Method, system and related devices for bridge life prediction based on deep learning

Abstract

This application relates to the field of data processing technology, and provides a method, system, and related equipment for bridge life prediction based on deep learning. The method involves acquiring and preprocessing raw bridge structural data, extracting modal features from the preprocessed data, determining dynamic parameters based on the correlation parameters between modal features and their contribution to bridge life prediction, and generating a fused feature vector by combining the modal features. This fused feature vector is then input into a spatiotemporally coupled hybrid model constructed based on an encoder and decoder, and its output is dynamically and adaptively corrected to generate a predicted value for the remaining bridge life. A bridge monitoring report is generated based on the raw data and the predicted value. The dynamic parameter adjustment mechanism enables the system to respond to environmental changes in real time, reducing prediction bias. The combination of the spatiotemporally coupled hybrid model and dynamic correction accurately reflects the evolution of the bridge's condition, enhancing the comprehensiveness, real-time performance, and accuracy of bridge structural life prediction.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and more specifically, to a method, system, and related equipment for predicting bridge lifespan based on deep learning. Background Technology

[0002] With the continuous advancement of infrastructure construction such as bridges worldwide, a large number of bridges are gradually entering their aging stage. Traditional bridge inspection methods mainly rely on manual inspections, which are not only inefficient and costly, but also greatly affected by the experience and subjective factors of the inspectors, making it difficult to comprehensively and promptly identify potential safety hazards in bridges.

[0003] Traditional methods rely on single-sensor data, such as strain and vibration, ignoring damage features hidden in multimodal data such as images and acoustic emissions, leading to high prediction errors. Furthermore, existing models are based on static assumptions and cannot capture the impact of dynamic factors such as the environment on fatigue life in real time, resulting in significant discrepancies between predicted and actual fatigue life. Summary of the Invention

[0004] This application provides a method, system, and related equipment for predicting bridge life based on deep learning, which can at least partially solve the problem of high error in predicting the life of bridge structures.

[0005] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.

[0006] According to one aspect of this application, a method for predicting bridge lifespan based on deep learning is provided, comprising: acquiring raw data of the bridge structure; preprocessing the raw data to extract modal features from the preprocessed data; determining dynamic parameters based on the correlation parameters between the modal features and the contribution parameters of each modal feature to the bridge lifespan prediction; combining the modal features according to the dynamic parameters to generate a fused feature vector; constructing a spatiotemporally coupled hybrid model based on an encoder and a decoder; inputting the fused feature vector into the hybrid model; dynamically and adaptively correcting the output of the hybrid model to generate a predicted value of the bridge's remaining lifespan; and generating a bridge monitoring report based on the raw data and the predicted value.

[0007] In this application, based on the aforementioned scheme, the raw data includes image data, vibration data, acoustic emission data, and environmental data.

[0008] In this application, based on the aforementioned scheme, the acquisition of raw data of the bridge structure includes: receiving crack images captured by a drone during inspection and corrosion images captured by a camera as the image data; acquiring time-domain signals of the bridge structure through a preset accelerometer as the vibration data; acquiring signals generated by cable breakage events through an acoustic emission sensor as the acoustic emission data; and measuring temperature, humidity, and corrosive gas concentration through an environmental sensor as the environmental data.

[0009] In this application, based on the aforementioned scheme, the modal features include image features, frequency band energy distribution features, time-frequency mode features, and environmental features.

[0010] In this application, based on the aforementioned scheme, the preprocessing of the original data and the extraction of modal features from the preprocessed data includes: preprocessing the image data to generate a high-resolution image based on a pre-trained super-resolution convolutional neural network model, and extracting image features from the high-resolution image; extracting the frequency band energy distribution features in the vibration data using a short-time Fourier transform algorithm; inputting the acoustic emission signal data into a pre-trained one-dimensional convolutional neural network model and outputting time-frequency mode features; and standardizing the environmental data and extracting environmental features from the standardized data.

[0011] In this application, based on the aforementioned scheme, the step of determining dynamic parameters based on the correlation parameters between modal features and the contribution parameters of each modal feature to bridge life prediction, and generating a fusion feature vector by combining the modal features according to the dynamic parameters, includes: combining the modal features to generate a one-dimensional data sequence; inputting the data sequence into a multi-layer encoder, generating correlation parameters between different modal features in the data sequence through a self-attention mechanism; determining the contribution parameters of the modal features in life prediction based on the statistical information of the modal features; determining dynamic parameters according to the correlation parameters and the contribution parameters; updating the data sequence based on the dynamic parameters to generate a fusion feature vector.

[0012] In this application, based on the aforementioned scheme, the step of constructing a spatiotemporally coupled hybrid model based on an encoder and decoder, inputting the fused feature vector into the hybrid model, and dynamically adaptively correcting the output of the hybrid model to generate a predicted value for the remaining life of the bridge includes: constructing a spatiotemporally coupled hybrid model based on an encoder and decoder, inputting the fused feature vector into the hybrid model, and outputting an initial predicted value; updating the crack propagation rate based on the relationship between environmental factors and crack propagation; correcting the initial predicted value based on the crack propagation rate, and outputting a predicted value for the remaining life of the bridge.

[0013] In this application, based on the aforementioned scheme, generating a bridge monitoring report based on the original data and the predicted value includes: obtaining a template for a bridge monitoring report; writing the original data and the predicted value into the template for the bridge monitoring report to generate a bridge monitoring report.

[0014] In this application, based on the aforementioned scheme, it further includes: comparing the predicted value of the bridge's remaining lifespan output by the model with a preset threshold; if the predicted value is less than or equal to the threshold, then triggering a corresponding early warning mechanism.

[0015] According to one aspect of this application, a system for predicting bridge lifespan based on deep learning is provided, comprising:

[0016] The acquisition module is used to acquire raw data about the bridge structure.

[0017] The extraction module is used to preprocess the original data and extract modal features from the preprocessed data;

[0018] The fusion module is used to determine dynamic parameters based on the correlation parameters between modal features and the contribution parameters of each modal feature to bridge life prediction, and to generate a fused feature vector by combining the modal features according to the dynamic parameters.

[0019] The prediction module is used to construct a spatiotemporally coupled hybrid model based on the encoder and decoder, input the fused feature vector into the hybrid model, and dynamically and adaptively correct the output of the hybrid model to generate a predicted value of the bridge's remaining life.

[0020] The reporting module is used to generate a bridge monitoring report based on the raw data and the predicted values.

[0021] According to one aspect of this application, a computer-readable medium is provided having a computer program stored thereon, which, when executed by a processor, implements the deep learning-based bridge life prediction method as described in the above embodiments.

[0022] According to one aspect of this application, an electronic device is provided, comprising: one or more processors; and a storage device for storing one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement the deep learning-based bridge life prediction method as described in the above embodiments.

[0023] According to one aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the deep learning-based bridge life prediction method provided in the various alternative implementations described above.

[0024] The main differences and technical effects of the technical solution of this application compared with the prior art are as follows:

[0025] On the one hand, traditional bridge inspection methods rely excessively on data from single types of sensors, such as strain or vibration data alone. This prevents them from comprehensively capturing the complex damage characteristics and actual condition of bridge structures. For example, a single sensor may fail to detect early surface cracks or internal micro-fractures. This technical solution, however, acquires multimodal raw data including images, vibration, acoustic emission, and environmental data, and performs comprehensive processing and feature extraction to achieve a holistic assessment of bridge condition. This multimodal data fusion approach can detect damage that a single data source cannot capture, significantly improving the accuracy and comprehensiveness of damage identification and effectively reducing prediction errors—a feat unmatched by traditional methods.

[0026] Secondly, existing models are often based on static assumptions, making it difficult to capture the impact of dynamic factors such as environmental changes on the fatigue life of bridges in real time. When environmental conditions change significantly, such as a sudden rise in temperature, a surge in humidity, or a sharp increase in the concentration of corrosive gases, the predictions given by these static models often deviate significantly from the actual lifespan. This technical solution, however, determines the correlation and contribution parameters between various modal features, and then dynamically adjusts the feature weights to generate a fused feature vector. This dynamic parameter adjustment mechanism allows the system to flexibly adjust according to real-time environmental conditions and bridge status, effectively reducing prediction bias caused by static assumptions, improving the consistency between prediction results and actual lifespan, and demonstrating significant advantages.

[0027] Thirdly, traditional models often struggle to effectively capture the complex temporal and spatial changes in bridge condition data. Damage to bridge structures can vary significantly at different locations and time points, and traditional methods lack effective analytical tools for this. This technical solution constructs a spatiotemporally coupled hybrid model, combined with a dynamic adaptive correction mechanism, enabling simultaneous processing of data across both time and space dimensions. This model not only captures the complex evolution of bridge conditions in time and space but also, through dynamic correction based on real-time environmental data and crack propagation rates, makes the prediction results closer to reality. This combination of spatiotemporally coupled analysis and dynamic correction significantly improves the prediction accuracy of bridge remaining life, providing a more reliable basis for bridge maintenance and management.

[0028] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0029] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0030] Figure 1 The flowchart illustrating a method for predicting bridge life based on deep learning in one embodiment of this application is shown.

[0031] Figure 2 The flowchart illustrating the generation of fused feature vectors in one embodiment of this application is shown schematically.

[0032] Figure 3 The illustration shows a schematic diagram of a deep learning-based bridge life prediction system in one embodiment of this application.

[0033] Figure 4 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown. Detailed Implementation

[0034] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art.

[0035] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this application. However, those skilled in the art will recognize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this application.

[0036] It should be noted that the data acquisition or information collection in this embodiment is performed after authorization by the user or the object of collection, and its process and purpose strictly follow the relevant regulations.

[0037] The block diagrams shown in the attached figures are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, or in one or more hardware modules composed of smart chips, smart integrated circuits, or application-specific integrated circuits (ASICs), or in different network and / or processor devices and / or microcontroller devices.

[0038] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0039] The implementation details of the technical solution of this application are described below:

[0040] Figure 1 A flowchart illustrating a deep learning-based bridge life prediction method according to an embodiment of this application is shown. (Refer to...) Figure 1 As shown, the deep learning-based bridge life prediction method includes at least steps S110 to S150, which are described in detail below:

[0041] S110, Obtain the raw data of the bridge structure.

[0042] In this embodiment, when acquiring raw data of the bridge structure, various sensors are carefully deployed at key parts of the bridge according to a predetermined monitoring plan. For example, strain sensors are installed at the piers to detect changes in structural stress, and accelerometers are set up on the bridge deck to capture vibration information. Simultaneously, drones are deployed to inspect the bridge's appearance along a planned route, capturing images of cracks, and cameras are used to continuously monitor the corrosion of bridge components. Subsequently, relevant information is continuously collected by sensors and equipment at a set sampling frequency. The drones and cameras transmit image data periodically. All data is transmitted in real-time and stably to the data storage center via wired or wireless transmission for subsequent processing and analysis, thereby comprehensively acquiring raw data covering structural characteristics, appearance, and other aspects.

[0043] In one embodiment of this application, the raw data includes image data, vibration data, acoustic emission data, and environmental data.

[0044] In one embodiment of this application, obtaining raw data of the bridge structure includes:

[0045] The image data includes crack images captured by the drone during inspection and corrosion images captured by the camera.

[0046] The vibration data is obtained by acquiring time-domain signals of the bridge structure using a preset accelerometer.

[0047] The signal generated by the cable breakage event is acquired by an acoustic emission sensor and used as the acoustic emission data.

[0048] The environmental data is obtained by measuring temperature, humidity, and corrosive gas concentrations using environmental sensors.

[0049] In this embodiment, an interface protocol is first established, for example, connecting sensor devices via a wired network or receiving image data transmitted from drones and cameras via a wireless network. For image data, real-time reception includes crack images captured by the drone during inspections and corrosion images captured by the camera. These images are stored in a specific file format. Upon reception, they are parsed according to corresponding encoding rules, converting binary data into visualized image information, and temporarily stored in the computer's memory or a designated storage area.

[0050] For vibration data, accelerometers continuously collect time-domain signals from the bridge structure. Data acquisition cards or other specialized equipment connected to the accelerometers read the electrical signals output by the sensors at a set sampling frequency and convert them into digital signals. These digital signals represent the bridge's vibration at different times and are received and stored as a data stream.

[0051] Regarding acoustic emission data, the signals generated by the cable breakage event are captured by acoustic emission sensors. A computer is also connected to the acoustic emission sensors via a corresponding interface to receive the signals in real time. These signals contain relevant information about the cable breakage event, and the computer performs preliminary processing and labeling of the signals during reception for subsequent processing.

[0052] Environmental data, such as temperature, humidity, and corrosive gas concentrations, are measured by corresponding environmental sensors. Computers interact with these sensors via specific communication protocols, periodically reading the data and storing it as structured data records for correlation analysis with other types of data.

[0053] The above process, utilizing drones to capture images of cracks, cameras to capture images of corrosion, and various devices such as accelerometers, acoustic emission sensors, and environmental sensors to collect data, comprehensively obtains information on the bridge's structural condition and environment from multiple dimensions. Image data visually presents the bridge's surface damage, vibration data reflects the structure's dynamic characteristics, acoustic emission data captures signals of minute internal damage, and environmental data demonstrates the impact of external factors on the bridge. This provides rich and detailed basic data for subsequent analysis, ensuring that there is no significant information gap in the assessment of the bridge's condition.

[0054] S120, preprocess the original data and extract modal features from the preprocessed data.

[0055] In this embodiment, when preprocessing the acquired raw bridge structure data and extracting modal features, processing is first performed according to the characteristics of different types of data. Image data undergoes enhancement and denoising operations to improve clarity and quality. Vibration data is filtered and denoised to eliminate interference signals; outliers and invalid segments in acoustic emission data are removed to ensure data validity. After preprocessing, various types of data are analyzed in depth. Crack morphology features are identified from images, structural vibration frequencies and mode shapes are extracted from vibration data, and features related to wire breakage events are obtained from acoustic emission data. Finally, these features are integrated as modal features for subsequent research.

[0056] In one embodiment of this application, the modal features include image features, frequency band energy distribution features, time-frequency mode features, and environmental features.

[0057] The process includes preprocessing the original data and extracting modal features from the preprocessed data, including:

[0058] Based on the pre-trained super-resolution convolutional neural network model, image data is preprocessed to generate high-resolution images, and image features are extracted from the high-resolution images.

[0059] The short-time Fourier transform algorithm is used to extract the frequency band energy distribution characteristics in the vibration data;

[0060] The acoustic emission signal data is input into a pre-trained one-dimensional convolutional neural network model, which outputs time-frequency pattern features.

[0061] The environmental data is standardized, and environmental features are extracted from the standardized data.

[0062] In this embodiment, the mapping relationship between low-resolution and high-resolution images is learned in advance through multi-layer convolutional operations to train the super-resolution convolutional neural network model. Its network structure consists of multiple convolutional layers stacked in an orderly manner, gradually learning the complex mapping relationship from low-resolution to high-resolution images through convolutional operations at different levels. The input low-resolution image first enters the network, undergoing convolutional operations at each layer using a series of convolutional kernels. Each convolutional layer is responsible for extracting and transforming image features. As the number of network layers increases, the abstraction level of the feature maps continuously improves, ultimately resulting in a model output with a resolution significantly increased to several times that of the original image, effectively enhancing image details and improving image quality. During the training phase, a large number of pairs of low-resolution and high-resolution images are used for learning, adjusting the model parameters to ensure the model can accurately reconstruct high-resolution images from low-resolution images.

[0063] In the application phase, a super-resolution convolutional neural network model is loaded, and the acquired low-resolution image data is input into the model. The model's convolutional layers extract and transform features from the image data, gradually enhancing the image's details and clarity through multi-layer neural network processing. Ultimately, the model outputs a high-resolution image with significantly improved resolution, enabling the clear identification of minute cracks with a width greater than or equal to a certain size.

[0064] The resulting high-resolution image is then saved, replacing the original low-resolution image. Image features are extracted from the high-resolution image to provide a more accurate data foundation for subsequent analysis.

[0065] In this embodiment, a short-time Fourier transform algorithm is used to process the acquired time-domain vibration signal (vibration data). First, the time-domain signal is divided into multiple short time segments, and a Fourier transform is performed on each time segment to convert the time-domain signal into a frequency-domain signal. This method allows for the extraction of the frequency band energy distribution characteristics of the signal within each time segment. These features are then organized and stored to form a frequency-domain feature dataset of the vibration signal, which is used for subsequent analysis of the vibration characteristics of the bridge structure. Extracting the frequency band energy distribution characteristics helps analyze the vibration characteristics of the bridge structure at different frequencies, identify the structure's natural frequencies and modal parameters, and provide a basis for determining whether the structure exhibits any anomalies.

[0066] A one-dimensional convolutional neural network model is pre-trained. The collected acoustic emission signal data is input into the model, which extracts features from the signal through operations such as convolution and pooling. It can automatically learn and output time-frequency pattern features that characterize the properties of the wire breakage event. Furthermore, statistical features, such as event count rate and energy accumulation curve, can be extracted from these time-frequency pattern features or the original acoustic emission signal.

[0067] For example, in a one-dimensional convolutional neural network model, convolutional layers slide convolutional kernels across the signal to extract local features; pooling layers downsample the features, reducing computation and enhancing the model's robustness; fully connected layers integrate the learned features to generate final statistical features, such as event count rates and energy accumulation curves. In this process, the model can automatically identify the time-frequency pattern characteristics of bridge cable breakage events and further generate relevant statistical features, such as event count rates and energy accumulation curves. These statistical features accurately describe the characteristics of bridge cable breakage events, providing crucial information for assessing the health status of bridge cables. For instance, a sudden increase in the event count rate may indicate an escalation of bridge cable breakage. These statistical features are extracted and stored as structured data for further analysis of bridge cable breakage events.

[0068] Environmental data such as temperature, humidity, and corrosive gas concentrations are standardized. First, the minimum and maximum values ​​of a given environmental parameter are identified. Then, for each data point of that parameter, a linear transformation is applied to map it to the interval between zero and one. Environmental features are then extracted from the standardized data. This normalization process eliminates dimensional differences between different environmental parameters, enabling the analysis and processing of different types of environmental data at the same scale, improving data comparability and model training effectiveness.

[0069] Through the above data acquisition and preprocessing steps, the data features of modalities such as images, vibrations, acoustic emissions, and environment are integrated, and the original multimodal data is converted into standardized multimodal feature vector sets. At the same time, the features of each modality are extracted. These feature vector sets will provide high-quality data input for subsequent feature fusion and model training.

[0070] The above process, for image data, utilizes a super-resolution convolutional neural network model to enhance resolution, effectively improving image details and making minute damages such as cracks more clearly visible, laying the foundation for accurate image feature extraction. For vibration data, appropriate algorithms remove noise interference, allowing the frequency band energy distribution characteristics to more realistically reflect the vibration characteristics of the structure. Acoustic emission data, after model processing, outputs time-frequency mode features, better revealing the characteristics of events such as cable breakage. Environmental features extracted after standardization of environmental data are more easily integrated with other features for comprehensive analysis. This series of preprocessing and feature extraction operations significantly improves data quality and usability, enabling subsequent analysis to be based on more accurate and representative features.

[0071] S130, determine dynamic parameters based on the correlation parameters between modal features and the contribution parameters of each modal feature to bridge life prediction, and generate a fusion feature vector by combining the modal features according to the dynamic parameters.

[0072] In this embodiment, when determining the fused feature vector, the correlation parameters, such as the degree of interrelation and the mode of action among the modal features, are first analyzed. Simultaneously, the contribution parameters of each modal feature to the bridge life prediction are evaluated. Based on these dynamically changing parameters, the weights and fusion methods of each modal feature in the combination process are adjusted. For example, higher weights are assigned to modal features with strong correlation and significant contributions. Through a specific fusion strategy, the modal features are organically integrated, ultimately generating a fused feature vector that comprehensively and accurately reflects the bridge's condition information.

[0073] like Figure 2 As shown, in one embodiment of this application, dynamic parameters are determined based on the correlation parameters between modal features and the contribution parameters of each modal feature to bridge life prediction. A fused feature vector is then generated by combining the modal features according to the dynamic parameters, including:

[0074] S210, combine the modal features to generate a one-dimensional data sequence;

[0075] S220, The data sequence is input into a multi-layer encoder, and the correlation parameters between different modal features in the data sequence are generated through a self-attention mechanism;

[0076] S230, Based on the statistical information of the modal features, determine the contribution parameters of the modal features in lifetime prediction;

[0077] S240, determine dynamic parameters based on the correlation parameters and the contribution parameters, update the data sequence based on the dynamic parameters, and generate a fusion feature vector.

[0078] In this embodiment, the acquired multiple modal features are first fused into a multimodal feature fusion. Feature vectors from different modalities, such as crack width and corrosion area information representing image features, frequency band energy information reflecting vibration features, and event count rate information characterizing acoustic emission features, are spliced ​​and combined in a preset order to form a one-dimensional data sequence.

[0079] Next, the concatenated data sequence is input into a multi-layer encoder. Within each encoder layer, a self-attention mechanism generates an indicator vector, an index vector, and a data vector for each position in the data sequence. The indicator vector represents the position or information that needs attention, the index vector serves as the index or identifier for each position in the input sequence, and the data vector contains the actual content information for each position in the input sequence.

[0080] Next, the similarity between the indicator vector and all index vectors is calculated and normalized to obtain the association parameters between each position and other positions. for:

[0081]

[0082] in, and Let represent the indicator vector and index vector of the i-th modal feature, respectively. represents the attention head dimension, j and N represent the identifier and total number of modal features, respectively, exp represents the exponentiation operation, and T represents the transpose operation.

[0083] For each modal feature, the above process first calculates a basic correlation parameter based on its own indicator vector, index vector, and the index vectors of all modal features. This correlation parameter reflects the correlation between different modal features; the stronger the correlation, the higher the correlation parameter, and the greater its weight in the decision-making process.

[0084] The above process reorganizes and redistributes information based on the degree of correlation between different locations. Through multiple layers of such operations, the inherent connections and long-distance dependencies between different modal features can be fully explored, effectively integrating the originally independent modal features into an organic whole. This enhances the information content and expressive power of the features, laying the foundation for subsequent dynamic weight allocation and lifetime prediction.

[0085] However, simply considering the correlation parameters between features is insufficient; it is also necessary to consider the changes in the features themselves. Therefore, a dynamic update term is introduced to obtain statistical information such as the historical mean and standard deviation of each modal feature. Based on the sensitivity of each modal feature to bridge damage and the difference between the current state and the historical state, their contribution parameters in life prediction are determined. for:

[0086]

[0087] in, The data vector representing the i-th modal feature. and It consists of its historical mean and standard deviation, i.e., statistical information; This represents a dynamically updated coefficient, which can be a learnable parameter or a fixed coefficient set according to the stability of the feature.

[0088] Through the above calculation process, if the current feature value deviates significantly from the historical distribution, it indicates that the feature may have undergone abnormal changes, making its indicative role in bridge damage more pronounced. In this case, the dynamic update term will increase, thereby increasing the weight of that modal feature. Conversely, if the current feature value is close to the historical distribution, the dynamic update term has a smaller impact on the weight. In this way, the dynamic parameters of each modal feature's real-time state can be used to determine its weight. for:

[0089]

[0090] Based on the calculated dynamic parameters of each modal feature, a weighted summation operation is performed on the fused features. Different weights are applied to the fused features, ensuring that features contributing significantly to bridge life prediction occupy a more prominent position in the final fused feature vector. The resulting fused feature vector integrates multimodal information and is optimized according to the dynamic weights of each feature, more accurately reflecting the current state of the bridge and its impact on life. This fused feature vector is then output to subsequent processing steps, providing high-quality input data for the training and prediction of the bridge life prediction model.

[0091] The above process, analyzing the correlation parameters between various modal features, allows for a deeper understanding of the interactions and relationships between different features, such as the potential intrinsic connection between image features and vibration features. Determining the contribution parameters of each modal feature to bridge life prediction clarifies the importance of each feature in the prediction process. Based on these two types of parameters, dynamic parameters are determined, and the data sequence is updated accordingly to generate a fused feature vector. This fully considers the importance and interrelationships of different features, organically integrating various features to form a comprehensive and representative feature set, more accurately representing the overall state of the bridge, and providing stronger feature support for subsequent life prediction.

[0092] S140, a spatiotemporally coupled hybrid model is constructed based on the encoder and decoder, the fused feature vector is input into the hybrid model, and the output of the hybrid model is dynamically and adaptively corrected to generate a predicted value of the bridge's remaining life.

[0093] In this embodiment, when constructing the spatiotemporally coupled hybrid model, the encoder first captures the spatial dependencies between different modal features in the fused feature vector, uncovering the intrinsic connections between the various structural components. The decoder models the temporal variation of features based on spatial dependency information. After the fused feature vector is input into the hybrid model, the model performs a preliminary analysis based on the learned spatiotemporal feature relationships. Subsequently, a dynamic adaptive correction mechanism is employed to adjust and optimize the model output in real time based on the bridge's real-time state changes and historical data feedback, thereby generating a more realistic prediction of the bridge's remaining lifespan.

[0094] In one embodiment of this application, a spatiotemporally coupled hybrid model is constructed based on an encoder and a decoder. The fused feature vector is input into the hybrid model, and the output of the hybrid model is dynamically and adaptively corrected to generate a predicted value for the remaining life of the bridge. This includes:

[0095] A spatiotemporally coupled hybrid model is constructed based on an encoder and a decoder. The fused feature vector is input into the hybrid model, and the initial prediction value is output.

[0096] The crack propagation rate is updated based on the relationship between environmental factors and crack propagation.

[0097] The initial prediction is corrected based on the crack propagation rate, and the predicted value of the bridge's remaining life is output.

[0098] In this embodiment, a hybrid model is pre-trained, which mainly includes a Transformer encoder and an LSTM (Long Short-Term Memory) decoder. Specifically, the Transformer encoder captures the spatial dependencies between multimodal features, inputting the fused feature vector into it. The encoder generates query, key, and value vectors for each feature location using a self-attention mechanism. Attention weights are obtained by calculating and normalizing the similarity between the query vector and all key vectors, and then a weighted sum is performed on the value vectors based on these weights. This process can uncover the degree of correlation between different features; for example, there may be a relationship between crack information in image features and frequency changes in vibration features.

[0099] Specifically, the output of the Transformer encoder serves as the input sequence for the LSTM decoder. As time progresses, the cracks in the bridge may gradually widen. The LSTM decoder is used to model the evolution of the damage over time. Based on a deep learning recurrent neural network, it controls the flow of information through input gates, forget gates, and output gates. When processing data at each time step, the LSTM can determine which information needs to be retained and which needs to be forgotten based on the current input and the hidden state of the previous time step, thus effectively learning the trend of damage change over time.

[0100] In the aforementioned model structure, the Transformer encoder can capture this spatial dependency, thereby better understanding the complex combination of features of the bridge state. By using the output of the Transformer encoder as the input sequence of the LSTM decoder, the LSTM can capture the long-term and short-term variation patterns over time, providing a temporal basis for lifetime prediction.

[0101] Optionally, at the bridge site, based on the compatibility of the hybrid model trained in the aforementioned steps and the computing resources of the equipment, the hybrid model can be deployed to edge computing devices at the bridge site. Optionally, during deployment, the model can be appropriately optimized and transformed to ensure that it can run efficiently on the edge devices. The advantage of edge deployment is that it can reduce data transmission latency, because real-time bridge monitoring data does not need to be transmitted to a remote server for processing; the prediction task can be completed directly on the edge devices at the site, thereby achieving low-latency (≤5 seconds) prediction and meeting real-time requirements.

[0102] First, the hybrid model is initialized by assigning initial values ​​to various parameters, including the self-attention weights in the encoder and the gating parameters in the decoder. After initialization, the fused feature vector generated in the preceding steps is passed as input to the hybrid model. This fused feature vector contains the key features of the bridge multimodal monitoring information after fusion and weight allocation, providing a comprehensive data foundation for subsequent model analysis.

[0103] Next, the input fused feature vector is fed into the Transformer encoder to capture the spatial dependencies between multimodal features. Under the self-attention mechanism, for each feature element in the fused feature vector, a corresponding query vector, key vector, and value vector are generated. By calculating the similarity between the query vector and all key vectors and performing normalization, the attention weights between each feature element and other elements are obtained. These weights reflect the degree of correlation between different features; for example, crack information in image features and specific frequency changes in vibration features may have an inherent relationship, and the self-attention mechanism can accurately capture this relationship. Subsequently, the corresponding value vectors are weighted and summed according to the calculated attention weights, allowing the model to focus on feature combinations more important for describing the bridge's state, thereby gaining a deeper understanding of the spatial relationships between multimodal features.

[0104] The output of the Transformer encoder serves as the input sequence for the LSTM decoder, which is then used to model the evolution of bridge damage over time. The LSTM uses a gating mechanism—input gate, forget gate, and output gate—to control the flow of information. The input gate determines how much of the input information at the current time step can enter the cell state; the forget gate determines how much information from the previous time step's cell state needs to be retained or forgotten; and the output gate determines how much information from the current cell state can be output to the next time step. Through this mechanism, the LSTM can effectively learn the changes in bridge damage at different locations at different time points, capturing both short-term fluctuations and long-term trends. For example, bridge cracks may experience slight expansion in the short term due to load changes, but in the long term, they will continue to develop due to factors such as material aging. The LSTM can effectively capture these temporal variations.

[0105] To enable the model to adapt in real time to the impact of environmental changes on bridge damage, the initial predicted values ​​output by the hybrid model are dynamically and adaptively corrected. During model operation, environmental data, such as changes in temperature, humidity, and corrosive gas concentration, are continuously monitored. Simultaneously, based on the model's current estimates of parameters such as crack length and load cycle count, and according to the intrinsic physical relationship between environmental factors and crack propagation, the crack propagation rate is dynamically and adaptively updated as follows:

[0106]

[0107] Where C represents the material crack propagation correlation constant, and m represents the stress intensity factor influence exponential constant; The stress intensity factor is an important parameter for measuring the intensity of the stress field at the crack tip, and it is related to factors such as external load, crack size, and structural shape. These represent dynamic adjustment coefficients, used to adjust the influence of environmental factors on the crack propagation rate; These represent the rates of change of current temperature and humidity, and the concentration of corrosive gases, respectively. Indicates reference temperature and humidity. This indicates the rate of change of the reference corrosive gas concentration.

[0108] The above process, by introducing an environmental change rate term, can adjust the prediction of crack propagation rate in real time according to changes in the current environment. Then, based on the predicted crack propagation rate, the initial prediction value output by the hybrid model is corrected, and the prediction value of the bridge's remaining life is output.

[0109] For example, when temperature and humidity suddenly increase or the concentration of corrosive gases rises, crack propagation may accelerate. Timely detection of such changes allows predictions to better reflect reality. For instance, when a sudden increase in the concentration of corrosive gases in the environment is detected, this could accelerate the corrosion and crack propagation of bridge materials. By calculating the crack propagation rate under the current conditions, the impact of such environmental changes on cracks in the spatiotemporal dimensions can be reflected in a timely manner, thereby improving the accuracy of bridge life prediction.

[0110] After processing using the aforementioned hybrid model and adjusting with a dynamic correction mechanism, the predicted remaining life of the bridge is calculated based on the damage evolution patterns learned by the model and the real-time corrected crack propagation predictions. Finally, this predicted value is output to provide crucial information for bridge maintenance decisions, helping relevant departments to promptly arrange repairs or reinforcement measures and ensure the safe operation of the bridge.

[0111] The above process, based on a spatiotemporally coupled hybrid model constructed by an encoder and decoder, allows the encoder to uncover the spatial dependencies between different modal features in the fused feature vector, while the decoder can model the temporal variation of features based on these spatial dependencies, thus better capturing the spatiotemporal evolution of the bridge's condition. Dynamic adaptive correction of the model output, incorporating the relationship between environmental factors and crack propagation to update the crack propagation rate, and accordingly revising the initial prediction, ensures that the prediction results fully consider the impact of actual environmental changes on the bridge's lifespan, further improving the accuracy and reliability of the predictions and making the generated remaining lifespan predictions closer to the bridge's actual condition.

[0112] S150, Based on the original data and the predicted values, generate a bridge monitoring report.

[0113] In this embodiment, when generating a bridge monitoring report, the acquired raw data is first systematically organized and presented according to different types such as images, vibrations, and acoustic emissions, showcasing the bridge's initial state information. Then, the predicted values ​​are integrated into the report in an intuitive and easy-to-understand manner, for example, by dedicating specific sections to explaining the bridge's remaining lifespan prediction results. Finally, a comprehensive analysis is performed based on the raw data and predicted values ​​to describe the bridge's current health status, potential risks, and future development trends, ultimately forming a monitoring report that provides a reliable basis for bridge maintenance and management.

[0114] In one embodiment of this application, the method further includes: comparing the predicted value of the remaining life of the bridge output by the model with a preset threshold; if the predicted value of the remaining life of the bridge is less than or equal to the threshold, a corresponding early warning mechanism is triggered.

[0115] Real-time monitoring data streams from the bridge, including multimodal data such as images, vibrations, acoustic emissions, and environmental data, are continuously transmitted to edge computing devices. At the edge devices, the input real-time data undergoes rapid preprocessing to ensure its format and features are consistent with the data used during model training. For example, image data undergoes resolution adjustment and normalization, while vibration signals are filtered and feature extracted. After preprocessing, the processed data is input into a deployed hybrid model. Based on the knowledge and patterns learned during training, the model rapidly analyzes and calculates the input real-time data, outputting a predicted remaining life of the bridge. This process requires edge computing nodes and devices to possess sufficient computing power to complete data processing and model inference within a short timeframe.

[0116] The predicted value of the bridge's remaining lifespan output by the model is compared with a preset threshold. If the predicted value of the bridge's remaining lifespan is less than or equal to the threshold, the corresponding early warning mechanism is triggered.

[0117] For example, relevant personnel can be notified via SMS that the bridge has a low remaining lifespan and poses a safety risk. Next, a warning window will pop up on the system's interface to prominently alert staff. Finally, if necessary, the bridge's load can be forcibly limited, such as reducing weight limits or prohibiting certain heavy vehicles from using the bridge, to ensure its safety. This series of early warning actions aims to promptly communicate the bridge's dangerous situation to relevant personnel and take effective measures to prevent accidents.

[0118] By deploying the model at the edge, rapid real-time predictions were achieved, ensuring timely monitoring of the bridge's condition. The reasonable setting and triggering mechanism of the early warning rules can quickly notify relevant personnel and take measures when safety hazards occur on the bridge, effectively improving the safety and reliability of the bridge and reducing potential safety risks.

[0119] The above process compares the predicted remaining lifespan of the bridge output by the model with a preset threshold. If the predicted remaining lifespan is less than or equal to the threshold, a corresponding early warning mechanism is triggered. This step can promptly detect abnormal changes in the bridge's lifespan, issue early warnings, and remind relevant personnel to take appropriate measures, such as strengthening monitoring and carrying out maintenance and reinforcement, effectively preventing safety accidents caused by bridge lifespan issues and ensuring the safe operation of the bridge and the safety of public life and property.

[0120] In one embodiment of this application, a bridge monitoring report is generated based on the raw data and the predicted values, including:

[0121] Obtain a template for a bridge monitoring report;

[0122] The original data and the predicted values ​​are written into the template of the bridge monitoring report to generate the bridge monitoring report.

[0123] To ensure the standardization and professionalism of bridge monitoring reports, the first step is to obtain suitable report templates. For example, standard report templates developed by relevant bridge engineering standards bodies or industry organizations are typically validated through extensive practical application and expert review, covering the key information and data presentation methods required for bridge monitoring. Alternatively, monitoring report templates successfully applied in similar large-scale bridge projects can be referenced and appropriately adjusted and optimized to suit the specific characteristics and monitoring requirements of this project. Furthermore, utilizing online resources to search for publicly available, representative bridge monitoring report examples, extracting common structural and content frameworks as template foundations, is crucial. When acquiring templates, focus should be placed on the clarity of the monitoring data presentation, the logical consistency of the analytical conclusions, and the overall readability of the report, ensuring that the template accurately conveys the bridge monitoring information.

[0124] After obtaining the bridge monitoring report template, accurately write the collected raw data and the predicted values ​​obtained through deep learning models into the template. For the raw data, fill in the relevant parameters of the image data, the time-domain and frequency-domain characteristics of the vibration data, the event statistics of the acoustic emission data, and the environmental data in the corresponding positions according to the format and classification set in the template, ensuring the completeness and accuracy of the data. Predicted values, such as the predicted remaining life of the bridge, should be presented in a prominent and clear manner in the report, along with the corresponding prediction basis and analysis explanations. During the writing process, pay attention to the consistency of data units and the standardization of format, and provide necessary explanations and clarifications so that report users can understand the meaning and importance of the data. After completing the data writing, conduct a comprehensive review and proofreading of the report, checking the accuracy of the data, the logical coherence, and the standardization of the format, ultimately generating a complete, accurate, and valuable bridge monitoring report.

[0125] The above process generates a report by acquiring a bridge monitoring report template and writing raw data and predicted values ​​into the template. This process presents complex analysis results in a standardized and organized manner. The raw data shows the initial state information of the bridge, while the predicted values ​​provide an assessment of the bridge's lifespan. The report makes this information readily available, allowing relevant personnel to quickly understand the overall condition of the bridge and providing a clear and intuitive reference for bridge maintenance, management, and decision-making.

[0126] This application's technical solution involves acquiring raw data of the bridge structure, preprocessing the raw data, and extracting modal features from the preprocessed data. Dynamic parameters are determined based on the correlation parameters between each modal feature and the contribution parameters of each modal feature to bridge life prediction. A fused feature vector is generated by combining the modal features according to the dynamic parameters. A spatiotemporally coupled hybrid model is constructed based on an encoder and decoder. The fused feature vector is input into the hybrid model, and the output of the hybrid model is dynamically and adaptively corrected to generate a predicted value for the remaining life of the bridge. A bridge monitoring report is generated based on the raw data and the predicted value. Through multimodal data fusion, the bridge state is comprehensively captured, reducing information loss. The dynamic parameter adjustment mechanism enables the system to respond to environmental changes in real time, reducing prediction bias. The combination of the spatiotemporally coupled hybrid model and dynamic correction accurately reflects the evolution of the bridge state, improving prediction accuracy and enhancing the comprehensiveness, real-time nature, and accuracy of bridge structural life prediction.

[0127] The following describes embodiments of the bridge life prediction system based on deep learning according to this application, which can be used to execute the bridge life prediction method based on deep learning in the above embodiments of this application. It is understood that the bridge life prediction system based on deep learning can be a computer program (including program code) running on a computer device. For example, the bridge life prediction system based on deep learning can install monitoring and management software to realize industrial cloud computing of big data generated during production operations through an industrial cloud platform therein; the bridge life prediction system based on deep learning can be used to execute the corresponding steps in the method provided in the embodiments of this application. For details not disclosed in the embodiments of the bridge life prediction system based on deep learning of this application, please refer to the embodiments of the bridge life prediction method based on deep learning described above in this application.

[0128] Figure 3 A block diagram of a deep learning-based bridge life prediction system according to an embodiment of this application is shown.

[0129] Reference Figure 3 As shown, a deep learning-based bridge life prediction system according to an embodiment of this application includes:

[0130] Module 310 is used to acquire raw data of the bridge structure;

[0131] Extraction module 320 is used to preprocess the original data and extract modal features from the preprocessed data;

[0132] The fusion module 330 is used to determine dynamic parameters based on the correlation parameters between each modal feature and the contribution parameters of each modal feature to the bridge life prediction, and to generate a fused feature vector by combining the modal features according to the dynamic parameters.

[0133] The prediction module 340 is used to construct a spatiotemporally coupled hybrid model based on the encoder and decoder, input the fused feature vector into the hybrid model, and dynamically and adaptively correct the output of the hybrid model to generate a predicted value of the bridge's remaining life.

[0134] Reporting module 350 is used to generate a bridge monitoring report based on the raw data and the predicted values.

[0135] In this application, based on the aforementioned scheme, the raw data includes image data, vibration data, acoustic emission data, and environmental data.

[0136] In this application, based on the aforementioned scheme, the acquisition of raw data of the bridge structure includes: receiving crack images captured by a drone during inspection and corrosion images captured by a camera as the image data; acquiring time-domain signals of the bridge structure through a preset accelerometer as the vibration data; acquiring signals generated by cable breakage events through an acoustic emission sensor as the acoustic emission data; and measuring temperature, humidity, and corrosive gas concentration through an environmental sensor as the environmental data.

[0137] In this application, based on the aforementioned scheme, the modal features include image features, frequency band energy distribution features, time-frequency mode features, and environmental features.

[0138] In this application, based on the aforementioned scheme, the preprocessing of the original data and the extraction of modal features from the preprocessed data includes: preprocessing the image data to generate a high-resolution image based on a pre-trained super-resolution convolutional neural network model, and extracting image features from the high-resolution image; extracting the frequency band energy distribution features in the vibration data using a short-time Fourier transform algorithm; inputting the acoustic emission signal data into a pre-trained one-dimensional convolutional neural network model and outputting time-frequency mode features; and standardizing the environmental data and extracting environmental features from the standardized data.

[0139] In this application, based on the aforementioned scheme, the step of determining dynamic parameters based on the correlation parameters between modal features and the contribution parameters of each modal feature to bridge life prediction, and generating a fusion feature vector by combining the modal features according to the dynamic parameters, includes: combining the modal features to generate a one-dimensional data sequence; inputting the data sequence into a multi-layer encoder, generating correlation parameters between different modal features in the data sequence through a self-attention mechanism; determining the contribution parameters of the modal features in life prediction based on the statistical information of the modal features; determining dynamic parameters according to the correlation parameters and the contribution parameters; updating the data sequence based on the dynamic parameters to generate a fusion feature vector.

[0140] In this application, based on the aforementioned scheme, the step of constructing a spatiotemporally coupled hybrid model based on an encoder and decoder, inputting the fused feature vector into the hybrid model, and dynamically adaptively correcting the output of the hybrid model to generate a predicted value for the remaining life of the bridge includes: constructing a spatiotemporally coupled hybrid model based on an encoder and decoder, inputting the fused feature vector into the hybrid model, and outputting an initial predicted value; updating the crack propagation rate based on the relationship between environmental factors and crack propagation; correcting the initial predicted value based on the crack propagation rate, and outputting a predicted value for the remaining life of the bridge.

[0141] In this application, based on the aforementioned scheme, generating a bridge monitoring report based on the original data and the predicted value includes: obtaining a template for a bridge monitoring report; writing the original data and the predicted value into the template for the bridge monitoring report to generate a bridge monitoring report.

[0142] In this application, based on the aforementioned scheme, it further includes: comparing the predicted value of the bridge's remaining lifespan output by the model with a preset threshold; if the predicted value is less than or equal to the threshold, then triggering a corresponding early warning mechanism.

[0143] This application's technical solution involves acquiring raw data of the bridge structure, preprocessing the raw data, and extracting modal features from the preprocessed data. Dynamic parameters are determined based on the correlation parameters between each modal feature and the contribution parameters of each modal feature to bridge life prediction. A fused feature vector is generated by combining the modal features according to the dynamic parameters. A spatiotemporally coupled hybrid model is constructed based on an encoder and decoder. The fused feature vector is input into the hybrid model, and the output of the hybrid model is dynamically and adaptively corrected to generate a predicted value for the remaining life of the bridge. A bridge monitoring report is generated based on the raw data and the predicted value. Through multimodal data fusion, the bridge state is comprehensively captured, reducing information loss. The dynamic parameter adjustment mechanism enables the system to respond to environmental changes in real time, reducing prediction bias. The combination of the spatiotemporally coupled hybrid model and dynamic correction accurately reflects the evolution of the bridge state, improving prediction accuracy and enhancing the comprehensiveness, real-time nature, and accuracy of bridge structural life prediction.

[0144] Figure 4 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown.

[0145] It should be noted that the computer system of the electronic device in this embodiment is only an example and should not impose any limitations on the function and scope of use of the embodiments of this application.

[0146] In this embodiment, the computer system includes a central processing unit 401, which can perform various appropriate actions and processes based on a program stored in read-only memory 402 or a program loaded from storage section 408 into random access memory 403, such as executing the deep learning-based bridge life prediction method described in the above embodiment. The random access memory 403 also stores various programs and data required for system operation, thereby realizing big data storage and big data management. The central processing unit 401, read-only memory 402, and random access memory 403 are interconnected via bus 404. Input / output interface 405 is also connected to bus 404.

[0147] The following components are connected to the input / output interface 405: an input section 406 including a keyboard, mouse, etc.; an output section 407 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 408 including a hard disk, etc.; and a communication section 409 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 409 performs communication processing via a network such as the Internet. A drive 410 is also connected to the input / output interface 405 as needed. A removable medium 411, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 410 as needed so that computer programs read from it can be installed into the storage section 408 as needed.

[0148] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including a computer program for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 409, and / or installed from removable medium 411. When the computer program is executed by central processing unit 401, it performs various functions defined in the system of this application.

[0149] It should be noted that the computer-readable medium shown in the embodiments of this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying a computer-readable computer program. The transmitted data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The computer program contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0150] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0151] The units described in the embodiments of this application can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

[0152] According to one aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various alternative implementations described above.

[0153] In another aspect, this application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to implement the deep learning-based bridge life prediction method described in the above embodiments.

[0154] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0155] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the methods according to the embodiments of this application.

[0156] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0157] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for predicting bridge lifespan based on deep learning, characterized in that, include: Obtain the raw data of the bridge structure; The original data is preprocessed, and modal features are extracted from the preprocessed data; Dynamic parameters are determined based on the correlation parameters between modal features and the contribution parameters of each modal feature to bridge life prediction. A fusion feature vector is generated by combining the modal features according to the dynamic parameters. The contribution parameters are calculated in real time based on the difference between the current state and the historical state of each modal feature, and the dynamic parameters are determined by the product of the correlation parameters and the contribution parameters. A spatiotemporally coupled hybrid model is constructed based on an encoder and a decoder. The fused feature vector is input into the hybrid model, and the output of the hybrid model is dynamically and adaptively corrected to generate a predicted value of the bridge's remaining life. Based on the physical relationship between environmental factors and crack propagation, the crack propagation rate is dynamically updated, and the initial prediction value is corrected accordingly. A bridge monitoring report is generated based on the raw data and the predicted values.

2. The method for predicting bridge life based on deep learning according to claim 1, characterized in that, The raw data includes image data, vibration data, acoustic emission data, and environmental data; This includes obtaining the original data of the bridge structure, including: The image data includes crack images captured by the drone during inspection and corrosion images captured by the camera. The vibration data is obtained by acquiring time-domain signals of the bridge structure using a preset accelerometer. The signal generated by the cable breakage event is acquired by an acoustic emission sensor and used as the acoustic emission data. The environmental data is obtained by measuring temperature, humidity, and corrosive gas concentrations using environmental sensors.

3. The method for predicting bridge life based on deep learning according to claim 2, characterized in that, The modal features include image features, frequency band energy distribution features, time-frequency pattern features, and environmental features; The process includes preprocessing the original data and extracting modal features from the preprocessed data, including: Based on the pre-trained super-resolution convolutional neural network model, image data is preprocessed to generate high-resolution images, and image features are extracted from the high-resolution images. The short-time Fourier transform algorithm is used to extract the frequency band energy distribution characteristics in the vibration data; The acoustic emission signal data is input into a pre-trained one-dimensional convolutional neural network model, which outputs time-frequency pattern features. The environmental data is standardized, and environmental features are extracted from the standardized data.

4. The method for predicting bridge life based on deep learning according to claim 1, characterized in that, Dynamic parameters are determined based on the correlation parameters between modal features and the contribution parameters of each modal feature to bridge life prediction. A fused feature vector is then generated by combining the modal features according to the dynamic parameters, including: The modal features are combined to generate a one-dimensional data sequence; The data sequence is input into a multi-layer encoder, and the correlation parameters between different modal features in the data sequence are generated through a self-attention mechanism. Based on the statistical information of the modal features, the contribution parameters of the modal features in lifetime prediction are determined; Dynamic parameters are determined based on the correlation parameters and the contribution parameters, and the data sequence is updated based on the dynamic parameters to generate a fused feature vector.

5. The method for predicting bridge lifespan based on deep learning according to claim 1, characterized in that, A spatiotemporally coupled hybrid model is constructed based on an encoder and decoder. The fused feature vector is input into the hybrid model, and the output of the hybrid model is dynamically and adaptively corrected to generate a predicted value for the remaining life of the bridge, including: A spatiotemporally coupled hybrid model is constructed based on an encoder and a decoder. The fused feature vector is input into the hybrid model, and the initial prediction value is output. The crack propagation rate is updated based on the relationship between environmental factors and crack propagation. The initial prediction is corrected based on the crack propagation rate, and the predicted value of the bridge's remaining life is output.

6. The method for predicting bridge life based on deep learning according to claim 1, characterized in that, Based on the raw data and the predicted values, a bridge monitoring report is generated, including: Obtain a template for a bridge monitoring report; The original data and the predicted values ​​are written into the template of the bridge monitoring report to generate the bridge monitoring report.

7. The method for predicting bridge life based on deep learning according to any one of claims 1-6, characterized in that, Also includes: The predicted value of the bridge's remaining lifespan output by the model is compared with a preset threshold. If the predicted value is less than or equal to the threshold, a corresponding early warning mechanism is triggered.

8. A system for predicting bridge lifespan based on deep learning, characterized in that, include: The acquisition module is used to acquire raw data about the bridge structure. The extraction module is used to preprocess the original data and extract modal features from the preprocessed data; The fusion module is used to determine dynamic parameters based on the correlation parameters between each modal feature and the contribution parameters of each modal feature to the bridge life prediction, and to generate a fused feature vector by combining the modal features according to the dynamic parameters; wherein, the contribution parameters are calculated in real time based on the difference between the current state and the historical state of each modal feature, and the dynamic parameters are determined by the product of the correlation parameters and the contribution parameters. The prediction module is used to construct a spatiotemporally coupled hybrid model based on the encoder and decoder, input the fused feature vector into the hybrid model, and dynamically and adaptively correct the output of the hybrid model to generate a predicted value of the bridge's remaining life. Based on the physical relationship between environmental factors and crack propagation, the crack propagation rate is dynamically updated, and the initial prediction value is corrected accordingly. The reporting module is used to generate a bridge monitoring report based on the raw data and the predicted values.

9. A computer-readable medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method for predicting bridge life based on deep learning as described in any one of claims 1 to 7.

10. An electronic device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the deep learning-based bridge life prediction method as described in any one of claims 1 to 7.

Images