Subsurface defects detection and localization
The integration of acoustic and video data with machine learning allows for precise localization and visualization of subsurface defects, addressing the inefficiencies of traditional methods by providing automated and comprehensive defect detection.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- NIRICSON SOFTWARE INC
- Filing Date
- 2025-05-06
- Publication Date
- 2026-07-23
AI Technical Summary
Existing methods for detecting and localizing subsurface defects in structures like bridges and buildings are labor-intensive, time-consuming, and lack comprehensive coverage, with a challenge in accurately mapping detected defects to the structure's surface.
An integrated system that combines acoustic response data collection with synchronized 360-degree video imaging, using a striker, acoustic sensor, and video camera to create a 3D model, process acoustic data with machine learning, and visualize defects on a 2D orthographic projection.
Enables efficient, accurate, and comprehensive localization and visualization of subsurface defects across various structural surfaces, improving detection accuracy and reducing manual effort.
Smart Images
Figure 00000021_0000 
Figure 00000022_0000 
Figure 00000023_0000
Abstract
Description
[0001] SUBSURFACE DEFECTS DETECTION AND LOCALIZATION
[0002] CROSS REFERENCE
[0003] This application claims priority to US Provisional Application No. 63 / 644,018, filed May 8, 2024, “Subsurface Defects Detection and Localization,” which is hereby incorporated by reference in its entirety.
[0004] TECHNICAL FIELD
[0005] The present invention relates generally to systems and methods for detecting and localizing subsurface defects in civil infrastructures.
[0006] BACKGROUND
[0007] Detecting subsurface defects such as delamination, low compressive strength, voids, honeycombing, rebar corrosion, and cracks in structures like bridges, buildings, and other infrastructures is crucial for assessing their structural integrity and planning maintenance. Traditional methods for subsurface defect detection often involve manual inspection techniques that can be time-consuming, labor-intensive, subjective and may not provide comprehensive coverage of the structure's surface.
[0008] Advances in acoustic analysis have enabled the development of automated systems for detecting subsurface defects. These systems typically employ acoustic sensors to capture the acoustic response of the structure when subjected to an impact or vibration with a striker, and then analyze the response to identify anomalies indicative of subsurface defects. However, accurately localizing the detected defects on the structure's surface remains a challenge.
[0009] SUMMARY
[0010] The present invention describes a subsurface defects inspection system and method that captures the surrounding images by a camera at each location where acoustic response is collected. The invention provides an automated system and method for detecting and localizing subsurface defects in civil infrastructure by integrating acoustic analysis with synchronized video imaging. The system comprises a striker module for generating acoustic signals on a structure’s surface, an acoustic sensor for capturingthe resulting acoustic response, and a video camera-preferably a 360-degree camera-for recording visual data of the inspection area. As the system moves along the structure, it periodically strikes the surface, and both the acoustic response and visual data are recorded at each strike location. The video data is used to construct a three-dimensional model of the structure, and the precise locations of each strike are determined by synchronizing the camera’s position with the striker’s impact timing. The acoustic data is processed to extract features indicative of subsurface defects, with background noise removed and machine learning classification applied to distinguish between defective and non-defective areas. The results are mapped onto the 3D model and visualized in a two-dimensional orthographic projection, where strike positions are color-coded to indicate the presence or absence of defects. This approach enables efficient, accurate, and comprehensive localization and visualization of subsurface defects across a variety of structural surfaces, including vertical walls, ceilings, and horizontal or sloped surfaces, and is adaptable for use with both unmanned aerial vehicles and wheeled rover platforms.
[0011] In an illustrative embodiment, the inventive method involves striking the surface at regular intervals while simultaneously recording the acoustic response and capturing 360-degree video images of the surrounding area. The strike positions are determined from the camera positions at the time of each strike. A 3D model of the structure is created from the video images using photogrammetry techniques, and the strike positions are determined from the camera positions and mapped onto the 3D model surface.
[0012] The acoustic response data is analyzed by segmenting it into "acoustic signal + background noise" and "background noise only" portions. Time-frequency spectra are generated for each segment, and the background noise spectrum is removed from the acoustic signal spectrum to create a feature vector. A machine learning model, trained on labeled hammer test data, classifies each feature vector as indicating either delamination or no delamination.
[0013] Finally, the classification results are visualized on a 2D orthographic projection of the 3D model, with the strike positions color-coded to indicate the presence or absence ofdelamination at each location. This provides a clear, localized map of subsurface defects on the structure's surface.
[0014] The system and method offer several advantages over existing techniques, including:
[0015] • Automated, efficient and precise subsurface defects localization by time synching acoustic response data collection and surrounding image capture by a built-in camera.
[0016] • Comprehensive image capture enabled by a video camera with wide-angle lens, such as a 360-degree camera.
[0017] • Improved subsurface detection accuracy through the combination of machine learning-based acoustic analysis and data localization.
[0018] • Applicability to various structure types and orientations (vertical walls, ceilings, horizontal / sloped surfaces).
[0019] BRIEF DESCRIPTION OF THE DRAWINGS
[0020] Figure 1 is a block diagram of a subsurface defects inspection system in accordance with the invention.
[0021] Figure 2 is a flow diagram of a method for detecting and localizing subsurface defects in accordance with the invention.
[0022] Figures 3-4 depict an illustrative embodiment mounted on an unmanned aerial vehicle (UAV).
[0023] Figure 5 depicts a subsurface defects inspection system and process for use on vertical walls.
[0024] Figures 6-7 illustrate a process for sensor localization and determining strike positions from 360-degree video camera positions, and Figure 8 shows the creation of a 2D orthographic map from the 3D model and camera positions.
[0025] Figures 9-11 depict an acoustic data analysis process, including segmentation, timefrequency spectrum generation, and feature vector creation.Figure 12 illustrates the classification of feature vectors using a trained machine learning model and the visualization of results on the orthographic map.
[0026] Figures 13-14 show an embodiment of the subsurface defects inspection system adapted for ceiling inspection using a UAV.
[0027] Figure 15 depicts the inventive inspection system adapted for horizontal / sloped surface inspection using a wheeled rover vehicle.
[0028] DETAILED DESCRIPTION OF ILLUSTRATIVE EMBODIMENTS
[0029] Overview
[0030] Applicant (Niricson Software, Inc.) has previously filed several provisional patent applications related to infrastructure inspection and surface and subsurface defect detection systems. These include:
[0031] 1. The provisional patent application titled "Contact-Based Infrastructure Inspection System" (Provisional Patent Application No. 63 / 512,627) describes a UAV-based system for conducting contact inspections of civil infrastructure. The disclosed system includes a payload with an adjustable end effector equipped with tools like an automated hammer and sensors for detecting defects, an onboard machine learning model for real-time defect inference, and a wall-following navigation system. The integrated system aims to improve safety and efficiency over manual inspections by enabling UAVs to perform contact-based assessments of hard-to-access areas, with the potential for further enhancements through computer vision techniques.
[0032] 2. "Toothed Roller for Contact-Based Infrastructure Inspection System" (Provisional Patent Application No. 63 / 529,105): This application describes a toothed roller mechanism for capturing high-resolution surface data from civil infrastructure. The toothed roller makes direct contact with the surface, allowing it to detect small defects.
[0033] 3. "Platform for Analyzing the Condition of Infrastructure Using Automated Defect and Change Detection" (Provisional Patent Application No. 63 / 547,283): This application outlines a platform that automates the inspection workflow from datacapture to defect detection and reporting. It includes modules for data management, 3D reconstruction, defect detection using Al, and a web-based interface for results visualization.
[0034] The present invention builds upon and extends the capabilities described in these prior provisional applications. While the previous applications laid the groundwork for subsurface defect detection, localization and visualization, the present invention introduces several novel improvements:
[0035] 1. Integration of acoustic sub-surface defect detection and 3D imaging techniques:
[0036] By time synching the acoustic response data collection and camara shutter timing , the invention enables the system to accurately localize the detected subsurface defects on the surface of a 3D model, and to visualize on a 2D orthoimage.
[0037] 2. End-to-end pipeline: The present invention provides a complete end-to-end pipeline for sub-surface defect detection and localization. It covers all stages from acoustic response data collection, machine learning-based defect detection, defects localization on a 3D model, 3D to 2D transformation, and result visualization. While the prior applications described some of these components, the present invention further integrates them into a cohesive workflow with well- defined interfaces between stages.
[0038] In summary, a goal of the present invention is to improve and extend the prior work by Applicant in the field of infrastructure inspection. By combining subsurface defect detection and localization techniques, utilizing machine learning, enabling intuitive defect visualization, and providing an end-to-end processing pipeline, Applicant intends to offer a more comprehensive and effective solution for automated surface and subsurface defect detection in civil infrastructure inspection applications. The novel techniques introduced below, including but not limited to enable automated localization of subsurface defects with high accuracy, extend the fundamental concepts from the prior applications to achieve improved detection accuracy, richer defect characterization, and easier interpretability of results.System Architecture
[0039] As shown in Figure 1 , the subsurface defects inspection system 100 comprises a striker module 102, an acoustic detector module 104, a 360-degree video camera module 106, and a mobility module 108. The system also includes a data processing module 110 and a visualization module 112.
[0040] The striker module 102 includes a contact surface for producing an acoustic signal when impacting the structure's surface. It is designed to generate consistent, repeatable acoustic excitation with minimal noise interference. The acoustic detector module 104, which may include a microphone or other acoustic sensor, captures the acoustic response of the structure to the striker impact. The 360-degree video camera module 106 records visual data of the surrounding area during the inspection process, providing comprehensive coverage and enabling the creation of a 3D model of the structure. The mobility module 108 allows the system to move along the surface being inspected. It may include wheels, tracks, or other locomotion mechanisms, depending on the specific application and surface type. For example, when inspecting vertical walls or ceilings, the system may be mounted on an unmanned aerial vehicle (UAV) as shown in Figures 3-5 and 11-12. For horizontal or sloped surfaces, a wheeled rover vehicle can be used, as depicted in Figure 1.
[0041] The data processing module 110 receives the acoustic and visual data from the detector and camera modules, respectively. It performs various signal processing and analysis techniques, including segmentation, time-frequency spectrum generation, feature extraction, and machine learning classification, to identify and localize subsurface defects.
[0042] The visualization module 112 creates a 2D orthographic map of the structure's surface from the 3D model and defects position data and overlays the defect classification results to provide a clear, localized representation of the detected defects.
[0043] Method of Operation
[0044] Figure 2 outlines the main steps of the subsurface defect inspection method 200. The process begins with the system moving along the structure's surface and periodicallystriking it with the striker module (step 202). At each strike position, the acoustic detector module records the resulting acoustic response (step 204), while the 360-degree video camera module captures visual data of the surrounding area (step 206). The strike positions are determined from the camera positions at the time of each strike (step 210). This is illustrated in more detail in Figures 5 and 6. The video images are used to create a 3D model of the structure using photogrammetry techniques (step 208), and the strike positions are mapped onto this model (step 212). Figure 8 shows the creation of a 2D orthographic projection from the 3D model and camera positions. The acoustic response data is processed by first segmenting it into "acoustic signal + background noise" and "background noise only" portions (step 214). Time-frequency spectra are generated for each segment using techniques like short-time Fourier transform (step 216), and the background noise spectrum is removed from the acoustic signal spectrum to create a feature vector (step 218). This process is depicted in Figures 9-11.
[0045] A machine learning model, previously trained on labeled training data, is used to classify each feature vector as indicating either delamination or no delamination (step 220). The classification results are then visualized on the 2D orthographic map, with the strike positions color-coded to represent the presence or absence of delamination (step 222). Figures 11-12 illustrate these steps.
[0046] Machine Learning Model
[0047] The machine learning model classifies the strike positions as delamination or no delamination based on features extracted from the acoustic response data collected at each strike position. The overall ML process can be broken down as follows:
[0048] 1. Data Collection:
[0049] • The acoustic detector module records the acoustic response at each strike position when the striker module impacts the surface (Figure 2, step 204).• The acoustic response data captures the vibrations and reflections generated by the strike, which can be affected by the presence of subsurface defects like delamination.
[0050] 2. Data Segmentation:
[0051] • The acoustic response data is segmented into two portions: "acoustic signal + background noise" and "background noise only" (Figure 2, step 214).
[0052] • This segmentation helps isolate the relevant acoustic information related to the strike from the ambient noise.
[0053] 3. Feature Extraction:
[0054] • Time-frequency spectra are generated for each segment of the acoustic response data, by a short-time Fourier transform (STFT) method (Figure 2, step 216).
[0055] • The background noise spectrum is removed from the acoustic signal spectrum to create a feature vector (Figure 2, step 218).
[0056] • The feature vector represents the unique characteristics of the acoustic response at each strike position, capturing information related to the presence or absence of delamination.
[0057] 4. Machine Learning Model Training:
[0058] • Prior to the actual inspection, the machine learning model is trained using labeled training data (Figure 2, step 220).
[0059] • The training data comprises acoustic response feature vectors along with their corresponding labels (delamination or no delamination).
[0060] • The model learns to associate specific patterns and characteristics in the feature vectors with the presence or absence of delamination.
[0061] 5. Classification:• During the inspection, the trained machine learning model is used to classify each strike position based on its corresponding feature vector (Figure 2, step 222).
[0062] • The model takes the feature vector as input and predicts whether it indicates delamination or no delamination.
[0063] • The classification is based on the learned patterns and decision boundaries from the training phase.
[0064] 6. Visualization:
[0065] • The classification results are visualized on the 2D orthographic projection of the 3D model (Figure 2, step 224).
[0066] • Each strike position is color-coded to represent the presence (red) or absence (green) of delamination (Figure 12).
[0067] • This visualization provides a clear, localized map of the detected subsurface defects on the structure's surface.
[0068] The machine learning model's ability to classify delamination is based on learning the distinctive patterns and characteristics in the acoustic response data that are indicative of subsurface defects. By training on labeled data, the model develops a decision boundary that allows it to differentiate between the acoustic signatures of delamination and solid, intact surfaces.
[0069] The specific machine learning algorithm used can vary depending on the complexity of the data and the desired performance. Common choices for this type of classification task include support vector machines (SVM), random forests, or even deep learning models like convolutional neural networks (CNNs) if a large amount of training data is available.
[0070] The accuracy and reliability of the machine learning model's classifications depend on the quality and representativeness of the training data, as well as the selection of appropriate features and the model's architecture. Proper validation and testing of the model are crucial to ensure its generalization performance on unseen data.Acoustic Data Processing
[0071] The acoustic data capture and processing, as well as the feature extraction for the machine learning model, are critical components of the subsurface defects inspection system. The acoustic detector module, which can include a microphone or other acoustic sensor, is responsible for capturing the acoustic response of the structure when the striker module impacts the surface (Figure 2, step 204). The acoustic sensor is positioned near the strike point to record the vibrations and reflections generated by the impact. The sensor should have a suitable frequency response range to capture the relevant acoustic information, typically in the audible to low ultrasonic range (20 Hz to 50 kHz). The acoustic data is digitized using an analog-to-digital converter (ADC) and stored for further processing.
[0072] The raw acoustic data undergoes several processing steps to extract meaningful information and prepare it for feature extraction. The first step is to segment the acoustic data into "acoustic signal + background noise" and "background noise only" portions (Figure 2, step 214). The "acoustic signal + background noise" segment contains the acoustic response directly related to the strike, along with any ambient noise present during the impact. The "background noise only" segment represents the ambient noise recorded without the strike, typically a short duration before or after the impact. Next, time-frequency analysis techniques, such as short-time Fourier transform (STFT), are applied to each segment (Figure 2, step 216). STFT divides the acoustic signal into short overlapping windows and computes the Fourier transform for each window. This process generates a spectrogram, which represents the frequency content of the signal over time. The background noise spectrum is then removed from the acoustic signal spectrum to isolate the strike-related information (Figure 2, step 218). This step helps to remove the influence of ambient noise and enhances the relevant acoustic features.
[0073] The machine learning model relies on discriminative features extracted from the processed acoustic data to classify the strike positions as delamination or no delamination. The choice of features depends on the characteristics of the acoustic signal and the nature of the subsurface defects. Some commonly used features inacoustic analysis for defect detection include spectral features, temporal features, cepstral features, and wavelet features. Spectral features capture the distribution of energy across different frequency bands and include examples such as spectral centroid (center of mass of the spectrum), spectral spread (variance of the spectrum), and spectral flux (change in the spectrum over time). Temporal features describe the temporal characteristics of the acoustic signal and include examples such as signal energy, root mean square (RMS) amplitude, and zero-crossing rate. Cepstral analysis involves computing the inverse Fourier transform of the logarithm of the spectrum, and cepstral coefficients, such as Mel-frequency cepstral coefficients (MFCCs), capture the shape of the spectral envelope and are widely used in speech and audio analysis. Wavelet transform decomposes the acoustic signal into different frequency bands while preserving temporal information, and wavelet coefficients and their statistical measures (e.g., mean, variance) can serve as features for defect detection. The selected features are combined to form a feature vector that represents each strike position (Figure 2, step 218). The feature vectors, along with their corresponding labels (delamination or no delamination), are used to train the machine learning model (Figure 2, step 220).
[0074] The acoustic data capture and processing pipeline aims to extract relevant information from the raw acoustic signals and transform it into a suitable representation for the machine learning model. The extracted features should be discriminative, meaning they should help differentiate between the acoustic signatures of delamination and solid, intact surfaces. The choice of features and the design of the feature extraction process require domain knowledge and experimentation to identify the most informative and robust features for the specific application. Feature selection techniques, such as statistical tests or wrapper methods, can be employed to select the most relevant features and reduce dimensionality.
[0075] It's important to note that the acoustic data capture and processing steps should be consistent across the training and testing phases to ensure the reliability and generalization of the machine learning model. By carefully designing the acoustic data capture and processing pipeline, along with selecting appropriate features, the subsurface defects inspection system can effectively leverage the information containedin the acoustic response data to detect and localize subsurface defects using machine learning techniques.
[0076] Feature Selection
[0077] Feature selection is an important step in the acoustic data processing pipeline, as it helps to identify the most relevant and informative features for the machine learning model, while reducing dimensionality and potentially improving computational efficiency. Several common techniques are used for feature selection in the context of acoustic data processing for defect detection. These include:
[0078] 1. Filter Methods: Filter methods assess the relevance of features based on their statistical properties, independently of the machine learning algorithm. Some popular filter methods include:
[0079] a. Variance Threshold: This method removes features with low variance, as they may not contribute significantly to the discrimination between classes. Features with variance below a specified threshold are considered irrelevant and are eliminated.
[0080] b. Correlation-based Feature Selection (CFS): CFS evaluates the correlation between features and the target variable (defect presence or absence), as well as the intercorrelation among features. It aims to select a subset of features that are highly correlated with the target variable but have low correlation with each other, reducing redundancy.
[0081] c. Information Gain (IG): IG measures the reduction in entropy or uncertainty achieved by using a particular feature to classify the data. Features with higher information gain are considered more informative and are selected.
[0082] 2. Wrapper Methods: Wrapper methods evaluate the performance of different feature subsets using a specific machine learning algorithm. They search for the optimal feature subset by iteratively training and testing the model with different combinations of features. Examples of wrapper methods include:
[0083] a. Recursive Feature Elimination (RFE): RFE recursively removes the least important features based on the coefficients or importance scores assigned by the machine learning model. It starts with all features and iteratively eliminates the least significant ones until a desired number of features is reached.b. Sequential Feature Selection: This method iteratively adds or removes features based on the model's performance. It can be implemented as Sequential Forward Selection (SFS), which starts with an empty feature set and gradually adds the most relevant features, or Sequential Backward Selection (SBS), which starts with all features and removes the least relevant ones.
[0084] 3. Embedded Methods: Embedded methods combine feature selection with the model training process, performing feature selection during the learning phase. Some examples include:
[0085] a. L1 Regularization (Lasso): Lasso is a regularization technique that adds a penalty term to the objective function, encouraging sparse feature representations. It tends to drive the coefficients of irrelevant features to zero, effectively performing feature selection.
[0086] b. Decision Tree-based Feature Importance: Decision tree-based algorithms, such as Random Forest or Gradient Boosting, can provide feature importance scores based on the contribution of each feature to the model's predictions. Features with higher importance scores are considered more relevant.
[0087] 4. Domain Knowledge and Expert Insight: In addition to the above data-driven methods, domain knowledge and expert insight play a crucial role in feature selection for acoustic defect detection. Experts in the field of acoustics and structural engineering can provide valuable guidance on the most relevant features based on their understanding of the physical phenomena and the characteristics of different types of defects.
[0088] When selecting features, it's important to consider the trade-off between the number of features and the model's performance. While a larger number of features may capture more information, it can also lead to increased computational complexity and the risk of overfitting. Therefore, it's often beneficial to start with a comprehensive set of features and then apply feature selection techniques to identify the most informative subset. Cross-validation and evaluation metrics, such as accuracy, precision, recall, and F1-score, can be used to assess the performance of different feature subsets and guide the selection process. It's also recommended to validate the selected features onindependent test data to ensure their generalization ability. By employing appropriate feature selection techniques and leveraging domain knowledge, the acoustic data processing pipeline can be optimized to extract the most relevant and discriminative features for subsurface defect detection using machine learning models.
[0089] Example Use Case
[0090] To illustrate one illustrative practical application of the subsurface defects inspection system, consider the following real-world scenario:
[0091] A civil engineering firm is tasked with assessing the structural integrity of a highway bridge overpass. The concrete surface of the bridge deck and underside exhibits signs of wear, and there are concerns about potential subsurface defects that could compromise the bridge's safety. The firm deploys the subsurface defects inspection system to perform a comprehensive assessment of the bridge. For the bridge deck (horizontal surface), a wheeled rover vehicle equipped with the striker, acoustic detector, and 360-degree video camera modules is used (Figure 15). The rover systematically traverses the deck surface, striking it at regular intervals and recording the acoustic response and visual data.
[0092] For the bridge underside (ceiling surface), an unmanned aerial vehicle (UAV) mounted with the inspection system is employed (Figures 13-14). The UAV flies along the underside, maintaining a consistent distance from the surface, and follows the same inspection process as the rover.
[0093] As the inspection progresses, the data processing module analyzes the acoustic response data to identify areas of potential delamination or other subsurface defects. The 360-degree video camera data is used to create a 3D model of the bridge using photogrammetry techniques (Figure 5). The strike positions are determined from the camera positions at each strike time and mapped onto the 3D model (Figures 6-7). A 2D orthographic projection of the model is generated, providing a top-down view of the bridge deck and a bottom-up view of the underside (Figure 8).
[0094] The machine learning model classifies each strike position as either indicating delamination (red) or no delamination (green) based on the extracted features from theacoustic response data (Figures 11-12). These classification results are then visualized on the orthographic projection, creating a clear, localized map of the detected subsurface defects (Figure 12). The civil engineering firm can use this information to identify areas of the bridge that require further investigation or repair. The localized defect map allows them to prioritize maintenance efforts and make informed decisions about the bridge's structural health.
[0095] By employing the subsurface defects inspection system, the firm can efficiently and accurately assess the condition of the bridge, covering a large surface area in a short amount of time. The integration of acoustic analysis and 360-degree video imaging provides a comprehensive understanding of the bridge's subsurface condition, enhancing the safety and reliability of the infrastructure. Importantly, this example demonstrates how the subsurface defects inspection system can be utilized to improve the efficiency and effectiveness of structural health monitoring for critical infrastructure assets.
[0096] Advantages and Benefits
[0097] The described system and method offer several key advantages over traditional subsurface defect inspection techniques:
[0098] 1. Efficiency and Automation: By combining acoustic analysis with 360-degree video imaging, the system enables automated subsurface defects inspection of large surface areas, reducing the time and labor required compared to manual methods.
[0099] 2. Precise Defect Localization: Mapping the acoustic response data onto the 3D model by the use of a 360-degree video camera, and orthographic projection allows for accurate localization of detected defects on the structure's surface. 3. Improved Accuracy: Integrating acoustic and visual data provides complementary information that enhances the accuracy of defect detection and classification. 4. Versatility: The modular design of the system allows it to be adapted for various structure types and orientations, including vertical walls, ceilings, and horizontal / sloped surfaces.Conclusion
[0100] The subsurface defects inspection system and method presented in this invention disclosure offer a novel, efficient, and accurate approach to detecting and localizing subsurface defects in structures. By combining acoustic response analysis with 360-degree video imaging and leveraging advanced signal processing and machine learning techniques, the system provides a comprehensive and automated solution for assessing the structural integrity of various infrastructure elements.
[0101] Some of the key advantages of the embodiments described herein include increased efficiency through automation, precise defect localization, improved accuracy by integrating acoustic and visual data, and versatility in adapting to different structure types and orientations. The modular design of the system, comprising the striker, acoustic detector, 360-degree video camera, and mobility modules, allows for flexible deployment in a range of inspection scenarios.
[0102] The subsurface defects inspection system presented in this invention offers a significant advancement in the field of structural health monitoring. By combining acoustic response analysis with 360-degree video imaging, the system enables accurate detection and precise localization of subsurface defects in a wide range of structures. The potential applications of this technology extend beyond the specific embodiments described in this patent. The principles and techniques outlined herein can be adapted and applied to other industries and domains where subsurface defect detection is critical. For example:
[0103] 1. Aerospace: The system could be modified to inspect aircraft structures, such as wings, fuselage, and composites, for subsurface damage caused by fatigue, impact, or manufacturing defects. This could enhance aviation safety and maintenance efficiency.
[0104] 2. Oil and Gas: Subsurface defect detection is crucial in the oil and gas industry, where pipelines, storage tanks, and offshore structures are subject to corrosion, erosion, and other forms of degradation. Adapting the system to theseenvironments could help prevent leaks, spills, and structural failures, minimizing environmental and economic risks.
[0105] 3. Manufacturing: In industrial manufacturing, quality control often relies on detecting internal defects in materials and products. The acoustic and imaging techniques described in this invention could be integrated into automated inspection systems for manufacturing lines, improving product quality and reducing waste.
[0106] As technology advances, the subsurface defects inspection system can be further enhanced by incorporating emerging technologies such as artificial intelligence, robotics, and advanced sensor systems. These advancements could enable even more intelligent, autonomous, and efficient inspection processes. Moreover, the data generated by this system could contribute to the development of predictive maintenance models and digital twins of structures. By analyzing the historical data and patterns of subsurface defects, it may be possible to predict the future deterioration of structures and optimize maintenance schedules accordingly.
Claims
CLAIMSWe claim:
1. A subsurface defects inspection system (100), comprising:a striker module (102) having a contact surface configured to produce an acoustic signal when impacting a structure surface;an acoustic detector module (104) configured to capture acoustic response data of the structure surface to the acoustic signal;a camera module (106) configured to capture visual data of an area surrounding the structure surface during inspection;a data processing module (110) configured to:create a three-dimensional model of the structure surface from the visual data;determine strike positions on the structure surface based on camera positions at times of acoustic signal production;analyze the acoustic response data to detect subsurface defects; and map detected subsurface defects to corresponding positions on the three- dimensional model.
2. The system of claim 1 , wherein the camera module (106) comprises a 360-degree video camera.
3. The system of claim 1 , wherein the data processing module (110) is further configured to:segment the acoustic response data into an acoustic signal with background noise portion and a background noise only portion;generate time-frequency spectra for each portion using short-time Fourier transform; andremove the background noise spectrum from the acoustic signal with background noise spectrum to create a feature vector.
4. The system of claim 3, wherein the data processing module (110) is further configured to classify the feature vector using a machine learning model trained onlabeled acoustic response data to determine presence or absence of subsurface defects.
5. The system of claim 1 , further comprising a mobility module (108) configured to move the system along the structure surface during inspection.
6. The system of claim 5, wherein the mobility module (108) comprises wheels for traversing horizontal or sloped surfaces.
7. The system of claim 5, wherein the system is mounted on an unmanned aerial vehicle for inspecting vertical walls or ceiling surfaces.
8. The system of claim 1 , wherein the data processing module (110) is further configured to create a two-dimensional orthographic projection from the three-dimensional model for visualizing the detected subsurface defects.
9. The system of claim 8, wherein the data processing module (110) is further configured to color-code positions on the two-dimensional orthographic projection to indicate presence or absence of subsurface defects.
10. A method (200) for detecting and localizing subsurface defects, comprising: striking a structure surface at intervals to produce acoustic signals (202); capturing acoustic response data of the structure surface using an acoustic detector (204);simultaneously capturing visual data of an area surrounding the structure surface using a camera (206);creating a three-dimensional model of the structure surface from the visual data (208);determining strike positions on the structure surface based on camera positions at times of acoustic signal production (210, 212);analyzing the acoustic response data to detect subsurface defects (214, 216, 218, 220); andmapping detected subsurface defects to corresponding positions on the three-dimensional model (222).
11. The method of claim 10, wherein capturing visual data comprises recording 360-degree video images of the area surrounding the structure surface.
12. The method of claim 10, wherein analyzing the acoustic response data comprises:segmenting the acoustic response data into an acoustic signal with background noise portion and a background noise only portion (214);generating time-frequency spectra for each portion (216); andremoving the background noise spectrum from the acoustic signal with background noise spectrum to create a feature vector (218).
13. The method of claim 12, further comprising classifying the feature vector using a machine learning model trained on labeled acoustic response data to determine presence or absence of subsurface defects (220).
14. The method of claim 10, further comprising creating a two-dimensional orthographic projection from the three-dimensional model for visualizing the detected subsurface defects (222).
15. The method of claim 14, further comprising color-coding positions on the two-dimensional orthographic projection to indicate presence or absence of subsurface defects.
16. The method of claim 10, wherein the structure surface comprises a vertical wall, and wherein the acoustic detector and camera are mounted on an unmanned aerial vehicle.
17. The method of claim 10, wherein the structure surface comprises a ceiling, and wherein the acoustic detector and camera are mounted on an unmanned aerial vehicle.
18. The method of claim 10, wherein the structure surface comprises a horizontal or sloped surface, and wherein the acoustic detector and camera are mounted on a wheeled rover vehicle.