Boiler heating surface crack detection method based on millimeter wave radar imaging

Abstract

According to the boiler heating surface crack detection method based on millimeter-wave radar imaging, the millimeter-wave radar is used for boiler heating surface detection, the limitation of a traditional visual means in smoke shielding and weak light environments is broken through, millimeter-wave radar detection equipment is integrated on a small unmanned aerial vehicle / robot platform, and the detection accuracy is improved. The compact mobile intelligent terminal is formed and can go deep into narrow or high-risk areas such as a boiler furnace, a horizontal flue and a tail flue which are difficult for personnel to enter to execute detection, and compared with a traditional method that equipment needs to be disassembled or scaffolds need to be erected for inspection, the mobile terminal scheme greatly reduces the difficulty of overhaul preparation and implementation, and the maintenance efficiency is improved. Through multi-angle and multi-frame radar image super-resolution fusion, presentation of cracks in the images is remarkably enhanced, tiny anomalies are clear and distinguishable in a complex background, defect features are automatically extracted by introducing the convolutional neural network, and the method is more objective and accurate compared with manual rule discrimination.

Description

Technical Field

[0001] The present invention relates to the technical field of thermal power plant equipment maintenance, and in particular to a method for detecting cracks on a boiler heating surface based on millimeter-wave radar imaging. Background Art

[0002] During long-term operation of thermal power generation boilers, cracks easily form on the heating surface tubes (such as water-wall tubes and superheater tubes) due to high temperatures, high pressures, and thermal cycles. This cracking, which leads to leakage, is a major cause of unplanned unit outages. This problem is exacerbated during deep peak load regulation: the frequent and significant increases and decreases in boiler load cause repeated thermal expansion and contraction of components, and thermal stress cycles exacerbate the formation and growth of cracks.

[0003] To prevent bursts and leaks in heating surface tubes, maintenance personnel traditionally conduct manual inspections during unit shutdowns for maintenance. These inspections include visual inspection, wall thickness measurement with a thickness gauge, and bulge detection with a plug gauge, supplemented by necessary non-destructive testing such as ultrasonic and eddy current testing. However, manual inspections have significant limitations: they require experienced personnel and are labor-intensive. Limited by environmental factors such as cramped on-site space, high-temperature residual heat, and insufficient light, the inspection scope is wide and comprehensive coverage is difficult to achieve in a timely manner. Consequently, defects often go undetected due to human error. Furthermore, camera-based machine vision inspections are often affected by flue gas, dust, and lighting inside the boiler, making it difficult to detect small cracks or defects covered by accumulated dust, resulting in insufficient reliability.

[0004] Millimeter-wave radar has excellent material penetration and environmental adaptability: it can operate in smoke, high temperatures, and darkness, and is unaffected by light. It can still obtain clear measurement data even in harsh environments with a lot of smoke and dust. Research has shown that millimeter-wave high-frequency signals are very sensitive to small discontinuities on metal surfaces. Even tiny cracks covered by coatings or oxide layers can cause changes in scattering characteristics and be detected by radar. By integrating millimeter-wave radar sensors into mobile vehicles such as drones, robots, or robotic arms, and building an integrated perception and motion control system, it is possible to overcome the limitations of spaces that are difficult for humans to reach, realize automatic inspections of large areas of heated surfaces, and significantly improve the coverage and safety of defect detection.

[0005] To address the above challenges, the present invention introduces millimeter-wave radar combined with an intelligent mobile vehicle to achieve remote and automatic detection of cracks on the boiler heating surface. Summary of the Invention

[0006] In a first aspect of the present disclosure, a method for detecting cracks in a boiler heating surface based on millimeter-wave radar imaging is provided, comprising the following steps: Control the autonomous mobile vehicle equipped with a millimeter-wave radar array to move along a preset trajectory, continuously transmitting frequency-modulated continuous wave signals to the boiler heating surface during movement and receiving its reflected echo signals; Calculating the radial distance of the boiler heating surface according to the phase difference and frequency difference of the reflected echo signal; fusing the motion trajectory of the autonomous mobile vehicle and the radial distance, performing synthetic aperture radar imaging processing on the fused data, and generating a radar image; Multimodal enhancement processing is performed on the radar image to obtain an enhanced image, the enhanced image is input into a pre-trained crack recognition model, and a crack recognition result is output.

[0007] In combination with the first aspect, calculating the radial distance of the boiler heating surface includes: performing a fast Fourier transform on a beat signal of the reflected echo signal; The radial distance of the boiler heating surface is calculated based on the beat frequency corresponding to the spectrum peak.

[0008] In combination with the first aspect, before calculating the radial distance of the boiler heating surface according to the phase difference and the frequency difference of the reflected echo signal, the method further includes: Estimating the real-time position of the autonomous mobile vehicle by fusing inertial measurement unit data and radar positioning data through a Kalman filter; Motion compensation is performed on the reflected echo signal based on the real-time posture.

[0009] In combination with the first aspect, the synthetic aperture radar imaging processing includes: performing range-direction pulse compression on the reflected echo signal to generate a range-compressed signal; performing range migration correction on the range compression signal; An azimuth phase focusing process is performed on the corrected signal to generate the radar image.

[0010] In combination with the first aspect, the multimodal enhancement processing includes: fusing an optical image captured by a visible light camera or an infrared camera with the radar image; Use the preset 3D model of the boiler structure to perform geometric distortion correction on the fused image; Improve image resolution through multi-frame super-resolution reconstruction.

[0011] In combination with the first aspect, the crack recognition model is trained in the following manner: Construct a training dataset containing millimeter-wave radar images and corresponding crack annotations; A transfer learning strategy is used to pre-train model weights using visible light pipeline defect images; The pre-trained model is fine-tuned using the training data set to optimize the crack recognition accuracy.

[0012] In combination with the first aspect, the method further includes specifying a preset trajectory of the autonomous mobile vehicle according to the crack identification result, specifically including: When a crack is identified, the crack's strike angle in the image is calculated; The movement direction of the autonomous mobile vehicle is adjusted according to the strike angle so that the radar beam azimuth forms a preset angle with the crack extension direction.

[0013] In combination with the first aspect, the method further comprises: Controlling the autonomous mobile vehicle to hover above the crack area; Collect at least two radar images with different viewing angles through multi-angle scanning; Compare the crack feature consistency of multi-view images to verify the recognition results.

[0014] According to a second aspect of the present disclosure, an electronic device is provided, comprising: one or more processors; A storage unit is used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors can implement the boiler heating surface crack detection method based on millimeter wave radar imaging.

[0015] According to a third aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method for detecting cracks in the heating surface of a boiler based on millimeter-wave radar imaging can be implemented.

[0016] Beneficial effects: The present disclosure provides a method for detecting cracks on the heating surface of a boiler based on millimeter-wave radar imaging. By using millimeter-wave radar for boiler heating surface detection, the limitations of traditional visual means in smoke and dust obstruction and weak light environments are broken through, and the millimeter-wave radar detection equipment is integrated into a small UAV / robot platform to form a compact mobile intelligent terminal. The terminal can perform detection in narrow or high-risk areas such as boiler furnaces, horizontal flues, and tail flues that are difficult for personnel to enter. Compared with traditional practices that require dismantling equipment or setting up scaffolding for inspection, the mobile terminal solution greatly reduces the difficulty of maintenance preparation and implementation. Through multi-angle and multi-frame radar image super-resolution fusion, the presentation of cracks in the image is significantly enhanced, making subtle anomalies clearly identifiable even in complex backgrounds. The introduction of convolutional neural networks to automatically extract defect features is more objective and accurate than manual rule judgment. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 The figure is a flow chart of a method for detecting cracks on a boiler heating surface based on millimeter-wave radar imaging according to an embodiment of the present disclosure.

[0018] Figure 2 This is an electronic device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0019] Exemplary embodiments are described in detail herein, with examples illustrated in the accompanying drawings. When the following description refers to the drawings, identical numerals in different drawings represent identical or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with the present disclosure.

[0020] The terms used in the embodiments of the present disclosure are for the purpose of describing specific embodiments only and are not intended to limit the embodiments of the present disclosure. The singular forms "a," "the," and "the" used in the embodiments of the present disclosure and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0021] It should be understood that although the terms first, second, third, etc. may be used to describe various information in the embodiments of the present disclosure, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of the embodiments of the present disclosure, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0022] like Figure 1 FIG. 1 is a flow chart of a method for detecting cracks on a boiler heating surface based on millimeter-wave radar imaging according to an embodiment of the present disclosure, comprising: S101: Controlling an autonomous mobile vehicle equipped with a millimeter-wave radar array to move along a preset trajectory, continuously transmitting a frequency-modulated continuous wave signal to the boiler heating surface during movement, and receiving its reflected echo signal; Specifically, an autonomous mobile vehicle (including a drone, wall-climbing robot or robotic arm) equipped with a millimeter-wave radar array is controlled to move along a preset trajectory, penetrate the smoke layer on the boiler's heating surface by transmitting a linear frequency-modulated continuous wave signal, and receive its reflected echo signal.

[0023] Assume that the frequency of the linear frequency modulation signal emitted by the radar changes with time The change is: from the starting frequency In the modulation cycle Internal linear sweep to Then the instantaneous frequency function can be expressed as .

[0024] in is the frequency modulation slope (Hz / s), is the sweep bandwidth, is the duration of a single Chirp. Integrating the frequency over time gives the instantaneous phase of the transmitted signal: .

[0025] in is the initial phase. So the transmitted signal can be expressed as: .

[0026] It can also be expressed in complex exponential form as .

[0027] The radar signal propagates to the surface of the object in front and is reflected. If the target distance is , then the echo signal has a round-trip time delay relative to the transmitted signal ( is the speed of light). Ignoring reflection attenuation, the received signal can be regarded as a delayed transmitted signal:

[0028] in Indicates that the signal is delayed due to round trip propagation And postpone.

[0029] S102: Calculating the radial distance of the boiler heating surface based on the phase difference and frequency difference of the reflected echo signal; The method for calculating the radial distance of the boiler heating surface includes: performing a fast Fourier transform on a beat signal of the reflected echo signal; The radial distance of the boiler heating surface is calculated based on the beat frequency corresponding to the spectrum peak.

[0030] Specifically, the received signal is mixed (multiplied) with the current transmitted signal and low-pass filtered to obtain the beat signal (intermediate frequency signal).

[0031] After low-pass filtering (LPF), the high-frequency items are filtered out, and the retained beat signal mainly contains the frequency Beat frequency component. Beat frequency Proportional to the distance to the target:

[0032] From this the target distance (radial distance) can be calculated .

[0033] The radar receiver performs FFT spectrum analysis on the mixed intermediate frequency signal and finds the beat frequency peak to obtain the target distance information.

[0034] For moving targets, multiple Chirp signal sequences are introduced, and the Doppler frequency shift can be extracted by using the phase difference of adjacent Chirp echoes to estimate the relative speed. (Doppler shift , is the wavelength).

[0035] S103: fusing the motion trajectory of the autonomous mobile vehicle and the radial distance, performing synthetic aperture radar imaging processing on the fused data to generate a radar image; The synthetic aperture radar imaging process includes: performing range-direction pulse compression on the reflected echo signal to generate a range-compressed signal; performing range migration correction on the range compression signal; An azimuth phase focusing process is performed on the corrected signal to generate the radar image.

[0036] Specifically, the goal of synthetic aperture radar imaging is to use the vehicle's motion to obtain multi-view echoes, coherently superimposing the energy of the same target point at different observation positions, and thus focusing to obtain a high-resolution image. To achieve this, the raw echo data needs to be decoupled and focused in the range and azimuth directions. The main steps include: For each echo, in fast time Matched filtering is performed on the echo (see step 4 of the implementation steps) to compress the echoes with long pulse width and improve the range resolution. This is equivalent to performing a range-direction FFT and phase correction on the echo at each azimuth position to align the pulses at different range units.

[0037] In squint geometry, the target's range migration (the curve projected onto different range bins over azimuth and time) varies with azimuth and position, requiring compensation. A range migration compensation phase factor can be designed to correct the range migration of all scatterers to a trajectory consistent with a reference range. For example, by multiplying a precalculated phase function in the range-Doppler domain, the range migration curves of all targets can be made uniform.

[0038] In azimuth (slow time ) Phase correction and superposition are performed on multiple observations of the same target. Since the motion of the vehicle introduces perspective differences, the echo data has phase differences and time delays in azimuth, which need to be corrected. Specifically, azimuth matched filtering / phase correction is introduced to eliminate the relative phase difference between data from different sub-apertures and focus them on the same target point. This step is usually achieved by transforming the data into the Doppler frequency domain through azimuth FFT, multiplying it by the compensation phase, and then inverse transforming it. In the classic SAR algorithm, this step is called azimuth compression, which achieves high azimuth resolution.

[0039] The processed data is projected to the image coordinates and accumulated to form the image intensity. Using back-projection, for each pixel in the image grid (assuming it corresponds to a scattering point ), calculate the theoretical distance between it and each radar position , extract the corresponding delay from the echo data The signal at the position is phase corrected (multiplied by factors), then for all Summation.

[0040] This process is equivalent to "projecting" the echo signal back to the scattering point position and accumulating it. After accumulation, the amplitude of the pixel is the reflection intensity imaging value of the point.

[0041] S104: performing multimodal enhancement processing on the radar image to obtain an enhanced image, inputting the enhanced image into a pre-trained crack recognition model, and outputting a crack recognition result.

[0042] Radar images can suffer from limited resolution and significant noise and speckle. To improve the visibility of anomalies (such as cracks) in images, image enhancement is required, combined with multimodal fusion of information from other sensors.

[0043] Super-resolution reconstruction aims to break through the original resolution limit of radar images and achieve higher clarity by fusing multiple frames. The implementation method is as follows.

[0044] Autonomous vehicles image the same area at different times, angles, or locations. Fusion of these images can enhance detail. Due to differences in perspective, image registration is required: spatially aligning multiple millimeter-wave radar images and calculating translation, rotation, and scale transformations between them. Registration can be performed using feature point matching (such as SURF / SIFT features) or correlation-based template matching. If the radar images have low resolution and few features, known changes in the autonomous vehicle's position can be used to infer the projective transformation relationship between the images.

[0045] After registering multiple images, pixel-level fusion can be performed using simple averaging to improve the signal-to-noise ratio, or Laplacian pyramid fusion to preserve high-frequency details. Information from multiple frames can complement each other's details: for example, noise in one frame may be smoothed out in another, resulting in clearer textures after fusion.

[0046] Super-resolution algorithms (such as those based on sparse representation or deep learning) take multiple low-resolution images as input and produce a high-resolution image as output. This requires that the images have been registered to a common high-resolution grid, and then the algorithm can infer details. Because millimeter-wave radar images contain speckle noise, denoising may be necessary before super-resolution.

[0047] Through the above method, an enhanced image with higher pixel resolution and lower noise than a single-frame original SAR image can be obtained.

[0048] After image reconstruction and enhancement are completed, intelligent algorithms are ultimately required to automatically detect and identify crack defects in the image.

[0049] This crack recognition model uses a convolutional neural network (CNN) for image anomaly detection and classification. To improve CNN's ability to identify cracks in millimeter-wave images, a targeted approach was implemented in the input feature design and training strategy.

[0050] Input feature design: Based on the characteristics of millimeter-wave radar images, we consider using the original radar intensity image and some derived features as the input channels of the CNN to enrich the information available to the model: Raw radar image: This is the basic input to the CNN, providing the distribution of the scene's reflective intensity. The enhanced and fused image contains the main visual clues of the crack (bright and dark lines, edges, etc.).

[0051] Edge / Gradient Map: Cracks appear as sudden changes in local grayscale in an image. Extracting a gradient magnitude map or a binary edge map after Canny edge detection can highlight the crack's outline. Using this as the second channel input allows the CNN to more easily learn linear edge features.

[0052] Multimodal channels: Aligned visible light images or other modal data can be fed into the CNN as additional channels. For example, cracks are often visible in visible light images (if lighting and dirt permit), complementing radar images.

[0053] Range and depth map: Radar imaging also provides range and depth information (such as a 3D point cloud projection map), which can be input into CNN to help it distinguish between noise and real targets.

[0054] CNN architecture design: Crack detection can be considered as an anomaly detection or semantic segmentation task. Possible network designs include: Global classification network: The image is divided into blocks or sliding windows and fed into a CNN for binary classification (presence or absence of cracks). This network structure is similar to conventional image classification CNNs, such as ResNet and VGG, requiring only a binary discrimination at the end. However, this method cannot precisely locate the crack location; it can only determine its presence, typically requiring post-processing to locate the detection area.

[0055] Object detection network: Using methods similar to Faster R-CNN or YOLO, the bounding box of the crack is directly regressed in the image. However, cracks are not closed objects, so using a rectangular box is not accurate.

[0056] Semantic segmentation networks are suitable for detecting thin, elongated areas such as cracks. Encoder-decoder structures such as U-Net and SegNet can be used to classify each pixel as either a crack pixel or background. This pixel-level annotation accurately depicts the crack shape, but the annotation process is time-consuming. For finer cracks, the problem can be transformed into edge detection, using networks such as Holistically-nested Edge Detection (HED) to detect crack edges in the image.

[0057] Given that cracks are typically thin and long, a CNN architecture that fuses multi-scale features can be used to capture the continuous morphology of cracks across the receptive fields of different convolutional layers. One effective approach is to add dilated convolutions or multi-scale pooling to the network to simultaneously capture local texture and global structural information. Another consideration is to introduce an attention mechanism to focus the network on high-contrast, fine-line areas, thereby increasing sensitivity to cracks.

[0058] Due to the limited actual crack sample data, this solution uses a series of enhancement and migration methods in CNN model training to improve model performance. The crack recognition model is trained in the following ways: Construct a training dataset containing millimeter-wave radar images and corresponding crack annotations; A transfer learning strategy is used to pre-train model weights using visible light pipeline defect images; The pre-trained model is fine-tuned using the training data set to optimize the crack recognition accuracy.

[0059] Specifically, various images of boiler tube wall surfaces were collected, including both normal surfaces and millimeter-wave radar images showing abnormalities such as cracks, rust, and wear. Because millimeter-wave images are difficult to obtain, we also collected a library of images of pipeline defects under normal visible light to assist in training (see step 6.3.3 for details). Each radar image was manually annotated with the crack location or category label. If segmentation was required, a mask was also required to annotate the crack pixels.

[0060] The dataset is cleaned and augmented. Augmentation methods include, but are not limited to, random rotation (any angle, which helps the network learn robustness to crack tilt in all directions), translation, scaling, mirror flipping, adding Gaussian noise, and contrast adjustment. These operations increase the diversity of training samples and alleviate the problem of insufficient training sets. For radar images, different noise levels or blur levels can also be simulated to make the model robust to variations in image quality.

[0061] Given the limited amount of radar crack image data, we employ a transfer learning strategy. We first initialize a CNN using a general image pre-trained model (such as ResNet pre-trained on ImageNet), or more specifically, a model pre-trained on visible light pipeline defect data. For example, we train the CNN using a large number of images of pipeline cracks and rust captured by a standard camera, allowing it to learn basic texture and shape features (including the fact that cracks typically appear as long, thin lines with high-contrast edges). We then transfer the model's weights to the millimeter-wave radar task and continue fine-tuning using millimeter-wave images. This expands the distribution of training data and accelerates convergence.

[0062] Perform supervised training on the CNN using the enhanced mmWave crack data. Use a cross-entropy loss (for classification) or a combined Dice / BCE loss (for segmentation), depending on the task. Monitor validation set performance during training to avoid overfitting. If the sample is extremely imbalanced (normal regions far outnumber crack pixels), increase the weight of the anomaly class in the loss function or employ a difficult sample mining strategy (such as focal loss).

[0063] Evaluate the model's crack detection accuracy, recall, and other metrics on an independent test set (or using cross-validation). Pay particular attention to the missed detection rate (the number of cracks missed as normal must be extremely low). If performance falls short, analyze the error cases and consider adding corresponding samples or adjusting the model (for example, if a certain type of rust is misclassified as a crack, add that type of sample or adjust the network's ability to distinguish textures).

[0064] Preferably, before calculating the radial distance of the boiler heating surface according to the phase difference and the frequency difference of the reflected echo signal, the method further includes: Estimating the real-time position of the autonomous mobile vehicle by fusing inertial measurement unit data and radar positioning data through a Kalman filter; Motion compensation is performed on the reflected echo signal based on the real-time posture.

[0065] Specifically, autonomous vehicles (such as drones) require precise attitude and motion parameter estimation during detection to stabilize movement and ensure accurate radar imaging geometry. To achieve this, the system utilizes multi-source sensor data fusion (such as an IMU, barometer, and odometer) and applies Kalman filtering to estimate the vehicle's state, including position, velocity, acceleration, and attitude angle. Kalman filtering provides an optimal recursive estimation method for system state in the presence of noisy measurements.

[0066] First, we need to establish a state space model of the vehicle, including the system's state equation (kinematic model) and observation equation (sensor measurement model): make For discrete moments The state vector of the vehicle contains information such as the vehicle's position, velocity, and attitude. For example, you can define $ , respectively represent the position coordinates ,speed( and attitude Euler angles (roll , pitch and roll ,yaw If acceleration deviation or sensor deviation is taken into account, it can also be included in the state.

[0067] Describe the evolution of the state over time, adding process noise to the linear system:

[0068] in is the state estimate at the previous moment, is the a priori estimate of the current state predicted by the model; is the state transfer matrix (determined by the system kinematics, such as the uniform velocity model The position will be added to the speed of the previous moment and multiplied by the sampling period); It is the control input (such as motor thrust command, etc.) combined with the control matrix effect; is the process noise, assumed to have zero mean and covariance .

[0069] The relationship between sensor observation and state can also be expressed as a linear form plus measurement noise:

[0070] in It is The sensor observation vector at each moment (such as acceleration and angular velocity measured by IMU, position features measured by vision / radar, etc.), is the measurement matrix (mapping the state to the observation space, such as when measuring the position directly may be a subset of the identity matrix), For measurement noise, assume zero mean and covariance .

[0071] The Kalman filter consists of two stages at each time step: prediction and update: According to the posterior state estimate of the previous moment and covariance , use the state equation to predict the current moment: , , Get the current measurement , and fuse it with the prediction to correct it: , in is a gain matrix that reflects the weighted ratio of prediction uncertainty to measurement uncertainty.

[0072] , That is, a correction amount weighted by adding the residual (the difference between the measurement and the prediction) to the prediction.

[0073] - , Get the covariance matrix of the posterior estimate (uncertainty usually decreases as more information is provided).

[0074] This recursive process continues, gradually revising the vehicle's attitude and position estimates from the initial state. By fusing IMU inertial data (high-frequency but subject to drift) with radar / visual data (low-frequency but absolute positioning), the Kalman filter can obtain accurate and smooth trajectories and attitudes. In this solution, the filtering results are used in two ways: 1) providing motion parameters (position, attitude, etc.) to the SAR imaging module for motion compensation and coordinate transformation; and 2) providing movement status to the autonomous control module, enabling precise hovering, fixed-point scanning, and other maneuvers.

[0075] Furthermore, the multimodal enhancement processing further includes: fusing an optical image captured by a visible light camera or an infrared camera with the radar image; Use the preset 3D model of the boiler structure to perform geometric distortion correction on the fused image; Improve image resolution through multi-frame super-resolution reconstruction.

[0076] Specifically, the autonomous vehicle is equipped with a visible light camera or infrared thermal imager to obtain optical images of the boiler's internal tube walls. The optical image is then aligned (registered) with the millimeter-wave radar image, for example by identifying common features such as pipe edges or flanges to align the coordinates of the two images. This can then be fused using layer overlay or feature stitching: texture details from the optical image are superimposed on the radar image to aid in crack identification. Another implementation algorithm uses data from both modalities as two input channels of a CNN, allowing the model to autonomously learn the fused features.

[0077] A CAD model or 3D point cloud model of the structure being inspected is obtained in advance and can be used for image correction and reconstruction. For example, if the spacing and radius of the boiler water-wall tubes are known, the geometric distortion of the radar image can be corrected: the radar imaging coordinates are mapped to the actual physical coordinates, flattening the tube wall surface. This mapping can be obtained through ray tracing simulation or calibration. Once the image is aligned with the real structure, the known structure can be used to eliminate certain artifacts and highlight real anomalies. For example, if the model indicates that a certain area should be smooth and continuous under normal circumstances, but the radar image shows a sudden change, an anomaly can be determined. The prior model can also be used for image reconstruction: by mapping the millimeter-wave radar echo onto the model surface (similar to texture mapping), a radar reflection map of the structure's surface can be obtained, making it easier for human inspection.

[0078] The resulting multimodal fusion image is more comprehensive and clearer than a single millimeter-wave radar image, visually enhancing contrast and detail, making linear anomalies like fine cracks more prominent. This provides higher-quality input for subsequent automatic recognition algorithms.

[0079] The method further includes specifying a preset trajectory of the autonomous mobile vehicle based on the crack identification result, specifically including: When a crack is identified, the crack's strike angle in the image is calculated; The movement direction of the autonomous mobile vehicle is adjusted according to the strike angle so that the radar beam azimuth forms a preset angle with the crack extension direction.

[0080] When a suspected crack is detected during cruising, the vehicle needs to adjust its movement to obtain the best imaging angle and more information to ensure that the crack is accurately imaged and verified. Specific strategies include: Angular imaging optimization of crack directionality: Cracks usually extend in a certain direction (such as vertical or horizontal). In order to image the crack more clearly, the azimuth resolution direction of the synthetic aperture radar should be consistent with the crack direction or at a certain angle. Empirically, SAR imaging has the highest resolution in the direction parallel to the moving track. Therefore, if a crack is detected to be roughly vertical, the moving path of the autonomous vehicle is controlled to scan the area as vertically as possible (the radar movement direction is parallel to the crack) to obtain higher resolution imaging along the crack direction. If the crack is very inclined, adjust the track angle so that the synthetic aperture takes the optimal imaging angle relative to the crack. This direction-adaptive scanning can achieve track angle by analyzing the azimuth angle of the crack in the image in real time. Adjustments, such as (Move the radar perpendicular to the crack direction and observe from the side) or (Parallel to the crack direction, depending on specific imaging requirements).

[0081] The method further comprises: Controlling the autonomous mobile vehicle to hover above the crack area; Collect at least two radar images with different viewing angles through multi-angle scanning; Compare the crack feature consistency of multi-view images to verify the recognition results.

[0082] Specifically, when CNN detects suspected crack features, the autonomous vehicle enters a hovering and fine scanning mode. It hovers at a fixed point near the crack area and collects radar views from multiple angles by changing its posture or circling in a small range. This is similar to "surround shooting" in photography: keeping the distance between the sensor and the target unchanged, changing the viewing angle $\phi$ to perform multiple imaging. On the one hand, multi-angle data can be used for the aforementioned super-resolution superposition (improving image details from different perspectives), and on the other hand, it can also verify the stability of crack features (real cracks still appear as continuous lines from different angles, while noise or false positives may be inconsistent). For example, the autonomous vehicle can be offset left and right around the center axis of the crack. Image twice, or move the image forward and backward by a few centimeters and image at slightly different incident angles. Then compare and fuse the multiple images to enhance the signal characteristics of the real crack.

[0083] The electronic device 200 may be a desktop computer, a notebook, a PDA, a cloud server, or other electronic device. The electronic device 200 may include but is not limited to a processor 201 and a memory 202. Those skilled in the art will appreciate that Figure 2 This is merely an example of the electronic device 200 and does not constitute a limitation of the electronic device 200. The electronic device 200 may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the electronic device may also include input and output devices, network access devices, buses, etc.

[0084] The processor 201 may 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, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0085] The memory 202 can be an internal storage unit of the electronic device 200, such as a hard drive or memory of the electronic device 200. The memory 202 can also be an external storage device of the electronic device 200, such as a plug-in hard drive, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash memory card, etc. equipped on the electronic device 200. Furthermore, the memory 202 can include both an internal storage unit of the electronic device 200 and an external storage device. The memory 202 is used to store the computer program 203 and other programs and data required by the electronic device. The memory 202 can also be used to temporarily store data that has been output or is about to be output.

[0086] The above embodiments are only used to illustrate the technical solutions of the present disclosure, rather than to limit them. Although the present disclosure has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present disclosure, and should all be included in the scope of protection of the present disclosure.

Claims

1. A method for detecting cracks on a boiler heating surface based on millimeter-wave radar imaging, characterized in that: The following steps are involved: Control the autonomous mobile vehicle equipped with a millimeter-wave radar array to move along a preset trajectory, continuously transmitting frequency-modulated continuous wave signals to the boiler heating surface during movement and receiving its reflected echo signals; Calculating the radial distance of the boiler heating surface according to the phase difference and frequency difference of the reflected echo signal; fusing the motion trajectory of the autonomous mobile vehicle and the radial distance, performing synthetic aperture radar imaging processing on the fused data, and generating a radar image; Multimodal enhancement processing is performed on the radar image to obtain an enhanced image, the enhanced image is input into a pre-trained crack recognition model, and a crack recognition result is output.

2. The method according to claim 1, characterized in that The method for calculating the radial distance of the boiler heating surface includes: performing a fast Fourier transform on a beat signal of the reflected echo signal; The radial distance of the boiler heating surface is calculated based on the beat frequency corresponding to the spectrum peak.

3. The method according to claim 2, characterized in that Before calculating the radial distance of the boiler heating surface according to the phase difference and the frequency difference of the reflected echo signal, the method further includes: Estimating the real-time position of the autonomous mobile vehicle by fusing inertial measurement unit data and radar positioning data through a Kalman filter; Motion compensation is performed on the reflected echo signal based on the real-time posture.

4. The method according to claim 1, wherein The synthetic aperture radar imaging process includes: performing range-direction pulse compression on the reflected echo signal to generate a range-compressed signal; performing range migration correction on the range compression signal; An azimuth phase focusing process is performed on the corrected signal to generate the radar image.

5. The method according to claim 1, characterized in that The multimodal enhancement processing further includes: fusing an optical image captured by a visible light camera or an infrared camera with the radar image; Use the preset 3D model of the boiler structure to perform geometric distortion correction on the fused image; Improve image resolution through multi-frame super-resolution reconstruction.

6. The method according to claim 1, characterized in that The crack recognition model is trained in the following way: Construct a training dataset containing millimeter-wave radar images and corresponding crack annotations; A transfer learning strategy is used to pre-train model weights using visible light pipeline defect images; The pre-trained model is fine-tuned using the training data set to optimize the crack recognition accuracy.

7. The method according to claim 1, characterized in that The method further includes specifying a preset trajectory of the autonomous mobile vehicle based on the crack identification result, specifically including: When a crack is identified, the crack's strike angle in the image is calculated; The movement direction of the autonomous mobile vehicle is adjusted according to the strike angle so that the radar beam azimuth forms a preset angle with the crack extension direction.

8. The method according to claim 7, characterized in that The method further comprises: Controlling the autonomous mobile vehicle to hover above the crack area; Collect at least two radar images with different viewing angles through multi-angle scanning; Compare the crack feature consistency of multi-view images to verify the recognition results.

9. An electronic device, characterized in that: include: one or more processors; A storage unit for storing one or more programs, which, when executed by the one or more processors, enables the one or more processors to implement the boiler heating surface crack detection method based on millimeter wave radar imaging according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it can implement the boiler heating surface crack detection method based on millimeter wave radar imaging according to any one of claims 1 to 8.

Images