A multi-algorithm collaborative optimization method for additive manufacturing of key components of a hydraulic turbine

Abstract

The application discloses a kind of multi-algorithm synergic optimization's water turbine key component additive manufacturing method, it is related to additive manufacturing technical field.The method of the present application constructs complete intelligent manufacturing system: in data acquisition stage, high-precision laser scanning technology (precision reaches 0.01mm) is used, the accurate three-dimensional reconstruction of damage area is realized;In the link of modeling analysis, through the finite element simulation of multi-physical field coupling, the thermodynamic behavior in the repair process is accurately predicted;Process optimization, innovatively combine adaptive differential evolution algorithm, improved NSGA-II multi-objective optimization algorithm and deep reinforcement learning (DRL), form the whole process intelligent decision system covering deviation analysis, path planning and parameter optimization.Especially the introduction of deep reinforcement learning algorithm, the millisecond level dynamic adjustment of key process parameters such as wire feeding speed, laser power is realized, and the stability and material utilization of manufacturing process are greatly improved.

Description

Technical Field

[0001] This invention relates to the field of additive manufacturing technology, specifically to an additive manufacturing method for key components of a water turbine that is optimized through multi-algorithm collaborative optimization. Background Technology

[0002] Key components of hydroelectric turbines (such as runner blades and guide vanes) are core components of hydroelectric power generation systems, and their performance directly affects the energy conversion rate, operational stability, and service life of the entire generator unit. These components operate under harsh conditions for extended periods, enduring intense alternating loads, cavitation wear, silt erosion, and chemical corrosion, making them highly susceptible to surface wear, cracks, and spalling. Traditional repair techniques mainly rely on conventional processes such as welding and thermal spraying, but these methods have inherent drawbacks such as high heat input, difficulty in controlling deformation, and low repair precision. The performance and service life of repaired components often fail to meet new product standards.

[0003] With the development of additive manufacturing technology, new digital manufacturing technologies, such as laser cladding and arc additive manufacturing, have provided new solutions for the remanufacturing of key components of hydraulic turbines. However, existing additive manufacturing methods still face a series of technical bottlenecks that urgently need to be addressed when applied to the repair of large, complex curved surface components such as hydraulic turbines.

[0004] Insufficient accuracy in data acquisition and modeling: The measurement accuracy of traditional 3D measurement technologies (such as contact coordinate measuring machines or low-precision laser scanning) is usually at the level of 0.1mm, which makes it difficult to accurately capture the microscopic damage morphology of complex geometric features such as blade surfaces and flow channels. This results in deviations between the initial model and the actual object, affecting the accuracy of the repair plan.

[0005] The level of intelligence in process planning is low: repair path planning relies heavily on manual experience and simple CNC programming, lacking a global optimization perspective. This planning method is inefficient and cannot simultaneously address multiple conflicting optimization objectives such as path efficiency, heat input uniformity, and deformation control.

[0006] Lagging deviation analysis and control capabilities: Existing methods mostly use static algorithms for post-manufacturing deviation analysis, lacking the ability to identify, predict and actively compensate for dimensional and shape deviations generated in the dynamic process of additive manufacturing in real time, resulting in unstable repair accuracy.

[0007] Rigid process parameter control strategies: During manufacturing, key process parameters such as laser power and wire / powder feeding speed are typically controlled using fixed parameter sets or simple PID feedback. This control strategy cannot adapt to dynamic disturbances such as heat accumulation and changes in boundary conditions during manufacturing, making it difficult to achieve real-time and precise control of the process window, and is prone to defects.

[0008] In-depth analysis reveals that the fundamental limitations of existing technologies lie in: fragmented processes, creating "information silos": each stage from data acquisition, modeling, path planning to process control operates independently, failing to effectively connect data and information flows, resulting in a lack of synergy and intelligence in the entire manufacturing system. Limited algorithmic capabilities: relying solely on single or traditional optimization algorithms is insufficient to address the complex multi-variable, strongly coupled, and nonlinear multi-objective optimization problems in additive manufacturing. Lack of real-time sensing and closed-loop control: the failure to deeply integrate real-time monitoring data (such as molten pool morphology and temperature field) with intelligent decision-making algorithms prevents the achievement of state-based adaptive real-time control. Therefore, there is an urgent need in this field for an intelligent additive manufacturing strategy that can span the entire process, integrate multi-source information, and possess self-sensing, self-decision-making, and self-optimization capabilities to fundamentally improve the quality, efficiency, and reliability of remanufacturing key components of hydraulic turbines. Summary of the Invention

[0009] The purpose of this invention is to address the aforementioned problems by providing an additive manufacturing method for key components of a hydro turbine through multi-algorithm collaborative optimization. By integrating laser scanning, finite element analysis, intelligent optimization algorithms, and advanced manufacturing technologies, this method aims to improve the manufacturing efficiency and quality of key components of a hydro turbine while reducing manufacturing costs and time.

[0010] The technical solution of the present invention is as follows:

[0011] A multi-algorithm collaborative optimization method for additive manufacturing of key components of a hydraulic turbine includes the following steps:

[0012] High-precision laser scanning technology was used to acquire three-dimensional point cloud data of key components of the water turbine, and the point cloud data was denoised and simplified to generate an accurate three-dimensional model.

[0013] A multi-physics coupled finite element model was established based on a three-dimensional model to simulate the stress-strain field of the component under working load and its thermodynamic behavior during the repair process.

[0014] An adaptive update differential evolution algorithm is used to identify and analyze geometric deviations in the additive manufacturing process in real time, and generate deviation correction schemes.

[0015] The improved NSGA-II multi-objective optimization algorithm is used to optimize the laser deposition path globally, with the optimization objectives being repair path length, heat input uniformity, and material utilization rate.

[0016] In the laser-arc hybrid additive manufacturing process, a deep reinforcement learning agent is introduced to dynamically adjust process parameters based on real-time monitoring data of the molten pool, thereby completing the remanufacturing of key components of the water turbine.

[0017] By integrating laser scanning, finite element modeling, adaptive update differential evolution algorithm, improved NSGA-II algorithm, and deep reinforcement learning (DRL) control in the laser-arc hybrid additive manufacturing (LHAM) process, a complete intelligent additive manufacturing system was constructed. This method achieves multi-algorithm collaborative optimization of the entire process from data acquisition, deviation analysis, path planning to process execution, ultimately significantly improving the overall efficiency, accuracy, and quality of remanufacturing key components of hydraulic turbines, while reducing manufacturing costs and time.

[0018] Furthermore, the three-dimensional data acquired by the high-precision laser scanning technology includes the curved geometric features of the blade, surface wear, scratch texture, and spatial attitude; the specific steps for generating the three-dimensional model include: point cloud registration, triangulation, and surface reconstruction; the measurement accuracy of the laser scanning technology reaches 0.01 mm.

[0019] High-precision 3D data acquisition provides a precise data foundation for all subsequent steps. The generated accurate 3D model ensures the reliability of finite element analysis, the accuracy of path planning, and a high degree of consistency between the final manufactured component and the design intent, guaranteeing manufacturing precision from the outset. Limiting the laser scanning precision to 0.01mm ensures the accurate capture of microscopic defects and morphologies of complex geometric features such as turbine blade surfaces and flow channels. This ultra-high precision lays an indispensable foundation for subsequent high-quality analysis, optimization, and manufacturing, and is a key prerequisite for achieving high-precision remanufacturing.

[0020] Furthermore, the establishment of the finite element model includes: defining temperature-dependent material properties corresponding to the substrate and the repair material, applying working condition loads and heat source models, and performing transient thermo-mechanical coupled field simulation to identify high stress concentration areas and potential deformation risk areas during the repair process.

[0021] Finite element analysis allows for the prediction and evaluation of component performance under actual operating conditions and thermodynamic behavior during repair processes before manufacturing. This enables design optimization to be implemented before the manufacturing stage, allowing for the early identification and mitigation of potential stress concentration and structural failure risks, thereby guiding the manufacture of components with superior performance and longer lifespan.

[0022] Furthermore, the adaptive update differential evolution algorithm identifies deviations by evaluating the differences between manufacturing process data and design models, and generates correction schemes; the algorithm has adaptive capabilities and can dynamically adjust its parameters and strategies according to changes in the manufacturing environment and real-time feedback.

[0023] The algorithm employs a dynamic linear ascending crossover probability scheme, with crossover probabilities... The calculation formula is:

[0024] ,

[0025] in, Set to 0.99, If set to 0.45, then This indicates the number of times the function is evaluated. This indicates the maximum number of function evaluations.

[0026] Using the methods described above, the algorithm and its dynamic crossover probability formula can adaptively balance global search and local exploitation capabilities. This enables the algorithm to identify complex deviations in the manufacturing process more quickly and accurately, and generate effective correction schemes, thereby significantly improving the efficiency and accuracy of deviation analysis and enhancing the stability and robustness of the manufacturing process.

[0027] Furthermore, the adaptive update differential evolution algorithm also employs a control update strategy, the formula of which is:

[0028] ,

[0029] in, Representation strategy The probability of being called in the next iteration. Representation strategy The score or integral value in the current iteration The number of strategies.

[0030] By introducing a strategy selection probability formula, the algorithm can dynamically allocate the call probability based on the actual performance of each strategy during iteration. This adaptive mechanism optimizes the algorithm's search behavior, avoids getting trapped in local optima, and ensures that it can continuously and efficiently approach the global optimum in complex manufacturing deviation optimization problems.

[0031] Furthermore, the improved NSGA-II algorithm introduces an adaptive tournament selection strategy, which dynamically adjusts the scale of tournament selection by evaluating the diversity and convergence state of the population to balance exploration and exploitation; the multi-objective functions include minimizing repair time, maximizing material utilization, and maximizing strength recovery rate.

[0032] By dynamically adjusting the selection pressure, this algorithm achieves an excellent balance between exploration (finding new paths) and development (optimizing existing paths). This enables it to provide a set of Pareto solutions that are superior in multiple objectives such as path length, heat input, and material utilization for repair path planning, thereby improving repair efficiency, molding quality, and material conservation.

[0033] Furthermore, the laser-arc hybrid additive manufacturing technology combines the advantages of laser melting and electric arc melting, and achieves the layer-by-layer construction of turbine blades by precisely controlling the power, moving speed, and material feeding rate of the laser beam and electric arc.

[0034] This method combines the advantages of both laser and electric arc heat sources, achieving a balance between high energy density, high deposition rate, and good molding quality. It is particularly suitable for the repair of large components in water turbines, significantly improving manufacturing efficiency while ensuring the mechanical properties and molding precision of the components.

[0035] Furthermore, the state space of the deep reinforcement learning algorithm includes the melt pool temperature, melt pool size, deposition height, and laser power deviation; the action space covers laser power adjustment, wire feed speed adjustment, and powder feed speed adjustment; the reward function adopts a segmented scoring mechanism, giving positive and negative rewards according to the degree of conformity of the deposition quality, guiding the algorithm to learn the optimal parameter combination.

[0036] By introducing the DRL algorithm into manufacturing process control using the methods described above, a leap from fixed parameter or simple feedback control to intelligent decision control is achieved. The algorithm can learn from historical and real-time data and autonomously optimize process parameters, enabling the manufacturing system to have adaptive and self-learning capabilities, and proactively respond to uncertainties and dynamic changes in the manufacturing process.

[0037] Furthermore, in the remanufacturing of the turbine blades, the three-dimensional geometry and defect data acquired by laser scanning support the establishment of the finite element model. The stress concentration distribution and strength benchmark value output by the model serve as constraints for the adaptive update differential evolution algorithm. The surface curvature and thin-wall dimensional deviation data identified by the algorithm further guide the improvement of the NSGA-II algorithm to plan a low-deviation repair path. The relevant optimized path is directly used as the execution instruction of the laser arc hybrid additive manufacturing (LHAM) equipment.

[0038] Furthermore, the real-time data on melt pool temperature, size, and deposition height deviation collected during the LHAM process are continuously input into the deep reinforcement learning (DRL) module. Based on this, the DRL dynamically adjusts the wire / powder feeding speed and simultaneously feeds the deviation improvement results back to the differential evolution algorithm to form a closed loop, thereby achieving intelligent collaborative optimization of the remanufacturing process.

[0039] Compared with existing technologies, the advantages of this invention are:

[0040] 1. Improve manufacturing efficiency and quality: Through multi-algorithm collaborative optimization, this invention can more effectively control the manufacturing process, thereby improving manufacturing efficiency and product quality;

[0041] 2. Reduce manufacturing costs and time: This invention reduces material waste and manufacturing time by optimizing the manufacturing path and process parameters, thereby reducing manufacturing costs;

[0042] 3. Improve the intelligence level of the manufacturing process: By introducing deep reinforcement learning algorithms, the manufacturing process can be dynamically adjusted in real time, thereby improving the intelligence level of the manufacturing process;

[0043] 4. Optimize material utilization efficiency: Through the precise control of wire and powder feeding speed by deep reinforcement learning algorithm, not only is the utilization efficiency of materials optimized, but the manufacturing quality is also improved, making the entire manufacturing process more flexible and efficient. Attached Figure Description

[0044] Figure 1 This is a flowchart of the method described in this application.

[0045] Figure 2 A more detailed flowchart of the method in this application is provided.

[0046] Figure 3 This is a flowchart of the adaptive update differential evolution algorithm of this application.

[0047] Figure 4 The flowchart shows the improved NSGA-Ⅱ algorithm of this application.

[0048] Figure 5 This is a schematic diagram of the laser arc hybrid additive manufacturing (LHAM) principle of simultaneous wire and powder feeding. Detailed Implementation

[0049] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0050] The features and performance of the present invention will be further described in detail below with reference to embodiments.

[0051] Please see Figure 1-5 A multi-algorithm collaborative optimization method for additive manufacturing of key components of a hydro turbine, such as... Figure 1As shown, a complete intelligent manufacturing system was constructed: In the data acquisition stage, high-precision laser scanning technology (accuracy up to 0.01mm) was used to achieve accurate three-dimensional reconstruction of the damaged area; in the modeling and analysis stage, multi-physics coupled finite element simulation was used to accurately predict the thermodynamic behavior during the repair process; in terms of process optimization, the adaptive differential evolution algorithm, the improved NSGA-II multi-objective optimization algorithm, and deep reinforcement learning (DRL) were innovatively combined to form a full-process intelligent decision-making system covering deviation analysis, path planning, and parameter optimization. In particular, the introduction of the deep reinforcement learning algorithm enabled millisecond-level dynamic adjustment of key process parameters such as wire and powder feeding speed and laser power, significantly improving the stability of the manufacturing process and material utilization. Specifically, the system includes the following steps:

[0052] Precise three-dimensional data of turbine blades were obtained using laser scanning technology, and a finite element model was established based on this data.

[0053] An adaptive update differential evolution algorithm is used for deviation analysis to identify potential deviations in the manufacturing process;

[0054] The repair path is optimized using an improved NSGA-II algorithm to ensure the efficiency and accuracy of the repair process;

[0055] In the laser arc hybrid additive manufacturing (LHAM) process, a deep reinforcement learning (DRL) algorithm is introduced to achieve real-time dynamic adjustment of process parameters, including controlling the speed of wire and powder feeding.

[0056] Laser scanning technology is used to acquire precise three-dimensional data of turbine blades, including their geometry, surface texture, and spatial location. Laser scanning generates three-dimensional point cloud data of the blades, allowing for the extraction of detailed geometric features such as length, width, height, curves, and angles. Simultaneously, laser scanning can analyze the reflectivity of the blade surface, obtaining surface texture information such as roughness, scratches, and wear. Furthermore, by calculating the three-dimensional coordinates of each point on the blade surface, the blade's specific position and orientation in space can be determined. This acquired information is integrated to form a precise three-dimensional model. This model is used not only for subsequent finite element analysis to predict the stress, strain, and other properties of the component under different conditions, but also to guide manufacturing activities in the laser-arc hybrid additive manufacturing (LHAM) process.

[0057] The establishment and application of the finite element model (FEM) is aimed at in-depth analysis of the stress, strain, and other properties of key components of a hydro turbine under different operating conditions. This process begins by constructing a detailed FEM model based on precise three-dimensional data obtained through laser scanning technology. This model accurately reflects the geometric characteristics and material properties of the blades. The physical and mechanical properties of the blade material, such as elastic modulus, Poisson's ratio, and yield strength, are defined in the model. Boundary conditions and load conditions are also set to simulate the stress state of the hydro turbine blades in the actual working environment. Next, the model is meshed using finite element analysis software to generate a mesh system suitable for numerical calculations. The analysis is then performed to obtain the stress distribution, strain distribution, and displacement within the blades. By analyzing these calculation results, the strength, stiffness, and stability of the hydro turbine blades can be evaluated, and potential stress concentration areas and weak points can be identified. Based on these analysis results, the design of the hydro turbine blades can be optimized, such as by adjusting the geometry and changing the material distribution, to improve the performance of the hydro turbine blades. Ultimately, the optimized turbine blade design will be used to guide the subsequent additive manufacturing process, ensuring that the manufactured components meet the design requirements and performance indicators, thereby achieving efficient and precise remanufacturing of key turbine components.

[0058] The adaptive update differential evolution algorithm is applied to accurately identify and effectively correct potential deviations in the additive manufacturing process of critical turbine components. This process first involves real-time analysis of data collected during manufacturing, which may include manufacturing parameters monitored by sensors, changes in material properties, and the real-time geometry of the turbine blades. The adaptive update differential evolution algorithm identifies deviations in the manufacturing process by evaluating the differences between this data and the design model. The algorithm uses a differential evolution strategy to generate a set of potential correction schemes, which may involve adjusting manufacturing parameters, changing material input rates, or modifying the component's manufacturing path. Then, the algorithm simulates the impact of these correction schemes on the manufacturing results, evaluates their effectiveness, and selects the optimal correction scheme for implementation in the actual manufacturing process. Furthermore, the algorithm possesses adaptive capabilities, dynamically adjusting its parameters and strategies based on changes in the manufacturing environment and real-time feedback during manufacturing to continuously optimize the manufacturing process, reduce deviations, and improve manufacturing accuracy and efficiency.

[0059] Crossover probability ( ) is a constant value, typically ranging from [0, 1]. This application implements a dynamic linear ascending crossover probability scheme to maintain population diversity while balancing local search reinforcement and global search exploration:

[0060] ,

[0061] in, Set to 0.99, If set to 0.45, then This indicates the number of times the function is evaluated. This indicates the maximum number of function evaluations.

[0062] Furthermore, the adaptive update differential evolution algorithm also employs a control update strategy, the formula of which is:

[0063] ,

[0064] in, Representation strategy The probability of being called in the next iteration. Representation strategy The score or integral value in the current iteration The number of strategies.

[0065] To address the complex curved surfaces and thin-walled structure of turbine blades, a dynamic linear ascending crossover probability strategy is introduced. (The value gradually increased from 0.45 to 0.99), maintaining population diversity in the early stage and strengthening local search in the later stage.

[0066] The improved NSGA-II algorithm introduces an adaptive tournament selection strategy, aiming to dynamically adjust the algorithm's selection pressure while maintaining population diversity. Specifically, the algorithm adaptively adjusts the scale of tournament selection by evaluating population diversity and convergence status, thereby balancing exploration and exploitation at different stages of evolution. When population diversity is high, i.e., individuals within the population exhibit significant differences, the algorithm increases the scale of tournament selection to promote a broader search and avoid premature convergence to local optima. Conversely, when population convergence is significant, i.e., individuals within the population tend to be similar, the algorithm reduces the scale of tournament selection to strengthen selection pressure, promote the rapid reproduction of superior individuals, and accelerate convergence to the global optimum. Through these improvements, the algorithm can more effectively handle complex multi-objective optimization problems, providing a better repair path for the additive manufacturing of key components of hydraulic turbines, thereby improving manufacturing efficiency and product quality. The multi-objective functions include minimizing repair time, maximizing material utilization, and maximizing strength recovery rate. By dynamically adjusting the selection pressure,

[0067] A balance is struck between population diversity and convergence to ensure that the repair path achieves an optimal trade-off among multiple objectives. In this process, the multi-objective function includes repair time (weight 0.3), material utilization rate (weight 0.4), and strength recovery rate (weight 0.3). The strength recovery rate is referenced to the blade strength baseline value from finite element analysis, with a target value set at ≥95%. An adaptive weighting mechanism is also implemented: when the proportion of individuals with repair time > a preset threshold exceeds 40% of the population, the repair time weight is automatically increased to 0.5; when the proportion of individuals with material utilization rate < 80% exceeds 30%, the material utilization rate weight is increased to 0.5, thereby dynamically balancing the optimization priorities of multiple objectives.

[0068] Laser-arc hybrid additive manufacturing (LHAM) technology is a key step in achieving efficient and precise manufacturing of critical components for hydro turbines. This technology combines the advantages of laser melting and electric arc melting, enabling layer-by-layer construction of turbine blades by precisely controlling the power, movement speed, and material feed rate of the laser beam and electric arc. In this process, the laser beam and electric arc work together to locally melt the material, while precise control of the wire or powder feeding speed ensures uniform and continuous material deposition on the component surface, forming the desired geometry. The advantages of LHAM technology lie in its ability to provide higher energy density and faster manufacturing speed while maintaining component manufacturing accuracy and surface quality. Furthermore, this technology allows the use of a variety of materials, including but not limited to metals and alloys, as well as composite materials, thus providing greater flexibility in the design and manufacturing of hydro turbine blades. In the method of this invention, the application of LHAM technology also includes integration with deep reinforcement learning (DRL) algorithms to achieve real-time dynamic adjustment of process parameters during manufacturing. This intelligent control strategy can automatically adjust the parameters of the laser and electric arc, as well as the material feed rate, based on real-time feedback during the manufacturing process, such as temperature, molten pool morphology, and material deposition. This optimizes the manufacturing process and improves the performance and quality of the components. Through this multi-algorithm collaborative optimization method, LHAM technology not only improves the manufacturing efficiency of key turbine components but also ensures the stability and reliability of the manufacturing process, providing an innovative and efficient solution for additive manufacturing of key turbine components.

[0069] The application of deep reinforcement learning (DRL) algorithms aims to achieve real-time feedback and dynamic adjustment of process parameters during the additive manufacturing of key components for hydraulic turbines. Specifically, DRL algorithms learn how to optimize process parameters, such as laser power, arc current, wire feed speed, and powder feed rate, based on the current manufacturing state and historical data, by simulating the behavior of an agent in its environment. DRL algorithms utilize deep neural networks to process and learn from complex manufacturing data, including 3D data acquired through laser scanning, finite element model analysis results, and real-time monitored manufacturing parameters. Through interaction with the manufacturing environment, the algorithm continuously tries and learns from experience to find the optimal combination of process parameters. During manufacturing, DRL algorithms can quickly make decisions and adjust process parameters based on real-time feedback information, such as temperature changes, molten pool morphology, and material deposition, to adapt to potential changes and uncertainties during manufacturing. Furthermore, DRL algorithms possess adaptive capabilities, dynamically adjusting their learning strategies and decision-making models based on changes in the manufacturing environment and real-time feedback during the manufacturing process to continuously optimize the manufacturing process. This intelligent control strategy not only improves the automation and intelligence of the manufacturing process, but also significantly improves the manufacturing precision and quality of key components of the turbine, while reducing manufacturing costs and time.

[0070] The algorithm's state space includes seven parameters: melt pool temperature (monitoring range 200-2000℃, accuracy ±5℃), melt pool size (length / width / height, accuracy ±0.1mm), deposition height (deviation from design value, accuracy ±0.05mm), and laser power (deviation from set value in real time). The action space covers three types of operations: laser power adjustment (step 5W, range 500-2000W), wire feed speed adjustment (step 0.1m / min, range 0.5-3m / min), and powder feed speed adjustment (step 0.5g / min, range 2-10g / min). The reward function uses a segmented scoring mechanism: +10 for deposition quality meeting design requirements, -2 for slight deviations, and -20 for severe deviations, thus guiding the algorithm to learn the optimal parameter combination.

[0071] Combination Figure 1 and Figure 2 This is the overall block diagram of the present invention, which is an additive manufacturing method for key components of a water turbine through multi-algorithm collaborative optimization.

[0072] Precise three-dimensional data of turbine blades were acquired using laser scanning technology, and a finite element model was established based on this data. An adaptive update differential evolution algorithm was employed for deviation analysis to identify potential deviations in the manufacturing process. An improved NSGA-II algorithm was used to optimize the repair path, ensuring the efficiency and accuracy of the repair process. In the laser-arc hybrid additive manufacturing (LHAM) process, a deep reinforcement learning (DRL) algorithm was introduced to achieve real-time dynamic adjustment of process parameters, including controlling the wire and powder feeding speed, thereby further improving the intelligence level of the manufacturing process. Precise control of the wire and powder feeding speed through the deep reinforcement learning algorithm not only optimized material utilization efficiency but also improved manufacturing quality, making the entire manufacturing process more flexible and efficient. Through the synergistic optimization of the above multiple algorithms, this study successfully achieved the efficient and precise remanufacturing of key turbine components.

[0073] Combination Figure 3 This is a flowchart of the adaptive updating differential evolution algorithm of the present invention, which is specifically designed to optimize the additive manufacturing process of key components of hydraulic turbines. The process starts from the "Start" node, first performing "Initialize Population" to create a set of potential solutions. The fitness of these solutions is evaluated, i.e., the degree to which they meet design requirements and performance indicators. The algorithm checks whether the termination condition is met, which may include reaching the maximum number of iterations, finding a satisfactory solution, or other preset criteria. If the termination condition is not met, the algorithm enters the adaptive update control parameter stage, dynamically adjusting the algorithm parameters based on the current population performance and search history to optimize the search process. A selection operation is performed, selecting individuals from the current population for subsequent operations. Then, a differential operation is performed to generate new individuals, followed by crossover and mutation operations to further explore the solution space and increase the diversity of the population. The fitness of the new population is evaluated, and the selection operation is performed again to update the population. This process is repeated until the termination condition is met. Once the termination condition is met, the algorithm outputs the optimal solution, i.e., the best solution found in the current search process, and the process ends. The entire process embodies an intelligent manufacturing strategy, utilizing the adaptive and iterative optimization capabilities of algorithms to achieve efficient and precise additive manufacturing of key components for water turbines.

[0074] Combination Figure 4This paper describes the application of the improved NSGA-II (Non-dominated sorting genetic algorithm II) algorithm in the additive manufacturing of key components for hydraulic turbines. First, an "initial population" is created, establishing a set of potential solutions. The fitness of these solutions is evaluated, i.e., the degree to which they meet design requirements and performance indicators. Non-dominated sorting and crowding distance calculation are performed to maintain population diversity and identify the optimal solution. After sorting and crowding distance calculation, the algorithm performs an "adaptive selection operation," dynamically adjusting the selection strategy based on the current population's fitness and diversity. New individuals are generated through "crossover" and "mutation," and the old and new populations are merged. The merged population undergoes "fast non-dominated sorting" and "elite strategy" screening to form a new generation of population. To further improve optimization performance, the algorithm dynamically adjusts crossover and mutation probabilities after updating the population and introduces a local search strategy to enhance population diversity and accelerate the finding of the optimal solution. The algorithm iterates continuously until a termination condition is met, such as reaching the maximum number of iterations or finding a satisfactory solution set. Once the termination condition is met, the algorithm outputs the optimal Pareto front solution set, which represents the solution that achieves the best trade-off among multiple objectives. The entire process embodies an intelligent manufacturing strategy, utilizing the algorithm's adaptability and iterative optimization capabilities to achieve efficient and precise additive manufacturing of key components for hydro turbines.

[0075] Combination Figure 5 This is a schematic diagram of the laser arc hybrid additive manufacturing (LHAM) wire and powder feeding principle of the present invention, specifically designed for additive manufacturing of key components of hydraulic turbines. In this system, the wire feeding system and the powder feeding system work together to achieve efficient material deposition and component forming. In the wire feeding system, the welding wire coil is guided to the end of the welding wire via a drive device and a wire feeding channel, while the laser source directs the laser beam to this end through a beam splitter to melt the welding wire and promote material deposition. Simultaneously, an atmosphere control system manages the gas inlet and outlet to ensure that an appropriate protective atmosphere is maintained during the melting process. On the other hand, in the powder feeding system, powder is guided from the powder container to the composite nozzle via a powder supply device and a powder feeding channel. Here, the powder is fed into the molten pool along with the molten welding wire to enhance material properties and forming quality. A robot or machine tool supports the composite nozzle and precisely controls its movement within the working area to achieve layer-by-layer construction of complex components. The formed part gradually forms in the molten pool and is monitored and adjusted in real time through a molten pool monitoring and control center. The control center manages key parameters such as laser power, wire feed speed, and powder feed rate to ensure the stability of the manufacturing process and the quality of the final product. This laser-arc hybrid additive manufacturing technology achieves efficient and precise manufacturing of key components for water turbines by precisely controlling material input and energy distribution, significantly improving manufacturing efficiency and product quality, and providing an innovative solution for the production of complex components.

[0076] Comparative Example:

[0077] Example 1: To verify the comprehensive advantages of the multi-algorithm collaborative optimization strategy proposed in this invention in improving the quality and efficiency of additive manufacturing of key components of water turbines, a systematic comparative analysis was conducted between the method described in this invention (multi-algorithm collaborative control), single-algorithm control (using only deep reinforcement learning), and traditional PID control methods, as shown in Table 1.

[0078] Table 1 Performance Comparison Analysis.

[0079]

[0080] Experimental data show that the multi-algorithm collaborative strategy adopted in this invention significantly outperforms the comparative methods in key performance indicators. Specifically, the repair accuracy reaches 0.05 mm, representing improvements of 58.3% and 80% compared to a single DRL algorithm (0.12 mm) and traditional PID control (0.25 mm), respectively; material utilization reaches 94.5%, representing improvements of 7.3% and 15.9%, respectively; repair time is shortened to 6.8 hours, with efficiency improvements of 23.6% and 45.2%; and strength recovery rate is increased to 98.2%, outperforming the comparative methods by 6.1% and 12.5%. These results fully demonstrate the synergistic advantages of multi-algorithm collaboration in deviation identification, path optimization, and dynamic parameter adjustment, achieving a comprehensive improvement in manufacturing accuracy, efficiency, and material utilization.

[0081] Example 2: To verify the optimization effect of the adaptive differential evolution algorithm used in this invention, the performance of the algorithm was compared with that of the standard differential evolution algorithm. The results are shown in Table 2.

[0082] Table 2 Performance Comparison Analysis.

[0083]

[0084] Experimental data show that the adaptive differential evolution algorithm of this invention, through a dynamic linear ascending crossover probability strategy (CR value gradually increasing from 0.45 to 0.99), enhances global exploration capabilities in the early stages of optimization and strengthens local search accuracy in the later stages, achieving a significant improvement in overall performance. Specifically, the algorithm improves the deviation recognition accuracy from 0.28mm to 0.15mm, an improvement of 23.6%; reduces the number of iterations required for convergence from 269 to 220, an efficiency improvement of 18.2%; at the same time, the population diversity maintenance index increases by 25.4%, and the local search success rate increases by 13.2%, effectively overcoming the optimization challenges brought by the complex curved surface and thin-walled structure of turbine blades.

[0085] Example 3: To verify the advantages of deep reinforcement learning (DRL) algorithms in dealing with dynamic disturbances such as thermal deformation in additive manufacturing, this study designed a comparative experiment between DRL intelligent control strategy and fixed parameter control. By simulating thermal accumulation and sudden disturbance scenarios in actual manufacturing, the two control methods were systematically compared in key indicators such as forming accuracy, process stability, and component performance. The specific experimental results are shown in Table 3.

[0086] Table 3 Performance Comparison Analysis.

[0087]

[0088] Experimental data shows that the DRL control strategy exhibits comprehensive advantages when facing dynamic disturbances such as thermal deformation: its thermal deformation compensation accuracy (0.07 mm) is improved by 66.7% compared to fixed parameter control (0.21 mm), the molten pool stability index is increased by 19.3 percentage points, the recovery time after disturbance is shortened by 68.8%, and the component forming dimension qualification rate is improved by 11.2 percentage points. This proves that DRL, through real-time sensing and intelligent adjustment, can significantly improve the anti-interference capability and forming quality of the manufacturing process.

[0089] The embodiments described above merely illustrate specific implementation methods of this application, and while the descriptions are detailed and specific, they should not be construed as limiting the scope of protection of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the technical solution of this application, and these modifications and improvements all fall within the scope of protection of this application.

Claims

1. A method for additive manufacturing of key components of a hydraulic turbine using multi-algorithm collaborative optimization, characterized in that, Includes the following steps: High-precision laser scanning technology was used to acquire three-dimensional point cloud data of key components of the water turbine, and the point cloud data was denoised and simplified to generate an accurate three-dimensional model. A multi-physics coupled finite element model was established based on a three-dimensional model to simulate the stress-strain field of the component under working load and its thermodynamic behavior during the repair process. An adaptive update differential evolution algorithm is used to identify and analyze geometric deviations in the additive manufacturing process in real time, and generate deviation correction schemes. The improved NSGA-II multi-objective optimization algorithm is used to optimize the laser deposition path globally, with the optimization objectives being repair path length, heat input uniformity, and material utilization rate. In the laser-arc hybrid additive manufacturing process, a deep reinforcement learning agent is introduced to dynamically adjust process parameters based on real-time monitoring data of the molten pool, thereby completing the remanufacturing of key components of the water turbine. The three-dimensional data obtained by the high-precision laser scanning technology includes the curved geometric features of the blade, surface wear, scratch texture, and spatial attitude. The specific steps for generating a 3D model include: point cloud registration, triangulation, and surface reconstruction; the measurement accuracy of laser scanning technology reaches 0.01 mm. The adaptive update differential evolution algorithm identifies deviations by evaluating the differences between manufacturing process data and design models, and generates correction schemes. The algorithm is adaptive and can dynamically adjust its parameters and strategies according to changes in the manufacturing environment and real-time feedback. The algorithm employs a dynamic linear ascending crossover probability scheme, with crossover probabilities... The calculation formula is: ， in, Set to 0.99, If set to 0.45, then This indicates the number of times the function is evaluated. This indicates the maximum number of function evaluations.

2. The additive manufacturing method for key components of a hydraulic turbine according to claim 1, characterized in that, The establishment of the finite element model includes: defining temperature-dependent material properties corresponding to the substrate and repair material, applying working condition loads and heat source models, and performing transient thermo-mechanical coupled field simulation to identify high stress concentration areas and potential deformation risk areas during the repair process.

3. The additive manufacturing method for key components of a hydraulic turbine according to claim 1, characterized in that, The adaptive update differential evolution algorithm also employs a control update strategy, the formula of which is: ， in, Representation strategy The probability of being called in the next iteration. Representation strategy The score or integral value in the current iteration The number of strategies.

4. The additive manufacturing method for key components of a hydraulic turbine according to claim 1, characterized in that, The improved NSGA-II algorithm introduces an adaptive tournament selection strategy, which dynamically adjusts the scale of tournament selection by evaluating the diversity and convergence state of the population to balance exploration and exploitation; the multi-objective functions include minimizing repair time, maximizing material utilization, and maximizing strength recovery rate.

5. The additive manufacturing method for key components of a hydraulic turbine according to claim 1, characterized in that, Laser-arc hybrid additive manufacturing technology combines the advantages of laser melting and electric arc melting. By precisely controlling the power, movement speed, and material feed rate of the laser beam and electric arc, it enables the layer-by-layer construction of turbine blades.

6. The additive manufacturing method for key components of a hydraulic turbine according to claim 1, characterized in that, The state space of the deep reinforcement learning algorithm includes the melt pool temperature, melt pool size, deposition height, and laser power deviation; the action space covers laser power adjustment, wire feed speed adjustment, and powder feed speed adjustment; the reward function adopts a segmented scoring mechanism, giving positive and negative rewards according to the degree of conformity of the deposition quality, guiding the algorithm to learn the optimal parameter combination.

7. The additive manufacturing method for key components of a hydraulic turbine according to claim 1, characterized in that, In the remanufacturing of turbine blades, the three-dimensional geometry and defect data acquired by laser scanning support the establishment of the finite element model. The stress concentration distribution and strength benchmark value output by the model serve as constraints for the adaptive update differential evolution algorithm. The surface curvature and thin-wall dimensional deviation data identified by the algorithm further guide the improvement of the NSGA-II algorithm to plan a low-deviation repair path. The relevant optimized path is directly used as the execution command of the laser arc hybrid additive manufacturing (LHAM) equipment.

8. The additive manufacturing method for key components of a hydraulic turbine according to claim 7, characterized in that, During the LHAM process, real-time data on molten pool temperature, size, and deposition height deviations are continuously input into the deep reinforcement learning (DRL) module. Based on this data, the DRL dynamically adjusts the wire / powder feeding speed and simultaneously feeds the deviation improvement results back to the differential evolution algorithm to form a closed loop, thereby achieving intelligent collaborative optimization of the remanufacturing process.

Images