Prestressed concrete pipe pile splicing quality on-line monitoring system and implementation method thereof

Abstract

The invention discloses a prestressed concrete pipe pile splicing quality online monitoring system which comprises a high-definition visual acquisition module, an infrared sensing thermal imaging acquisition module, a welding machine working parameter acquisition module, a data processing module, an algorithm analysis module, a man-machine interaction screen display module and a cloud storage presentation training module. The invention further discloses an implementation method of the prestressed concrete pipe pile splicing quality online monitoring system. Through the high-definition visual acquisition module, the infrared thermal imaging acquisition module and the welding machine working parameter acquisition module, comprehensive monitoring of the welding process is realized; the cloud storage presentation training module receives data from the data processing module through a 5G communication network, efficient storage, display and model iteration training are carried out, quantitative evaluation and diagnosis of the pile splicing quality of the prestressed concrete pipe pile are achieved, the reliability of the welding quality is ensured, and the construction quality of the prestressed concrete pipe pile is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of monitoring and controlling pile foundation mechanical equipment, and in particular relates to a system and method for online monitoring of the quality of prestressed concrete pipe pile connections. Background Art

[0002] Prestressed concrete pipe piles have been widely used in various projects such as construction, railways, highways, bridges, ports and docks due to their wide range of design applicability, stable pile quality, superior seismic performance, high bearing capacity, simple construction and economy. However, due to the limitations of production, transportation and construction, the length of a single prestressed concrete pipe pile usually cannot meet the design requirements at one time, especially when the bearing layer depth at the pile end is large, and it is often necessary to connect the piles on site to achieve the designed pile length. Therefore, the connection method and quality of prestressed concrete pipe piles directly affect the construction quality of the project and the bearing capacity of the pile foundation.

[0003] With the widespread application of prestressed concrete pipe piles in various projects, they face more and more challenges in the design, production and construction process. Prestressed concrete pipe piles are usually prefabricated in factories, and the length of a single pile is generally 6-12 meters. However, when the depth of the bearing layer at the pile end exceeds the length of a single pile, multiple pipe piles need to be connected to meet the design pile length requirements. During the construction process, welding pile connection is widely used in actual projects due to its simple construction and high efficiency. Welding pile connection requires full and continuous welds, and the weld root must be fully penetrated, and must be naturally cooled after welding before continuing to sink the pile. However, the welding quality is easily affected by the welding skills and sense of responsibility of the construction personnel, resulting in quality defects such as cold welding, cracking, and desoldering during the welding process, which in turn poses a hidden danger to the quality of the pipe piles.

[0004] Although welding quality has a crucial impact on the quality of butt-jointed piles, most prestressed concrete pipe piles currently do not adequately monitor welding quality during construction. The lack of effective intelligent monitoring means makes it impossible to accurately monitor welding quality in real time during the pile connection process, and it is also impossible to identify and repair defects in the welding process in a timely manner. In actual construction, welding quality usually relies on the experience and responsibility of on-site operators to control, and there is a lack of effective quality review and remedial measures after the pile is sunk. This monitoring blind spot seriously restricts the construction quality of prestressed concrete pipe piles, which in turn affects the safety and reliability of the entire project. Summary of the invention

[0005] The purpose of the present invention is to provide an online monitoring system for the quality of prestressed concrete pipe pile connections to solve the problems raised in the above-mentioned background technology. The online monitoring system for the quality of prestressed concrete pipe pile connections provided by the present invention has the characteristics of being able to monitor the welding process in real time during the construction of pipe pile connections, accurately evaluate the welding quality, ensure that defects in the welding process are discovered and repaired in time, thereby improving the construction quality of prestressed concrete pipe pile connections and ensuring the overall safety and stability of the project.

[0006] Another object of the present invention is to provide a method for realizing an online monitoring system for the connection quality of prestressed concrete pipe piles.

[0007] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: an online monitoring system for the quality of prestressed concrete pipe pile connection, comprising a high-definition visual acquisition module, an infrared sensing thermal imaging acquisition module, a welding machine working parameter acquisition module, a data processing module, an algorithm analysis module, a human-computer interaction screen display module and a cloud storage presentation training module, wherein the high-definition visual acquisition module is used to collect high-definition continuous frame images of the welding area of ​​two prestressed concrete pipe piles before and after the connection in real time, and extract the characteristic information of the welding area through image processing technology; the infrared sensing thermal imaging acquisition module is used to obtain the surface thermal imaging image of the welding area of ​​two prestressed concrete pipe piles in real time, and extract the temperature information of each pixel point to monitor the temperature anomaly during the welding process; the welding The machine working parameter acquisition module is used to collect the working parameters generated by the welding machine during the welding process; the data processing module is used to fuse and preprocess the data collected by the high-definition visual acquisition module, the infrared sensor thermal imaging camera acquisition module and the welding machine working parameter acquisition module to generate a welding quality data set; the algorithm analysis module uses a deep learning algorithm to analyze the welding quality data set to determine whether the welding quality meets the requirements. When there are defects or abnormalities in the welding, the abnormal information is displayed in the human-computer interaction screen module, and a re-welding decision is made based on the welding defects; the cloud storage presentation training module uses the 5G communication network to receive data from the data processing module for storage, display and model iterative training to achieve continuous optimization of the quality monitoring algorithm.

[0008] Furthermore, in the present invention, the human-computer interaction screen display module is used to input work design information or select work design information that has been saved. The work design information includes but is not limited to the pile diameter of prestressed concrete pipe piles, type of shielding gas, type of welding wire, pile length and number of connected piles.

[0009] Furthermore, in the present invention, the data processing module has a built-in 5G wireless communication unit for uploading the system working data to the cloud storage and presentation training module in real time; the system working data includes a multimodal data set collected by a high-definition visual acquisition module, an infrared sensor thermal imaging acquisition module, and a welding machine working parameter acquisition module, and also includes welding quality assessment results and repair welding decision recommendations calculated by an algorithm analysis module.

[0010] Further in the present invention, the welding machine working parameter acquisition module includes a voltage sensor, a current sensor, a welding wire feeding speed sensor, a gas flow sensor and a welding gun inclination sensor, which are used to monitor various working parameters during the welding process in real time.

[0011] Further in the present invention, a voltage sensor and a current sensor are installed at the welding machine equipment to collect and monitor the changes in voltage and current during welding in real time; a wire feeding speed sensor is installed at the wire outlet of the wire reel to monitor the wire feeding speed of the welding wire during welding and ensure a stable supply of welding wire during welding; a gas flow sensor is installed at the gas outlet of the gas storage bottle to monitor the flow of the protective gas to ensure that the gas supply during welding meets the requirements; a welding gun inclination sensor is installed at the welding gun to accurately monitor the inclination angle of the welding gun nozzle, thereby ensuring that the angle of the welding gun during welding meets the specified process requirements.

[0012] Furthermore, in the present invention, a deep learning model is provided inside the algorithm analysis module, which performs welding quality analysis through multi-modal data fusion to improve the feature expression and combination capabilities of multi-source data; the deep learning model receives and processes pre-processed high-definition images, thermal images and welding machine working parameter data, adopts a multi-input architecture, extracts low-level features through a convolutional neural network with shared weights, and then enhances the correlation and complementarity of each modal feature through a self-attention mechanism; the algorithm analysis module also integrates an attention mechanism to dynamically adjust the feature weights between each modality to ensure that the network effectively pays attention to key features.

[0013] Furthermore, in the present invention, the deep learning model also introduces the Transformer architecture to capture the global dependencies between cross-modal data, and further extracts the graph structure relationship between data through the graph neural network.

[0014] In the present invention, further, a method for implementing the online monitoring system for the quality of prestressed concrete pipe pile connection comprises the following steps:

[0015] S1. Installation initialization: fix the high-definition visual acquisition module and the infrared sensor thermal imaging acquisition module at the prestressed concrete pile machine, adjust the shooting angle so that the acquisition module can accurately align with the welding area of ​​the prestressed concrete pile, and install the welding machine working parameter acquisition module at the welding machine equipment, initialize and configure the data processing module, algorithm analysis module and human-computer interaction screen display module, and ensure that each module is in standby state and ready to start working;

[0016] S2, data acquisition, when the operator is welding the joint of two prestressed concrete pipe piles, the welding machine working parameter acquisition module collects parameters of the welding machine equipment, and the high-definition visual acquisition module and infrared sensor thermal imaging acquisition module collect high-definition images and thermal images of the welding in real time;

[0017] S3, data processing, the data processing module is used to cut, identify and sort the continuous frame images collected by the high-definition camera acquisition module and the infrared sensor thermal imaging camera acquisition module according to the welding time sequence to obtain the basic image sequence of the welding area, and perform position fusion calculation processing based on the relative position information between each high-definition visual acquisition module, infrared sensor thermal imaging acquisition module and the welding area to obtain the overall scene sequence of the welding area, and generate a welding quality data set according to the time sequence combined with the working parameters collected by the welding machine working parameter acquisition module;

[0018] S4. Output of welding quality monitoring results. The algorithm analysis module receives the welding quality data set transmitted by the data processing module, and uses the deep learning algorithm to perform real-time analysis on the welding quality data set to determine whether the welding quality meets the set standards. When there are defects or abnormalities in the welding, the algorithm analysis module will generate and transmit abnormal information to the human-computer interaction screen display module, display the abnormal situation in real time, and propose repair welding decisions based on the type of welding defects;

[0019] S5. Data upload: Through the 5G wireless communication network, the original data obtained by the high-definition visual acquisition module, infrared thermal imaging sensor module, and welding machine working parameter acquisition module, as well as the welding quality data set processed by the data processing module and the welding quality evaluation results generated by the algorithm analysis module are uploaded to the cloud storage presentation training module to realize data storage, display and model iterative training, thereby promoting the continuous optimization of the quality monitoring algorithm.

[0020] In the present invention, further, in step S3, the method for processing the welding quality data set by the data processing module comprises the following steps:

[0021] S31: Before welding, the data processing module pre-processes the high-definition images of the prestressed concrete pipe pile to-be-welded area acquired by the high-definition visual acquisition module at different angles, including filtering, edge recognition, tilt correction, denoising, smoothing and enhancement processing; extracts the gap area to be welded through the trained gap recognition model to be welded, and extracts the alignment degree of the gap, the gap size and the characteristics of the edge impurity degree from the gap, and generates the shape feature value of the area to be welded; at the same time, the data processing module calibrates the contour of the welding area in the thermal imaging image according to the shape feature value of the welding area extracted from the high-definition image, and forms a preset welding area boundary; in the subsequent welding process, the data processing module monitors the temperature area in the thermal imaging image in real time, and determines whether the temperature area is located within the calibrated welding area contour; calculates the spatial overlap or similarity measurement of the two through the image area matching algorithm, so as to quantitatively evaluate whether the welding area is accurately located in the preset welding area;

[0022] S32: During the welding process, the data processing module performs denoising on the thermal image, and uses a histogram equalization algorithm to enhance the global and local contrast of the image and suppress background noise; further uses adaptive binarization and Canny edge detection algorithms to perform image threshold segmentation, and uses a signal modulation algorithm to calculate welding defect characteristic values ​​and welding area contour characteristic values; at the same time, the data processing module generates welding machine working characteristic values ​​based on the current, voltage, welding wire feeding speed, shielding gas flow rate and welding gun angle parameters acquired in real time by the welding machine working parameter acquisition module;

[0023] S33: After welding is completed, the data processing module pre-processes the high-definition image of the welding end area again, performs filtering, edge recognition, tilt correction, denoising, smoothing and enhancement processing, and identifies and extracts weld shape feature values ​​such as weld width, shape and texture.

[0024] In the present invention, further, in step S4, the training method of the deep learning model includes the following steps:

[0025] S41: Preliminary training: collect high-definition welding images, thermal images and welding machine working data with different connection qualities, annotate them, construct a multimodal data set, and divide them into training set and validation set according to a certain ratio. The deep learning model uses the training set for parameter optimization, uses fitting learning to gradually adjust the model parameters, and combines self-supervised learning methods to reduce the reliance on manual annotation. During the model training process, the validation set is used to regularly evaluate the performance of the model, and multiple validation samples are randomly selected to calculate the loss function. Early stopping or cross-validation is used to ensure the stability of the model and avoid overfitting. After reaching the predetermined convergence conditions, the optimal parameter configuration of the preliminary training model is determined.

[0026] S42: Cloud training, where the cloud storage presentation training module receives data uploaded from the on-site human-machine interactive screen display module in real time through the 5G communication network; incremental learning and online learning technologies are used in the cloud to regularly sample and annotate data, and new annotated data is added to the training set; cloud training not only accelerates the model training process through parallel computing, but also combines transfer learning to fine-tune the previously trained model, optimize model parameters and weights, and adapt it to a wider range of practical application scenarios;

[0027] S43: Edge deployment: deploy the updated model to the on-site algorithm analysis module through the 5G communication network to complete the deployment of edge computing. With the advantages of edge computing, the on-site algorithm analysis module can quickly process data locally and output welding quality assessment results in real time.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] 1. The present invention uses a high-definition visual acquisition module, an infrared thermal imaging acquisition module and a welding machine working parameter acquisition module to respectively collect static and dynamic images, thermal imaging data and welding machine process parameters of the welding area to achieve comprehensive monitoring of the welding process;

[0030] 2. The high-definition visual information of the present invention can provide the characteristics of gap alignment, size, impurities, and weld width, shape, texture, etc. before and after welding, providing a standard reference for quality assessment;

[0031] 3. The thermal imaging data of the present invention can record temperature changes in real time, form a thermal map, and provide evidence of thermal changes;

[0032] 4. The present invention can provide real-time process support by comparing images and thermal imaging data, and the welding machine working parameter acquisition module can provide real-time process support;

[0033] 5. The present invention can not only monitor the welding process in real time, but also conduct a comprehensive quality assessment, avoiding single reliance on weld images and improving the accuracy of quality analysis;

[0034] 6. The cloud storage presentation training module of the present invention utilizes the 5G communication network to receive data from the data processing module, and performs efficient storage, display and model iteration training, thereby achieving quantitative evaluation and diagnosis of the quality of prestressed concrete pipe pile connection, ensuring the reliability of welding quality, and improving the construction quality of prestressed concrete pipe piles;

[0035] 7. In the present invention, the operator can conveniently input or call the imported design information through the human-computer interactive screen display module, so as to make corresponding settings and adjustments according to different construction requirements during the welding process;

[0036] 8. The deep learning model of the present invention introduces the Transformer architecture to capture the global dependencies between cross-modal data. Through the self-supervised learning method of training data, the deep learning model can automatically learn the potential rules in the data without manual labeling, thereby improving the robustness and generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a system block diagram of the present invention;

[0038] Figure 2 It is a schematic diagram of the method flow of the present invention;

[0039] Figure 3 A block diagram of a training method for a deep learning model of the present invention;

[0040] Figure 4 It is a structural schematic diagram of the static pile driver in the present invention.

[0041] In the figure: 100, static pile driver; 101, pile driving platform; 102, pile clamping platform; 103, lifting mechanism; 104, long ship; 105, short ship; 106, welding quality monitoring module; 107, pull rope sensor; 108, prestressed concrete pipe pile. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] Example 1

[0044] See also Figure 1-Figure 4The present invention provides the following technical solutions: an online monitoring system for the quality of prestressed concrete pipe pile connection, comprising a high-definition visual acquisition module, an infrared sensing thermal imaging acquisition module, a welding machine working parameter acquisition module, a data processing module, an algorithm analysis module, a human-computer interaction screen display module and a cloud storage presentation training module, wherein the high-definition visual acquisition module is used to collect high-definition continuous frame images of the welding area of ​​two prestressed concrete pipe piles before and after the connection in real time, and extract key feature information of the welding area through image processing technology; the infrared sensing thermal imaging acquisition module is used to obtain the surface thermal imaging image of the welding area of ​​the two prestressed concrete pipe piles in real time, and extract the temperature information of each pixel point to monitor possible temperature anomalies during the welding process; the welding machine working parameter acquisition module is used to collect the welding machine during welding The key working parameters generated in the process include current, voltage, wire feeding speed, shielding gas flow rate and welding gun angle; the data processing module is used to fuse and preprocess the data collected by the high-definition visual acquisition module, the infrared sensor thermal imaging camera acquisition module and the welding machine working parameter acquisition module to generate a welding quality data set; the algorithm analysis module uses a deep learning algorithm to analyze the welding quality data set to determine whether the welding quality meets the requirements. When there are defects or abnormalities in the welding, the abnormal information is displayed in the human-computer interaction screen module, and a repair welding decision is made based on the welding defects; the cloud storage presentation training module uses the 5G communication network to receive data from the data processing module for efficient storage, display and model iterative training, so as to achieve continuous optimization of the quality monitoring algorithm.

[0045] Specifically, the high-definition visual acquisition module, infrared sensor thermal imaging acquisition module and welding machine working parameter acquisition module are connected to the data processing module respectively. The data processing module is connected to the algorithm analysis module for two-way interaction, and is connected to the cloud storage presentation training module through the 5G communication module. The algorithm analysis module is also connected to the human-computer interaction screen display module in a two-way manner.

[0046] Specifically, the human-computer interaction screen display module is used to input work design information or select work design information that has been saved. The work design information includes but is not limited to the diameter of prestressed concrete pipe piles, type of shielding gas, type of welding wire, pile length and number of connected piles; the operator can conveniently input or call the imported design information through this module, so as to make corresponding settings and adjustments according to different construction requirements during the welding process.

[0047] Specifically, the data processing module has a built-in 5G wireless communication unit, which is used to upload the system working data to the cloud storage presentation training module in real time; the system working data includes multimodal data sets collected by the high-definition visual acquisition module, the infrared sensor thermal imaging acquisition module and the welding machine working parameter acquisition module, and also includes the welding quality assessment results and repair welding decision recommendations calculated by the algorithm analysis module; the cloud storage presentation training module can perform more efficient data storage, processing and model iteration training to achieve continuous optimization and accurate assessment of welding quality.

[0048] Specifically, the welding machine working parameter acquisition module includes a voltage sensor, a current sensor, a wire feeding speed sensor, a gas flow sensor and a welding gun inclination sensor, which are used to monitor various key working parameters during the welding process in real time.

[0049] Specifically, the voltage sensor and the current sensor are installed at the power connection of the welding machine to monitor the voltage and current data of the welding machine in real time during the welding process; the wire feeding speed sensor is installed at the wire outlet of the wire reel, and adopts a roller device structure with an encoder. When the welding wire is displaced, the roller rotates accordingly, and the encoder records the number of revolutions of the roller, thereby calculating the displacement of the welding wire, and further converting it into the wire feeding speed of the welding wire during the welding process, thereby ensuring a stable supply of welding wire during the welding process; the gas flow sensor is installed at the gas outlet of the gas storage bottle, and is used to measure the flow rate of the shielding gas, specifically the amount of gas flowing out per unit time. Its accurate measurement of the gas flow rate helps to ensure a stable supply of shielding gas during the welding process, thereby preventing problems such as insufficient or excessive gas during the welding process, and further ensuring the welding quality; the welding gun inclination sensor is installed at the welding gun handle, and measures the inclination angle of the welding gun nozzle during the welding process through the principle of geometric conversion, and monitors the relative angle change between the welding gun and the welding area in real time, to ensure that the operating angle of the welding gun meets the process requirements and avoid welding defects caused by improper angles.

[0050] Specifically, a deep learning model is set up inside the algorithm analysis module, which performs welding quality analysis through multi-modal data fusion to improve the feature expression and combination capabilities of multi-source data; the deep learning model receives and processes pre-processed high-definition images, thermal images and welding machine working parameter data, adopts a multi-input architecture, extracts low-level features through a convolutional neural network (CNN) with shared weights, and then enhances the correlation and complementarity of each modal feature through a self-attention mechanism. The self-attention mechanism allows the network to automatically learn the dependencies between different modal features and dynamically adjust the weight of each feature according to its importance. For each input feature, they are first mapped to a space of the same dimension through a linear transformation, and then the similarity between the query (Query) and the key (Key) is calculated, and the weighted feature representation is obtained through weighted summation; to further improve the performance of the model, the algorithm analysis module also integrates an attention mechanism (SE) to dynamically adjust the feature weights between the modalities to ensure that the network effectively pays attention to key features.

[0051] Specifically, in order to optimize the feature fusion process, the deep learning model also introduced the Transformer architecture to capture the global dependencies between cross-modal data, and further extract the graph structure relationship between data through the graph neural network (GNN); especially in the welding quality assessment process, the interaction between different data sources has an important impact on the final assessment results; the feature vector is further vectorized in the fully connected layer to generate a multi-modal comprehensive feature representation; through the self-supervised learning method of training data, the deep learning model can automatically learn the potential rules in the data without manual labeling, thereby improving the robustness and generalization ability of the model.

[0052] Example 2

[0053] One engineering application background in this embodiment is a residential construction project. The pressure pile used is a prestressed concrete pipe pile (PHC), with a pile diameter of 800mm and a single pile total length of 43m, divided into three pile sections, with pile section lengths of 13m, 15m and 15m respectively. During the construction process, the static pile pressure method is used for pile foundation construction, and the pile section connection method is welding pile connection, and the welding process uses carbon dioxide gas shielded welding. The pile head of the lower pile section should be kept 0.5 to 1.0m above the ground to meet the construction and connection requirements.

[0054] Before construction begins, the operator inputs or selects the saved work design information through the human-computer interaction screen module. This information includes but is not limited to: the pile diameter of the prestressed concrete pipe pile (800mm), the type of protective gas (CO2), the length of the pile section (13m, 15m, 15m), the total pile length (43m) and the number of pile connections (2 times). Through this module, the operator can easily input a new design or call up pre-stored design data. After receiving the design information, the system will initialize the data processing module, the algorithm analysis module and the human-computer interaction screen module to ensure that each module is in standby mode and ready for construction operations at any time.

[0055] After the construction begins, the pile-pressing platform of the static pile driver presses the first 15m long prestressed concrete pipe pile vertically into the soil. When the top of the first pipe pile is lower than the working range of the pile clamping platform, the lifting mechanism on the static pile driver places the second 15m long pipe pile into the pile clamping platform. The operator docks and aligns the bottom of the second pipe pile with the top of the first pipe pile to ensure the connection accuracy of the two pipe piles. The static pile driver applies downward pressure until the connection surface of the two prestressed concrete pipe piles is about 0.8-1.1 meters from the ground. The pile-pressing platform stops applying pressure and enters the pile connection preparation stage.

[0056] Before preparing for welding, the system adjusts the shooting angles of multiple high-definition visual acquisition modules and infrared thermal imaging acquisition modules to align them with the welding area of ​​the prestressed concrete piles, ensuring that all welding viewing angles can be acquired from all angles. The operator checks the welding machine working parameter acquisition module installed on the welding machine equipment to confirm that it is working properly. After completing the inspection, the data processing module, algorithm analysis module and human-computer interaction screen display module are initialized again to prepare for the start of the welding operation.

[0057] During the welding process, the high-definition visual acquisition module starts to collect continuous high-definition frame images of the welding area of ​​the two prestressed concrete pipe piles in real time, and extracts the key feature information of the welding area through image processing technology. When the operator performs the pile welding, the welding machine working parameter acquisition module collects the process parameters of the welding machine such as current, voltage, wire feeding speed, shielding gas flow and welding gun angle in real time. At the same time, the infrared sensor thermal imaging module monitors and collects the thermal imaging image of the welding area in real time, and the high-definition visual acquisition module collects the high-definition image of the weld after welding in real time. The collected image data and welding machine working parameters are processed by the data processing module to generate a welding quality data set. The data set is transmitted to the algorithm analysis module, which uses a deep learning algorithm to analyze the data set in real time to determine whether the welding quality meets the preset standards. When defects or abnormalities are detected in the welding process, the system will display the abnormal information in the human-computer interaction screen display module, and make corresponding repair welding decisions according to the defect type until the human-computer interaction screen shows that the welding is qualified, wait for the weld to cool naturally for 8 minutes, and then continue the pile pressing work.

[0058] When the static pile driver drives the second 15m long prestressed concrete pipe pile into the soil until the top of the pile is below the working range of the clamping platform, the lifting mechanism on the static pile driver puts the third 13m long pipe pile into the clamping platform. The operator docks and aligns the bottom of the third pipe pile with the top of the second pipe pile to ensure the connection accuracy of the two pipe piles. The static pile driver applies downward pressure until the connection surface of the two prestressed concrete pipe piles is about 0.8-1.1 meters from the ground. The above welding monitoring process is repeated to complete the welding work and pass the welding quality inspection. After the weld cools down, the pile driving work continues, and the additional pipe piles are used to assist the pile driving. After the three prestressed concrete pipe piles are driven to the specified depth, the entire construction is completed.

[0059] Example 3

[0060] Furthermore, the method for implementing the online monitoring system for the quality of prestressed concrete pipe pile connection described in the present invention comprises the following steps:

[0061] S1. Installation initialization: fix the high-definition visual acquisition module and the infrared sensor thermal imaging acquisition module at the prestressed concrete pile machine, adjust the shooting angle so that the acquisition module can accurately align with the welding area of ​​the prestressed concrete pile, and install the welding machine working parameter acquisition module at the welding machine equipment, initialize and configure the data processing module, algorithm analysis module and human-computer interaction screen display module, and ensure that each module is in standby state and ready to start working;

[0062] S2, data acquisition, when the operator is welding the joint of two prestressed concrete pipe piles, the welding machine working parameter acquisition module collects parameters of the welding machine equipment, and the high-definition visual acquisition module and infrared sensor thermal imaging acquisition module collect high-definition images and thermal images of the welding in real time;

[0063] S3, data processing, the data processing module is used to cut, identify and sort the continuous frame images collected by the high-definition camera acquisition module and the infrared sensor thermal imaging camera acquisition module according to the welding time sequence to obtain the basic image sequence of the welding area, and perform position fusion calculation processing based on the relative position information between each high-definition visual acquisition module, infrared sensor thermal imaging acquisition module and the welding area to obtain the overall scene sequence of the welding area, and generate a welding quality data set according to the time sequence combined with the working parameters collected by the welding machine working parameter acquisition module;

[0064] S4. Output of welding quality monitoring results. The algorithm analysis module receives the welding quality data set transmitted by the data processing module, and uses the deep learning algorithm to perform real-time analysis on the welding quality data set to determine whether the welding quality meets the set standards. When there are defects or abnormalities in the welding, the algorithm analysis module will generate and transmit abnormal information to the human-computer interaction screen display module, display the abnormal situation in real time, and propose repair welding decisions based on the type of welding defects;

[0065] S5. Data upload: Through the 5G wireless communication network, the original data obtained by the high-definition visual acquisition module, infrared thermal imaging sensor module, and welding machine working parameter acquisition module, as well as the welding quality data set processed by the data processing module and the welding quality evaluation results generated by the algorithm analysis module are uploaded to the cloud storage presentation training module to achieve efficient data storage, display and model iterative training, thereby promoting the continuous optimization of the quality monitoring algorithm.

[0066] Specifically, in step S3, the method for the data processing module to process the welding quality data set includes the following steps:

[0067] S31: Before welding, the data processing module pre-processes the high-definition images of the prestressed concrete pipe pile to-be-welded area acquired by the high-definition visual acquisition module at different angles, including filtering, edge recognition, tilt correction, denoising, smoothing and enhancement processing; extracts the gap area to be welded through the trained gap recognition model to be welded, and extracts the alignment degree of the gap, the gap size and the characteristics of the edge impurity degree from the gap, and generates the shape feature value of the area to be welded; at the same time, the data processing module calibrates the contour of the welding area in the thermal imaging image according to the shape feature value of the welding area extracted from the high-definition image, and forms a preset welding area boundary; in the subsequent welding process, the data processing module monitors the temperature area in the thermal imaging image in real time, and determines whether the temperature area is located within the calibrated welding area contour; calculates the spatial overlap or similarity measurement of the two through the image area matching algorithm, so as to quantitatively evaluate whether the welding area is accurately located in the preset welding area;

[0068] S32: During the welding process, the data processing module performs denoising on the thermal image, and uses a histogram equalization algorithm to enhance the global and local contrast of the image and suppress background noise; further uses adaptive binarization and Canny edge detection algorithms to perform image threshold segmentation, and uses a signal modulation algorithm to calculate welding defect characteristic values ​​and welding area contour characteristic values; at the same time, the data processing module generates welding machine working characteristic values ​​based on the current, voltage, welding wire feeding speed, shielding gas flow rate and welding gun angle parameters acquired in real time by the welding machine working parameter acquisition module;

[0069] S33: After welding is completed, the data processing module pre-processes the high-definition image of the welding end area again, performs filtering, edge recognition, tilt correction, denoising, smoothing and enhancement processing, and identifies and extracts weld shape feature values ​​such as weld width, shape and texture.

[0070] Specifically, the data processing module fuses the shape characteristic values ​​of the area to be welded, the contour characteristic values ​​of the area to be welded, the thermal imaging welding defect characteristic values, the welding area contour characteristic values, the welding machine working characteristic values ​​and the weld shape characteristic values ​​in time sequence to generate a complete welding quality data set.

[0071] Specifically, the filtering step uses a Gaussian filtering algorithm to smooth the image and remove high-frequency noise. The specific formula is as follows:

[0072]

[0073] Among them, I smoothed (x, y) represents the output image, G(i, j) represents the Gaussian filter kernel, and k is the size of the filter kernel; edge recognition uses the Canny edge detection algorithm, and its steps include image gradient calculation, non-maximum suppression and double threshold processing, and finally the edge information of the welding area is obtained; tilt correction uses Hough transform to correct the tilt angle in the image.

[0074] Specifically, the similarity measurement uses the normalized mutual information (NMI) algorithm, and the formula is as follows:

[0075]

[0076] Among them, image A and image B represent the area to be welded and the actual welding area in the thermal imaging image respectively; p(A i , B j ) is the joint probability distribution; H(A) and H(B) are the entropies of image A and image B, respectively; this method can effectively evaluate the accuracy of the welding area by accurately calculating the overlap between the welding area and the preset area, and then determine whether the welding process meets the predetermined requirements.

[0077] Specifically, the histogram equalization algorithm adjusts the distribution of the grayscale of the image to make the brightness distribution of the image more uniform. The specific algorithm steps are: 1. Calculate the grayscale histogram of the image; 2. Calculate the cumulative distribution function of the grayscale; 3. Use CDF to map the original image to generate an equalized image. Further use the adaptive and Canny edge detection algorithm to perform threshold segmentation on the welding thermal image. The adaptive binarization can dynamically select the threshold according to the grayscale histogram of the thermal image and divide the image into two parts: the target and the background. The formula is:

[0078]

[0079] Where T is the threshold value, is the inter-class variance; the grayscale level of the image is [0, L-1]; the Canny edge detection algorithm further extracts the contour of the welding area, and finally combines the signal modulation algorithm to calculate the welding defect characteristic value and the welding area contour characteristic value. At the same time, the data processing module generates the welding machine working characteristic value based on the current, voltage, wire feeding speed, shielding gas flow and welding gun angle working parameters collected in real time by the welding machine working parameter acquisition module; the calculation of the welding working characteristic value can be obtained by the weighted average method, and the formula is as follows:

[0080] f machine =α1I+α2V+α3S+α4F+α5θ;

[0081] Among them, I, V, S, F, and θ represent the current, voltage, wire feeding speed, gas flow rate, and welding angle of the welding machine, respectively, and α1, α2, α3, α4, and α5 are the corresponding weight coefficients.

[0082] Specifically, after welding is completed, the data processing module filters, identifies edges, corrects tilt, denoises, smoothes and enhances the high-definition image of the welding end area again, extracts the weld width, shape, texture and other features to generate weld shape feature values; the weld width can be extracted by the contour extraction algorithm after edge detection, and the geometric parameters of the weld are obtained by using methods such as Hough transform; the weld shape and texture can be extracted by using texture analysis methods (such as grayscale co-occurrence matrix) to extract the texture features of the weld surface, and the formula is as follows:

[0083]

[0084] Among them, p(x, y) is the gray level co-occurrence matrix of the high-definition image; (i, j) is the moving distance, which is used to calculate the texture features of the weld.

[0085] Specifically, in step S4, the training method of the deep learning model includes the following steps:

[0086] S41: Preliminary training: collect high-definition welding images, thermal images and welding machine working data with different connection qualities, annotate them, construct a multimodal data set, and divide them into training set and validation set according to a certain ratio. The deep learning model uses the training set for parameter optimization, uses fitting learning to gradually adjust the model parameters, and combines self-supervised learning methods to reduce the reliance on manual annotation. During the model training process, the validation set is used to regularly evaluate the performance of the model, and multiple validation samples are randomly selected to calculate the loss function. Technical means such as early stopping or cross-validation are used to ensure the stability of the model and avoid overfitting. After reaching the predetermined convergence conditions, the optimal parameter configuration of the preliminary training model is determined.

[0087] S42: Cloud training, where the cloud storage presentation training module receives data uploaded from the on-site human-machine interactive screen display module in real time through the 5G communication network; incremental learning and online learning technologies are used in the cloud to regularly sample and annotate data, and new annotated data is added to the training set; cloud training not only accelerates the model training process through parallel computing, but also combines transfer learning to fine-tune the previously trained model, optimize model parameters and weights, and adapt it to a wider range of practical application scenarios;

[0088] S43: Edge deployment: deploy the updated model to the on-site algorithm analysis module through the 5G communication network to complete the deployment of edge computing. With the advantages of edge computing, the on-site algorithm analysis module can quickly process data locally and output welding quality assessment results in real time.

[0089] In summary, the present invention uses a high-definition visual acquisition module, an infrared thermal imaging acquisition module and a welding machine working parameter acquisition module to respectively collect static and dynamic images, thermal imaging data and welding machine process parameters of the welding area to achieve comprehensive monitoring of the welding process; the high-definition visual information of the present invention can provide the alignment, size, impurities and weld width, shape, texture and other characteristics of the gap before and after welding, providing a standard reference for quality assessment; the thermal imaging data of the present invention can record temperature changes in real time, form a thermal map, and provide evidence of thermal changes; by comparing images and thermal imaging data, the welding machine working parameter acquisition module can provide process support in real time; the present invention can not only monitor the welding process in real time, but also conduct a comprehensive quality assessment, avoid relying solely on weld images, and improve the accuracy of quality analysis; the cloud storage of the present invention The storage presentation training module utilizes the 5G communication network to receive data from the data processing module for efficient storage, display and model iteration training, thereby achieving quantitative evaluation and diagnosis of the quality of prestressed concrete pipe pile connections, ensuring the reliability of welding quality, and improving the construction quality of prestressed concrete pipe piles. In the present invention, operators can conveniently input or call imported design information through the human-computer interaction screen display module so as to make corresponding settings and adjustments according to different construction requirements during the welding process. The deep learning model of the present invention introduces the Transformer architecture to capture the global dependencies between cross-modal data. Through the self-supervised learning method of training data, the deep learning model can automatically learn the potential rules in the data without manual labeling, thereby improving the robustness and generalization ability of the model.

[0090] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An online monitoring system for the quality of prestressed concrete pipe pile connections, characterized in that: It includes high-definition visual acquisition module, infrared sensor thermal imaging acquisition module, welding machine working parameter acquisition module, data processing module, algorithm analysis module, human-computer interaction screen display module and cloud storage presentation training module, among which, The high-definition visual acquisition module is used to collect high-definition continuous frame images of the welding area of ​​two prestressed concrete pipe piles before and after the piles are connected in real time, and extract the characteristic information of the welding area through image processing technology; The infrared sensing thermal imaging acquisition module is used to obtain the surface thermal imaging images of the welding area of ​​two prestressed concrete pipe piles in real time and extract the temperature information of each pixel point to monitor the temperature anomaly during the welding process; The welding machine working parameter acquisition module is used to collect the working parameters generated by the welding machine during the welding process; The data processing module is used to fuse and preprocess the data collected by the high-definition visual acquisition module, the infrared sensor thermal imaging camera acquisition module and the welding machine working parameter acquisition module to generate a welding quality data set; The algorithm analysis module uses a deep learning algorithm to analyze the welding quality data set to determine whether the welding quality meets the requirements. When there are defects or abnormalities in the welding, the abnormal information is displayed in the human-computer interaction screen display module, and repair welding decisions are made based on the welding defects. The cloud storage presentation training module uses the 5G communication network to receive data from the data processing module for storage, display and model iterative training to achieve continuous optimization of the quality monitoring algorithm.

2. The online monitoring system for prestressed concrete pipe pile connection quality according to claim 1 is characterized by: The human-computer interaction screen display module is used to input work design information or select work design information that has been saved. The work design information includes but is not limited to the diameter of the prestressed concrete pipe pile, the type of shielding gas, the type of welding wire, the pile length and the number of connected piles.

3. The online monitoring system for prestressed concrete pipe pile connection quality according to claim 1 is characterized by: The data processing module has a built-in 5G wireless communication unit, which is used to upload the system working data to the cloud storage and presentation training module in real time; the system working data includes a multimodal data set collected by a high-definition visual acquisition module, an infrared sensor thermal imaging acquisition module, and a welding machine working parameter acquisition module, and also includes welding quality assessment results and repair welding decision recommendations calculated by an algorithm analysis module.

4. The online monitoring system for prestressed concrete pipe pile connection quality according to claim 1 is characterized by: The welding machine working parameter acquisition module includes a voltage sensor, a current sensor, a welding wire feeding speed sensor, a gas flow sensor and a welding gun inclination sensor, which are used to monitor various working parameters during the welding process in real time.

5. The online monitoring system for prestressed concrete pipe pile connection quality according to claim 4 is characterized by: The voltage sensor and current sensor are installed at the welding machine equipment to collect and monitor the changes in voltage and current during the welding process in real time; the wire feeding speed sensor is installed at the wire outlet of the wire reel to monitor the wire feeding speed of the welding process and ensure the stable supply of welding wire during the welding process; the gas flow sensor is installed at the gas outlet of the gas storage bottle to monitor the flow of the protective gas to ensure that the gas supply during the welding process meets the requirements; the welding gun inclination sensor is installed at the welding gun to accurately monitor the inclination angle of the welding gun nozzle, so as to ensure that the angle of the welding gun during welding meets the specified process requirements.

6. The online monitoring system for prestressed concrete pipe pile connection quality according to claim 1 is characterized by: The algorithm analysis module is internally provided with a deep learning model, which performs welding quality analysis through multi-modal data fusion to improve the feature expression and combination capabilities of multi-source data; the deep learning model receives and processes pre-processed high-definition images, thermal images and welding machine working parameter data, adopts a multi-input architecture, extracts low-level features through a convolutional neural network with shared weights, and then enhances the correlation and complementarity of each modal feature through a self-attention mechanism; the algorithm analysis module also integrates an attention mechanism to dynamically adjust the feature weights between each modality to ensure that the network effectively pays attention to key features.

7. The online monitoring system for prestressed concrete pipe pile connection quality according to claim 6 is characterized by: The deep learning model also introduces the Transformer architecture to capture the global dependencies between cross-modal data, and further extracts the graph structure relationships between data through graph neural networks.

8. A method for implementing an online monitoring system for prestressed concrete pipe pile connection quality according to any one of claims 1 to 7, characterized in that: The following steps are involved: S1. Installation initialization: fix the high-definition visual acquisition module and the infrared sensor thermal imaging acquisition module at the prestressed concrete pile machine, adjust the shooting angle so that the acquisition module can accurately align with the welding area of ​​the prestressed concrete pile, and install the welding machine working parameter acquisition module at the welding machine equipment, initialize and configure the data processing module, algorithm analysis module and human-computer interaction screen display module, and ensure that each module is in standby state and ready to start working; S2, data acquisition, when the operator is welding the joint of two prestressed concrete pipe piles, the welding machine working parameter acquisition module collects parameters of the welding machine equipment, and the high-definition visual acquisition module and infrared sensor thermal imaging acquisition module collect high-definition images and thermal images of the welding in real time; S3, data processing, the data processing module is used to cut, identify and sort the continuous frame images collected by the high-definition camera acquisition module and the infrared sensor thermal imaging camera acquisition module according to the welding time sequence to obtain the basic image sequence of the welding area, and perform position fusion calculation processing based on the relative position information between each high-definition visual acquisition module, infrared sensor thermal imaging acquisition module and the welding area to obtain the overall scene sequence of the welding area, and generate a welding quality data set according to the time sequence combined with the working parameters collected by the welding machine working parameter acquisition module; S4. Output of welding quality monitoring results. The algorithm analysis module receives the welding quality data set transmitted by the data processing module, and uses the deep learning algorithm to perform real-time analysis on the welding quality data set to determine whether the welding quality meets the set standards. When there are defects or abnormalities in the welding, the algorithm analysis module will generate and transmit abnormal information to the human-computer interaction screen display module, display the abnormal situation in real time, and propose repair welding decisions based on the type of welding defects; S5. Data upload: Through the 5G wireless communication network, the original data obtained by the high-definition visual acquisition module, infrared thermal imaging sensor module, and welding machine working parameter acquisition module, as well as the welding quality data set processed by the data processing module and the welding quality evaluation results generated by the algorithm analysis module are uploaded to the cloud storage presentation training module to realize data storage, display and model iterative training, thereby promoting the continuous optimization of the quality monitoring algorithm.

9. The method for realizing an online monitoring system for prestressed concrete pipe pile connection quality according to claim 8, characterized in that: In step S3, the method for processing the welding quality data set by the data processing module comprises the following steps: S31: Before welding, the data processing module pre-processes the high-definition images of the prestressed concrete pipe pile to-be-welded area acquired by the high-definition visual acquisition module at different angles, including filtering, edge recognition, tilt correction, denoising, smoothing and enhancement processing; extracts the gap area to be welded through the trained gap recognition model to be welded, and extracts the alignment degree of the gap, the gap size and the characteristics of the edge impurity degree from the gap, and generates the shape feature value of the area to be welded; at the same time, the data processing module calibrates the contour of the welding area in the thermal imaging image according to the shape feature value of the welding area extracted from the high-definition image, and forms a preset welding area boundary; in the subsequent welding process, the data processing module monitors the temperature area in the thermal imaging image in real time, and determines whether the temperature area is located within the calibrated welding area contour; calculates the spatial overlap or similarity measurement of the two through the image area matching algorithm, so as to quantitatively evaluate whether the welding area is accurately located in the preset welding area; S32: During the welding process, the data processing module performs denoising on the thermal image, and uses a histogram equalization algorithm to enhance the global and local contrast of the image and suppress background noise; further uses adaptive binarization and Canny edge detection algorithms to perform image threshold segmentation, and uses a signal modulation algorithm to calculate welding defect characteristic values ​​and welding area contour characteristic values; at the same time, the data processing module generates welding machine working characteristic values ​​based on the current, voltage, welding wire feeding speed, shielding gas flow rate and welding gun angle parameters acquired in real time by the welding machine working parameter acquisition module; S33: After welding is completed, the data processing module pre-processes the high-definition image of the welding end area again, performs filtering, edge recognition, tilt correction, denoising, smoothing and enhancement processing, and identifies and extracts weld shape feature values ​​such as weld width, shape and texture.

10. The method for realizing an online monitoring system for prestressed concrete pipe pile connection quality according to claim 8, characterized in that: In step S4, the training method of the deep learning model includes the following steps: S41: Preliminary training: collect high-definition welding images, thermal images and welding machine working data with different connection qualities, annotate them, construct a multimodal data set, and divide them into training set and validation set according to a certain ratio. The deep learning model uses the training set for parameter optimization, uses fitting learning to gradually adjust the model parameters, and combines self-supervised learning methods to reduce the reliance on manual annotation. During the model training process, the validation set is used to regularly evaluate the performance of the model, and multiple validation samples are randomly selected to calculate the loss function. Early stopping or cross-validation methods are used to ensure the stability of the model and avoid overfitting. After reaching the predetermined convergence conditions, the optimal parameter configuration of the preliminary training model is determined. S42: Cloud training, where the cloud storage presentation training module receives data uploaded from the on-site human-machine interactive screen display module in real time through the 5G communication network; incremental learning and online learning technologies are used in the cloud to regularly sample and annotate data, and new annotated data is added to the training set; cloud training not only accelerates the model training process through parallel computing, but also combines transfer learning to fine-tune the previously trained model, optimize model parameters and weights, and adapt it to a wider range of practical application scenarios; S43: Edge deployment: deploy the updated model to the on-site algorithm analysis module through the 5G communication network to complete the deployment of edge computing. With the advantages of edge computing, the on-site algorithm analysis module can quickly process data locally and output welding quality assessment results in real time.

Images