Welding Repair Method for Cracks and Wear Defects of Hydraulic Turbine Runner Blades

Abstract

The present invention relates to the technical field of welding repair of hydroturbine runner blades, specifically a welding repair method for cracks and wear defects of hydroturbine runner blades. The present invention extracts the first welding repair feature of the welding repair blade information through a welding repair analysis model, classifies the same welding repair categories based on a clustering algorithm, obtains the corresponding welding repair parameters and welding repair steps, generates a time series of welding repair parameters in chronological order, performs a primary parameter optimization through the welding repair analysis model, performs a secondary optimization of the welding repair parameters through a particle swarm model, optimizes the welding repair steps, and associates the welding repair blade information, welding repair parameters, and welding repair steps; through the above method, the welding repair work of runner blades with different wear and cracks can be classified, so as to optimize the existing work content, reduce the time of welding repair work, and reduce the economic loss of hydropower station power generation.

Description

Technical Field

[0001] The present invention relates to the technical field of welding repair of turbine runner blades, and specifically to a welding repair method for cracks and wear defects of turbine runner blades. Background Art

[0002] As an important facility for current hydropower generation, a water turbine drives the runner blades of the water turbine to rotate through water power, and then converts the mechanical energy that drives the runner blades to rotate into electrical energy through the internal structure of the water turbine, so as to achieve the purpose of converting mechanical energy into electrical energy.

[0003] Since the runner blades of the water turbine are easily affected by water pressure, high-speed water flow and impurities such as sediment in the water during use, the runner blades are prone to cracks and wear. In order to ensure the normal operation of hydropower generation, these damaged runner blades are usually welded and repaired.

[0004] Due to the different causes of cracks and wear, different welding repair methods need to be adopted for different situations during the welding repair process. For example, the material should be selected to be the same as that of the damaged runner blade, and at the same time, corresponding welding process parameters such as welding current, voltage and temperature should be adopted according to the crack and wear positions of the runner blade. Non-destructive testing should also be carried out after welding repair; since the welding repair of the runner blade involves too many processes and related parameters, a large amount of time will be spent during the welding repair process, and the shutdown of the water turbine during the welding repair process is also likely to cause a large amount of power generation economic losses to the hydropower plant.

[0005] In order to improve the efficiency of welding repair and reduce power generation economic losses, the present invention proposes a welding repair method for cracks and wear defects of turbine runner blades. Summary of the Invention

[0006] The purpose of the present invention is to provide a welding repair method for cracks and wear defects of turbine runner blades. Through the welding repair blade information of the historical welding repair plan, the corresponding first welding repair features are extracted according to the welding repair analysis model and associated with the historical welding repair plan; based on the clustering algorithm, the same welding repair categories are classified to obtain the corresponding welding repair parameters and welding repair steps; and a time series of welding repair parameters is generated in chronological order, and the welding repair parameters are optimized once through the welding repair analysis model, and the welding repair parameters are optimized twice through the particle swarm model, and the welding repair steps are optimized, and the welding repair blade information, welding repair parameters and welding repair steps are associated; through the above method, the welding repair work of turbine runner blades with different wear and cracks can be classified, so as to optimize the existing work content, reduce the time of welding repair work, and reduce the power generation economic losses of hydropower stations.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A welding repair method for cracks and wear defects of a water turbine runner blade, comprising:

[0009] Obtain a historical welding repair plan, and extract the welding repair blade information corresponding to the historical welding repair plan; the welding repair blade information is the image information and video information of the runner blade that needs to be welded and repaired;

[0010] Extract the first welding repair feature of the welding repair blade information according to the welding repair analysis model;

[0011] The welding repair analysis model includes a welding repair information processing unit, a welding repair feature recognition unit, a welding repair feature analysis unit, and a first welding repair feature output unit;

[0012] The welding repair information processing unit performs image preprocessing on the welding repair blade information;

[0013] The welding repair feature recognition unit recognizes welding repair defect features based on the preprocessed welding repair blade information; the welding repair feature recognition unit uses yolov5 to recognize the welding repair defect features;

[0014] The welding repair feature analysis unit analyzes the global welding repair factor and the local welding repair factor according to the welding repair defect features;

[0015] Further, the welding repair feature analysis unit generates the global welding repair factor according to the overall pixel ratio of the welding repair defect features recognized by yolov5 on the blade; the welding repair feature analysis unit obtains the defect lines of the welding repair defect features through DeepLSD and calculates the pixel number of the defect lines to generate the local welding repair factor;

[0016] The first welding repair feature output unit integrates the global welding repair factor and the local welding repair factor to generate the first welding repair feature;

[0017] Associate the corresponding historical welding repair plan based on the first welding repair feature;

[0018] Cluster through the first welding repair feature, call the historical welding repair plans of the same welding repair category according to the clustered first welding repair features, and obtain the corresponding welding repair parameters and welding repair steps;

[0019] Further, divide the global welding repair factor into multiple level tags; take the level tags of the global welding repair factor as the abscissa and the local welding repair factor as the ordinate to generate welding repair clustering coordinates, and use the k-means algorithm to process the multiple welding repair clustering coordinates to obtain multiple welding clustering sets;

[0020] Mark welding clustering tags for the multiple welding clustering sets; call the welding repair parameters and the welding repair steps of the corresponding historical welding repair plan by the same welding clustering tag, and label the welding repair parameters and the welding repair steps with the corresponding welding clustering tag;

[0021] Based on the clustered welding repair parameters and the welding repair steps, generate an initial welding repair plan for each welding repair category; in the chronological order of the initial welding repair plan, form a welding repair parameter sequence with the welding repair parameters, and identify abnormal welding repair parameters in the welding repair parameter sequence according to the welding repair analysis model for the first parameter optimization;

[0022] Generate the welding repair parameter sequence with the same welding repair parameters in chronological order, and the welding repair parameter sequence includes multiple welding repair parameter sequences; input each welding repair parameter sequence into the welding repair analysis model;

[0023] The welding repair analysis model further includes a welding parameter standardization unit, a welding parameter abnormality identification unit, and a welding parameter abnormality marking unit;

[0024] The welding parameter standardization unit performs time window division and standardization processing on each welding repair parameter sequence; the welding parameter abnormality identification unit identifies welding parameter abnormalities in the standardized welding repair parameter sequence;

[0025] The welding parameter abnormality identification unit includes 1 layer of transformer layer, 4 layers of time convolutional layers, 6 layers of LSTM layers, and 1 layer of feature fusion layer;

[0026] The welding parameter abnormality marking unit marks and outputs the welding repair parameter sequences identified as welding parameter abnormalities;

[0027] Perform secondary parameter optimization on the welding repair parameters after the first parameter optimization through a particle swarm model;

[0028] Further, initialize the velocity and position of the particles with the repair time, repair cost, and risk of the welding repair parameters optimized by the primary parameters. Taking the minimum welding repair time as the target condition and the welding repair cost and welding repair risk as penalty coefficients, optimize the welding repair parameters after the primary parameter optimization to generate the optimal welding repair parameters;

[0029] Screen and eliminate the corresponding welding repair steps according to the abnormal welding repair parameters after the primary parameter optimization; optimize the welding repair steps through the particle swarm model for the eliminated welding repair steps;

[0030] Further, digitalize the welding repair steps into welding step data, and initialize the repair time, repair cost, and risk of the digitalized welding repair steps as the velocity and position of the particles. Taking the minimum welding repair time as the target condition and the welding repair cost and welding repair risk as penalty coefficients, optimize the welding repair steps to generate the optimal welding repair steps;

[0031] Associate the first welding repair feature, the optimized welding repair steps, and the welding repair parameters optimized by the secondary parameters to form a welding repair scheme network for welding repair work invocation;

[0032] Further, the first welding repair feature, the optimized welding repair steps, and the welding repair parameters optimized by the secondary parameters generate an initial welding association weight, and reduce or enhance the welding association weight according to the invocation result.

[0033] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0034] 1. Through the welding repair analysis model, the present invention obtains the first welding repair feature of each welding repair blade information in the historical welding repair scheme and the abnormal data in the welding process. By this method, various existing crack and wear defects can be classified based on historical data, so as to obtain the welding treatment process for the same type of crack and wear defects. At the same time, some welding repair parameters that are obviously not conducive to reducing the welding repair time can also be screened out, which is convenient for the subsequent optimization process of the welding scheme, provides a stable data basis for the optimization of subsequent welding parameters, reduces the data interference between the welding repair parameters and steps of different wear and crack problems, improves the accuracy of the welding repair scheme optimization, and is also convenient for improving the efficiency of the welding repair scheme.

[0035] 2. Through the clustering method, the present invention can, based on the identified first welding repair features, cluster these feature data to obtain various categories of defects and cracks, and associate corresponding welding repair parameters and welding repair steps. Through this method, corresponding welding repair parameters and welding repair steps can be quickly provided based on the existing data features, providing welding repair data for different problem situations for subsequent optimization of welding repair time, facilitating the optimization and classification of subsequent welding repair plans, and improving the efficiency of welding repair.

[0036] 3. The present invention uses the particle swarm algorithm with time data as the objective and cost data and repair risk data as penalty coefficients to optimize the welding repair parameters and welding repair steps, obtaining the optimal welding repair parameters and welding repair steps. Through this method, optimization can be carried out not only from the perspective of time to improve the welding repair efficiency for different crack and defect problems, but also from the perspectives of cost and the danger of the repair process, reducing costs and enhancing the safety of welding repair personnel while ensuring the welding repair efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is the flowchart of the method of the present invention;

[0038] Figure 2 is the virtual device diagram of the welding repair information processing unit, welding repair feature recognition unit, welding repair feature analysis unit, and first welding repair feature output unit of the welding repair analysis model of the present invention;

[0039] Figure 3 is the virtual device diagram of the welding parameter standardization unit, welding parameter anomaly recognition unit, and welding parameter anomaly marking unit of the welding repair analysis model of the present invention;

[0040] Figure 4 is the defect line diagram of the runner blade crack of DeepLSD of the present invention;

[0041] Figure 5 is the network structure diagram of the welding repair analysis model of the present invention for identifying the welding repair parameter sequence. DETAILED DESCRIPTION OF THE INVENTION

[0042] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0043] As an important structure for hydroturbine power generation, the runner blades of a hydroturbine have a significant impact on the power generation efficiency and safety of hydroelectric power generation during their use. Therefore, once the above problems occur, the runner blades need to be welded and repaired. Since different welding repair schemes are required for different problems, the welding repair takes a long time, and the hydroturbine needs to be shut down during the repair process, which also greatly reduces the power generation economic loss. For this reason, the present invention provides a welding repair method for cracks and wear defects of hydroturbine runner blades, referring to Figure 1 as shown, the technical solution is as follows:

[0044] Obtain historical welding repair schemes, and extract the welding repair blade information corresponding to the historical welding repair schemes;

[0045] Extract the first welding repair feature of the welding repair blade information according to the welding repair analysis model, and associate the corresponding historical welding repair scheme based on the first welding repair feature;

[0046] Cluster through the first welding repair feature, call the historical welding repair schemes of the same welding repair category according to the clustered first welding repair feature, and obtain the corresponding welding repair parameters and welding repair steps;

[0047] Based on the clustered welding repair parameters and the welding repair steps, generate the initial welding repair schemes for each welding repair category; according to the time sequence of the initial welding repair schemes, form a welding repair parameter sequence with the welding repair parameters, and identify the abnormal welding repair parameters in the welding repair parameter sequence according to the welding repair analysis model for primary parameter optimization; perform secondary parameter optimization on the welding repair parameters of the primary parameter optimization through the particle swarm model;

[0048] Screen and eliminate the corresponding welding repair steps according to the abnormal welding repair parameters after the primary parameter optimization; perform welding repair step optimization on the eliminated welding repair steps through the particle swarm model;

[0049] Associate the first welding repair feature, the optimized welding repair steps, and the welding repair parameters of the secondary parameter optimization to form a welding repair scheme network for welding repair work invocation.

[0050] Based on the welding repair blade information for different wear and crack problems in historical welding repair solutions, different first welding repair characteristics of blade wear and cracks are extracted through a welding repair analysis model. Clustering is performed based on the extracted first welding repair characteristics. Through this method, various existing wear and crack problems can be classified in the first stage. Based on various existing problems, the welding repair parameters and welding repair steps corresponding to the same crack and wear problems can be called for further optimization, providing a more accurate data basis for subsequent optimization of welding repair work, achieving the purpose of improving welding repair efficiency and reducing power generation economic losses;

[0051] Based on the clustered welding repair steps and welding repair parameters, the welding repair parameters are formed into a time series in chronological order, and their abnormal data is extracted for primary optimization, and the corresponding welding repair steps are screened out according to the abnormal data. The particle swarm model is used to perform secondary optimization on the welding repair parameters and welding repair steps after primary optimization. Through primary optimization, data different from other welding repair parameters can be excluded. At the same time, considering the possible misjudgment problems caused by technological improvement or equipment improvement, the feature learning based on the time series can exclude some accidentally occurring abnormal welding repair parameters and retain the normal welding repair parameters that have been retained for a long time to minimize the possibility of misjudgment;

[0052] At the same time, based on the particle swarm algorithm, comprehensive optimization can be carried out in terms of efficiency and other aspects for each welding repair parameter and each welding repair step to extract the optimal welding repair parameters and welding repair steps under the same crack and wear problems, and generate a welding repair network in combination with the corresponding welding repair blade information. This can not only facilitate welding personnel to call welding repair solutions to maximize the improvement of welding repair efficiency and reduce power generation economic losses, but also can perform actual optimization work according to the actual situation of the actual repair work.

[0053] Embodiment 1

[0054] For specific illustration, the following content is combined to elaborate on the induction process of welding repair for the same crack problem and wear problem:

[0055] Obtain a historical welding repair solution and extract the welding repair blade information corresponding to the historical welding repair solution; the welding repair blade information is the image information and video information of the rotating blades that need to be welded and repaired; if it is video information, the image information in the video is extracted frame by frame as input data;

[0056] Based on the image data, the crack problems and wear problems of the existing runner blades can be understood more clearly, which is convenient for subsequent data training, data analysis, and data feature storage. At the same time, it is also convenient for subsequent archiving and problem visualization, and can ensure the accuracy of the information analysis of the welded repair blades to the greatest extent, so as to provide accurate problem data for optimizing the subsequent welded repair work process and improving the efficiency of the welded repair work;

[0057] Extract the first welding repair feature of the welded repair blade information according to the welding repair analysis model;

[0058] The welding repair analysis model includes a welding repair information processing unit, a welding repair feature recognition unit, a welding repair feature analysis unit, and a first welding repair feature output unit, as shown in Figure 2 shown;

[0059] The welding repair information processing unit performs image preprocessing on the welding repair blade information; the preprocessing includes Gaussian filtering for denoising, Laplacian operator edge detection, and normalization of the image data;

[0060] The welding repair feature recognition unit identifies the welding repair defect features based on the preprocessed welding repair blade information; the welding repair feature recognition unit uses yolov5 to identify the welding repair defect features;

[0061] The welding repair feature analysis unit analyzes the global welding repair factor and the local welding repair factor according to the welding repair defect features; in order to ensure that the overall image of the runner blade is relatively complete and the images can be unified, each image only contains one runner blade and after operations such as image cutting, the resolution of each image is ensured to be the same;

[0062] Furthermore, the welding repair feature analysis unit generates the global welding repair factor according to the overall pixel ratio of the welding repair defect features identified by yolov5 on the blade; the calculation of the global welding repair factor is as follows:

[0063] ;

[0064] where, is the global welding repair factor, is the total number of pixel points of the yolov5 candidate box, is the total number of pixel points of the whole picture;

[0065] The welding repair feature analysis unit obtains the defect lines of the welding repair defect features through DeepLSD and calculates the number of pixels of the defect lines to generate the local welding repair factor, as shown in Figure 4As shown; the number of pixels of the defective line is the total number of pixel points of the defective line generated after recognition passing through the image data;

[0066] Based on yolov5, feature extraction can be performed on the crack and wear data in the image data and video data of the runner blade. At the same time, the feature extraction box based on yolov5 can also determine the pixel range of the crack and wear data, which is convenient for obtaining the overall characteristics of the crack and wear data on the entire runner blade; based on DeepLSD, the trend data in the crack and wear data can be obtained. Through the line features in these image data, the local data of these problem data can be extracted. Combining the overall data and local data can more fully obtain the recognition result, improve the recognition accuracy of the crack and wear data, and facilitate subsequent data induction and summary;

[0067] The first welding repair feature output unit integrates the global welding repair factor and the local welding repair factor to generate the first welding repair feature;

[0068] Through the model, unified data feature extraction can be performed on the welding repair blade information provided in the historical welding repair plan. At the same time, based on the extracted problem data features, it can also provide a data basis for the subsequent classification work of the data with the same problem, so as to achieve targeted optimization for different crack and wear problems, improve the accuracy of welding repair work optimization, and realize a more efficient and accurate welding repair optimization process;

[0069] Associate the corresponding historical welding repair plan based on the first welding repair feature;

[0070] Perform clustering through the first welding repair feature, call the historical welding repair plans of the same welding repair category according to the clustered first welding repair feature, and obtain the corresponding welding repair parameters and welding repair steps;

[0071] Furthermore, divide the global welding repair factor into multiple grade labels; use the grade label of the global welding repair factor as the abscissa and the local welding repair factor as the ordinate to generate a welding repair clustering coordinate. Through the k-means algorithm for multiple welding repair clustering coordinates, obtain multiple welding clustering sets. In this embodiment, the grade label The calculation formula is: ; The calculation expression of is not unique, and it can also be expressed according to expert experience and other calculation formulas;

[0072] Label multiple of the welding clustering sets with welding clustering labels; call the welding repair parameters (such as, but not limited to, welding current, welding voltage, welding speed, etc.) and the welding repair steps (such as, but not limited to, preheating, segmented welding, post-weld heat treatment, non-destructive testing, etc.) of the corresponding historical welding repair plan through the same welding clustering label, and label the welding repair parameters and the welding repair steps with the corresponding welding clustering labels;

[0073] By performing data conversion on all data, the features recognized by yolov5 can be data-quantified, and combined with DeepLSD to form corresponding data coordinates. Through these data coordinates, multiple categories can be obtained based on the clustering algorithm to summarize the same crack problems and wear problems. Through this method, corresponding clustering labels can be generated for all coordinate data, facilitating the acquisition of corresponding category recognition results from the data feature dimension of the blade, so as to obtain the corresponding welding repair parameters and welding repair processes for subsequent induction work, providing a solid data foundation for improving the welding repair efficiency.

[0074] In this embodiment, the recognition results of several currently commonly used object detection algorithms for cracks and wear of runner blades are compared. The same image dataset is used for multiple detections, and the highest recognition accuracy is selected as the final result. The recognition results are shown in Table 1:

[0075] Table 1 Recognition accuracies of multiple object detection algorithms for cracks and wear of runner blades

[0076] Target detection algorithm Recognition accuracy yolov5 92.84% R-FCN 92.61% Fast R-CNN 92.06% Faster R-CNN 92.67%

[0077] It can be seen from the results in Table 1 that yolov5 has the highest recognition accuracy for cracks and wear of runner blades;

[0078] Embodiment 2

[0079] For specific illustration, the optimization process of the welding repair plan is described in combination with the following content:

[0080] Based on the welding repair parameters and the welding repair steps after clustering, generate the initial welding repair plans for each welding repair category; according to the time sequence of the initial welding repair plans, form the welding repair parameter sequences of the welding repair parameters, and identify the abnormal welding repair parameters in the welding repair parameter sequences according to the welding repair analysis model for the first parameter optimization;

[0081] Generate the welding repair parameter sequences of the same welding repair parameters according to the time sequence. The welding repair parameter sequences include multiple welding repair parameter sequences; input each welding repair parameter sequence into the welding repair analysis model;

[0082] The welding repair analysis model further includes a welding parameter standardization unit, a welding parameter anomaly identification unit, and a welding parameter anomaly marking unit, as shown in Figure 3 ;

[0083] The welding parameter standardization unit performs time window division and standardization processing on each welding repair parameter sequence; the time window divides the data according to the time window length set by expert experience, and the standardization processing uses Z-score for standardization;

[0084] The welding parameter anomaly identification unit identifies welding parameter anomalies in the standardized welding repair parameter sequence;

[0085] As shown in Figure 5 , the welding parameter anomaly identification unit includes 1 layer of transformer layer, 4 layers of time convolutional layers, 6 layers of LSTM layers, and 1 layer of feature fusion layer; the transformer layer corresponds to the Figure 5 converter in, and the LSTM layer corresponds to the Figure 5 long short-term memory layer in;

[0086] Through the transformer layer and the time convolutional layer, the extraction process of abnormal features can be strengthened. At the same time, the LSTM layer can also ensure learning based on the time features of time-series data to reduce the abnormal recognition error caused by the upgrade of welding equipment or technology. The welding processing parameters are screened from different dimensions of abnormal features to ensure the accuracy of the screened data, and it also ensures that the subsequent data optimization work reduces the efficiency of welding repair work;

[0087] In this embodiment, some welding repair parameter sequences containing anomalies are generated and labeled by combining expert data processing. A total of 5 welding repair parameter sequences are formed. Each welding repair parameter sequence is not trained with data to simulate data anomaly recognition in a real environment. The anomaly recognition results of the welding repair analysis model are shown in Table 2:

[0088] Table 2 Recognition accuracy and precision of the welding repair analysis model for 5 welding repair parameter sequences

[0089] Welding repair parameter sequence Abnormal recognition accuracy Abnormal recognition precision Sequence 1 90.17% 92.12% Sequence 2 89.88% 91.53% Sequence 3 90.52% 91.67% Sequence 4 90.03% 92.66% Sequence 5 89.79% 91.85%

[0090] It can be seen from the results in Table 1 that the anomaly recognition accuracy is around 90%, the anomaly recognition precision is between 91% and 93%, and the recognition results for the welding repair parameter sequence are good;

[0091] The welding parameter anomaly marking unit marks and outputs the welding repair parameter sequence identified as the welding parameter anomaly; the abnormal welding repair parameter is an obvious data increase or obvious data decrease in a short time in the time series, and these abnormal data will be output; the output result is the data label of this section of data after time window segmentation. For example, the data label of normal data is 0, and the data label of abnormal data is 1; by arranging the welding repair parameters in chronological order and extracting the abnormal characteristics of the welding repair parameters through the welding repair analysis model, it is possible to screen out some welding repair parameters with obvious anomalies based on the same crack problems and wear problems. At the same time, it can also reduce the anomaly recognition error caused by welding equipment upgrades or technology upgrades, improve the accuracy of these abnormal parameters, reduce the interference of these abnormal data on the particle swarm model optimization process during the optimization process, improve the accuracy of data optimization, facilitate more accurately reducing the welding repair time, and reducing the downtime of the water turbine welding repair project; to ensure the accuracy of data screening, after output, relevant technical personnel will conduct a secondary identification of the marked abnormal data to exclude the data that is beneficial to the welding plan optimization but is identified as abnormal;

[0092] The particle swarm model is used to perform secondary parameter optimization on the welding repair parameters optimized by the primary parameters;

[0093] Further, the velocity and position of the particle are initialized with the repair time, repair cost, and risk of the welding repair parameters optimized by the primary parameters. Taking the minimum welding repair time as the target condition and the welding repair cost and welding repair risk as the penalty coefficients, the welding repair parameters after the primary parameter optimization are optimized to generate the optimal welding repair parameters; specifically, the corresponding weights of the welding repair parameters are generated based on the repair time, repair cost, and risk corresponding to the welding repair parameters. Based on the weights of the welding repair parameters, the velocity and position of the particle are initialized. Among them, the repair time generates the corresponding weight according to the length of the repair time, the repair cost generates the corresponding weight according to the size of the cost input, and the risk generates the corresponding weight according to the degree of risk. The setting of the weight values is not unique and can be set according to the existing data combined with expert experience;

[0094] Taking the minimum welding repair time as the target condition can ensure that the welding repair parameters used for the crack problems and wear problems of the same type of runner blades reach the minimum welding repair time. At the same time, considering the cost problem and the danger in the welding repair process, taking these two parameters as the penalty coefficients can ensure the safe implementation of the welding repair process while minimizing the welding repair time and reducing the cost input to ensure the maximization of the water turbine power generation economy;

[0095] Screen and eliminate the corresponding welding repair steps according to the abnormal welding repair parameters optimized by the primary parameters; optimize the welding repair steps through the particle swarm model for the eliminated welding repair steps;

[0096] Further, digitalize the welding repair steps into welding step data. Initialize the repair time, repair cost, and risk level of the digitalized welding repair steps as the velocity and position of the particles. With the minimum welding repair time as the target condition and the welding repair cost and welding repair risk level as penalty coefficients, optimize the welding repair steps to generate the optimal welding repair steps; The content of digitalizing the welding steps is to generate corresponding data labels according to the steps. For example, preheating is 1, segmented welding is 2, etc. The welding repair steps also initialize the velocity and position of the particles with the weights of the welding repair steps for repair time, repair cost, and risk level. The method for generating the weights of the welding repair steps is the same as that of the welding repair parameter weights;

[0097] After digitalizing and quantifying the welding steps, adopt the same optimization process as the welding repair parameters. With the minimum welding repair time as the target and the cost and risk level as penalty coefficients, adopting the same optimization process can ensure that the optimization data analysis dimension of the welding repair steps is consistent with that of the welding repair parameters, guarantee the accuracy of the data in the optimization dimension, and also obtain the best welding optimization steps based on the same optimization goal. This can not only improve the efficiency of welding repair from the aspect of welding repair parameters but also from the welding repair work, reducing the downtime of the water turbine during the welding repair process;

[0098] Associate the first welding repair feature, the optimized welding repair steps, and the welding repair parameters optimized by the secondary parameters to form a welding repair scheme network for welding repair work invocation;

[0099] Further, generate an initial welding association weight for the first welding repair feature, the optimized welding repair steps, and the welding repair parameters optimized by the secondary parameters, and reduce or enhance the welding association weight according to the invocation result;

[0100] Through the association weight, corresponding data can be fed back for the welding repair result in the actual welding repair project to understand whether the optimization scheme can achieve the purpose of complete welding repair and reduce the welding repair time. By reducing or increasing the welding association weight, the optimization results of the optimized welding repair parameters and the welding repair scheme can be understood, and subsequent adjustment and re-optimization of the welding repair scheme can be carried out.

[0101] Although embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A welding repair method for cracks and wear defects of turbine runner blades, characterized in that: include: Obtaining a historical welding repair scheme, and extracting welding repair blade information corresponding to the historical welding repair scheme; The welding repair blade information is image information and video information of the runner blade that needs welding repair; Extracting a first welding repair feature of the welded repair blade information according to the welding repair analysis model, and associating the corresponding historical welding repair scheme based on the first welding repair feature; The first welding repair feature of the welding repair blade information extracted according to the welding repair analysis model includes: The welding repair analysis model includes a welding repair information processing unit, a welding repair feature recognition unit, a welding repair feature analysis unit and a first welding repair feature output unit; The welding repair information processing unit performs image preprocessing on the welding repair blade information; The welding repair feature recognition unit identifies welding repair defect features based on the preprocessed welding repair blade information; the welding repair feature analysis unit analyzes global welding repair factors and local welding repair factors according to the welding repair defect features; The first welding repair feature output unit integrates the global welding repair factor and the local welding repair factor to generate the first welding repair feature; Clustering is performed through the first welding repair feature, calling the historical welding repair scheme of the same welding repair category according to the clustered first welding repair feature, and obtaining corresponding welding repair parameters and welding repair steps; Based on the clustered welding repair parameters and welding repair steps, an initial welding repair scheme for each welding repair category is generated; according to the time sequence of the initial welding repair scheme, the welding repair parameters are formed into a welding repair parameter sequence, and the abnormal welding repair parameters of the welding repair parameter sequence are identified according to the welding repair analysis model to perform a primary parameter optimization; and the welding repair parameters of the primary parameter optimization are secondary optimized by a particle swarm model; According to the abnormal welding repair parameters after the primary parameter optimization, corresponding welding repair steps are screened and eliminated; and the welding repair steps after elimination are optimized by using the particle swarm model; The first welding repair feature, the optimized welding repair step and the welding repair parameters optimized by the secondary parameters are associated to form a welding repair solution network for welding repair work calls.

2. The welding repair method for cracks and wear defects of turbine runner blades according to claim 1 is characterized in that: The welding repair feature recognition unit uses yolov5 to identify the welding repair defect feature; the welding repair feature analysis unit generates the global welding repair factor based on the overall pixel ratio of the welding repair defect feature identified by yolov5 on the blade; the welding repair feature analysis unit obtains the defect line of the welding repair defect feature through DeepLSD and calculates the number of pixels of the defect line to generate the local welding repair factor.

3. The welding repair method for cracks and wear defects of turbine runner blades according to claim 1, characterized in that: Clustering is performed through the first welding repair feature, calling the historical welding repair scheme of the same welding repair category according to the clustered first welding repair feature, and obtaining corresponding welding repair parameters and welding repair steps, including: The global welding repair factor is divided into multiple level labels; the level label of the global welding repair factor is used as the horizontal coordinate and the local welding repair factor is used as the vertical coordinate to generate welding repair clustering coordinates, and the multiple welding repair clustering coordinates are obtained by k-means algorithm to obtain multiple welding clustering sets; The plurality of welding cluster sets are marked with welding cluster labels; the welding repair parameters and the welding repair steps corresponding to the historical welding repair scheme are called by the same welding cluster label, and the welding repair parameters and the welding repair steps are marked with the corresponding welding cluster labels.

4. The welding repair method for cracks and wear defects of turbine runner blades according to claim 1, characterized in that: According to the time sequence of the initial welding repair scheme, the welding repair parameters are formed into a welding repair parameter sequence, and the abnormal welding repair parameters of the welding repair parameter sequence are identified according to the welding repair analysis model to perform a parameter optimization, which includes: The same welding repair parameters are used to generate the welding repair parameter sequence according to the time sequence, wherein the welding repair parameter sequence includes a plurality of welding repair parameter sequences; each welding repair parameter sequence is input into the welding repair analysis model; The welding repair analysis model also includes a welding parameter standardization unit, a welding parameter abnormality identification unit and a welding parameter abnormality marking unit; The welding parameter standardization unit divides each welding repair parameter sequence into time windows and performs standardization processing; the welding parameter abnormality identification unit identifies welding parameter abnormalities in the standardized welding repair parameter sequence; and the welding parameter abnormality marking unit marks and outputs the welding repair parameter sequence identified as the welding parameter abnormality.

5. The welding repair method for cracks and wear defects of turbine runner blades according to claim 4 is characterized in that: The welding parameter anomaly recognition unit includes 1 transformer layer, 4 time convolution layers, 6 LSTM layers and 1 feature fusion layer.

6. The welding repair method for cracks and wear defects of turbine runner blades according to claim 1, characterized in that: The secondary parameter optimization of the welding repair parameters optimized by the primary parameter optimization through the particle swarm model includes: initializing the speed and position of the particles with the repair time, repair cost and hazard of the welding repair parameters optimized by the primary parameter optimization, taking the minimum welding repair time as the target condition, taking the welding repair cost and the welding repair hazard as the penalty coefficients, optimizing the welding repair parameters after the primary parameter optimization, and generating the optimal welding repair parameters.

7. The welding repair method for cracks and wear defects of turbine runner blades according to claim 1, characterized in that: The optimization of the welding repair step after elimination by using the particle swarm model includes: digitizing the welding repair step, initializing the repair time, repair cost and hazard of the digitized welding repair step as the speed and position of the particle, taking the minimum welding repair time as the target condition, taking the welding repair cost and the welding repair hazard as the penalty coefficient, optimizing the welding repair step, and generating the optimal welding repair step.

8. The welding repair method for cracks and wear defects of turbine runner blades according to claim 1, characterized in that: Associating the first welding repair feature, the optimized welding repair step and the welding repair parameters optimized by the secondary parameters includes: generating an initialized welding association weight using the first welding repair feature, the optimized welding repair step and the welding repair parameters optimized by the secondary parameters, and enhancing or reducing the welding association weight according to the calling result.

Images