An intelligent structural damage detection method and system

Abstract

The present invention relates to the field of damage identification technology, and discloses an intelligent structural damage detection method and system, the method comprising: acquiring multiple structural sensor monitoring data, and performing multi-dimensional fusion processing on the multiple structural sensor monitoring data to obtain monitoring fusion data; based on the environment in which the monitoring structure is located, adaptively adjusting the pre-set environmental condition indicators to obtain environmental feature extraction parameters; according to the monitoring fusion data and the environmental feature extraction parameters, using a pre-trained multi-stage deep learning model to perform structural damage identification to obtain a structural damage identification result; according to the structural damage identification result, using an incremental transfer learning network to evaluate the structural damage to obtain a structural damage evaluation result. The present invention can significantly improve the accuracy and efficiency of structural damage identification through multi-source data fusion, environmental adaptive processing, incremental transfer learning and comprehensive damage degree evaluation.

Description

Technical Field

[0001] The present invention relates to the technical field of damage identification, and in particular to an intelligent structural damage detection method and system. Background Art

[0002] In the field of modern engineering management and safety monitoring, timely and accurate identification and assessment of structural damage is crucial. With the development of technology, especially the advancement of Internet of Things and artificial intelligence technology, structural health monitoring systems are gradually moving towards intelligence and automation. However, existing damage identification technologies face many challenges, including how to effectively process and fuse data from different sensors, how to adapt to changing environmental conditions, and how to quickly and accurately identify and assess emerging damage types. These challenges have prompted researchers and engineers to seek more efficient and intelligent solutions to improve the accuracy and efficiency of structural health monitoring and ensure the safety of human life and property.

[0003] The existing patent CN202311581511.2 discloses a method for identifying damage of prefabricated structures based on Bayesian theory, including: conducting a seismic excitation test on the prefabricated structure, collecting acceleration response data through sensors; constructing a log-likelihood function of the acceleration response data based on modal force information; optimizing the log-likelihood function through the EM algorithm to obtain measured modal parameters; establishing an objective function based on the difference between the measured modal parameters and the finite element model parameters, the finite element model is a theoretical structural model for the design of the prefabricated structure; obtaining a modified finite element model that is closest to the measured modal parameters through continuous iteration of the MCMC algorithm; and locating the damage position and damage size by performing a loading simulation test on the modified finite element model through simulated moving loads. This patent improves computational efficiency, saves computational time, and can realize visualization of damage warning and real-time monitoring of prefabricated structures.

[0004] However, the patent has the following technical problems during its implementation: the existing technology is difficult to comprehensively capture damage characteristics from multiple dimensions, resulting in insufficient recognition accuracy and reliability, ignoring the impact of environmental conditions on damage recognition results, and lacking an effective environmental adaptation mechanism, resulting in large fluctuations in recognition results under different environmental conditions; the existing technology is difficult to adapt to new damage types or new data distributions, lacks the ability to continuously learn and optimize, and cannot comprehensively evaluate the degree, scope and specific location of the damage.

[0005] Therefore, how to provide a structural damage detection technology to achieve comprehensive structural identification and / or evaluation is an urgent problem to be solved. Summary of the invention

[0006] The embodiments of the present invention provide an intelligent structural damage detection method and system to solve the above-mentioned technical problems in the prior art.

[0007] In order to have a basic understanding of some aspects of the disclosed embodiments, a brief summary is given below. This summary is not a general review, nor is it intended to identify key / important components or to describe the scope of protection of these embodiments. Its only purpose is to present some concepts in a simple form as a prelude to the detailed description that follows.

[0008] According to a first aspect of an embodiment of the present invention, an intelligent structural damage detection method is provided.

[0009] In one embodiment, the intelligent structural damage detection method includes:

[0010] Acquire multiple structural sensor monitoring data, and perform multi-dimensional fusion processing on the multiple structural sensor monitoring data to obtain monitoring fusion data;

[0011] Based on the environment where the monitoring structure is located, the pre-set environmental condition indicators are adaptively adjusted to obtain environmental feature extraction parameters;

[0012] According to the monitoring fusion data and the environmental feature extraction parameters, a pre-trained multi-stage deep learning model is used to perform structural damage identification to obtain a structural damage identification result.

[0013] In one embodiment, the structure sensor includes: a spectral imaging sensor, an infrared thermal imager, and a vibration sensor.

[0014] In one embodiment, the intelligent structural damage detection method further includes: preprocessing the structural sensor monitoring data before performing multi-dimensional fusion processing on the multiple structural sensor monitoring data; wherein the preprocessing includes: denoising processing, normalization processing and / or data format and size standardization processing.

[0015] In one embodiment, the data fusion formula when performing multi-dimensional fusion processing on multiple structural sensor monitoring data is: In the formula, represents monitoring fusion data, is a multidimensional data fusion function, It is a weight parameter pre-set according to the importance of each structural sensor monitoring data. It is for i A preprocessing function for structured sensor data applications; Monitoring data for structure sensors.

[0016] In one embodiment, the calculation formula of the environmental condition index is: In the formula, Indicates environmental condition indicators, Represents the ambient light intensity, represents the background noise level, is the sensor temperature, and Used to adjust the contribution of light intensity, background noise level, and temperature to the environmental condition index, is the exponential influence coefficient of light intensity;

[0017] The calculation formula for adaptive adjustment of the environmental condition index is: In the formula, represents the environmental feature extraction parameters dynamically adjusted based on environmental condition indicators, , , , They are parameters set according to experimental data to adjust the response sensitivity to changes in environmental conditions during the feature extraction process.

[0018] In one embodiment, the multi-stage deep learning model has three stages, namely, an environment perception adjustment stage, a deep convolutional neural network stage, and a residual network fusion stage; wherein,

[0019] The environmental perception adjustment stage uses environmental feature extraction parameters to enhance the monitoring fusion data. The calculation formula is: In the formula, It represents the output of the environmental perception adjustment stage, that is, the enhanced data; represents the normalization factor, which is used to ensure The scope of the function's internal parameters; represents monitoring fusion data; represents the environmental condition index; e represents the natural base; represents pi;

[0020] The deep convolutional neural network stage includes multiple convolutional layers, each of which uses a set of filters to extract the features output by the environment perception adjustment stage. The calculation formula of the deep convolutional neural network stage is:

[0021]

[0022] In the formula, Represents the output features of the last convolutional layer; represents the output of the deep convolutional neural network stage; Representing characteristics of outputs of the environment perception adjustment stage; represents a filter; Represents the convolution operation; Indicates bias; represents the scaling parameter; represents the regularization coefficient; represents a linear rectification function;

[0023] The residual network fusion stage adopts a residual network to fuse the outputs of the environmental perception adjustment stage and the deep convolutional neural network stage, and performs deep learning processing through convolutional layers, activation layers, pooling layers and fully connected layers, and obtains the final damage identification result through mixed activation functions and sparsity processing; the calculation formula of the residual network fusion stage is: In the formula, represents the output of the residual network fusion stage, that is, the damage identification result; Represents the SoftMax activation function; Indicates the splicing symbol; represents the parameter that adjusts the strength of the linear relationship between the deep convolutional neural network output and the damage identification result; Represents a parameter that adjusts the strength of the impact of sparsity handling in the output of a deep convolutional neural network.

[0024] In one embodiment, the intelligent structural damage detection method further includes: evaluating the structural damage using an incremental transfer learning network according to the structural damage identification result to obtain a structural damage evaluation result.

[0025] In one embodiment, the incremental transfer learning network includes: an incremental learning preprocessing layer, an adaptive transfer learning layer, an incremental learning update layer and a fully connected layer; wherein,

[0026] The incremental learning preprocessing layer is used to perform nonlinear transformation on the structural damage identification result and adjust the scale and distribution of the data. The formula of the nonlinear transformation is:

[0027] In the formula, represents the preprocessed data, that is, the output of the incremental learning preprocessing layer, Represents the activation function, which increases the nonlinearity of the processed data; to represents the tuning parameters, which are optimized through experiments and are used to control the specific behavior and intensity of data transformation; represents the output of the residual network fusion stage; e represents the natural base;

[0028] The adaptive transfer learning layer is used to perform transfer learning optimization on the output of the incremental learning preprocessing layer using a preset adaptive weight parameter. The transfer learning optimization formula is: In the formula, Represents the output of the adaptive transfer learning layer, which represents the data features adjusted and optimized by transfer learning. The data features are the output of the incremental learning preprocessing layer. Refined and optimized; Represents the optimization function of the entire transfer learning process; represents the adjustment coefficient of the nth layer; n represents the layer index in the transfer learning layer, Represents the processing function of the nth layer, the output of the incremental learning preprocessing layer Perform transformation processing; represents the adaptive weight parameter; represents the Hadamard product, which is used for element-wise weighting of eigenvectors;

[0029] The incremental learning update layer is used to update the output of the adaptive transfer learning layer, and the update formula is:

[0030] In the formula, represents the updated incremental transfer learning network model parameters, Represents the model parameters before updating; is the learning rate, which controls the size of the update step; Represents the loss function The gradient of model parameters directly guides the update direction of model parameters; is the target output; is the introduced sinusoidal adjustment factor;

[0031] The fully connected layer is used to generate a structural damage assessment result according to the updated incremental transfer learning network model parameters; the calculation formula of the fully connected layer is:

[0032] In the formula, Represents the assessment output of the damage degree, is the activation function; and They are the weight and bias of the fully connected layer respectively; It indicates that the updated data features are obtained by forward propagating the updated incremental transfer learning network model parameters through the incremental transfer learning network.

[0033] According to a second aspect of an embodiment of the present invention, an intelligent structural damage detection system is provided.

[0034] In one embodiment, the intelligent structural damage detection system comprises:

[0035] A data acquisition and processing module is used to obtain monitoring data of multiple structural sensors and perform multi-dimensional fusion processing on the monitoring data of multiple structural sensors to obtain monitoring fusion data;

[0036] An environmental feature extraction module is used to adaptively adjust the pre-set environmental condition indicators based on the environment in which the monitoring structure is located to obtain environmental feature extraction parameters;

[0037] The structural damage identification module is used to extract parameters according to the monitoring fusion data and the environmental characteristics, and to perform structural damage identification using a pre-trained multi-stage deep learning model to obtain a structural damage identification result.

[0038] In one embodiment, the structure sensor includes: a spectral imaging sensor, an infrared thermal imager, and a vibration sensor.

[0039] In one embodiment, the data acquisition and processing module preprocesses the structure sensor monitoring data before performing multi-dimensional fusion processing on the multiple structure sensor monitoring data; wherein the preprocessing includes: denoising processing, normalization processing and / or data format and size standardization processing.

[0040] In one embodiment, the data fusion formula of the data acquisition and processing module when performing multi-dimensional fusion processing on the monitoring data of multiple structural sensors is: In the formula, represents monitoring fusion data, is a multidimensional data fusion function, It is a weight parameter pre-set according to the importance of each structural sensor monitoring data. It is for A preprocessing function for structured sensor data applications; Monitoring data for structure sensors.

[0041] In one embodiment, the calculation formula of the environmental condition index is: In the formula, Indicates environmental condition indicators, Represents the ambient light intensity, represents the background noise level, is the sensor temperature, and Used to adjust the contribution of light intensity, background noise level, and temperature to the environmental condition index, is the exponential influence coefficient of light intensity;

[0042] The calculation formula for adaptive adjustment of the environmental condition index is: In the formula, represents the environmental feature extraction parameters dynamically adjusted based on environmental condition indicators, , , , They are parameters set according to experimental data to adjust the response sensitivity to changes in environmental conditions during the feature extraction process.

[0043] In one embodiment, the multi-stage deep learning model has three stages, namely, an environment perception adjustment stage, a deep convolutional neural network stage, and a residual network fusion stage; wherein,

[0044] The environmental perception adjustment stage uses environmental feature extraction parameters to enhance the monitoring fusion data. The calculation formula is: In the formula, It represents the output of the environmental perception adjustment stage, that is, the enhanced data; represents the normalization factor, which is used to ensure The scope of the function's internal parameters; represents monitoring fusion data; represents the environmental condition index; e represents the natural base; represents pi;

[0045] The deep convolutional neural network stage includes multiple convolutional layers, each of which uses a set of filters to extract the features output by the environment perception adjustment stage. The calculation formula of the deep convolutional neural network stage is:

[0046] In the formula, Represents the output features of the last convolutional layer; represents the output of the deep convolutional neural network stage; Representing characteristics of outputs of the environment perception adjustment stage; represents a filter; Represents the convolution operation; Indicates bias; represents the scaling parameter; represents the regularization coefficient; represents a linear rectification function;

[0047] The residual network fusion stage adopts a residual network to fuse the outputs of the environmental perception adjustment stage and the deep convolutional neural network stage, and performs deep learning processing through convolutional layers, activation layers, pooling layers and fully connected layers, and obtains the final damage identification result through mixed activation functions and sparsity processing; the calculation formula of the residual network fusion stage is: In the formula, represents the output of the residual network fusion stage, that is, the damage identification result; Represents the SoftMax activation function; Indicates the splicing symbol; represents the parameter that adjusts the strength of the linear relationship between the deep convolutional neural network output and the damage identification result; Represents a parameter that adjusts the strength of the impact of sparsity handling in the output of a deep convolutional neural network.

[0048] In one embodiment, the intelligent structural damage detection system further includes: a structural damage assessment module, which is used to evaluate the structural damage based on the structural damage identification result using an incremental transfer learning network to obtain a structural damage assessment result.

[0049] In one embodiment, the incremental transfer learning network includes: an incremental learning preprocessing layer, an adaptive transfer learning layer, an incremental learning update layer and a fully connected layer; wherein,

[0050] The incremental learning preprocessing layer is used to perform nonlinear transformation on the structural damage identification result and adjust the scale and distribution of the data. The formula of the nonlinear transformation is:

[0051] In the formula, represents the preprocessed data, that is, the output of the incremental learning preprocessing layer, Represents the activation function, which increases the nonlinearity of the processed data; to represents the tuning parameters, which are optimized through experiments and are used to control the specific behavior and intensity of data transformation; represents the output of the residual network fusion stage; e represents the natural base;

[0052] The adaptive transfer learning layer is used to perform transfer learning optimization on the output of the incremental learning preprocessing layer using a preset adaptive weight parameter. The transfer learning optimization formula is: In the formula, Represents the output of the adaptive transfer learning layer, which represents the data features adjusted and optimized by transfer learning. The data features are the output of the incremental learning preprocessing layer. Refined and optimized; Represents the optimization function of the entire transfer learning process; represents the adjustment coefficient of the nth layer; n represents the layer index in the transfer learning layer, Represents the processing function of the nth layer, the output of the incremental learning preprocessing layer Perform transformation processing; represents the adaptive weight parameter; represents the Hadamard product, which is used for element-wise weighting of eigenvectors;

[0053] The incremental learning update layer is used to update the output of the adaptive transfer learning layer, and the update formula is: In the formula, represents the updated incremental transfer learning network model parameters, Represents the model parameters before updating; is the learning rate, which controls the size of the update step; Represents the loss function The gradient of model parameters directly guides the update direction of model parameters; is the target output; is the introduced sinusoidal adjustment factor;

[0054] The fully connected layer is used to generate a structural damage assessment result according to the updated incremental transfer learning network model parameters; the calculation formula of the fully connected layer is: In the formula, Represents the assessment output of the damage degree, is the activation function; and They are the weight and bias of the fully connected layer respectively; It indicates that the updated data features are obtained by forward propagating the updated incremental transfer learning network model parameters through the incremental transfer learning network.

[0055] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0056] 1. By combining data collected by multiple sensor systems (such as spectral imaging sensors, infrared thermal imagers, and vibration sensors), the present invention can capture characteristic information of structural damage from different angles and dimensions, greatly improving the accuracy and reliability of damage identification. By introducing environmental condition indicators and adaptively adjusting feature extraction parameters based on these indicators, the present invention can effectively cope with the variability of environmental conditions, such as light intensity, background noise level, and sensor temperature, ensuring that the damage identification process remains efficient and accurate under different environmental conditions.

[0057] 2. By adopting an incremental transfer learning network, the present invention can not only quickly adapt to new types of damage identification, but also continuously optimize and update existing identification models, improve the recognition accuracy of known damage types, and enable the model to have the ability of continuous learning and evolution, and be more adaptable; by further processing the output of the deep learning model through a fully connected layer, the present invention can comprehensively evaluate multiple dimensions such as the severity, scope and location of the damage, providing a scientific basis for the repair and maintenance of the structure; at the same time, different damage levels are divided by preset thresholds, making the evaluation results more intuitive and easy to understand.

[0058] 3. The present invention can effectively solve the problems existing in the prior art, such as the difficulty in comprehensively capturing damage features from multiple dimensions, resulting in insufficient recognition accuracy and reliability, ignoring the impact of environmental conditions on damage recognition, lacking an effective environmental adaptive mechanism, causing large fluctuations in recognition effects under different environmental conditions and difficulty in adapting to new damage types or new data distributions, lacking the ability to continuously learn and optimize, and being unable to comprehensively evaluate the extent, scope, and specific location of damage. The present invention can significantly improve the accuracy and efficiency of structural damage recognition through multi-source data fusion, environmental adaptive processing, incremental transfer learning, and comprehensive damage degree evaluation.

[0059] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0061] Figure 1 is a flow chart of an intelligent structural damage detection method according to an exemplary embodiment;

[0062] Figure 2 is a structural block diagram of an intelligent structural damage detection system according to an exemplary embodiment;

[0063] Figure 3 The figure is a schematic diagram showing the structure of a computer device according to an exemplary embodiment. DETAILED DESCRIPTION

[0064] Figure 1 An embodiment of an intelligent structural damage detection method of the present invention is shown.

[0065] In this optional embodiment, the intelligent structural damage detection method includes:

[0066] Step S101, acquiring multiple structural sensor monitoring data, and performing multi-dimensional fusion processing on the multiple structural sensor monitoring data to obtain monitoring fusion data;

[0067] Step S103, based on the environment in which the monitoring structure is located, adaptively adjust the preset environmental condition indicators to obtain environmental feature extraction parameters;

[0068] Step S105, extracting parameters based on the monitoring fusion data and the environmental characteristics, and using a pre-trained multi-stage deep learning model to perform structural damage identification to obtain a structural damage identification result;

[0069] Step S107, based on the structural damage identification result, using an incremental transfer learning network, assessing the structural damage to obtain a structural damage assessment result.

[0070] Figure 2 An embodiment of an intelligent structural damage detection system of the present invention is shown.

[0071] In this optional embodiment, the intelligent structural damage detection system includes:

[0072] The data acquisition and processing module 201 is used to obtain multiple structural sensor monitoring data and perform multi-dimensional fusion processing on the multiple structural sensor monitoring data to obtain monitoring fusion data;

[0073] The environmental feature extraction module 203 is used to adaptively adjust the preset environmental condition index based on the environment in which the monitoring structure is located to obtain environmental feature extraction parameters;

[0074] The structural damage identification module 205 is used to extract parameters according to the monitoring fusion data and the environmental characteristics, and to perform structural damage identification using a pre-trained multi-stage deep learning model to obtain a structural damage identification result;

[0075] The structural damage assessment module 207 is used to assess the structural damage according to the structural damage identification result by using an incremental transfer learning network to obtain a structural damage assessment result.

[0076] In order to facilitate understanding of the above technical solutions of the present invention, the above technical solutions of the present invention are described in detail from the perspective of principle and idea as follows.

[0077] In specific applications, the present invention captures characteristic information of structural damage from different angles and dimensions by combining data collected by multiple sensor systems (such as spectral imaging sensors, infrared thermal imagers, and vibration sensors), thereby greatly improving the accuracy and reliability of damage identification. The introduction of environmental condition indicators and adaptive adjustment of feature extraction parameters based on this indicator can effectively cope with the variability of environmental conditions, such as light intensity, background noise level, and sensor temperature, to ensure that the damage identification process can remain efficient and accurate under different environmental conditions. By adopting an incremental transfer learning network, it can not only quickly adapt to new types of damage identification, but also continuously optimize and update existing recognition models to improve the recognition accuracy of known damage types, so that the model has the ability of continuous learning and evolution and is more adaptable. By further processing the output of the deep learning model through a fully connected layer, it can comprehensively evaluate multiple dimensions such as the severity, scope, and location of the damage, providing a scientific basis for the repair and maintenance of the structure. At the same time, different damage levels are divided by preset thresholds, making the evaluation results more intuitive and easy to understand.

[0078] In practical application, the intelligent structural damage detection process of the present invention can be as follows:

[0079] S1. Collect data, pre-process and multi-dimensionally fuse the data, introduce environmental condition indicators, adaptively adjust feature extraction parameters, and use a multi-stage deep learning model for damage identification;

[0080] Data is collected from multiple sensor systems, including spectral imaging sensors, infrared thermal imagers, and vibration sensors, which are fixed to key structures to monitor the health of the structure in real time. The collected data includes spectral imaging, infrared thermal imaging, vibration data, etc. Each sensor data is represented as, where the subscript represents a different sensor type, providing a unique perspective and information for damage detection.

[0081] In order to unify the format and scale of the collected multi-source data, each type of collected data is preprocessed, and the preprocessing includes steps such as denoising and normalization to standardize the data format and size for subsequent processing. The preprocessing method can use existing technology, which is relatively mature.

[0082] The information provided by different sensors is integrated to obtain a comprehensive description of the structural damage, and the pre-processed data is integrated into a unified data representation using multi-dimensional data fusion technology. The formula for data fusion is: in, represents the fused data, is a multidimensional data fusion function, It is a weight parameter pre-set according to the importance of each sensor data, which ensures the reasonable weight distribution of different source data in the data fusion process. It is for i Preprocessing functions for various sensor data applications.

[0083] In view of the variability of environmental conditions, an environmental condition index is introduced, and the calculation formula of the environmental condition index is: in, Indicates environmental condition indicators, Represents the ambient light intensity, represents the background noise level, is the sensor temperature, and Used to adjust the contribution of light intensity, background noise level, and temperature to the environmental condition index, is the exponential influence coefficient of light intensity.

[0084] Based on the environmental condition index, the feature extraction parameters are adaptively adjusted to respond to the current environment in the best way. The adjustment formula of the feature extraction parameters is: middle, Represents the feature extraction parameters dynamically adjusted based on environmental condition indicators, which are used to adjust the feature extraction strategy to adapt to the current environmental conditions. , , , The response sensitivity to changes in environmental conditions during the feature extraction process is adjusted according to the parameters set according to the experimental data.

[0085] The fused data And feature extraction parameters adjusted according to environmental condition indicators The data are input into the deep learning model, and a multi-stage deep learning model design is used to further improve the accuracy of damage identification.

[0086] The first stage of the multi-stage deep learning model is the environment perception adjustment layer, which adjusts the processing strategy of the subsequent network layer according to the feature extraction parameters. The environment perception adjustment layer uses the feature extraction parameters to enhance the fused data, including dynamically adjusting feature scaling, applying different activation function parameters, etc. The implementation process of the environment perception adjustment layer is as follows: in, The output of the first stage, i.e., the enhanced data, will be input into the subsequent stages of the deep learning model for further damage identification and classification. is a normalization factor to ensure The range of the function's internal parameters is appropriate to accommodate different intensities of environmental conditions.

[0087] The second stage of the deep learning model is a deep convolutional neural network, which contains multiple convolutional layers. Each convolutional layer uses a set of filters to extract the features of the output of the first stage. Input to the first convolutional layer, the convolution operation uses filters and bias , output features Each convolutional layer is followed by an activation layer, which uses the ReLU function to increase the nonlinear ability of the deep convolutional neural network. After the convolutional layer and the activation layer, the pooling layer is used to reduce the spatial dimension of the features, reduce the amount of calculation, and retain important feature information. The first pooling layer passes the output to the second convolutional layer until the last pooling layer outputs the result, which is passed to the fully connected layer to map the deep features to the final output. , the second stage formula is expressed as:

[0088] in, is the output feature of the last convolutional layer, is the output of the second stage, i.e. The data features extracted from is the filter, is the convolution operation, is the bias, is the scaling parameter, is the regularization coefficient.

[0089] The third stage further processes the outputs of the first and second stages, further deepens the learning of damage features, and finally outputs damage identification results. The third stage uses a residual network to fuse the outputs of the first and second stages, and then performs deep learning processing through convolutional layers, activation layers, pooling layers, and fully connected layers. The final damage identification results are obtained through mixed activation functions and sparsity processing. The implementation formula of the third stage is: in, is the output of the third stage, i.e., the damage identification result, is the SoftMax activation function, which converts the output of the third stage into a probability distribution, where each element represents the probability of a specific damage type. It is a splicing symbol. is a parameter that adjusts the strength of the linear relationship between the output of the deep convolutional neural network model and the damage identification result. It is a parameter that adjusts the strength of the sparsity processing part in the output of the deep convolutional neural network model. The damage identification results are presented in the form of probability distribution, and the probability of each category represents the possibility of the existence of damage in that category. For example, to identify multiple types of structural damage designs (such as cracks, corrosion, peeling, etc.), each damage type will have a corresponding probability value.

[0090] S2. Based on the damage identification results, an incremental transfer learning network is proposed to improve the recognition accuracy of known damage types and realize damage assessment.

[0091] According to the damage identification results of the deep learning model, an incremental transfer learning network is proposed, including an incremental learning preprocessing layer, an adaptive transfer learning layer, an incremental learning update layer, and a fully connected layer, aiming to deepen the accuracy of damage identification and quickly adapt to the identification of new types of damage. The incremental transfer learning network combines the strategies of transfer learning and incremental learning to achieve continuous optimization and updating of existing recognition models to adapt to new types of damage, improve the recognition accuracy of known damage types, and realize damage assessment.

[0092] In the incremental learning preprocessing layer, considering the complexity of the data fusion collected from multiple sensor systems, the damage identification results need to be reprocessed to make them more suitable as the input of the incremental transfer learning network model. By comprehensively considering the feature distribution and potential information content of the data, a nonlinear transformation is designed to adjust the scale and distribution of the data, thereby enhancing the sensitivity of the incremental transfer learning network model to the features. The nonlinear transformation formula is: in, represents the preprocessed data, that is, the output of the incremental learning preprocessing layer, It is an activation function that increases the nonlinearity of the processed data; to To adjust the parameters, experimental optimization is performed to control the specific behavior and intensity of the data transformation.

[0093] The adaptive transfer learning layer performs in-depth analysis on preprocessed data to achieve rapid adaptation to new damage types and accurate identification of known damage types. A composite weight adjustment and dynamic linking between layers are used to dynamically adjust the weight distribution during the learning process through the following formula to optimize the overall performance of the model: in, Represents the output of the adaptive transfer learning layer, which represents the data features adjusted and optimized by transfer learning. The data features are obtained from the damage identification results after preprocessing. It is extracted and optimized from the original data to better capture the characteristics of damage and improve the accuracy and efficiency of damage identification; Represents the optimization function of the entire transfer learning process, which is responsible for optimizing the parameters in the adjustment process; is the adjustment coefficient of the nth layer, ensuring that the contribution of each layer is properly evaluated and utilized. n is the layer index in the transfer learning layer. is the processing function of the nth layer, which processes the input data Perform transformation processing; It is an adaptive weight parameter, which adjusts the weight of the input data through a logarithmic function, improving the model's ability to automatically identify the importance of features. is the Hadamard product (element-wise multiplication) for element-wise weighting of the eigenvectors.

[0094] In the incremental learning update layer, the following formula is used to achieve rapid learning of new data and effective memory of old knowledge, ensuring that the incremental transfer learning network model can quickly adapt to new situations in the process of continuous learning without losing the grasp of the learned content: in, represents the updated incremental transfer learning network model parameters, represents the model parameters before updating, is the learning rate, which controls the size of the update step, Represents the loss function The gradient of model parameters directly guides the update direction of model parameters; is the target output; It is a sinusoidal adjustment factor introduced to achieve fine-tuning of model parameters through periodic adjustment, so as to balance the learning and memory of new and old knowledge.

[0095] Use the updated model parameters for the forward propagation of the incremental transfer learning network to obtain updated data features , the updated data features are input into the fully connected layer to generate the final damage assessment result. The calculation formula of the fully connected layer is: in, Represents the assessment output of the damage degree, is the activation function, and are the weights and biases of the fully connected layer respectively.

[0096] The assessment of the degree of damage comprehensively reflects information on multiple dimensions such as the severity, scope, and location of the damage. Multiple thresholds are preset according to actual needs, and the damage assessment is divided into different intervals, each of which corresponds to a damage level.

[0097] Figure 3 An embodiment of a computer device of the present invention is shown, which may be a server, and includes a processor, a memory and a network interface connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store static information and dynamic information data. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the steps in the above method embodiment are implemented.

[0098] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0099] In addition, the present invention also provides a computer device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiment when executing the computer program.

[0100] In addition, the present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method embodiment are implemented.

[0101] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0102] The present invention is not limited to the structures which have been described above and shown in the drawings, and various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. An intelligent structural damage detection method, characterized in that: include: Acquire multiple structural sensor monitoring data, and perform multi-dimensional fusion processing on the multiple structural sensor monitoring data to obtain monitoring fusion data; Based on the environment where the monitoring structure is located, the pre-set environmental condition indicators are adaptively adjusted to obtain environmental feature extraction parameters; Extracting parameters based on the monitoring fusion data and the environmental features, using a pre-trained multi-stage deep learning model to perform structural damage identification to obtain a structural damage identification result; The multi-stage deep learning model has three stages, namely, the environment perception adjustment stage, the deep convolutional neural network stage and the residual network fusion stage; wherein, The environmental perception adjustment stage uses environmental feature extraction parameters to enhance the monitoring fusion data. The calculation formula is: ； In the formula, represents the output of the environmental perception adjustment stage, i.e., the enhanced data; represents the normalization factor, which is used to ensure The scope of the function's internal parameters; represents monitoring fusion data; Indicates environmental condition index; e indicates natural base; indicates pi ; The deep convolutional neural network stage includes multiple convolutional layers, each of which uses a set of filters to extract the features output by the environment perception adjustment stage. The calculation formula of the deep convolutional neural network stage is: ； ； In the formula, Represents the output features of the last convolutional layer; represents the output of the deep convolutional neural network stage; Representing characteristics of outputs of the environment perception adjustment stage; represents a filter; Represents the convolution operation; Indicates bias; represents the scaling parameter; represents the regularization coefficient; represents the linear rectification function; The residual network fusion stage adopts a residual network to fuse the outputs of the environmental perception adjustment stage and the deep convolutional neural network stage, and performs deep learning processing through convolutional layers, activation layers, pooling layers and fully connected layers, and obtains the final damage identification result through mixed activation functions and sparsity processing; the calculation formula of the residual network fusion stage is: ； In the formula, represents the output of the residual network fusion stage, that is, the damage identification result; Represents the SoftMax activation function; Indicates the splicing symbol; represents the parameter that adjusts the strength of the linear relationship between the deep convolutional neural network output and the damage identification result; Represents a parameter that adjusts the strength of the impact of sparsity handling in the output of a deep convolutional neural network.

2. The intelligent structural damage detection method according to claim 1, characterized in that: Structural sensors include: spectral imaging sensors, infrared thermal imagers and vibration sensors.

3. The intelligent structural damage detection method according to claim 1, characterized in that: Also includes: Before performing multi-dimensional fusion processing on the monitoring data of multiple structural sensors, the monitoring data of the structural sensors are preprocessed; The preprocessing includes: noise removal, normalization and / or data format and size standardization.

4. The intelligent structural damage detection method according to claim 1, characterized in that: The data fusion formula for multi-dimensional fusion processing of multiple structural sensor monitoring data is: ； In the formula, represents monitoring fusion data, is a multidimensional data fusion function, It is a weight parameter pre-set according to the importance of each structural sensor monitoring data. It is for i A preprocessing function for structured sensor data applications; Monitoring data for structure sensors.

5. The intelligent structural damage detection method according to claim 1, characterized in that: The calculation formula of the environmental condition index is: ； In the formula, Indicates environmental condition indicators, Represents the ambient light intensity, represents the background noise level, is the sensor temperature, and Used to adjust the contribution of light intensity, background noise level, and temperature to the environmental condition index, is the exponential influence coefficient of light intensity; The calculation formula for adaptive adjustment of the environmental condition index is: ； In the formula, represents the environmental feature extraction parameters dynamically adjusted based on environmental condition indicators, , , , They are parameters set according to experimental data to adjust the response sensitivity to changes in environmental conditions during the feature extraction process.

6. The intelligent structural damage detection method according to claim 1, characterized in that: Also includes: According to the structural damage identification result, the structural damage is evaluated using an incremental transfer learning network to obtain a structural damage evaluation result.

7. The intelligent structural damage detection method according to claim 6, characterized in that: The incremental transfer learning network includes: an incremental learning preprocessing layer, an adaptive transfer learning layer, an incremental learning update layer and a fully connected layer; wherein, The incremental learning preprocessing layer is used to perform nonlinear transformation on the structural damage identification result and adjust the scale and distribution of the data. The formula of the nonlinear transformation is: ； In the formula, represents the preprocessed data, that is, the output of the incremental learning preprocessing layer, Represents the activation function, which increases the nonlinearity of the processed data; to represents the tuning parameters, which are optimized through experiments and used to control the specific behavior and intensity of data transformation; represents the output of the residual network fusion stage; e represents the natural base; The adaptive transfer learning layer is used to perform transfer learning optimization on the output of the incremental learning preprocessing layer using a preset adaptive weight parameter. The transfer learning optimization formula is: ； In the formula, Represents the output of the adaptive transfer learning layer, which represents the data features adjusted and optimized by transfer learning. The data features are the output of the incremental learning preprocessing layer. Refined and optimized; Represents the optimization function of the entire transfer learning process; Indicates The adjustment coefficient of the layer; represents the layer index in the transfer learning layer, Indicates The processing function of the layer is the output of the incremental learning preprocessing layer Perform transformation processing; represents the adaptive weight parameter; represents the Hadamard product, which is used for element-wise weighting of eigenvectors; The incremental learning update layer is used to update the output of the adaptive transfer learning layer, and the update formula is: ； In the formula, represents the updated incremental transfer learning network model parameters, Represents the model parameters before updating; is the learning rate, which controls the size of the update step; Represents the loss function The gradient of model parameters directly guides the update direction of model parameters; is the target output; is the introduced sinusoidal adjustment factor; The fully connected layer is used to generate a structural damage assessment result according to the updated incremental transfer learning network model parameters; the calculation formula of the fully connected layer is: ； In the formula, Represents the assessment output of the damage degree, is the activation function; and They are the weights and biases of the fully connected layer respectively; It indicates that the updated data features are obtained by forward propagating the updated incremental transfer learning network model parameters through the incremental transfer learning network.

8. An intelligent structural damage detection system, characterized in that: include: A data acquisition and processing module is used to obtain monitoring data of multiple structural sensors and perform multi-dimensional fusion processing on the monitoring data of multiple structural sensors to obtain monitoring fusion data; An environmental feature extraction module is used to adaptively adjust the pre-set environmental condition indicators based on the environment in which the monitoring structure is located to obtain environmental feature extraction parameters; A structural damage identification module is used to extract parameters according to the monitoring fusion data and the environmental characteristics, and to perform structural damage identification using a pre-trained multi-stage deep learning model to obtain a structural damage identification result; The multi-stage deep learning model has three stages, namely, the environment perception adjustment stage, the deep convolutional neural network stage and the residual network fusion stage; wherein, The environmental perception adjustment stage uses environmental feature extraction parameters to enhance the monitoring fusion data. The calculation formula is: ； In the formula, represents the output of the environmental perception adjustment stage, i.e., the enhanced data; represents the normalization factor, which is used to ensure The scope of the function's internal parameters; represents monitoring fusion data; represents the environmental condition index; e represents the natural base; represents pi; The deep convolutional neural network stage includes multiple convolutional layers, each of which uses a set of filters to extract the features output by the environment perception adjustment stage. The calculation formula of the deep convolutional neural network stage is: ； ； In the formula, Represents the output features of the last convolutional layer; represents the output of the deep convolutional neural network stage; Representing characteristics of outputs of the environment perception adjustment stage; represents a filter; Represents the convolution operation; Indicates bias; represents the scaling parameter; represents the regularization coefficient; represents the linear rectification function; The residual network fusion stage adopts a residual network to fuse the outputs of the environmental perception adjustment stage and the deep convolutional neural network stage, and performs deep learning processing through convolutional layers, activation layers, pooling layers and fully connected layers, and obtains the final damage identification result through mixed activation functions and sparsity processing; the calculation formula of the residual network fusion stage is: ； In the formula, represents the output of the residual network fusion stage, that is, the damage identification result; Represents the SoftMax activation function; Indicates the splicing symbol; represents the parameter that adjusts the strength of the linear relationship between the deep convolutional neural network output and the damage identification result; Represents a parameter that adjusts the strength of the impact of sparsity handling in the output of a deep convolutional neural network.

9. The intelligent structural damage detection system according to claim 8, characterized in that: Also includes: The structural damage assessment module is used to assess the structural damage based on the structural damage identification result by using an incremental transfer learning network to obtain a structural damage assessment result.

Images