An underwater high-pressure dry welding process weld forming quality evaluation method, welding process parameter optimization method and device for Q345 steel plate fillet weld
By combining experimental and numerical simulation methods, a BP neural network model was constructed, which solved the reliance on human experience in underwater high-pressure dry welding, and realized efficient quality evaluation and parameter optimization of Q345 steel plate fillet welds, thereby improving the controllability and efficiency of welding quality.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- DALIAN MARITIME UNIVERSITY
- Filing Date
- 2026-01-23
- Publication Date
- 2026-06-09
AI Technical Summary
In the process of underwater high-pressure dry welding of Q345 steel plate fillet welds, there is a lack of process parameter optimization and forming quality evaluation methods that can get rid of dependence on manual experience, improve quality controllability and work efficiency, and existing technologies are difficult to meet the needs of on-site real-time decision-making.
A combination of experimental and numerical simulation methods was adopted. By conducting experiments with different welding process parameters on Q345 steel test plates, an initial sample library was established. The data was expanded using a welding numerical simulation model, a BP neural network was trained, a weld formation quality evaluation and prediction model was constructed, and a human-computer interaction APP was embedded for real-time evaluation and parameter optimization.
It enables accurate prediction of weld formation quality and rapid optimization of process parameters, improves the controllability of welding quality and operational efficiency, reduces reliance on manual experience, and ensures the stability of welding quality.
Smart Images
Figure CN122165083A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of welding technology, and more specifically, to a method for evaluating weld formation quality, a method for optimizing welding process parameters, and an apparatus for underwater high-pressure dry welding of fillet welds in Q345 steel plates. Background Technology
[0002] With the continuous increase in global maritime transport volume, maritime accidents such as ship collisions and groundings are frequent, resulting in a large number of sunken ships and marine structures urgently needing salvage and repair. If structural connections cannot be completed efficiently and reliably in an underwater environment, it will have long-term adverse effects on marine ecology and the shipping economy. Therefore, underwater welding technology has become a key supporting means for marine engineering rescue and maintenance.
[0003] Underwater welding is mainly classified into three categories according to environmental conditions: wet welding, partial dry welding, and dry welding. Among them, underwater high-pressure dry welding constructs a sealed dry chamber in the area to be welded, isolating the welding zone from seawater. This allows the weld joint performance to approach that of land-based welding, significantly reducing the diving depth, environmental pressure, and safety risks. It also has the advantages of flexibility, strong adaptability, and relatively low cost, and has been widely used in high-value-added scenarios such as ship damage rescue, underwater shipwreck and debris salvage, and emergency repair of offshore platforms.
[0004] Inside dry compartments, welding operations are typically performed by automated or remotely controlled welding robots. For low-alloy high-strength steels like Q345, fillet welds are the most common joint type connecting the hull frame, reinforcing ribs, and deck, and their forming quality directly determines the load-bearing capacity and service safety of the repaired structure. However, underwater high-pressure dry welding has a narrow process window: process parameters such as welding heat input, welding speed, and weld leg size are strongly coupled with environmental factors such as compartment pressure, humidity, and cooling rate. Any slight fluctuation can lead to defects such as incomplete fusion, undercut, excessive angular deformation, or residual stress concentration. Currently, on-site parameter adjustments still mainly rely on the personal experience of underwater welders or technicians. However, highly skilled underwater welding technicians are scarce, parameter settings are highly subjective, and quality stability significantly decreases once personnel changes occur.
[0005] To reduce reliance on on-site welding tests, numerical simulation technology has been introduced into underwater welding research in recent years. By establishing a thermo-mechanical coupled finite element model, the temperature field, residual stress, and angular deformation distribution can be predicted on a computer, thus partially replacing physical testing. However, relying solely on simulation for a single calculation is time-consuming and cannot meet the needs of real-time decision-making on-site.
[0006] Furthermore, artificial intelligence methods, particularly backpropagation (BP) neural networks, have been reported in the field of welding quality prediction, enabling nonlinear mapping between input and output. However, existing research is mostly based on conventional butt or corner joints on land, with limited training datasets and a lack of coupling with the environmental variables unique to underwater high-pressure dry welding. This results in insufficient model generalization ability, making it difficult to directly transfer to the underwater fillet weld scenario for Q345 steel.
[0007] Therefore, in terms of optimizing process parameters and evaluating forming quality for underwater high-pressure dry welding of Q345 steel plate fillet welds, there is a lack of an integrated solution that can reduce over-reliance on manual experience and improve the controllability and efficiency of underwater welding quality. Summary of the Invention
[0008] In view of the shortcomings of the existing technology, this application provides a method for evaluating the weld formation quality, a method for optimizing welding process parameters, and an apparatus for underwater high-pressure dry welding of fillet welds in Q345 steel plates. This invention uses a small number of real experiments to calibrate numerical simulations, then uses simulated big data to train a neural network, and finally achieves real-time evaluation of welding quality and rapid verification and optimization of welding process parameters through application encapsulation.
[0009] The technical means employed in this invention are as follows: On one hand, this invention discloses a method for evaluating the weld formation quality of underwater high-pressure dry welding process for fillet welds of Q345 steel plates, including the following steps: Fillet weld tests with different welding process parameters were carried out on Q345 steel test plates. The residual stress and penetration depth of different welds were obtained based on the test results and used as weld quality evaluation indicators to establish an initial process parameter-welding quality sample library. The process parameters and welding quality evaluation index data in the initial process parameter-welding quality sample library are substituted into the welding numerical simulation model for accuracy verification. After the verification is qualified, the process parameters and welding quality evaluation index data are expanded and sampled using the welding numerical simulation model to generate the expanded process parameter-welding quality sample library. Using the process parameters in the expanded process parameter-welding quality sample library as input and the weld quality evaluation index as output, a BP neural network is trained to obtain a weld formation quality evaluation prediction model. The weld formation quality evaluation and prediction model is used to evaluate and predict the weld formation quality based on the process parameters to be evaluated.
[0010] Furthermore, the method also includes: The weld formation quality evaluation and prediction model is embedded into a human-computer interaction APP. The welding process parameters are input into the APP in real time, and the corresponding residual stress and penetration depth prediction values are output.
[0011] Furthermore, the welding process parameters include weld leg size, heat input, and welding speed.
[0012] Furthermore, the welding numerical simulation model is a thermo-mechanical coupled finite element model, used to calculate the temperature field, residual stress, and weld penetration distribution.
[0013] On the one hand, this invention also discloses a method for optimizing underwater high-pressure dry welding process parameters for fillet welds of Q345 steel plates, comprising the following steps: Input the welding process parameters to be optimized into the weld formation quality evaluation and prediction model constructed above to obtain the corresponding weld quality evaluation index. Determine whether the obtained weld quality evaluation index meets the preset acceptable range. If it is within the acceptable range, the optimization process ends. Otherwise, readjust the input welding process parameters until the obtained new weld quality evaluation index is acceptable, then the optimization process ends.
[0014] On one hand, the present invention also discloses a weld formation quality evaluation device for underwater high-pressure dry welding process of Q345 steel plate fillet welds, used to implement the above-mentioned underwater high-pressure dry welding process of Q345 steel plate fillet weld quality evaluation method, including: The initial sample acquisition unit is used to carry out fillet weld tests with different welding process parameters on Q345 steel test plates. Based on the test results, the measured values of residual stress and penetration depth of different welds are obtained as weld quality evaluation indicators, and an initial process parameter-welding quality sample library is established. The expanded sample acquisition unit is used to substitute the process parameters and welding quality evaluation index data in the initial process parameter-welding quality sample library into the welding numerical simulation model for accuracy verification. After verification, the welding numerical simulation model is used to expand the sampling of process parameters and welding quality evaluation index data to generate an expanded process parameter-welding quality sample library. The model training unit is used to train a BP neural network with the process parameters in the expanded process parameter-welding quality sample library as input and the weld quality evaluation index as output, so as to obtain a weld formation quality evaluation prediction model. The quality evaluation unit is used to evaluate and predict the quality of weld formation based on the weld formation quality evaluation and prediction model.
[0015] On the other hand, the present invention also discloses an underwater high-pressure dry welding process parameter optimization device for Q345 steel plate fillet welds, used to implement the above-mentioned underwater high-pressure dry welding process parameter optimization method for Q345 steel plate fillet welds, including: The quality evaluation unit is used to input the welding process parameters to be optimized into the weld formation quality evaluation and prediction model constructed in claim 1, and obtain the corresponding weld quality evaluation index. The parameter optimization unit is used to determine whether the obtained weld quality evaluation index meets the preset qualified range. If it is within the qualified range, the optimization process ends. Otherwise, the input welding process parameters are readjusted until the new weld quality evaluation index is qualified, and then the optimization process ends.
[0016] Compared with the prior art, the present invention has the following advantages: 1. This invention employs a combination of experimental and numerical simulation methods. First, welding experiments are designed to test the weld blocks under different welding parameters. Various weld seam information is obtained from the resulting weld blocks, and the variation law of process parameters affecting weld quality is analyzed. The results are then compared with numerical simulations to prove their accuracy. The numerical simulation system is used to expand the input set of weld seam process parameters to obtain a neural network input set. Finally, a BP neural network model is developed in MATLAB, and the parameter set is used as the training set to train the model. This model is then integrated with APP DESIGNER to form a human-computer interface APP prediction system.
[0017] 2. Before predicting the process parameters of the welding robot, the welding process parameters can be preset using the weld quality evaluation system designed in this invention. Based on the corresponding working conditions, the weld leg size that meets the actual welding requirements is determined. The weld leg size, preset values of welding process parameters, etc., are used as system input values. Through model prediction, the corresponding weld penetration depth, residual stress, and other predicted results are obtained, thus acquiring accurate predicted values.
[0018] 3. This invention has a one-click training function, which can perform group training on welds under different welding conditions in real time, making it convenient to set different welding process parameters for different welding needs.
[0019] This invention offers high prediction accuracy, effectively saving time in presetting welding parameters, providing technical support for new salvage methods, offering a reasonable design method, and ensuring project schedule. Attached Figure Description
[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0021] Figure 1This is a flowchart illustrating the process of evaluating the weld formation quality of an underwater high-pressure dry welding process for fillet welds of Q345 steel plates according to the present invention.
[0022] Figure 2 This is the system architecture for evaluating the weld formation quality of underwater high-pressure dry welding process for fillet welds of Q345 steel plates, as described in this embodiment of the invention.
[0023] Figure 3 This is a schematic diagram of the human-computer interaction APP operation interface for the embedded weld formation quality evaluation and prediction model in an embodiment of the present invention. Detailed Implementation
[0024] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.
[0025] Example 1 like Figure 1 As shown, this invention provides a method for evaluating the weld formation quality of underwater high-pressure dry welding process for fillet welds of Q345 steel plates, comprising the following steps: S1. Perform fillet weld tests with different welding process parameters on Q345 steel test plates. Obtain the measured values of residual stress and penetration depth of different welds based on the test results, and use them as weld quality evaluation indicators to establish an initial process parameter-welding quality sample library. Preferably, the welding process parameters include weld leg size, heat input, and welding speed.
[0026] like Figure 2 The diagram shows the experimental system built in this embodiment. The welding method used in the experiment was CO2 gas shielded welding, and flux-cored welding wire was used. The welding machine used was a Finnish Kemppi welding machine. By adjusting the wire feed speed knob on the welding machine, the corresponding welding current and voltage were indirectly changed. After welding, a cutting machine was used to cut and sample the weld at a better point. The weld formation area was then ground and polished with sandpaper of different grits (200-5000). A VHX-7000 digital microscope was used to observe the microstructure, and the weld penetration, weld width, and weld leg dimensions were marked on the weld pool using a dimensioning tool. The welding process parameters for the better weld under three sets of weld leg dimensions were selected based on the weld cross-sectional formation. Here, the line energy was obtained according to an existing line energy formula.
[0027] S2. Substitute the process parameters and welding quality evaluation index data from the initial process parameter—welding quality sample library into the welding numerical simulation model for accuracy verification. The accuracy is verified by comparing the molten pool size obtained from the experiment with that obtained from the welding numerical simulation. After successful verification, the process parameters and welding quality evaluation index data are expanded using the welding numerical simulation model to generate an expanded process parameter—welding quality sample library.
[0028] Welding process parameters for three sets of welds with optimal weld leg sizes were selected and applied to the welding simulation software SYSWELD. The temperature and stress field distributions of the resulting welds were analyzed, and the weld formation quality under different weld leg sizes was summarized. The accuracy of the numerical simulation results was verified by comparing the post-weld pool size obtained from the SYSWELD numerical simulation software with the weld pool size in the welding experiment. During the numerical simulation, the welding line energy needs to be set; the line energy formula is as follows:
[0029] Where: E is the welding line energy (J / mm); I is the welding current (A); U is the welding voltage (V); V is the welding speed (mm / s); This is a constant term, usually taken as 0.75; This study summarizes the influence of different process parameters on weld quality. Based on the experimental results of each group of weld leg sizes, the set of welding heat input parameters is expanded, the interval between heat inputs is shortened, and extensive simulations are performed in welding numerical simulation software to analyze the corresponding weld parameters such as residual stress and molten pool size. The process parameter input set is extended and generalized, and the welding numerical simulation results are expanded and extensively analyzed using numerical simulation software to construct an expanded process parameter-weld quality sample library.
[0030] S3. Using the process parameters in the expanded process parameter-welding quality sample library as input and the weld quality evaluation index as output, train a BP neural network to obtain a weld formation quality evaluation prediction model.
[0031] To train a BP neural network model using MATLAB, the welding parameter set is first organized. Different parameters often have different values, so the welding parameters are normalized to obtain values within the same range. The normalized input set is then used as the input set for the BP neural network. First, the hidden layers of the neural network need to be selected. The number of hidden layers can be set in the training software, and the error magnitude is observed. Through multiple comparisons of the hidden layer errors, a hidden layer count of 6 was selected, resulting in an error of 0.017. Model training is then completed based on this. Predictive analysis is performed within the model. The input set is divided into a training set and a prediction set, and the model's accuracy is verified by comparison. The obtained model can be used to obtain the predicted results. Inverse normalization of the predicted values yields the weld formation quality parameters.
[0032] S4. Based on the weld formation quality evaluation and prediction model, perform weld formation quality evaluation and prediction on the process parameters to be evaluated.
[0033] Furthermore, the method also includes: S5. Embed the weld formation quality evaluation and prediction model into the human-computer interaction APP, and input the welding process parameters to be welded into the APP in real time, that is, output the corresponding residual stress and penetration depth prediction values.
[0034] In MATLAB, the App Designer toolbox is used for human-computer system design. Interactive pages are designed, including drag-and-drop buttons, text boxes, tables, charts, and other components. Input boxes receive predicted features, and buttons trigger model training and prediction. Finally, the backpropagation (BP) model logic is linked in the App's callback functions. The "Train" button callback calls preprocessing and training code, while the "Predict" button callback loads the trained network and outputs the results. Simultaneously, the interface components provide real-time feedback on training progress, error values, and other key information. Users input the welding process parameters (weld leg size, heat input, and welding speed) on the window. Clicking the "Predict" button allows for result prediction. First, the input process parameters (weld leg size, heat input, and welding speed) are read and arrayed into x2. Then, the trained neural network model app.trained_net is called, x2 is transposed, and prediction is performed to obtain y2. Finally, the prediction results are assigned to the corresponding display components: app.ResidualStress.Value displays the predicted residual stress, and app.depth.Value displays the predicted weld depth, thus completing the process from process parameter input to forming index prediction. The specific display interface is shown below. Figure 3 As shown.
[0035] The specific process for quality evaluation of the APP constructed through this embodiment includes the following steps: 1. Select a TXT format data file using uigetfile.
[0036] 2. Read the file path and display it in the txt_name input box.
[0037] 3. Use importdata to read the file content, take the first 3 columns as input features (x1), and the last 2 columns as target output (y1).
[0038] 4. Call the previously defined train_fitting_network function to train the neural network, store the trained model in trained_net, and store the training results in results; 5. Users can input the welding process parameters for this welding operation on the window: weld leg size, heat input, and welding speed.
[0039] 6. Clicking the prediction button will predict the outcome of this welding operation. First, the input process parameters (weld leg size, line energy, welding speed) are read and formed into an array x2; then, the trained neural network model app.trained_net is called, and after transposing x2, the prediction is performed to obtain y2.
[0040] 7. Assign the prediction results to the corresponding display components, where app.ResidualStress.Value displays the predicted residual stress and app.depth.Value displays the predicted penetration depth.
[0041] Example 2 Based on the evaluation method disclosed in Example 1, this example discloses a method for optimizing underwater high-pressure dry welding process parameters for fillet welds of Q345 steel plates, including the following steps: A1. Input the welding process parameters to be optimized into the weld formation quality evaluation and prediction model constructed above, and obtain the corresponding weld quality evaluation index. A2. Determine whether the obtained weld quality evaluation index meets the pre-set qualified range. If it is within the qualified range, the optimization process ends. Otherwise, readjust the input welding process parameters until the new weld quality evaluation index is qualified, then the optimization process ends.
[0042] For specific examples in this embodiment, please refer to the examples described in Embodiment 1. This embodiment will not repeat them here.
[0043] Example 3 Based on the evaluation method disclosed in Example 1, this example also discloses a weld formation quality evaluation device for underwater high-pressure dry welding process of Q345 steel plate fillet welds, used to implement the quality evaluation method in Example 1, including: The initial sample acquisition unit is used to carry out fillet weld tests with different welding process parameters on Q345 steel test plates. Based on the test results, the measured values of residual stress and penetration depth of different welds are obtained as weld quality evaluation indicators, and an initial process parameter-welding quality sample library is established. The expanded sample acquisition unit is used to substitute the process parameters and welding quality evaluation index data in the initial process parameter-welding quality sample library into the welding numerical simulation model for accuracy verification. After verification, the welding numerical simulation model is used to expand the sampling of process parameters and welding quality evaluation index data to generate an expanded process parameter-welding quality sample library. The model training unit is used to train a BP neural network with the process parameters in the expanded process parameter-welding quality sample library as input and the weld quality evaluation index as output, so as to obtain a weld formation quality evaluation prediction model. The quality evaluation unit is used to evaluate and predict the quality of weld formation based on the weld formation quality evaluation and prediction model.
[0044] For specific examples in this embodiment, please refer to the examples described in Embodiment 1. This embodiment will not repeat them here.
[0045] Example 4 Based on the evaluation method disclosed in Example 2, this example also discloses an underwater high-pressure dry welding process parameter optimization device for Q345 steel plate fillet welds, used in the welding process parameter optimization method of Example 2, including: The quality evaluation unit is used to input the welding process parameters to be optimized into the weld formation quality evaluation and prediction model constructed in claim 1, and obtain the corresponding weld quality evaluation index. The parameter optimization unit is used to determine whether the obtained weld quality evaluation index meets the preset qualified range. If it is within the qualified range, the optimization process ends. Otherwise, the input welding process parameters are readjusted until the new weld quality evaluation index is qualified, and then the optimization process ends.
[0046] For specific examples in this embodiment, please refer to the examples described in Embodiment 2. This embodiment will not repeat them here.
[0047] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for evaluating the weld formation quality of underwater high-pressure dry welding process for fillet welds of Q345 steel plates, characterized in that, Includes the following steps: Fillet weld tests with different welding process parameters were carried out on Q345 steel test plates. The residual stress and penetration depth of different welds were obtained based on the test results and used as weld quality evaluation indicators to establish an initial process parameter-welding quality sample library. The process parameters and welding quality evaluation index data in the initial process parameter-welding quality sample library are substituted into the welding numerical simulation model for accuracy verification. After the verification is qualified, the process parameters and welding quality evaluation index data are expanded and sampled using the welding numerical simulation model to generate the expanded process parameter-welding quality sample library. Using the process parameters in the expanded process parameter-welding quality sample library as input and the weld quality evaluation index as output, a BP neural network is trained to obtain a weld formation quality evaluation prediction model. The weld formation quality evaluation and prediction model is used to evaluate and predict the weld formation quality based on the process parameters to be evaluated.
2. The method for evaluating the weld formation quality of underwater high-pressure dry welding process for fillet welds of Q345 steel plates according to claim 1, characterized in that, The method further includes: The weld formation quality evaluation and prediction model is embedded into a human-computer interaction APP. The welding process parameters are input into the APP in real time, and the corresponding residual stress and penetration depth prediction values are output.
3. The method for evaluating the weld formation quality of underwater high-pressure dry welding process for fillet welds of Q345 steel plates according to claim 1, characterized in that, The welding process parameters include weld leg size, heat input, and welding speed.
4. The method for evaluating the weld formation quality of underwater high-pressure dry welding process for fillet welds of Q345 steel plates according to claim 1, characterized in that, The welding numerical simulation model is a thermo-mechanical coupled finite element model, used to calculate the temperature field, residual stress, and weld penetration distribution.
5. A method for optimizing underwater high-pressure dry welding process parameters for fillet welds of Q345 steel plates, characterized in that, Includes the following steps: Input the welding process parameters to be optimized into the weld formation quality evaluation and prediction model constructed in claim 1 to obtain the corresponding weld quality evaluation index. Determine whether the obtained weld quality evaluation index meets the preset acceptable range. If it is within the acceptable range, the optimization process ends. Otherwise, readjust the input welding process parameters until the obtained new weld quality evaluation index is acceptable, then the optimization process ends.
6. A device for evaluating the weld formation quality of underwater high-pressure dry welding process for fillet welds of Q345 steel plates, used to implement the method as described in claim 1, characterized in that... include: The initial sample acquisition unit is used to carry out fillet weld tests with different welding process parameters on Q345 steel test plates. Based on the test results, the measured values of residual stress and penetration depth of different welds are obtained as weld quality evaluation indicators, and an initial process parameter-welding quality sample library is established. The expanded sample acquisition unit is used to substitute the process parameters and welding quality evaluation index data in the initial process parameter-welding quality sample library into the welding numerical simulation model for accuracy verification. After verification, the welding numerical simulation model is used to expand the sampling of process parameters and welding quality evaluation index data to generate an expanded process parameter-welding quality sample library. The model training unit is used to train a BP neural network with the process parameters in the expanded process parameter-welding quality sample library as input and the weld quality evaluation index as output, so as to obtain a weld formation quality evaluation prediction model. The quality evaluation unit is used to evaluate and predict the quality of weld formation based on the weld formation quality evaluation and prediction model.
7. A device for optimizing underwater high-pressure dry welding process parameters for fillet welds of Q345 steel plates, used to implement the method as described in claim 5, characterized in that... include: The quality evaluation unit is used to input the welding process parameters to be optimized into the weld formation quality evaluation and prediction model constructed in claim 1, and obtain the corresponding weld quality evaluation index. The parameter optimization unit is used to determine whether the obtained weld quality evaluation index meets the preset qualified range. If it is within the qualified range, the optimization process ends. Otherwise, the input welding process parameters are readjusted until the new weld quality evaluation index is qualified, and then the optimization process ends.