Next Article in Journal
Software Defect Prediction Based on Machine Learning and Deep Learning Techniques: An Empirical Approach
Previous Article in Journal
Improving Distantly Supervised Relation Extraction with Multi-Level Noise Reduction
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Hypervolume-Based Multi-Objective Optimization Method Applying Deep Reinforcement Learning to the Optimization of Turbine Blade Shape

Department of Systems Innovation, The University of Tokyo, Tokyo 113-8654, Japan
*
Author to whom correspondence should be addressed.
AI 2024, 5(4), 1731-1742; https://doi.org/10.3390/ai5040085
Submission received: 19 August 2024 / Revised: 10 September 2024 / Accepted: 13 September 2024 / Published: 24 September 2024
(This article belongs to the Section AI Systems: Theory and Applications)

Abstract

:
A multi-objective turbine shape optimization method based on deep reinforcement learning (DRL) is proposed. DRL-based optimization methods are useful for repeating optimization tasks that arise in applications such as the design of turbines and automotive parts. In conventional research, DRL is applied only to single-optimization tasks. In this study, a multi-objective optimization method using improvements in hypervolume is proposed. The proposed method is applied to a benchmark problem and a turbine optimization problem. It succeeded in efficiently solving the problems, and Pareto optimal solutions are obtained.

1. Introduction

Deep reinforcement learning (DRL) utilizes deep neural networks (DNNs) in reinforcement learning algorithms. DRL is utilized in various industrial fields, mainly for control problems such as robotics [1,2,3], path planning [4,5], and automobile control [6,7]. DRL is also utilized in mechanical design applications [8], including the designs of airfoil shapes [9,10,11], turbine blades [12], and flow controls [13,14], as well as involving topology optimization [15].
In mechanical optimization, designers must find solutions to multi-objective optimization problems. In multi-objective optimization, it is necessary to obtain Pareto solutions [16]. Genetic algorithms, including the multi-objective GA (MOGA) [17], non-dominated sorting GA (NSGA) [18], and NSGA-II [19], are among the major methods used to perform multi-objective optimization [20]. Moreover, reinforcement learning has recently been utilized in mechanical design applications because of its high generalization capability. A properly trained RL agent has good generalization capability [9,12,21]. Hence, once trained, an RL agent is applicable to slightly different optimization problems. However, multi-objective DRL (MODRL) methods for mechanical designs have not been extensively studied.
Therefore, this study focuses on MODRL for use in multi-objective optimization problems. Several MODRL methods have been proposed for use in the field of computer science [5,22,23]. The weighted-sum method (WSM) is often used [24,25,26]. The WSM takes the weighted sum of the objective functions and treats the problem as a single-objective optimization problem. However, in using the WSM to obtain the Pareto front, we have to solve multiple weighted-sum problems by changing the weights. The computational cost of this procedure is high, and a non-convex Pareto front cannot be obtained using this procedure. Another commonly used method is the ε -constrained method [27,28,29], which uses only one primary objective function and uses the others as constraints.
The hypervolume, which is the volume of the hypercubes generated by the Pareto solutions, is often considered to obtain good Pareto solutions. Moreover, it is often used as both an indicator to measure the goodness of the obtained Pareto solutions and an objective function to be maximized [30,31,32,33]. Good Pareto solutions can be obtained by maximizing the hypervolumes.
In DRL, an extensive computational cost is incurred in the model training. Once trained, the agent can adapt to various problems, but if the objective function changes, then the agent must be retrained. Therefore, if DRL is coupled with the WSM or ε -constrained method, then the DRL agent must be trained multiple times to change the weights or constraints. In such cases, the computational costs are extremely high. In contrast, if the hypervolume-based method is coupled with DRL, then only one training phase is required, as the objective function and constraints do not change. Therefore, in this study, we propose a novel MODRL method that couples the hypervolume method with DRL.
The remainder of this paper is organized as follows. In Section 2, we formulate the proposed method by using the hypervolume. In Section 3, we describe the application of the proposed method to a benchmark problem. In Section 4, we formulate and solve the turbine optimization problem by using the proposed method. Finally, in Section 5, we conclude the paper.

2. Hypervolume-Based Multi-Objective Deep Reinforcement Learning

2.1. Deep Reinforcement Learning for Turbine Blades

RL obtains the optimal strategy for a Markov decision process (MDP). An MDP is a stochastic process that consists of states s t , actions a t , and rewards r t , where t is the time step. In MDP, the state s t and action a t are related only to those values of the previous time step t 1 . They are not related to the values of the other steps.
The proximal policy optimization (PPO) [34] algorithm was used in this study. PPO uses an actor network and a critic network. The actor network performs actions, whereas the critic network estimates the values of the actions and evaluates the policy of the actor network.
Ref. [12] proposed a DRL-based turbine optimization method that uses computational fluid dynamics (CFD) as an environment. In this method, an agent takes actions to modify turbine shapes. In this study, we coupled this DRL-based optimization method [12] with the hypervolume method.

2.2. Hypervolumes of Pareto Solutions

In multi-objective optimization, the solutions are divided into non-dominated solutions (i.e., Pareto solutions) and dominated solutions. We assume that the number of objective functions is d and that all functions are minimized. Let z i d ( i = 1 , 2 , , n ) be a solution, where i is the suffix and n is the number of solutions. A solution z * d is the dominant solution if
z k * > z k i , k ( 1 k d )
holds. Pareto solutions are defined as non-dominated solutions.
Let P t be a set of Pareto solutions at time step t and P t = { p 1 , p 2 , , p n } , where n is the number of Pareto solutions. In the two-dimensional case, p i = x i , y i . A hypercube C ( p i ) of solution p i is defined as a hypercube with p i and reference point p 0 as its diagonal points. In the two-dimensional case, C ( p i ) is defined as
C ( p i ) = x , y x i x x 0 , y i y y 0 .
The hypervolume V of the Pareto solutions P is defined as the hypervolume of the union of sets C ( p i ) :
V ( P ) = i C ( p i ) ,
where · denotes the hypervolume.

2.3. Reward Function

The reward function is defined according to three patterns: having the CFD computation fail, obtaining dominated solutions, and successfully finding a Pareto solution. The agent receives penalties in the first two patterns and rewards in the final pattern. An overview of rewards and penalties is provided in Figure 1, where C 1 , C 2 , C 3 , and C 4 are the constants and we let C 1 = 100 , C 2 = 10 , C 3 = 20 , and C 4 = 50 .
  • Pattern 1: Failed CFD
If the CFD computation does not converge, a fixed penalty is assigned as a reward.
r t = C 1 .
  • Pattern 2: Dominated solutions
If the CFD computation converges but is a dominated solution, then the agent obtains a penalty, according to the distance between the solution and the Pareto front.
r t = log C 2 d C 3 , if d < C 2 C 4 if d C 2 ,
where d is the distance between the solution and the Pareto front.
  • Pattern 3: Pareto solutions
If the obtained solution is a Pareto solution, the agent obtains a reward, according to the improvement in the hypervolume. The improvement is defined as follows:
r t = C 1 2 V ( P t ) V ( P t 1 ) 2 .

2.4. Optimization Algorithm

The optimization algorithm consists of two phases: training and optimization. The agent is trained in the training phase and used in the optimization phase. An overview of the training phase is provided in Figure 2. Each case corresponds to a single-flow condition and consists of 10 episodes, each of which corresponds to one initial solution. A CFD analysis was conducted on the initial shape. Subsequently, the state is input to the agent, and the agent modifies its shape. However, CFD computations sometimes fail because of unsuitable shape modifications. For example, a shape sometimes contains a self-intersection. The termination condition of each episode is the number of shape modifications reaching 50 or the CFD computation failing to converge. The trained agent was used in the optimization phase.

3. Benchmark Problem

3.1. Problem Definition

To demonstrate the proposed method, we solve the benchmark optimization problem defined in [35]. The benchmark problem is expressed as
min . f 1 = x 1 ,
min . f 2 = g h ,
subject   to g = 1 + 10 i = 2 N x i N 1 ,
h = 1 f 1 g α , if f 1 g , 0 , otherwise ,
x i 1 , 1 , i = 1 , , N .
The Pareto front of the problem is analytically solved and written as
f 1 = x 1 , f 2 = 1 f 1 α .
The Pareto front is upward concave if α > 1 , and we use α = 2 . If we use the reference point p 0 = ( 1 , 1 ) , then the hypervolume is V ( P ) = 1 / 3 .

3.2. Model Architecture

The state s t of the model is defined as follows:
s t = f 1 ( x t ) , f 2 ( x t ) , v ( x t , P t ) ,
where v ( x t , P t ) 2 denotes the vector from the current solution x t to the point nearest to the Pareto front. Therefore, the state vector is a four-dimensional vector. The action a t 2 is a vector consisting of a continuous variable within the range of 0.1 , 0.1 for each design variable x 1 and x 2 .
The architectures of the actor and critic networks are presented in Figure 3. The dropout rate was 0.2 . The weights were initialized using HeNormal [36] for the hidden layers and GlorotNormal [37] for the last layer. Table 1 lists the hyperparameters. Default settings were used for hyperparameters except that the learning rate was selected to achieve convergence.
The reward function is the same as that defined in Section 2.3, except for that of “Pattern 1”. Since CFD computation is not included in the benchmark problem, “Pattern 1: Failed CFD” (Equation (1)) is not considered.

3.3. Results

We trained the agent for 1000 episodes, as shown in Table 1. In each episode, the total hypervolume obtained by the agent was calculated, as shown in Figure 4. The moving average of the 10 episodes is also shown in red in Figure 4. The hypervolume after 1000 episodes of training was 0.324 , which corresponded to 97.2 % of the analytical solution ( 1 / 3 ). Moreover, Figure 4 shows that the score and hypervolume converged within 100 episodes.
The obtained solutions are presented in Figure 5. Without training (episode 0), the agent moved randomly, and Pareto solutions could not be obtained. In episode 1000, the agent succeeded in obtaining Pareto solutions. Although the analytical solution of the Pareto front was upwardly concave, the agent succeeded in obtaining a Pareto front.

4. Turbine Optimization Problem

4.1. Problem Definition

We optimized the turbine shape for cascaded turbine blades. We assumed repeated optimization tasks for the problem [12], which is a series of optimization problems whose flow conditions were slightly different from each other. Such optimization tasks are often present in industrial fields. In this study, we assume the flow conditions listed in Table 2. We randomly choose one flow condition from Table 2 and use it in one case. The objective function is to minimize the pressure loss and maximize the torque. The pressure loss is defined as
P loss = P T in P T ex P T ex P S ex ,
where P T in and P T ex are the total pressures at the inlet and outlet boundaries, respectively, and P S ex is the static pressure at the outlet boundary. The torque generated by the turbine blade is denoted as F. The optimization problem is formulated as follows:
min . P loss ,
min . F ,
subject   to | ϕ out ϕ target | 0.5 [ deg ] .
The turbine blade shape is defined by adding thickness to the camber line. That is, the upper and lower sides of the blade are defined as y = y c + y t and y = y c y t , respectively. The camber line is defined as follows
y c = m p 2 2 p x x 2 , [ 0 x p ] , m 1 p 2 1 2 p + 2 p x x 2 , [ p < x 1 ] ,
where m and p denote the maximum camber and its position, respectively. Subsequently, the shape was rotated according to the origin and scaled such that the chord length became 1 to set a positive stagger angle. The thickness y t is defined as
y t = t 20 ( 0.29690 x 1 2 0.12600 x 0 . 35160 2 + 0.28430 x 3 0.10150 x 4 ) ,
where t is the maximum thickness ratio.
The camber line and thickness distribution are then approximated using B-spline curves. The B-spline curves and control points are presented in Figure 6. Eventually, the blade shape is represented by y = y c B + y t B and y = y c B y t B , where y c B and y t B are the B-spline curves of the camber line and thickness distribution, respectively. The control points of the B-spline curves are used as the design variables.

4.2. CFD Computation

The CFD computation is carried out using the MISES software suite, which is a collection of programs for cascade analysis and design [38]. An example computational grid is presented in Figure 7. The pressure distribution is obtained by using MISES. The pressure values are sampled from the pressure distributions on the pressure and suction sides, which are represented by p p and p s , respectively. In total, 120 samples are collected from each side. The coordinates of the sampled points are represented by s , which are the coordinates of the streamline. The pressure values p p and p s are normalized using the peak pressure at the leading edge p 0 . Hence, p p / p 0 and p s / p 0 are used as state variables, as explained in Section 4.3. Figure 8 shows the pressure distributions on the pressure and suction side surfaces.

4.3. Model Architecture

The state variable s t is constructed using the results of the CFD analysis. The elements of s t are as follows. In total, the number of dimensions of s t was 1093.
1.
Information of each case:
  • Inlet Mach number,
  • Inlet flow angle,
  • Pitch,
  • Target outlet flow angle.
2.
Geometric information:
  • Coordinates of the camber line ( x c , y c ) ,
  • Coordinates of the thickness distribution ( x t , y t ) ,
  • Metal angle of the leading edge.
3.
Information of the flow field:
  • Pressure distribution on the pressure side p p / p 0 ,
  • Pressure distribution on the suction side p s / p 0 ,
  • Outlet flow angle.
4.
Information of the Pareto front:
  • Vector from the current point to the nearest Pareto front.
If the CFD computation failed to converge, information on the flow field could not be obtained. In this case, the following values were used: s = 0 , p p / p 0 = s p / p 0 = 0 , and an outlet flow angle of 20 [deg].
The action a t is a vector in 17 that corresponds to the
  • Difference between the x and y coordinates of the four control points of a camber line,
  • Difference between the x and y coordinates of the four control points of a thickness distribution,
  • Difference of the stagger angle.
The range of a t was [ 1 , 1 ] , and a t was scaled for each design variable. The scale factors for the camber line were 0.003 , 0.001 for the thickness distribution, and 0.2 for the stagger angle. As explained in Section 4.1, the turbine shape was generated using the control points and then rotated according to the stagger angle. The shape was then scaled such that the chord line became 1.
Figure 9 illustrates the DNN architectures of the actor and critic networks.

4.4. Results

The agent was optimized after training for 1000 episodes. One case was randomly chosen, and 10 initial solutions were used. The results of the objective values are shown in Figure 10, which indicates that the agent successfully finds Pareto solutions. The hypervolume history is also shown in Figure 11. When a new initial solution is set, the agent succeeds in finding new Pareto solutions. Hence, the hypervolume improves drastically.
The Pareto solutions obtained depend on the initial solution. Therefore, it is necessary to use multiple initial solutions.

5. Conclusions

A novel multi-objective optimization method using the hypervolume of the Pareto front is proposed for a turbine optimization problem. This method uses the improvement in hypervolumes as a reward. The proposed method was validated using a benchmark problem and then applied to a turbine optimization problem.
In the hypervolume-based method, the objective function and constraint conditions are not changed. Hence, only one DRL agent is required to obtain Pareto solutions. If we use the WSM or ε -constrained method, then different DRL agents must be trained when the weights or constraints are changed. This generalization capability reduces the computational cost. Moreover, because DRL has a rich generalization capability, it can be used for multiple tasks.

Author Contributions

Conceptualization, K.Y.; methodology, K.Y. and R.Y.; software, K.Y. and R.Y.; validation, K.Y. and R.Y.; formal analysis, K.Y. and R.Y.; investigation, K.Y. and R.Y.; resources, K.Y.; data curation, R.Y.; writing—original draft preparation, K.Y.; writing—review and editing, K.Y.; visualization, R.Y.; supervision, K.Y., S.O. and K.S.; project administration, K.Y.; funding acquisition, K.Y. All authors have read and agreed to the published version of the manuscript.

Funding

This work was supported by JSPS KAKENHI Grant Number JP23K13239.

Informed Consent Statement

Not applicable.

Data Availability Statement

Data are available upon request.

Conflicts of Interest

The first author is a board member of MJOLNIR SPACEWORKS.

References

  1. Morales, E.F.; Murrieta-Cid, R.; Becerra, I.; Esquivel-Basaldua, M.A. A survey on deep learning and deep reinforcement learning in robotics with a tutorial on deep reinforcement learning. Intell. Serv. Robot. 2021, 14, 773–805. [Google Scholar] [CrossRef]
  2. Zhu, K.; Zhang, T. Deep reinforcement learning based mobile robot navigation: A review. Tsinghua Sci. Technol. 2021, 26, 674–691. [Google Scholar] [CrossRef]
  3. Zhao, W.; Queralta, J.P.; Westerlund, T. Sim-to-Real Transfer in Deep Reinforcement Learning for Robotics: A Survey. In Proceedings of the 2020 IEEE Symposium Series on Computational Intelligence (SSCI), Canberra, ACT, Australia, 1–4 December 2020; pp. 737–744. [Google Scholar] [CrossRef]
  4. Liu, R.; Qu, Z.; Huang, G.; Dong, M.; Wang, T.; Zhang, S.; Liu, A. DRL-UTPS: DRL-Based Trajectory Planning for Unmanned Aerial Vehicles for Data Collection in Dynamic IoT Network. IEEE Trans. Intell. Veh. 2023, 8, 1204–1218. [Google Scholar] [CrossRef]
  5. Li, K.; Zhang, T.; Wang, R. Deep Reinforcement Learning for Multiobjective Optimization. IEEE Trans. Cybern. 2021, 51, 3103–3114. [Google Scholar] [CrossRef] [PubMed]
  6. Liao, J.; Liu, T.; Tang, X.; Mu, X.; Huang, B.; Cao, D. Decision-Making Strategy on Highway for Autonomous Vehicles Using Deep Reinforcement Learning. IEEE Access 2020, 8, 177804–177814. [Google Scholar] [CrossRef]
  7. Wang, H.; Yuan, S.; Guo, M.; Li, X.; Lan, W. A deep reinforcement learning-based approach for autonomous driving in highway on-ramp merge. Proc. Inst. Mech. Eng. Part D J. Automob. Eng. 2021, 235, 2726–2739. [Google Scholar] [CrossRef]
  8. Ramu, P.; Thananjayan, P.; Acar, E.; Bayrak, G.; Park, J.W.; Lee, I. A survey of machine learning techniques in structural and multidisciplinary optimization. Struct. Multidiscip. Optim. 2022, 65, 266. [Google Scholar] [CrossRef]
  9. Yonekura, K.; Hattori, H. Framework for design optimization using deep reinforcement learning. Struct. Multidiscip. Optim. 2019, 60, 1709–1713. [Google Scholar] [CrossRef]
  10. Lou, J.; Chen, R.; Liu, J.; Bao, Y.; You, Y.; Chen, Z. Aerodynamic optimization of airfoil based on deep reinforcement learning. Phys. Fluids 2023, 35, 037128. [Google Scholar] [CrossRef]
  11. Du, Q.; Liu, T.; Yang, L.; Li, L.; Zhang, D.; Xie, Y. Airfoil design and surrogate modeling for performance prediction based on deep learning method. Phys. Fluids 2022, 34, 015111. [Google Scholar] [CrossRef]
  12. Yonekura, K.; Hattori, H.; Shikada, S.; Maruyama, K. Turbine blade optimization considering smoothness of the Mach number using deep reinforcement learning. Inf. Sci. 2023, 642, 119066. [Google Scholar] [CrossRef]
  13. Vignon, C.; Rabault, J.; Vinuesa, R. Recent advances in applying deep reinforcement learning for flow control: Perspectives and future directions. Phys. Fluids 2023, 35, 031301. [Google Scholar] [CrossRef]
  14. Tang, H.; Rabault, J.; Kuhnle, A.; Wang, Y.; Wang, T. Robust active flow control over a range of Reynolds numbers using an artificial neural network trained through deep reinforcement learning. Phys. Fluids 2020, 32, 053605. [Google Scholar] [CrossRef]
  15. Jang, S.; Yoo, S.; Kang, N. Generative Design by Reinforcement Learning: Enhancing the Diversity of Topology Optimization Designs. Comput.-Aided Des. 2022, 146, 103225. [Google Scholar] [CrossRef]
  16. Deb, K. Multi-Objective Optimization Using Evolutionary Algorithms; Wiley: Hoboken, NJ, USA, 2001. [Google Scholar]
  17. Murata, T.; Ishibuchi, H. MOGA: Multi-objective genetic algorithms. In Proceedings of the 1995 IEEE International Conference on Evolutionary Computation, Perth, WA, Australia, 29 November–1 December 1995; Volume 1, p. 289. [Google Scholar] [CrossRef]
  18. Srinivas, N.; Deb, K. Muiltiobjective Optimization Using Nondominated Sorting in Genetic Algorithms. Evol. Comput. 1994, 2, 221–248. [Google Scholar] [CrossRef]
  19. Deb, K.; Pratap, A.; Agarwal, S.; Meyarivan, T. A fast and elitist multiobjective genetic algorithm: NSGA-II. IEEE Trans. Evol. Comput. 2002, 6, 182–197. [Google Scholar] [CrossRef]
  20. Goldberg, D.E. Genetic Algorithms in Search, Optimization, and Machine Learning; Addison-Wesley: Hoboken, NJ, USA, 1989. [Google Scholar] [CrossRef]
  21. Dussauge, T.P.; Sung, W.J.; Fischer, O.J.P.; Mavris, D.N. A reinforcement learning approach to airfoil shape optimization. Sci. Rep. 2023, 13, 9753. [Google Scholar] [CrossRef]
  22. Keat, E.Y.; Sharef, N.M.; Yaakob, R.; Kasmiran, K.A.; Marlisah, E.; Mustapha, N.; Zolkepli, M. Multiobjective Deep Reinforcement Learning for Recommendation Systems. IEEE Access 2022, 10, 65011–65027. [Google Scholar] [CrossRef]
  23. Al-Jumaily, A.; Mukaidaisi, M.; Vu, A.; Tchagang, A.; Li, Y. Examining multi-objective deep reinforcement learning frameworks for molecular design. Biosystems 2023, 232, 104989. [Google Scholar] [CrossRef]
  24. Yang, X.S. Chapter 14—Multi-Objective Optimization. In Nature-Inspired Optimization Algorithms; Yang, X.S., Ed.; Elsevier: Oxford, UK, 2014; pp. 197–211. [Google Scholar] [CrossRef]
  25. Kaim, A.; Cord, A.F.; Volk, M. A review of multi-criteria optimization techniques for agricultural land use allocation. Environ. Model. Softw. 2018, 105, 79–93. [Google Scholar] [CrossRef]
  26. Kalayci, C.B.; Ertenlice, O.; Akbay, M.A. A comprehensive review of deterministic models and applications for mean-variance portfolio optimization. Expert Syst. Appl. 2019, 125, 345–368. [Google Scholar] [CrossRef]
  27. Mesquita-Cunha, M.; Figueira, J.R.; Barbosa-Póvoa, A.P. New ϵ-constraint methods for multi-objective integer linear programming: A Pareto front representation approach. Eur. J. Oper. Res. 2023, 306, 286–307. [Google Scholar] [CrossRef]
  28. Yang, Z.; Cai, X.; Fan, Z. Epsilon constrained method for constrained multiobjective optimization problems: Some preliminary results. In Proceedings of the Companion Publication of the 2014 Annual Conference on Genetic and Evolutionary Computation, New Vancouver, BC, Canada, 12–16 July 2014; GECCO Comp’14. pp. 1181–1186. [Google Scholar] [CrossRef]
  29. Becerra, R.L.; Coello, C.A.C. Solving Hard Multiobjective Optimization Problems Using ϵ-Constraint with Cultured Differential Evolution. In Proceedings of the Parallel Problem Solving from Nature—PPSN IX, 9th International Conference, Reykjavik, Iceland, 9–13 September 2006; Runarsson, T.P., Beyer, H.G., Burke, E., Merelo-Guervós, J.J., Whitley, L.D., Yao, X., Eds.; Springer: Berlin/Heidelberg, Germany, 2006; pp. 543–552. [Google Scholar]
  30. Shang, K.; Ishibuchi, H.; He, L.; Pang, L.M. A Survey on the Hypervolume Indicator in Evolutionary Multiobjective Optimization. IEEE Trans. Evol. Comput. 2021, 25, 1–20. [Google Scholar] [CrossRef]
  31. Auger, A.; Bader, J.; Brockhoff, D.; Zitzler, E. Hypervolume-based multiobjective optimization: Theoretical foundations and practical implications. Theor. Comput. Sci. 2012, 425, 75–103. [Google Scholar] [CrossRef]
  32. Shang, K.; Ishibuchi, H. A New Hypervolume-Based Evolutionary Algorithm for Many-Objective Optimization. IEEE Trans. Evol. Comput. 2020, 24, 839–852. [Google Scholar] [CrossRef]
  33. Guerreiro, A.P.; Fonseca, C.M.; Paquete, L. The Hypervolume Indicator: Computational Problems and Algorithms. ACM Comput. Surv. 2021, 54, 119. [Google Scholar] [CrossRef]
  34. Schulman, J.; Wolski, F.; Dhariwal, P.; Radford, A.; Klimov, O. Proximal Policy Optimization Algorithms. arXiv 2017, arXiv:1707.06347. [Google Scholar]
  35. Deb, K.; Pratap, A.; Meyarivan, T. Constrained Test Problems for Multi-objective Evolutionary Optimization. In Proceedings of the Evolutionary Multi-Criterion Optimization, First International Conference, EMO 2001, Zurich, Switzerland, 7–9 March 2001; Zitzler, E., Thiele, L., Deb, K., Coello Coello, C.A., Corne, D., Eds.; Springer: Berlin/Heidelberg, Germany, 2001; pp. 284–298. [Google Scholar]
  36. He, K.; Zhang, X.; Ren, S.; Sun, J. Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification. In Proceedings of the 2015 IEEE International Conference on Computer Vision (ICCV), Santiago, Chile, 7–13 December 2015; pp. 1026–1034. [Google Scholar] [CrossRef]
  37. Glorot, X.; Bengio, Y. Understanding the difficulty of training deep feedforward neural networks. In Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics, Sardinia, Italy, 13–15 May 2010; Teh, Y.W., Titterington, M., Eds.; Proceedings of Machine Learning Research. Volume 9, pp. 249–256. [Google Scholar]
  38. Drela, M.; Youngren, H. A User’s Guide to MISES 2.63; MIT Aerospace Computational Design Laboratory: Cambrige, MA, USA, 2008. [Google Scholar]
Figure 1. Rewards and penalties.
Figure 1. Rewards and penalties.
Ai 05 00085 g001
Figure 2. Overview of the DRL-based optimization method.
Figure 2. Overview of the DRL-based optimization method.
Ai 05 00085 g002
Figure 3. Actor and critic networks for the benchmark problem.
Figure 3. Actor and critic networks for the benchmark problem.
Ai 05 00085 g003
Figure 4. Score and hypervolume of the benchmark problem. (The red line represents the moving average).
Figure 4. Score and hypervolume of the benchmark problem. (The red line represents the moving average).
Ai 05 00085 g004
Figure 5. Solutions obtained after different episodes.
Figure 5. Solutions obtained after different episodes.
Ai 05 00085 g005
Figure 6. Example of an airfoil shape.
Figure 6. Example of an airfoil shape.
Ai 05 00085 g006
Figure 7. Computation grids for CFD computation.
Figure 7. Computation grids for CFD computation.
Ai 05 00085 g007
Figure 8. Example of a pressure distribution on the pressure and suction side surfaces.
Figure 8. Example of a pressure distribution on the pressure and suction side surfaces.
Ai 05 00085 g008
Figure 9. Actor and critic networks for the turbine problem.
Figure 9. Actor and critic networks for the turbine problem.
Ai 05 00085 g009
Figure 10. Objective value of the turbine problem.
Figure 10. Objective value of the turbine problem.
Ai 05 00085 g010
Figure 11. History of the hypervolume (HV) of a turbine problem.
Figure 11. History of the hypervolume (HV) of a turbine problem.
Ai 05 00085 g011
Table 1. Hyperparameters of the benchmark problem.
Table 1. Hyperparameters of the benchmark problem.
HyperparameterValueHyperparameterValue
Episodes1000Batch size512
Learning rate 10 3 to 10 5 Epochs10
Discount rate γ 0.99GAE discount rate λ 0.95
Dropout ratio 0.2 Optimization algorithmAdam
Table 2. Flow conditions of repeated optimization tasks of turbine blades.
Table 2. Flow conditions of repeated optimization tasks of turbine blades.
HyperparameterValue
Inlet flow angle0.3–0.5 [rad]
Inlet Mach number0.2–0.35 [Mach]
Pitch1.2–1.5 [rad]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Yonekura, K.; Yamada, R.; Ogawa, S.; Suzuki, K. Hypervolume-Based Multi-Objective Optimization Method Applying Deep Reinforcement Learning to the Optimization of Turbine Blade Shape. AI 2024, 5, 1731-1742. https://doi.org/10.3390/ai5040085

AMA Style

Yonekura K, Yamada R, Ogawa S, Suzuki K. Hypervolume-Based Multi-Objective Optimization Method Applying Deep Reinforcement Learning to the Optimization of Turbine Blade Shape. AI. 2024; 5(4):1731-1742. https://doi.org/10.3390/ai5040085

Chicago/Turabian Style

Yonekura, Kazuo, Ryusei Yamada, Shun Ogawa, and Katsuyuki Suzuki. 2024. "Hypervolume-Based Multi-Objective Optimization Method Applying Deep Reinforcement Learning to the Optimization of Turbine Blade Shape" AI 5, no. 4: 1731-1742. https://doi.org/10.3390/ai5040085

APA Style

Yonekura, K., Yamada, R., Ogawa, S., & Suzuki, K. (2024). Hypervolume-Based Multi-Objective Optimization Method Applying Deep Reinforcement Learning to the Optimization of Turbine Blade Shape. AI, 5(4), 1731-1742. https://doi.org/10.3390/ai5040085

Article Metrics

Back to TopTop