Non-destructive testing device and method for detecting construction engineering structures

Abstract

The application provides a nondestructive testing device and method for construction engineering structure detection, and relates to the technical field of construction engineering structure detection. The method synchronously collects physical characteristic data and environment-related data of the construction engineering structure through a multi-modal detection unit; pre-processes and fuses the collected multi-source data to extract structure defect-related features; intelligently identifies the defect-related features based on a deep learning model to locate the type, position and size parameters of the structure defects; combines a structure mechanics analysis model and historical detection data to predict the defect development trend and structure safety risk level; and finally outputs the detection results and risk warning report through a visualization unit. The application realizes accurate detection, positioning and risk prediction of construction engineering structure defects through multi-modal data fusion and intelligent algorithm cooperation, and significantly improves the detection efficiency and accuracy.

Description

Technical Field

[0001] This invention relates to the field of structural testing technology for construction projects, and specifically to a non-destructive testing device and method for structural testing in construction projects. Background Technology

[0002] The safety of construction structures (such as concrete structures, steel structures, and masonry structures) directly affects the service life of the project and the safety of people and property. Therefore, structural inspections are necessary during the construction, completion and acceptance, and operation and maintenance phases to promptly detect defects such as cracks, holes, corrosion, and deformation. Non-destructive testing (NDT) technology has become the mainstream technology for structural inspection in construction projects due to its advantages of not requiring damage to the structure and its high inspection efficiency.

[0003] However, existing nondestructive testing technologies still have many limitations: First, traditional testing methods often rely on a single testing means (such as using only ultrasonic testing or infrared testing), and the data collected is limited in dimension, making it difficult to comprehensively reflect the true state of structural defects, resulting in a high rate of false positives and false negatives. Second, data processing often uses traditional signal analysis methods, which are insufficient for extracting features from complex defects and rely on human experience for defect identification, making it highly subjective and significantly affecting the accuracy of testing due to the skill level of the personnel. Third, existing technologies mainly focus on real-time defect identification and lack the ability to predict the development trend of defects, making it impossible to provide early warnings of structural safety hazards and hindering preventive maintenance of structures. Fourth, testing devices are mostly fixed structures, making it difficult to adapt to the structural testing needs of different types and locations of construction projects, resulting in poor operational flexibility.

[0004] Therefore, how to break through the limitations of traditional non-destructive testing technology and provide a multi-dimensional, high-precision, intelligent, and predictable non-destructive testing method and device to achieve accurate detection and risk warning of structural defects in construction projects has become a technical problem that the industry urgently needs to solve. Summary of the Invention

[0005] This invention provides a non-destructive testing device and method for structural inspection of construction projects, which can realize multi-dimensional accurate detection, intelligent identification and risk prediction of structural defects in construction projects, and solve the problems of single data, low accuracy and no prediction capability of traditional technology.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0007] In a first aspect, the present invention provides a non-destructive testing method for structural inspection of construction projects, comprising the following steps:

[0008] Collect multimodal detection data of the construction project structure, including structural physical property data and environmental correlation data;

[0009] The multimodal detection data is preprocessed and feature fusion is performed to extract structural defect-related features;

[0010] The structural defect association features are analyzed based on the intelligent recognition model to identify the type, location, and size parameters of the structural defect;

[0011] By combining the structural mechanics analysis model with historical detection data, the development trend of the structural defects and the structural safety risk level are predicted.

[0012] Output structured detection results and risk warning reports.

[0013] As a further improvement to the technical solution of this invention, the collection of multimodal detection data of construction engineering structures specifically includes:

[0014] A multimodal detection unit composed of ultrasonic sensors, infrared thermal imaging sensors, fiber optic strain sensors, and electromagnetic induction sensors is used to simultaneously collect data on the internal density of the structure, surface temperature field, strain distribution, and material uniformity as data on the physical properties of the structure.

[0015] Temperature and humidity data, as well as external vibration data, are collected from the detection environment using temperature and humidity sensors and vibration sensors, serving as environmental correlation data.

[0016] The structural physical characteristic data and environmental correlation data are time-stamped and format-standardized to obtain multimodal detection data.

[0017] As a further improvement to the technical solution of the present invention, the preprocessing and feature fusion of the multimodal detection data to extract structural defect correlation features specifically includes:

[0018] The multimodal detection data is denoised by using a wavelet threshold denoising algorithm to remove environmental interference noise and sensor inherent noise.

[0019] Feature extraction was performed on the denoised single-modal data. Temporal, frequency and morphological features were extracted from the structural physical property data, and influencing factor features were extracted from the environmental correlation data.

[0020] A weighted fusion algorithm is used to fuse the features of each single mode. Weight coefficients are assigned according to the importance of features in different detection scenarios to generate a structural defect-related feature vector.

[0021] As a further improvement to the technical solution of the present invention, the analysis of the associated features of the structural defects based on the intelligent recognition model, and the identification of the type, location, and size parameters of the structural defects, specifically includes:

[0022] The intelligent recognition model is a convolutional neural network model with an embedded attention mechanism. The model is trained using historical detection data labeled with defect type, location coordinates, and size parameters.

[0023] The defect-associated feature vector is input into the trained intelligent recognition model, and the weights of the key defect features are strengthened through an attention mechanism.

[0024] The model output layer outputs the type of structural defects (cracks, holes, corrosion, deformation), their three-dimensional coordinates, and dimensional parameters (length, width, depth).

[0025] As a further improvement to the technical solution of this invention, by combining the structural mechanics analysis model with historical detection data, the prediction of the development trend of the structural defects and the structural safety risk level specifically includes:

[0026] Construct a structural mechanics analysis model that matches the object being tested, input the defect parameters and structural design parameters, and calculate the influence coefficient of the defect on the structural bearing capacity.

[0027] Historical inspection data of the same structure are retrieved to establish a defect development time series model and fit the change pattern of defect parameters over time.

[0028] By combining the impact coefficient with the defect development time series model, the defect development status within a future preset period can be predicted.

[0029] Based on the preset safety assessment standards, the structural safety risk levels are divided into (low risk, medium risk, high risk, and extremely high risk).

[0030] As a further improvement to the technical solution of this invention, the output of structured detection results and risk warning reports specifically includes:

[0031] The defect type, location, size parameters, development trend and risk level are integrated into structured inspection data;

[0032] Defect location annotation maps, size quantification charts, and risk level heat maps are generated using visualization algorithms.

[0033] Structured detection data and visual charts are integrated into risk warning reports, which support local storage, remote transmission to terminal devices, and printing output.

[0034] As a further improvement to the technical solution of the present invention, the construction engineering structure includes concrete structure, steel structure, masonry structure and composite structure.

[0035] A second aspect of the present invention provides a non-destructive testing device for structural inspection of construction projects, comprising:

[0036] The multimodal detection unit is used to collect structural physical property data and environmental correlation data of construction engineering structures to form multimodal detection data;

[0037] The data processing unit is used to preprocess the multimodal detection data, fuse features, and extract structural defect-related features.

[0038] The intelligent identification unit is used to analyze the associated features of the structural defects based on the intelligent identification model, and to identify the type, location and size parameters of the structural defects.

[0039] The risk prediction unit is used to combine the structural mechanics analysis model with historical inspection data to predict the development trend of defects and the level of structural safety risk.

[0040] The output unit is used to output structured detection results and risk warning reports.

[0041] As a further improvement to the technical solution of the present invention, the multimodal detection unit includes a retractable detection bracket, on which a multi-degree-of-freedom adjustment seat is provided. An ultrasonic sensor, an infrared thermal imaging sensor, a fiber optic strain sensor, an electromagnetic induction sensor, a temperature and humidity sensor, and a vibration sensor are all installed on the multi-degree-of-freedom adjustment seat, which can realize adaptive adjustment of the detection angle and the detection distance.

[0042] A third aspect of the present invention provides a computer device comprising a memory and a processor, the memory storing code, wherein the processor is configured to acquire the code and execute the non-destructive testing method for structural inspection of construction projects as described above.

[0043] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the non-destructive testing method for structural inspection of construction projects as described above.

[0044] The technical solution of the present invention has the following advantages over the prior art:

[0045] This invention overcomes the limitations of traditional single-detection methods by simultaneously collecting multimodal detection data on the physical properties of structures and their environmental correlations. It combines preprocessing and feature fusion techniques to comprehensively extract defect-related features, and then uses an intelligent recognition model to accurately locate defect types, locations, and dimensional parameters. Simultaneously, it integrates structural mechanics analysis and historical data to predict defect development trends and safety risk levels, ultimately outputting structured detection results and risk warning reports. This not only effectively solves the technical pain points of traditional non-destructive testing methods, such as low defect identification accuracy, high false positive and false negative rates, and lack of risk prediction capabilities, but also achieves multi-dimensional coverage, intelligent analysis, and forward-looking early warning for structural defect detection in construction projects. The detection process does not require damage to the structure itself, making it suitable for the entire lifecycle detection of various types of construction structures. It provides scientific data support for structural maintenance decisions, significantly improves detection efficiency and maintenance targeting, effectively avoids structural safety accidents, and possesses outstanding technical practicality and engineering application value. Attached Figure Description

[0046] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0047] Figure 1 This is a schematic diagram of the framework of a non-destructive testing method for structural inspection in construction projects, according to an embodiment of the present invention.

[0048] Figure 2 This is a schematic diagram of the module frame of a non-destructive testing device for structural testing in construction projects, according to an embodiment of the present invention.

[0049] Figure 3 This is a schematic diagram of the composition of a computing device according to an embodiment of the present invention. Detailed Implementation

[0050] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0051] The present invention will be further described in detail below with reference to the accompanying drawings.

[0052] Reference Figure 1 In a first aspect, the present invention provides a non-destructive testing method for structural inspection of construction projects, comprising the following steps:

[0053] Collect multimodal detection data of the construction project structure, including structural physical property data and environmental correlation data;

[0054] The multimodal detection data is preprocessed and feature fusion is performed to extract structural defect-related features;

[0055] The structural defect association features are analyzed based on the intelligent recognition model to identify the type, location, and size parameters of the structural defect;

[0056] By combining the structural mechanics analysis model with historical detection data, the development trend of the structural defects and the structural safety risk level are predicted.

[0057] Output structured detection results and risk warning reports.

[0058] In practice, the process begins by simultaneously acquiring physical characteristic data of the structure itself and associated data of the detection environment through a multimodal detection unit, achieving a comprehensive perception of the object being inspected. Next, the collected multi-source data is preprocessed to remove noise interference, and then key information from different dimensions is integrated using feature fusion technology to form associated features that comprehensively reflect the state of structural defects. Following this, a trained intelligent recognition model is used to conduct in-depth analysis of the defect-related features, accurately locating the core parameters of the defects. Then, combining structural mechanics analysis and historical inspection data, a defect development law model is constructed to predict future trends and the degree of impact on structural safety. Finally, the inspection results and risk assessment information are integrated into a structured report, available to users through various output methods. This entire process forms a complete technical chain of acquisition, processing, identification, prediction, and output, ensuring the comprehensiveness, accuracy, and foresight of the inspection work.

[0059] This invention fundamentally solves the technical pain points of traditional non-destructive testing (NDT) techniques, such as reliance on a single data source, low defect identification accuracy, and lack of risk prediction capabilities. By collaboratively acquiring multimodal data, it overcomes the information limitations of single detection methods, ensuring comprehensive capture of structural defects. The combination of feature fusion and intelligent recognition models significantly improves the accuracy and stability of defect identification, avoiding subjective interference from human judgment. The introduction of structural mechanics analysis and time-series prediction extends the detection work from simple current status identification to future risk warning, providing a scientific basis for preventative maintenance of construction structures. Structured reports and multi-channel output methods enhance the practicality and dissemination efficiency of the detection results. This method is applicable to the entire lifecycle inspection of construction structures, without damaging the structure itself, ensuring the safety of the inspection while significantly improving inspection efficiency and the targeted nature of maintenance decisions, thus possessing broad engineering application value.

[0060] In some embodiments, collecting multimodal detection data of construction engineering structures specifically includes:

[0061] A multimodal detection unit composed of ultrasonic sensors, infrared thermal imaging sensors, fiber optic strain sensors, and electromagnetic induction sensors is used to simultaneously collect data on the internal density of the structure, surface temperature field, strain distribution, and material uniformity as data on the physical properties of the structure.

[0062] Temperature and humidity data, as well as external vibration data, are collected from the detection environment using temperature and humidity sensors and vibration sensors, serving as environmental correlation data.

[0063] The structural physical characteristic data and environmental correlation data are time-stamped and format-standardized to obtain multimodal detection data.

[0064] In practical implementation, a multimodal detection unit composed of various functional sensors is first constructed. Each sensor collects one type or category of key data. Ultrasonic sensors detect internal structural parameters such as density and crack depth; infrared thermal imaging sensors capture temperature field differences on the structural surface caused by defects; fiber optic strain sensors monitor the micro-strain distribution of the structure; and electromagnetic induction sensors identify the uniformity of the structural material and the degree of corrosion. Temperature and humidity sensors and vibration sensors collect data on temperature and humidity changes and external vibration interference in the detection environment, providing environmental references for subsequent data correction and defect analysis. During data acquisition, all sensors operate synchronously at a preset acquisition frequency. After acquisition, various data types are timestamped to ensure consistency in time across different sources. Then, format standardization converts different types and formats of data into a unified data structure, facilitating efficient execution of subsequent preprocessing and feature fusion steps.

[0065] This invention achieves comprehensive perception of the internal and external, static and dynamic characteristics of a structure through the collaborative work of multiple sensors. Compared with traditional single-sensor acquisition methods, it can acquire richer and more comprehensive detection data, providing sufficient information support for defect identification. Timestamp synchronization ensures the spatiotemporal matching of multi-source data, avoiding feature misalignment caused by asynchronous data acquisition and improving the accuracy of subsequent data processing. Format standardization simplifies the subsequent data processing workflow, reduces the difficulty of integrating different types of data, and improves the overall efficiency of the detection method. Simultaneously, targeted sensor selection ensures the professionalism and accuracy of various data acquisition methods. For example, the high sensitivity of infrared thermal imaging sensors to surface defects and the precise capture of micro-strain by fiber optic strain sensors lay a data foundation for accurate defect identification.

[0066] In some embodiments, preprocessing and feature fusion of the multimodal detection data to extract structural defect-related features specifically includes:

[0067] The multimodal detection data is denoised by using a wavelet threshold denoising algorithm to remove environmental interference noise and sensor inherent noise.

[0068] Feature extraction was performed on the denoised single-modal data. Temporal, frequency and morphological features were extracted from the structural physical property data, and influencing factor features were extracted from the environmental correlation data.

[0069] A weighted fusion algorithm is used to fuse the features of each single mode. Weight coefficients are assigned according to the importance of features in different detection scenarios to generate a structural defect-related feature vector.

[0070] In practice: The first step is data denoising. A wavelet threshold denoising algorithm is used to purify the acquired multimodal data. This algorithm decomposes the signal into low-frequency approximate components and high-frequency detail components by performing wavelet transform on the data. The low-frequency components contain the main information of the signal, while the high-frequency components are mainly noise. An adaptive threshold is set to threshold the high-frequency components, and then the signal is reconstructed through inverse wavelet transform, thereby removing environmental interference noise and sensor inherent noise. The second step is single-modal feature extraction. For different types of data after denoising, feature parameters that can reflect their core characteristics are extracted. Time-domain features, frequency-domain features, and morphological features are extracted from the structural physical characteristic data. These features can quantitatively describe the amplitude changes, frequency distribution, and waveform morphology of the data. Influencing factor features are extracted from environmental correlation data to characterize the degree of influence of environmental factors on the structural state and detection results. The third step is feature fusion. A weighted fusion algorithm is used to integrate the features of each single mode. According to the importance of various features to defect identification in different detection scenarios, corresponding weight coefficients are assigned through scientific methods, and all features are integrated into a unified defect-related feature vector, realizing the organic fusion of multi-dimensional information.

[0071] Wavelet thresholding denoising algorithm effectively removes noise interference while preserving key signal information, improving the signal-to-noise ratio and providing a high-quality data foundation for subsequent feature extraction and recognition analysis. Targeted extraction of single-modal features ensures that the core information of each data type is not overlooked, and the combination of time-domain, frequency-domain, and morphological features comprehensively characterizes the signal features of structural defects. Weighted fusion algorithm, through reasonable weight allocation, highlights the role of key features and suppresses the influence of secondary or interfering features. The resulting defect-related feature vector comprehensively and accurately reflects the overall state of structural defects. Compared to single-modal features, it contains richer information and has higher discriminative power, providing strong support for the accurate analysis of subsequent intelligent recognition models. Furthermore, this processing flow has good adaptability, allowing adjustment of feature extraction methods and weight allocation according to different structural types and defect types, improving the method's versatility.

[0072] The preferred wavelet thresholding denoising algorithm uses the db4 wavelet basis, which possesses good temporal localization characteristics and frequency domain resolution, making it suitable for denoising structural detection signals. The wavelet decomposition level is set to three layers, ensuring denoising effectiveness while avoiding signal distortion caused by over-decomposition. Temporal features specifically include parameters such as peak value, mean, variance, peak factor, and kurtosis, reflecting the signal's amplitude distribution and fluctuation characteristics. Frequency domain features specifically include parameters such as dominant frequency, spectral energy, and spectral entropy, characterizing the signal's frequency composition and energy distribution. Morphological features specifically include parameters such as waveform slope, number of abrupt changes, and waveform duration, describing the signal's changing trends and morphological characteristics. Environmental influence factor features specifically include parameters such as temperature and humidity change rate, peak vibration amplitude, and vibration frequency. The preferred method for determining the weighting coefficients is the analytic hierarchy process (AHP). By constructing a hierarchical model, the importance of each type of feature is compared pairwise, and the normalized weighting coefficients are calculated. For example, in the scenario of detecting cracks in concrete structures, the feature weight corresponding to the data acquired by the ultrasonic sensor can be set to 0.4, the feature weight corresponding to the data acquired by the infrared thermal imaging sensor is 0.3, the feature weight corresponding to the data acquired by the fiber optic strain sensor is 0.2, and the feature weight corresponding to the environmental data is 0.1, ensuring that the features of key detection data are given sufficient attention.

[0073] In some embodiments, analyzing the associated features of the structural defects based on an intelligent recognition model to identify the type, location, and size parameters of the structural defects specifically includes:

[0074] The intelligent recognition model is a convolutional neural network model with an embedded attention mechanism. The model is trained using historical detection data labeled with defect type, location coordinates, and size parameters.

[0075] The defect-associated feature vector is input into the trained intelligent recognition model, and the weights of the key defect features are strengthened through an attention mechanism.

[0076] The model output layer outputs the type of structural defects (cracks, holes, corrosion, deformation), their three-dimensional coordinates, and dimensional parameters (length, width, depth).

[0077] It should be noted that the core of the intelligent structural defect recognition method is based on a convolutional neural network model with an embedded attention mechanism to identify defect parameters. The training process of the intelligent recognition model requires a large amount of historical detection data labeled with defect types, location coordinates, and size parameters. The model parameters are continuously adjusted through backpropagation to enable the model to accurately learn and recognize defect features. In the actual detection process, the defect-related feature vector obtained through feature fusion is first input into the trained model. The model's attention mechanism automatically focuses on key features related to the defect, strengthening the weights of these features while suppressing interference from irrelevant features, thus improving the targeting of feature recognition. Subsequently, convolutional layers perform depth extraction and dimensionality transformation on the feature vectors, pooling layers perform dimensionality reduction on the extracted features to reduce computation while retaining key information, and fully connected layers map the processed features to a preset output space. Finally, the output layer outputs the specific parameters of the structural defect, including defect type, three-dimensional coordinate position, and size parameters, achieving accurate identification and quantitative representation of the defect.

[0078] Compared to traditional convolutional neural networks, convolutional neural network models with embedded attention mechanisms can more accurately capture key defect features, effectively improving the accuracy and robustness of defect identification and avoiding misjudgments and missed judgments caused by irrelevant features. Trained on a large amount of historical data, the model possesses the ability to identify defects of different types and scales, exhibiting strong adaptability and being unaffected by human experience and subjective judgment, ensuring the objectivity and consistency of detection results. The quantified output of defect parameters (3D position and size data) provides accurate basic data for subsequent risk prediction, making risk assessment more scientific and reliable. Simultaneously, the model has strong parallel computing capabilities, enabling rapid processing of multimodal fused feature data, shortening the identification response time, and improving the efficiency of detection work.

[0079] The network structure of this intelligent recognition model is specifically configured as follows: the dimension of the input layer is consistent with the dimension of the defect-associated feature vector, determined based on the actual feature extraction results; a total of 3 convolutional layers are set, with the first layer having a kernel size of [missing information]. The number of convolutional layers is 64, the stride is 1, the padding method is "same", and the activation function is ReLU; the kernel size of the second layer is... The number of convolutional layers is 128, the stride is 1, the padding method is "same", and the activation function is ReLU; the kernel size of the third layer is... The convolutional layer has 256 elements, a stride of 1, and uses the same padding method. The activation function is ReLU. The attention layer employs a channel attention mechanism, which enhances effective channel features by assigning weights to the channel dimensions of the feature maps output by the convolutional layers. The pooling layer uses max pooling with a kernel size of [missing value]. The model employs a step size of 2. It has two fully connected layers: the first layer has 1024 neurons with ReLU activation, and the second layer has 512 neurons with ReLU activation. The number of neurons in the output layer is determined based on the number of defect types, the dimension of the location coordinates, and the number of size parameters. Softmax activation is used for classification tasks, and Linear activation is used for regression tasks. During model training, the ratio of training, validation, and test sets is set to 7:2:1, and cross-validation is used to avoid overfitting. The loss function is a combination of cross-entropy loss (for defect type classification) and mean squared error loss (for location and size regression). The optimizer is the Adam optimizer, with an initial learning rate of 0.001, dynamically adjusted using a learning rate decay strategy. Training stops when the recognition accuracy on the validation set is no less than 95%, ensuring model performance. In practical applications, the model's recognition response time is less than 1 second, meeting the requirements for real-time on-site detection.

[0080] In some embodiments, combining structural mechanics analysis models with historical detection data to predict the development trend of structural defects and the level of structural safety risks specifically includes:

[0081] Construct a structural mechanics analysis model that matches the object being tested, input the defect parameters and structural design parameters, and calculate the influence coefficient of the defect on the structural bearing capacity.

[0082] Historical inspection data of the same structure are retrieved to establish a defect development time series model and fit the change pattern of defect parameters over time.

[0083] By combining the impact coefficient with the defect development time series model, the defect development status within a future preset period can be predicted.

[0084] Based on the preset safety assessment standards, the structural safety risk levels are divided into (low risk, medium risk, high risk, and extremely high risk).

[0085] In practice, the first step involves selecting a matching structural mechanics analysis model based on the structural type of the object being inspected. For example, a finite element analysis model is used for concrete structures, and a truss mechanics model is used for steel structures. The defect parameters obtained from intelligent identification and the structural design parameters are input into the model. Mechanical calculations are then used to analyze the impact of defects on the structural bearing capacity, yielding an influence coefficient. The second step involves retrieving historical inspection data of the same construction project from a database, including defect parameters, environmental data, and structural performance data at different time points. Based on this time-series data, a defect development time-series model is constructed to fit the changing patterns of defect parameters over time. The third step combines the influence coefficients obtained from the structural mechanics analysis with the defect development time-series model, substituting preset time period parameters to calculate the expected development state of defects in different future time periods, including trends in defect size and location. The fourth step, referring to preset structural safety assessment standards, and considering the current state, development trend, and impact on bearing capacity of the defects, the structural safety risk is classified into different levels, completing the risk assessment.

[0086] This invention's predictive method combines structural mechanics analysis with historical data time-series analysis to achieve a scientific prediction of defect development trends. This overcomes the limitations of traditional detection methods, which can only identify the current defect state, providing a forward-looking basis for preventative structural maintenance. The calculation of the influence coefficient quantifies the degree of impact of defects on structural safety, making risk assessment more scientific. The risk level classification makes the detection results more intuitive, facilitating users to formulate targeted maintenance strategies based on risk levels and effectively preventing structural safety accidents. The construction of the time-series model fully utilizes historical detection data to uncover the inherent laws governing defect development, improving the accuracy and reliability of the predictive results.

[0087] The selection of a structural mechanics analysis model needs to be determined based on the structural stress characteristics and construction form. For concrete beams, slabs, columns, and other components, the finite element analysis model can use solid elements or shell elements to model and simulate the stress and strain distribution of the components. For steel trusses, frames, and other structures, the truss mechanics model can calculate the internal forces of the components through nodal force equilibrium equations and analyze the impact of defects on the distribution of internal forces. The preferred defect development time series model is the Long Short-Term Memory (LSTM) network model, which can effectively capture long-term dependencies in time series data and is suitable for predicting defect development trends. The model's input features include historical defect parameters, historical environmental data, and the structure's service life, while the output is the predicted values ​​of defect parameters at different future time points. The preset time period can be set to 1 year, 3 years, or 5 years, covering short-term, medium-term, and long-term maintenance needs. The classification of structural safety risk levels references the technical standards for building structure testing, specifically dividing them into four levels: Low risk indicates that the defect has no obvious development trend, has no impact on the structural load-bearing capacity, and requires no special maintenance; Medium risk indicates that the defect has a slow development trend, has a minor impact on the structural load-bearing capacity, and requires regular monitoring; High risk indicates that the defect is developing rapidly and has already had a certain impact on the structural load-bearing capacity, requiring reinforcement measures; Extremely high risk indicates that the defect is approaching a critical state, seriously affecting structural safety, and requires immediate cessation of use and emergency reinforcement. The calculation of the influence coefficient must consider the type, size, and location of the defect, as well as the structural design safety factor. For example, the influence coefficient of crack-type defects is related to the crack length, width, depth, and concrete strength grade; the influence coefficient of corrosion-type defects is related to the corrosion area, corrosion depth, and steel strength.

[0088] In some embodiments, the output of structured detection results and risk warning reports specifically includes:

[0089] The defect type, location, size parameters, development trend and risk level are integrated into structured inspection data;

[0090] Defect location annotation maps, size quantification charts, and risk level heat maps are generated using visualization algorithms.

[0091] Structured detection data and visual charts are integrated into risk warning reports, which support local storage, remote transmission to terminal devices, and printing output.

[0092] In practical implementation, the first step is to integrate the information obtained from intelligent identification, such as defect type, 3D coordinates of location, and size parameters, with the defect development trend and safety risk level obtained from risk prediction. This information is then organized into structured inspection data according to a preset data structure to ensure the data's systematic and standardized nature. The second step involves using visualization algorithms to process the structured inspection data into charts, generating defect location annotation maps, size quantification charts, and risk level heatmaps. The defect location annotation map overlays the 3D location of defects onto the 3D model of the structure, visually displaying the distribution of defects. The size quantification chart uses bar charts, line charts, and other formats to show the specific size and development trend of defects. The risk level heatmap uses different color intensities to represent the risk level of different areas of the structure, facilitating the rapid identification of high-risk areas. The third step integrates the structured inspection data and visualization charts into a standardized risk warning report, supporting multiple output methods, including local storage to the inspection device's storage module, remote transmission to regulatory terminal equipment via a communication module, and printing out paper reports, meeting the needs of different users.

[0093] Structured data processing makes test results more organized, facilitating data storage, retrieval, and subsequent analysis. The generation of visual charts transforms abstract data into intuitive graphics, reducing the difficulty for users to understand the test results and facilitating a quick grasp of defect status and risk distribution. Multiple output methods enhance the practicality and dissemination efficiency of test results; testing personnel can view results on-site, and regulatory authorities can obtain reports remotely, achieving closed-loop management of the testing process. Standardized report formats ensure the uniformity and professionalism of test results, providing a clear and reliable basis for maintenance decisions.

[0094] The pre-defined structure of the structured inspection data includes four parts: basic inspection information, structural information, defect information, and risk assessment information. Basic inspection information includes inspection time, location, personnel, and equipment number; structural information includes structure type, name, construction year, and design parameters; defect information includes defect number, type, 3D coordinates of location, dimensional parameters, and inspection method; and risk assessment information includes defect development trend prediction, impact coefficient, safety risk level, and maintenance recommendations. The visualization algorithm preferably uses a professional data visualization library. The defect location annotation map is built based on the structure's BIM model, associating defect parameters with components in the BIM model to achieve accurate defect location annotation. In the dimensional quantification charts, bar charts are used to display the size comparison of different defects, and line charts are used to display the development trend of individual defects. The risk level heatmap uses the following color schemes: green for low risk, yellow for medium risk, orange for high risk, and red for extremely high risk, providing clear color contrast for easy identification. The risk warning report should preferably be in PDF format, as this format offers good compatibility and immutability, facilitating long-term preservation and dissemination. The report should include a cover, table of contents, inspection overview, basic structural information, detailed defect information, risk assessment results, maintenance recommendations, and appendices. The appendices may contain supplementary information such as original inspection data, sensor parameters, and model training reports. The output method is as follows: the local storage module uses a solid-state drive with a capacity of at least 128GB to support fast data read and write; the communication module preferably uses a 5G communication module to ensure the speed and stability of remote transmission, while also being compatible with 4G, WiFi, and other communication methods; the printing interface uses a standard USB interface or wireless printing protocol, supporting connection to mainstream printers on the market.

[0095] In some embodiments, the construction structure includes concrete structure, steel structure, masonry structure, and composite structure.

[0096] It should be noted that for concrete structures, the focus of inspection is on defects such as internal cracks, voids, and insufficient density. Ultrasonic sensors are used to detect internal defects, infrared thermal imaging sensors to capture surface cracks, and fiber optic strain sensors to monitor structural strain. For steel structures, the focus is on corrosion, weld defects, and deformation. Electromagnetic induction sensors are used to identify the degree of corrosion, ultrasonic sensors to detect weld defects, and fiber optic strain sensors to monitor deformation. For masonry structures, the focus is on mortar fullness, masonry cracks, and insufficient strength. Ultrasonic sensors are used to detect mortar fullness and internal cracks, and infrared thermal imaging sensors to capture surface defects. For composite structures, the inspection strategies of different sensors are flexibly combined according to the characteristics of their components (such as concrete-steel structure composites and masonry-concrete structure composites). The weights of feature extraction and the parameters of the intelligent recognition model are adjusted to ensure the adaptability of the inspection method to various types of structures.

[0097] This invention overcomes the limitations of traditional non-destructive testing (NDT) techniques for specific structural types, achieving comprehensive testing of various construction structures. It eliminates the need for developing separate testing schemes for different structural types, reducing testing costs and technical barriers. By optimizing sensor combinations and parameter settings for different structural types, it ensures targeted and accurate testing, enhancing the method's versatility and practicality. Applicable to the entire lifecycle of construction projects, it meets the diverse testing needs of the construction, completion and acceptance, and operation and maintenance phases, providing comprehensive technical support for structural safety management.

[0098] Concrete structures include plain concrete, reinforced concrete, and prestressed concrete structures. Common applications include building floor slabs, beams, columns, bridge main beams, and tunnel linings. During inspection, the detection frequency of the ultrasonic sensor can be adjusted to 2-3MHz, focusing on detecting the depth of internal cracks and the size of pores. The detection distance of the infrared thermal imaging sensor can be controlled between 0.5-2m to ensure clear capture of surface cracks. Steel structures include structures made of carbon structural steel and low-alloy high-strength structural steel. Common applications include steel structures in factories, bridges, and towers. The detection frequency of the electromagnetic induction sensor can be set to 1-5kHz, focusing on identifying the area and depth of corrosion. When using the ultrasonic sensor for weld inspection, the probe angle can be adjusted to 45°, 60°, or 70° depending on the weld type to ensure the detection of internal weld defects. Masonry structures include brick masonry, stone masonry, and block masonry, with common applications including building walls, fences, and small bridge piers. The ultrasonic sensor's detection point spacing is set to 100-200mm, using a grid-like detection method to cover the entire detection area. The infrared thermal imaging sensor's detection time is preferably selected during periods of stable ambient temperature to avoid the impact of temperature fluctuations on the detection results. Composite structures include steel-concrete composite structures, masonry-concrete composite structures, and steel-wood composite structures. During detection, the sensor layout needs to be adjusted according to the composition ratio and connection method of the composite structure. For example, in the detection of steel-concrete composite beams, electromagnetic induction sensors and ultrasonic sensors are used for the steel structure, while ultrasonic sensors and infrared thermal imaging sensors are used for the concrete structure. The sensor installation positions must avoid interference from structural connection points. This method maintains a consistent core technology chain across different structural types, adjusting only sensor parameters, feature weights, and fine-tuning parameters of the recognition model based on structural characteristics to ensure the method's universality and adaptability.

[0099] Reference Figure 2 The second aspect of the present invention provides a non-destructive testing device for structural inspection of construction projects, comprising:

[0100] The multimodal detection unit is used to collect structural physical property data and environmental correlation data of construction engineering structures to form multimodal detection data;

[0101] The data processing unit is used to preprocess the multimodal detection data, fuse features, and extract structural defect-related features.

[0102] The intelligent identification unit is used to analyze the associated features of the structural defects based on the intelligent identification model, and to identify the type, location and size parameters of the structural defects.

[0103] The risk prediction unit is used to combine the structural mechanics analysis model with historical inspection data to predict the development trend of defects and the level of structural safety risk.

[0104] The output unit is used to output structured detection results and risk warning reports.

[0105] It should be noted that the multimodal detection unit, acting as the data input, synchronously collects structural physical characteristic data and environmental correlation data through various sensors, transmitting the raw data to the data processing unit. Upon receiving the raw data, the data processing unit sequentially performs denoising, single-modal feature extraction, and weighted feature fusion operations to generate a defect-related feature vector, which is then transmitted to the intelligent recognition unit. The intelligent recognition unit calls upon its built-in, trained intelligent recognition model to analyze the defect-related feature vector, outputting the defect type, location, and size parameters, which are then transmitted to the risk prediction unit. The risk prediction unit retrieves a preset structural mechanics analysis model and a historical detection database, calculates the influence coefficient based on the defect parameters, predicts the defect development trend, and classifies the safety risk level, transmitting the risk assessment results to the output unit. The output unit receives the structured data from the data processing unit, the defect parameters from the intelligent recognition unit, and the risk assessment results from the risk prediction unit, integrating them to generate a structured report and visual charts, presenting them to the user through various output methods. All units are connected via an internal high-speed bus to ensure real-time and stable data transmission.

[0106] This invention integrates all aspects of the inspection process into a cohesive whole through modular design, achieving automation and integration of inspection work, significantly reducing manual labor intensity and improving inspection efficiency. The functional division of each unit is clear: the hardware architecture of the data processing unit ensures real-time processing of multi-source data; the hardware acceleration module of the intelligent identification unit improves the speed and accuracy of defect identification; the model library of the risk prediction unit supports risk assessment for various structural types; and the multi-channel output of the output unit enhances ease of use. The modular structure of the device facilitates later maintenance and upgrades, allowing for the replacement or addition of sensors and updates to algorithm models according to technological advancements and actual needs, extending the device's lifespan. The overall design balances comprehensiveness, accuracy, and practicality in inspection, adapting to the inspection needs of different scenarios and structural types, providing an integrated technical solution for structural inspection in construction projects.

[0107] The multimodal detection unit features pluggable sensors that connect to the detection unit via standardized interfaces, facilitating sensor replacement, calibration, and expansion. Other types of sensors, such as gas sensors and acoustic sensors, can be added according to detection requirements. The data processing unit employs a heterogeneous architecture of FPGA and ARM. The FPGA handles parallel computation for data preprocessing, achieving a processing speed of at least 1GB / s, enabling rapid denoising and format conversion of multi-source data. The ARM processor, with a multi-core architecture and a clock speed of at least 1.8GHz, is responsible for running feature fusion algorithms and control logic, ensuring real-time data processing. The intelligent recognition unit incorporates a hardware acceleration module, preferably using a neural network processor, which accelerates the computation of convolutional neural network models, reducing defect recognition response time to less than one second. The unit internally stores multiple trained defect recognition models, automatically selecting the appropriate model based on the structural and defect types. The risk prediction unit has a built-in structural mechanics analysis model library, including mechanical models for various structural types such as concrete structures, steel structures, masonry structures, and composite structures, supporting dynamic updates and expansions of the models. The unit connects to a historical monitoring database via a data interface; the database can be stored locally or deployed in the cloud, supporting rapid retrieval and updates of historical data. The output unit includes a touchscreen display, a communication module, a storage module, and an interface module, with the touchscreen display having a resolution of at least [insert resolution here]. It supports multi-touch operation for easy viewing and operation; the communication module integrates multiple communication methods such as 5G, 4G, WiFi, and Bluetooth to ensure remote data transmission in different environments; the storage module uses a solid-state drive with a capacity of no less than 128GB to support long-term storage of raw data, processing results, and report files; the interface module includes USB, HDMI, printer, and sensor interfaces to meet the needs of data transmission and external device connection. The overall size of the device is designed for portability, weighing no more than 5kg, and equipped with a rechargeable lithium battery with a battery life of no less than 8 hours to meet the needs of mobile on-site testing.

[0108] In some embodiments, the multimodal detection unit includes a retractable detection bracket, on which a multi-degree-of-freedom adjustment seat is provided. An ultrasonic sensor, an infrared thermal imaging sensor, a fiber optic strain sensor, an electromagnetic induction sensor, a temperature and humidity sensor, and a vibration sensor are all mounted on the multi-degree-of-freedom adjustment seat, enabling adaptive adjustment of the detection angle and detection distance.

[0109] It should be noted that the telescopic testing bracket, as a supporting structure, adjusts its overall length via a telescopic mechanism to adapt to testing scenarios of different heights and distances. The multi-degree-of-freedom adjustment seat on the bracket adjusts the detection angle of the sensors mounted on it through rotation and pitch mechanisms, enabling omnidirectional detection of different parts of the structure. Various sensors are fixed to the adjustment seat via magnetic interfaces, ensuring the stability of the sensor installation and facilitating quick disassembly and replacement. The laser positioning module emits a laser beam to assist the sensors in accurately aligning with the detection area, ensuring the spatial positioning accuracy of the detection data. During the detection process, the bracket provides stable support, the adjustment seat adjusts the sensor posture, the laser positioning module calibrates the detection position, and the sensors synchronously collect data, jointly completing the accurate acquisition of multimodal detection data.

[0110] The telescopic bracket is made of carbon fiber, combining high strength and lightweight characteristics. This ensures the bracket's stability while also making it easy to carry and operate. The telescopic length ranges from 1-3 meters, meeting the testing needs of most building structures at different heights and distances. The multi-degree-of-freedom adjustable base allows for 360-degree rotation and ±90-degree pitch adjustment, enabling the sensor to flexibly adjust the detection angle without blind spots, adapting to the testing of complex structures. The magnetic sensor interface allows for convenient installation and quick disassembly, facilitating sensor replacement, calibration, and combination, enhancing the flexibility and adaptability of the testing unit. The laser positioning module has a positioning accuracy of at least ±2mm, ensuring the sensor is accurately aligned with the detection area, avoiding data distortion caused by detection position deviations, and improving the accuracy of the detection data. The overall structural design balances stability, flexibility, and convenience, enabling the multimodal testing unit to adapt to complex on-site testing environments, improving the efficiency and quality of testing work.

[0111] The telescopic bracket features a multi-stage sleeve design for its telescopic mechanism. The sleeves are secured by locking devices, employing either threaded or snap-locking methods to ensure stability after length adjustment. Anti-slip pads at the bottom of the bracket increase friction with the ground, enhancing its stability under varying ground conditions. The multi-degree-of-freedom adjustable base includes a rotation mechanism and a pitch mechanism. The rotation mechanism is driven by a stepper motor, achieving a rotation angle accuracy of 0.1 degrees and enabling continuous or fixed-point rotation. The pitch mechanism is also driven by a stepper motor, achieving a pitch angle accuracy of 0.1 degrees and supporting both manual and automatic adjustment modes. The magnetic interface incorporates a strong magnet and a positioning pin. The strong magnet ensures a secure connection between the sensor and the adjustable base, while the positioning pin guarantees the sensor's positional accuracy. The interface's contact resistance is less than 10 milliohms, ensuring stable data transmission. The standardized interface design adapts to the installation requirements of different sensor types, automatically completing electrical connection and data communication after the sensor connects to the interface. The laser positioning module uses a semiconductor laser with a wavelength of 650nm and an output power not exceeding 5mW, meeting safety standards. The laser beam divergence angle is less than 0.1 milliradians, ensuring accurate positioning even at long distances. The positioning module and sensor are relatively fixed in position, and a calibration algorithm eliminates positioning deviations, ensuring consistency between the laser positioning point and the sensor's detection center. The detection unit is also equipped with a sensor attitude monitoring module, which monitors the sensor's tilt angle and horizontal state in real time. The data is fed back to the control unit, allowing users to adjust the sensor's attitude and ensuring the accuracy of the detection data. The module's measurement accuracy is ±0.05 degrees, providing real-time feedback of the sensor's attitude information.

[0112] Reference Figure 3 A third aspect of the present invention provides a computer device comprising a memory and a processor, the memory storing code, wherein the processor is configured to acquire the code and execute the non-destructive testing method for structural inspection of construction projects as described above.

[0113] It should be noted that the memory, as a data storage medium, is used to store the program code implementing the non-destructive testing method, the collected raw test data, intermediate data during processing, the trained intelligent recognition model and structural mechanics analysis model, the historical test database, and the generated test reports. The processor, as the core computing unit, obtains the program code stored in the memory via the communication bus and executes operations such as multimodal data acquisition control, data preprocessing, feature fusion, intelligent recognition, risk prediction, and result output sequentially according to the instructions in the program code. During the data acquisition phase, the processor sends control commands to the multimodal detection unit of the non-destructive testing device through the communication interface, controlling the sensor's startup, acquisition frequency, and data transmission. During the data processing phase, the processor runs preset denoising algorithms, feature extraction algorithms, and fusion algorithms to process the collected multi-source data. During the intelligent recognition and risk prediction phase, the processor calls the model file stored in the memory, inputs the processed feature data, performs model calculations, and obtains defect parameters and risk assessment results. During the result output phase, the processor controls the communication interface to transmit the test results to the output device or store them in the memory. The communication bus is used to realize data transmission and instruction exchange between the processor, memory, and communication interface, ensuring the coordinated operation of all components.

[0114] This computer equipment provides a stable and efficient hardware platform for non-destructive testing (NDT) methods. The high-performance processor ensures the rapid execution of complex algorithms (such as deep learning models and structural mechanics calculations), improving the efficiency of testing. Large-capacity storage meets the storage needs of program code, model files, and massive amounts of testing data, supporting the querying and reuse of historical data and providing data support for predicting defect development trends. Standardized communication interfaces and bus architecture ensure compatibility and stable communication between the computer equipment and various units of the NDT apparatus, facilitating equipment integration and expansion. The computer equipment can serve as the core control unit of the NDT apparatus or as an independent data analysis device, receiving data transmitted from the testing apparatus for offline analysis, enhancing the flexibility and applicability of the testing methods. Through the hardware support of the computer equipment, the automation and intelligentization of NDT methods become possible, ensuring the accuracy and reliability of testing results.

[0115] In some embodiments, the non-destructive testing method for structural testing of construction projects described above can be implemented by a computer device, which includes at least one processor, a communication bus, a memory, and at least one communication interface.

[0116] A processor can be a general-purpose central processing unit (CPU) or an application-specific integrated circuit (ASIC).

[0117] A communication bus can be used to transmit information between the aforementioned components.

[0118] The memory can be read-only memory (ROM) or other types of static storage devices capable of storing static information and instructions, random access memory (RAM) or other types of dynamic storage devices capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, universal optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited to these. The memory can exist independently and be connected to the processor via a communication bus. The memory can also be integrated with the processor.

[0119] The memory stores program code for executing the present invention, and its execution is controlled by a processor. The processor executes the program code stored in the memory. The program code may include one or more software modules. The non-destructive testing method for structural inspection of construction projects in the above embodiments can be implemented by a processor and one or more software modules in the program code in the memory.

[0120] A communication interface is a device that uses any transceiver or similar device to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0121] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0122] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This embodiment of the invention does not limit the type of computer device.

[0123] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the non-destructive testing method for structural inspection of construction projects as described above.

[0124] To better understand the technical solution of the present invention, the present invention will be described in detail below with reference to specific implementation examples.

[0125] refer to Figure 1 The figure is an exemplary flowchart of the nondestructive testing method according to an embodiment of the present invention. The method mainly includes the following steps:

[0126] Step 1: Multimodal Testing Data Acquisition: Move the multimodal testing unit of the nondestructive testing device to the testing area. Adjust the length of the telescopic support (e.g., to 2m) and the angle of the multi-degree-of-freedom adjustable seat according to the structural type of the construction project (e.g., concrete beam). Align the laser positioning module with the testing area. Activate the ultrasonic sensor (operating frequency 3MHz), infrared thermal imaging sensor (temperature resolution 0.05℃), and fiber optic strain sensor (accuracy...). The system uses an electromagnetic induction sensor (detection range 0-200mm) to simultaneously collect data on the internal density, surface temperature field, strain distribution, and material uniformity of the structure; it also activates a temperature and humidity sensor and a vibration sensor to collect ambient temperature and humidity (e.g., 25℃, 60%RH) and external vibration data (e.g., 0.1g); and obtains multimodal detection data through timestamp synchronization (accuracy 1ms) and JSON format standardization processing.

[0127] Step 2: Data Preprocessing and Feature Fusion: The FPGA module of the data processing unit uses the db4 wavelet threshold denoising algorithm to denoise the multimodal detection data, removing environmental interference and sensor noise; feature extraction is performed on the denoised data: peak value, dominant frequency, waveform slope, and other features are extracted from ultrasonic data; temperature difference, temperature gradient, and other features are extracted from infrared thermal imaging data; strain peak value, coefficient of variation, and other features are extracted from strain data; temperature and humidity change rate, vibration amplitude, and other features are extracted from environmental data; the weight of each feature is determined using the analytic hierarchy process (e.g., ultrasonic feature weight 0.4, infrared feature weight 0.3, strain feature weight 0.2, environmental feature weight 0.1), and a defect-related feature vector is generated through a weighted fusion algorithm.

[0128] Step 3: Intelligent Defect Recognition: Input the defect-related feature vector into the Attention-CNN model of the intelligent recognition unit. The model strengthens the weights of key features such as cracks and holes through the channel attention mechanism and suppresses interference from irrelevant features. The model output layer outputs the defect type (e.g., concrete crack), three-dimensional coordinate position (e.g., X=500mm, Y=300mm, Z=100mm), and size parameters (e.g., length 200mm, width 0.2mm, depth 50mm). The recognition response time is ≤1s.

[0129] Step 4: Risk Prediction: The risk prediction unit calls the finite element analysis model of the concrete structure, inputs the defect parameters (crack length 200mm, width 0.2mm, depth 50mm) and structural design parameters (C30 concrete strength, beam cross-section size 300×500mm, span 6m), and calculates the influence coefficient of the defect on the bearing capacity as 0.15; retrieves the historical inspection data of the beam (inspection records of the past 2 years), fits the crack length growth law through the LSTM time series prediction model, and predicts that the crack length will increase to 250mm in the next 3 years; combining the influence coefficient and the prediction results, according to the "Technical Standard for Building Structure Inspection", the safety risk level of the beam is determined to be medium risk, and regular monitoring is required every 6 months.

[0130] Step 5: Output Results: The output unit integrates data such as defect type, location, size, development trend, and risk level into structured inspection data. It generates a defect location annotation map (overlaid on the 3D beam model), a crack length growth trend line graph, and a risk level heat map (medium risk is marked in yellow) using Matplotlib. The structured data and visualization charts are integrated into a PDF format risk warning report, which can be viewed in real time by inspection personnel through a touch screen, transmitted to the regulatory department's terminal via a 5G module, or printed as a paper report via a printer interface.

[0131] The technical solutions provided by the embodiments of the present invention have the following beneficial effects:

[0132] Multimodal data fusion enhances the comprehensiveness of detection: This invention uses multiple sensors such as ultrasonic waves, infrared thermal imaging, and fiber optic strain to simultaneously collect structural physical property data and environmental correlation data, breaking through the limitations of traditional single detection methods. Multi-dimensional data collaboration can comprehensively reflect the state of structural defects and reduce the rate of false and false detections.

[0133] Intelligent algorithm optimization improves recognition accuracy: a convolutional neural network model with embedded attention mechanism is used for defect recognition. Combined with the fusion analysis of time domain, frequency domain and morphological features, the weight of key features is strengthened. Compared with traditional manual recognition and simple algorithms, the defect recognition accuracy is improved to more than 95%, and it is not affected by subjective factors of personnel.

[0134] Risk prediction enables preventive maintenance: Through structural mechanics analysis models and time-series prediction models, not only can current defects be identified, but also the development trend of defects and the structural safety risk level can be predicted, providing data support for the preventive maintenance of construction structures and avoiding safety accidents.

[0135] The device is flexible and adaptable to various testing scenarios: the telescopic bracket and multi-degree-of-freedom adjustable seat design, combined with the magnetic sensor interface, can adapt to various construction engineering structures such as concrete and steel structures, as well as testing needs at different heights and locations, significantly improving operational flexibility;

[0136] Real-time visualization output enhances ease of use: The combination of structured inspection results and visual charts supports local viewing, remote transmission and printing, allowing inspection personnel and regulatory authorities to monitor the structural status in real time, thereby improving the efficiency and management level of inspection work.

[0137] In summary, this invention achieves precise, intelligent, and predictive non-destructive testing of construction structures through the collaborative innovation of multimodal data acquisition, intelligent feature fusion, deep learning recognition, and mechanical model prediction. It solves many pain points of traditional technologies and has significant technological advancements and practical application value.

[0138] The technical solutions provided by the embodiments of the present invention have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the embodiments of the present invention. The descriptions of the embodiments above are only for helping to understand the principles of the embodiments of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the embodiments of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A non-destructive testing method for structural inspection in construction projects, characterized in that, Includes the following steps: Collect multimodal detection data of the construction project structure, including structural physical property data and environmental correlation data; The multimodal detection data is preprocessed and feature fusion is performed to extract structural defect-related features; The structural defect association features are analyzed based on the intelligent recognition model to identify the type, location, and size parameters of the structural defect; By combining the structural mechanics analysis model with historical detection data, the development trend of the structural defects and the structural safety risk level are predicted. Output structured detection results and risk warning reports.

2. The non-destructive testing method for structural inspection of construction projects according to claim 1, characterized in that: The collection of multimodal testing data for construction structures specifically includes: A multimodal detection unit composed of ultrasonic sensors, infrared thermal imaging sensors, fiber optic strain sensors, and electromagnetic induction sensors is used to simultaneously collect data on the internal density of the structure, surface temperature field, strain distribution, and material uniformity as data on the physical properties of the structure. Temperature and humidity data, as well as external vibration data, are collected from the detection environment using temperature and humidity sensors and vibration sensors, serving as environmental correlation data. The structural physical characteristic data and environmental correlation data are time-stamped and format-standardized to obtain multimodal detection data.

3. The non-destructive testing method for structural inspection of construction projects according to claim 1, characterized in that: Preprocessing and feature fusion of the multimodal detection data to extract structural defect-related features specifically includes: The multimodal detection data is denoised by using a wavelet threshold denoising algorithm to remove environmental interference noise and sensor inherent noise. Feature extraction was performed on the denoised single-modal data. Temporal, frequency and morphological features were extracted from the structural physical property data, and influencing factor features were extracted from the environmental correlation data. A weighted fusion algorithm is used to fuse the features of each single mode. Weight coefficients are assigned according to the importance of features in different detection scenarios to generate a structural defect-related feature vector.

4. The non-destructive testing method for structural inspection of construction projects according to claim 1, characterized in that: The analysis of the structural defect association features based on the intelligent recognition model, specifically identifying the type, location, and size parameters of the structural defects, includes: The intelligent recognition model is a convolutional neural network model with an embedded attention mechanism. The model is trained using historical detection data labeled with defect type, location coordinates, and size parameters. The defect-associated feature vector is input into the trained intelligent recognition model, and the weights of the key defect features are strengthened through an attention mechanism. The model output layer outputs the type, three-dimensional coordinates, and size parameters of structural defects.

5. The non-destructive testing method for structural inspection of construction projects according to claim 1, characterized in that: Combining structural mechanics analysis models with historical inspection data, the prediction of the development trend of structural defects and the level of structural safety risk specifically includes: Construct a structural mechanics analysis model that matches the object being tested, input the defect parameters and structural design parameters, and calculate the influence coefficient of the defect on the structural bearing capacity. Historical inspection data of the same structure are retrieved to establish a defect development time series model and fit the change pattern of defect parameters over time. By combining the impact coefficient with the defect development time series model, the defect development status within a future preset period can be predicted. Based on the preset safety assessment standards, the structural safety risk level is classified.

6. The non-destructive testing method for structural inspection of construction projects according to claim 1, characterized in that: The output of structured detection results and risk warning reports specifically includes: The defect type, location, size parameters, development trend and risk level are integrated into structured inspection data; Defect location annotation maps, size quantification charts, and risk level heat maps are generated using visualization algorithms. Structured detection data and visual charts are integrated into risk warning reports, which support local storage, remote transmission to terminal devices, and printing output.

7. The non-destructive testing method for structural inspection of construction projects according to claim 1, characterized in that: The construction project structures include concrete structures, steel structures, masonry structures, and composite structures.

8. A non-destructive testing device for structural inspection in construction projects, characterized in that, include: The multimodal detection unit is used to collect structural physical property data and environmental correlation data of construction engineering structures to form multimodal detection data; The data processing unit is used to preprocess the multimodal detection data, fuse features, and extract structural defect-related features. The intelligent identification unit is used to analyze the associated features of the structural defects based on the intelligent identification model, and to identify the type, location and size parameters of the structural defects. The risk prediction unit is used to combine the structural mechanics analysis model with historical inspection data to predict the development trend of defects and the level of structural safety risk. The output unit is used to output structured detection results and risk warning reports.

9. A computer device comprising a memory and a processor, the memory storing code, characterized in that, The processor is configured to acquire the code and execute the non-destructive testing method for structural inspection of construction projects as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the non-destructive testing method for structural testing of construction projects as described in any one of claims 1 to 7.

Images