crack severity, and generate maintenance recommendations through the intelligent decision support module. S6. Generate a structural health assessment report, providing detailed crack characteristics, 3D models, and maintenance plans.

Images