A nondestructive detection device and method for concrete reinforcement corrosion in complex environments

Abstract

The present invention relates to a device and method for nondestructive detection of concrete reinforcement corrosion in a complex environment. The device includes two drone devices, and drone No. 1 is equipped with WiFi signals, Bluetooth signals and 5G signal transmitting devices, and these signals are closely attached to the concrete surface and transmitted. Drone No. 2 is equipped with a wireless signal receiving device, which is responsible for receiving a variety of wireless signals after propagation through the concrete medium. Non-destructive detection is performed using the propagation characteristics of wireless signals in concrete. The device converts the measured signal attenuation data into decibel form, combines environmental data and concrete batching ratio data, removes the error value by Laida criterion, and uses Z-Score algorithm for normalization. For each signal, a gradient boosting decision tree prediction model is used to quantify the content of chloride, sulfate and calcium carbonate components in concrete; based on the Gaussian excitation function, the quantification results of multiple signals are integrated to finally generate the prediction results of concrete reinforcement corrosion.

Description

Technical Field

[0001] The present invention relates to the fields of civil engineering structure health monitoring, drone technology application and machine learning, and more particularly to a device and method for nondestructive detection of concrete reinforcement corrosion in a complex environment. Background Art

[0002] With the continuous development and extended service life of civil engineering structures, steel corrosion is one of the main reasons for the degradation of concrete structure performance. Traditional steel corrosion detection methods usually require destructive sampling or drilling, which is not only complicated and costly, but also causes secondary damage to the structure.

[0003] Wireless signal nondestructive testing technology is a non-contact, non-destructive testing method that uses wireless signals to extract information about the internal structure and performance of the object being tested by analyzing the transmission characteristics of wireless signals. Its core principle lies in the interaction between wireless signals and the object being tested. When wireless signals pass through or reflect from the object being tested, their transmission characteristics will change. These changes are closely related to the internal structure and material properties of the object being tested. By analyzing these changes, internal defects, damage or other performance problems of the object being tested can be inferred.

[0004] Gradient boosting decision tree is an optimized distributed gradient boosting library designed to implement efficient, flexible and portable machine learning algorithms. The Gradient Boosting framework solves many data science problems quickly and accurately through parallel tree boosting. Its basic components are decision trees, which together form a gradient boosting decision tree model as "weak learners". In the gradient boosting decision tree, the generation of the latter decision tree will take into account the prediction results of the previous decision tree, that is, the deviation of the previous decision tree is taken into account, so that the training samples that the previous decision tree made mistakes will receive more attention in the future. The inventor can realize the rapid and comprehensive detection of civil engineering structures by carrying different sensors and detection devices on drones, and proposes a non-destructive detection device and method for concrete steel corrosion based on the collaboration of two drones. Summary of the invention

[0005] In order to solve the destructive detection of traditional methods, the purpose of the present invention is to provide a nondestructive detection device and method for concrete steel bar corrosion in a complex environment to solve the above problems. A nondestructive detection device for concrete steel bar corrosion in a complex environment includes:

[0006] UAV No. 1, wherein the UAV No. 1 is equipped with a wireless signal transmitting device, and transmits WiFi signals, Bluetooth signals, and 5G signals by adjusting the transmission power and frequency so that the UAV closely adheres to the concrete surface;

[0007] UAV No. 2, wherein UAV No. 2 is equipped with a wireless signal receiving device that matches the wireless signal transmitting device of UAV No. 1, and the wireless signal receiving device is used to receive a variety of wireless signals after being propagated through the concrete medium, and extract signal strength, delay, and phase difference;

[0008] The flight control unit is used to control the flight trajectory and coordinated work of UAV 1 and UAV 2;

[0009] Positioning devices are installed on UAV 1 and UAV 2 to determine their relative positions and heights, ensuring accuracy and stability during the inspection process;

[0010] The signal conversion device converts the received wireless signal into decibel form for subsequent processing.

[0011] A nondestructive detection method for concrete reinforcement corrosion in a complex environment comprises the following steps:

[0012] Control UAV 1 and UAV 2 and guide them to fly to the target area where they need to stick close to the concrete surface.

[0013] Fine-tune the attitudes of UAV 1 and UAV 2 so that they can approach the concrete surface at a suitable angle and speed.

[0014] Control drone No. 1 to slowly descend until the wireless signal transmitter is close to the concrete surface.

[0015] When drone No. 1 is stably attached to the concrete surface, the wireless signal transmitter is turned on to transmit Bluetooth signals, 5G signals, and WIFI signals in sequence. At the same time, drone No. 2 starts to receive wireless signals to ensure good signal quality without interference or interruption.

[0016] The signal processing module of the No. 2 drone accurately converts the received wireless signal into decibel form for subsequent processing.

[0017] After the data processing module, the one-hot encoding classification features are converted into binary features, the Laida criterion is used to eliminate outliers in the data, and the Z-Score value of each data is calculated to replace the original data point to achieve normalization.

[0018] The gradient boosting decision tree model is used for training to select the decision tree as the basic model. The formula is:

[0019]

[0020] Use grid search to screen important hyperparameters, obtain important hyperparameter values ​​by searching and verifying all values ​​within the set value range, optimize important hyperparameters, and determine their optimal values.

[0021] The dataset is divided into input features (X) and target variables (y), where the target variable represents the predicted percentage of chloride, sulfate, and calcium carbonate.

[0022] The data set is divided into a training set and a test set in a ratio of 80:20. The residual is continuously reduced by iterating the error results of the previous decision tree, and a new decision tree is established in the direction of residual reduction to generate predictions for the chloride, sulfate, and calcium carbonate content in concrete. The objective function formula is:

[0023]

[0024] Where Obj is the objective function, y i and Represent the true value and the predicted value respectively, L is the difference between the predicted value and the true value, and Ω is the regularization function;

[0025] In order to prevent the gradient boosting decision tree model from overfitting the training data, resulting in a decrease in performance on the test data, a regular term that suppresses complexity is set, and its formula is:

[0026]

[0027] Where T is the number of leaf nodes in the tree, w j is the weight of the jth leaf node, which represents the contribution of the leaf node to the final prediction. γ and λ are regularization coefficients used to control the strength of regularization. γ controls the influence of the number of leaf nodes, while λ controls the influence of the leaf node weight.

[0028] Furthermore, the gradient boosting decision tree training model used uses a second-order Taylor expansion to approximate the objective function to optimize the objective function, and its formula is:

[0029]

[0030] Among them, G j and H j They respectively represent the sum of the first-order partial derivatives and second-order partial derivatives of the samples contained in the leaf node j.

[0031] The root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R 2 ) is used as an indicator to measure the performance of the gradient boosting decision tree model. RMSE is more sensitive to large errors, while MAE treats all errors equally and may be more stable in some cases. Considering the two error measurement methods comprehensively to avoid the misleading of a single indicator, the formula is:

[0032]

[0033] Among them, y i is the true value of the i-th sample, is the model's predicted value for the i-th sample, and n is the total number of samples.

[0034] The importance of the attribute is calculated by the amount by which each attribute split point improves the performance metric in a single decision tree, weighted and counted by the node, to quantify the content of chloride, sulfate, and calcium carbonate in concrete for each signal.

[0035] Considering the contents of chloride, sulfate, calcium carbonate and other components x1, x2, x3 in concrete generated based on Bluetooth signals, 5G signals and WiFi signals, a new input x is generated, and a Gaussian function is used as the excitation function, and its formula is:

[0036] x=γ1*x1+γ2*x2+γ3*x3

[0037]

[0038] Among them, G represents the output value of the Gaussian kernel function; ||xk|| is the Euclidean norm, x is the input, k is the center of the Gaussian function, and σ is the variance of the Gaussian function.

[0039] Based on the Gaussian function, the training model has the following objective function formula:

[0040]

[0041] Among them, w i is the connection weight.

[0042] Based on the output of the objective function, a prediction of concrete reinforcement corrosion is generated.

[0043] Compared with traditional methods, the nondestructive detection device and method for concrete steel bar corrosion based on the collaboration of two drones avoids the damage to concrete caused by extracting samples through wireless signal detection. At the same time, drones can quickly cover large areas and can operate on complex or inaccessible terrain, which can improve detection efficiency compared to manual methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is a flow chart of a nondestructive detection method for concrete reinforcement corrosion in a complex environment. DETAILED DESCRIPTION

[0045] Example 1

[0046] This embodiment describes a nondestructive detection device for concrete reinforcement corrosion in a complex environment, and the device includes:

[0047] Control UAV 1 and UAV 2 and guide them to fly to the target area where they need to stick close to the concrete surface.

[0048] Fine-tune the attitudes of UAV 1 and UAV 2 so that they can approach the concrete surface at a suitable angle and speed.

[0049] Control drone No. 1 to slowly descend until the wireless signal transmitter is close to the concrete surface.

[0050] When drone No. 1 is stably attached to the concrete surface, the wireless signal transmitter is turned on to transmit Bluetooth signals, 5G signals, and WIFI signals in sequence. At the same time, drone No. 2 starts to receive wireless signals to ensure good signal quality without interference or interruption.

[0051] The signal processing module of the No. 2 drone accurately converts the received wireless signal into decibel form for subsequent processing.

[0052] Guide the drone back to its take-off point or a preset landing area.

[0053] The present invention provides a nondestructive detection device for concrete reinforcement corrosion in a complex environment. The device performs nondestructive detection by using the propagation characteristics of different wireless signals in concrete through the collaborative work of two machines. The device is implemented in combination with the method of Example 2.

[0054] Example 2

[0055] like Figure 1 As shown, this embodiment describes a method for nondestructive detection of concrete reinforcement corrosion in a complex environment, and the method includes the following steps:

[0056] S1: Control UAV No. 1 to make it close to the concrete surface to be tested; operate UAV No. 2 to make it located on the other side of the concrete to be tested;

[0057] S2: Operate UAV 1 to transmit multiple wireless signals and ensure that UAV 2 can successfully receive these signals;

[0058] S3: Convert the WiFi signal, Bluetooth signal and 5G signal received by the second drone into decibel form;

[0059] S4: Collect environmental data, concrete data, and aggregate type data;

[0060] S5: For the collected data, the error values ​​are eliminated by using the Laida criterion, the missing values ​​are filled by the median filling method, and finally the Z-Score algorithm is used for normalization to eliminate the influence of different dimensions and units on the prediction results;

[0061] S6: Through the gradient boosting decision tree, learn and implement the quantitative analysis of multiple components of concrete, and generate the corresponding chloride, sulfate and calcium carbonate content data for WiFi signals, Bluetooth signals and 5G signals;

[0062] S7: Use weighted accumulation to aggregate inputs between different signals and use Gaussian activation;

[0063] S8: Generate a prediction of concrete reinforcement corrosion based on the Gaussian function activation results.

[0064] Specifically, data such as environmental factors and concrete composition as well as wireless signals in decibel form are extracted as input.

[0065] Use one-hot encoding to convert categorical features into binary features.

[0066] Calculate the mean (μ) and standard deviation (σ) of the input data.

[0067] For each data point, the deviation from the mean is calculated (Δ=data point-μ).

[0068] Data points with deviations exceeding 3σ (|Δ|>3σ) are considered error values ​​and are removed.

[0069] For each data point, calculate its Z-Score value, Z = (data point - μ) / σ, use the Z-Score value to replace the original data point to complete the normalization process.

[0070] The gradient boosting decision tree model is used for training to select the decision tree as the base model. For each data point x i The predicted value can be expressed as:

[0071]

[0072] Where F is the set of all possible decision trees, f i Represents the i-th decision tree.

[0073] Use grid search to screen important hyperparameters, obtain important hyperparameter values ​​by searching and verifying all values ​​within the set value range, optimize important hyperparameters, and determine their optimal values.

[0074] The dataset is divided into input features (X) and target variables (y), where the target variable represents the predicted percentage of chloride, sulfate, and calcium carbonate.

[0075] The gradient boosting decision tree training model with the optimal hyperparameter combination uses a ratio of 80:20 to divide the data set into a training set and a test set. By iterating the erroneous results of the previous decision tree, the residual is continuously reduced, and a new decision tree is established in the direction of residual reduction to generate predictions for the chloride, sulfate, and calcium carbonate content in concrete. The objective function formula is:

[0076]

[0077] in is the loss function for the gradient boosted decision tree model.

[0078] In order to prevent the gradient boosting decision tree model from overfitting the training data, resulting in a decrease in performance on the test data, a regular term that suppresses complexity is set, and its formula is:

[0079]

[0080] Where T is the number of leaf nodes in the tree, w j is the weight of the jth leaf node, which represents the contribution of the leaf node to the final prediction. γ and λ are regularization coefficients used to control the strength of regularization. γ controls the influence of the number of leaf nodes, while λ controls the influence of the leaf node weight.

[0081] The gradient boosting decision tree training model used uses a second-order Taylor expansion to approximate the objective function to optimize the objective function, and its formula is:

[0082]

[0083] Among them, G j and H j They respectively represent the sum of the first-order partial derivatives and second-order partial derivatives of the samples contained in the leaf node j.

[0084] The root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R 2 ) is used as an indicator to measure the performance of the gradient boosting decision tree model. RMSE is more sensitive to large errors, while MAE treats all errors equally and may be more stable in some cases. Considering the two error measurement methods comprehensively to avoid the misleading of a single indicator, the formula is:

[0085]

[0086] Among them, y i is the true value of the i-th sample, is the model's predicted value for the i-th sample, and n is the total number of samples.

[0087] The importance of the attribute is calculated by the amount by which each attribute split point improves the performance metric in a single decision tree, weighted and counted by the node, to quantify the content of chloride, sulfate, and calcium carbonate in concrete for each signal.

[0088] Considering the contents of chloride, sulfate, calcium carbonate and other components x1, x2, x3 in concrete generated based on Bluetooth signals, 5G signals and WiFi signals, a new input x is generated, and a Gaussian function is used as the excitation function. The formula is:

[0089] x=γ1*x1+γ2*x2+γ3*x3

[0090]

[0091] Among them, G represents the output value of the Gaussian kernel function; y represents the response value based on the input x and the center of the Gaussian function; x is the input, ||xk|| is the Euclidean norm, k is the center of the Gaussian function, and σ is the variance of the Gaussian function.

[0092] Based on the Gaussian function, the training model has the following objective function formula:

[0093]

[0094] Among them, w i is the connection weight.

[0095] Finally, based on the output of the objective function, a prediction of concrete reinforcement corrosion is generated.

[0096] The present invention converts the measured signal attenuation data into decibel form, and extracts environmental data and concrete mix ratio data at the same time. After removing the error value of the above-mentioned attenuation signal, environmental data, and concrete mix ratio data using the Laida criterion, the Z-Score algorithm is used for normalization. For each signal, a gradient boosting decision tree prediction model is used to quantify the content of components such as chloride, sulfate, and calcium carbonate in concrete. Based on the content of multiple components in concrete, the quantification results generated by multiple signals are integrated, and the Gaussian function is used as the excitation function to finally generate the prediction results of concrete steel bar corrosion. Compared with the traditional method, the non-destructive detection device and method for concrete steel bar corrosion based on the cooperation of two unmanned aerial vehicles avoids the damage to the concrete caused by the extracted samples through wireless signal detection. At the same time, the unmanned aerial vehicle can quickly cover a large area and can operate on complex or inaccessible terrain, which can improve the detection efficiency compared with the manual method.

[0097] The implementation modes of the present invention are described in detail above in conjunction with the embodiments, but the present invention is not limited to the above-mentioned implementation modes. For ordinary technicians in this technical field, after knowing the contents recorded in the present invention, they can make several equivalent changes and substitutions thereto without departing from the principle of the present invention, and these equivalent changes and substitutions should also be regarded as belonging to the protection scope of the present invention.

Claims

1. A nondestructive detection method for concrete reinforcement corrosion in a complex environment, characterized in that: The steps include: S1: Control UAV No. 1 to make it close to the concrete surface to be tested; operate UAV No. 2 to make it located on the other side of the concrete to be tested; S2: Operate UAV 1 to transmit multiple wireless signals and ensure that UAV 2 can successfully receive these signals; S3: Convert the WiFi signal, Bluetooth signal and 5G signal received by the second drone into decibel form; S4: Collect environmental data, concrete data, and aggregate type data; S5: For the collected data, the error values ​​are eliminated by using the Laida criterion, the missing values ​​are filled by the median filling method, and finally the Z-Score algorithm is used for normalization to eliminate the influence of different dimensions and units on the prediction results; S6: Through the gradient boosting decision tree, learn and implement the quantitative analysis of multiple components of concrete, and generate the corresponding chloride, sulfate and calcium carbonate content data for WiFi signals, Bluetooth signals and 5G signals; S7: Use weighted accumulation to aggregate inputs between different signals and use Gaussian activation; S8: Generate a prediction of concrete reinforcement corrosion based on the Gaussian function activation results; Step S5 includes the following steps: S5-1: The collected data are encoded by one-hot, the categorical features are converted into binary features, and the dataset is divided into input features and target variables. The target variables represent the predicted percentage of chloride, sulfate, and calcium carbonate content. S5-2: Use the Laida criterion to process the error values ​​in the data set, and remove the data that is less than the mean minus three times the standard deviation and the data that is greater than the mean plus three times the standard deviation; S5-3: Standardize the data processed in step S5-2, and use the Z-Score algorithm to convert data of different magnitudes into a unified Z value for comparison and analysis. The formula is as follows: ； Where X represents the original data, μ represents the mean, and σ represents the standard deviation; Step S6 includes the following steps: S6-1: The gradient boosting decision tree algorithm is used to predict the content of multiple components of concrete using the decibel and frequency attenuation of wireless signals after propagation in concrete, ambient temperature, humidity, exposure time, water-cement ratio, aggregate type and size, and admixture data as inputs; S6-2: The gradient boosting decision tree prediction model is as follows: ； The data set is n-dimensional and m-dimensional. is a set of decision tree structures, q is the tree structure of samples mapped to leaf nodes, and w is the real number score of leaf nodes; S6-3: The objective function consists of two parts: the loss function and the regularization term. The loss function is used to measure the difference between the model prediction value and the true value. The regularization term is used to control the complexity of the model. The formula is as follows: ； in is the objective function, and represent the true value and the predicted value respectively. is the difference between the predicted value and the true value, is the regularization function; S6-4: Use the second-order Taylor expansion to approximate the target function. For a small change in the independent variable x, the Taylor series expansion provides an accurate approximation. for exist The slope at for exist The slope at: ； S6-5: Using root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R 2 ) is used as an indicator to measure the performance of the gradient boosting decision tree prediction model.

2. A nondestructive detection method for concrete reinforcement corrosion in a complex environment according to claim 1, characterized in that: Step S8 uses a Gaussian function as the excitation function to ultimately generate a prediction result of concrete reinforcement corrosion: ； ； In the formula, G represents the output value of the Gaussian kernel function; y represents the response value based on the input x and the center of the Gaussian function; For input, is the Euclidean norm, is the center of the Gaussian function, is the Gaussian function variance, is the connection weight.

Images