Multi-objective Design Exploration Research Articles

Multi-objective robustness indicators for evaluation and exploration of design margins

ABSTRACT Specification of design margins serves the purpose of mitigating the ubiquitous uncertainty of the engineering design process. This paper addresses the twofold challenge of establishing a suitable concept of margins early in the design process, before the design is mature enough for detailed modelling, and evaluating margin inter-dependencies. As margins are often specified in an unstructured manner, there is a risk that one margin might lead to a need for more margins, e.g. tight tolerances, downstream. To avoid adverse margin dependencies, five Multi-Objective Robustness Indicators based on multi-objective design exploration are presented: The Relative Utopia Index, the Relative Pareto Shape Index, the Relative Compromise Index, the Relative Average Sensitivity Index, and the Average Feasible Variable Range. They aggregate optimality, trade-offs and sensitivity of design configurations. The indicators are applied to a case example, a flat head tank design, to illustrate the effect and inter-dependency of margin specification. Three margin types were identified. The first type affects only the Objective Space size, the second type affects only the Pareto set size and the third type affects Pareto set shape. The third margin type is the most critical as it leads to harder compromises between design objectives later in the design process.

IronMan-Pro: Multiobjective Design Space Exploration in HLS via Reinforcement Learning and Graph Neural Network-Based Modeling

Despite the great success of high-level synthesis (HLS) tools, we observe several unresolved challenges: 1) the high-level abstraction of HLS programming styles sometimes conceals optimization opportunities; 2) the actual quality of resulting RTL designs is hard to predict; and 3) existing HLS tools do not provide flexible tradeoff (Pareto) solutions among different objectives and constraints. To this end, we propose an end-to-end framework, namely, IronMan-Pro. The primary goal is to enable a flexible and automated design space exploration (DSE), to provide either optimized solutions under user-specified constraints or Pareto tradeoffs among different objectives (such as resource types, area, and latency). IronMan-Pro consists of three components: 1) GPP, a highly accurate graph-neural-network-based performance and resource predictor; 2) RLMD, a reinforcement-learning-based multiobjective design exploration engine for optimal resource allocation strategies, aiming to provide Pareto solutions among different objectives; and 3) CT, a code transformer to assist RLMD and GPP, which extracts the data flow graphs from original HLS C/C++ and automatically generates synthesizable code with optimized HLS directives. Experimental results show that, 1) GPP achieves high prediction accuracy, reducing the prediction errors of HLS tools by <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$10.9\times $ </tex-math></inline-formula> in resource utilization and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$5.7\times $ </tex-math></inline-formula> in critical path (CP) timing; 2) compared with meta-heuristic-based techniques, IronMan-Pro generates superior solutions improving resource utilization by <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$16.0\% \sim 29.5\%$ </tex-math></inline-formula> and CP timing by 7.6%–16.5%; and 3) under user-specified constraints, IronMan-Pro can find satisfying solutions over 96% of the cases, more than twice as many as that of meta-heuristic-based techniques and with a speedup of up to <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$400\times $ </tex-math></inline-formula> . This work demonstrates the great potential of applying machine learning algorithms in the electronic design automation domain, especially for the hard-to-solve problems, such as timing estimation and optimization. IronMan-Pro is available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/lydiawunan/IronMan</uri> .

A study on multi-objective design exploration based on deep clustering and logistic regression

Multi-objective design exploration is a framework to aim to extract knowledge to make rational decisions in multi-objective design problems. Since a set of Pareto solutions obtained by multi-objective topology optimization contains various types implicitly, it is difficult to extract the useful design knowledge by conventional data mining methods based on the objective function values. This study focuses on a framework for multi-objective design exploration that reveals the structure of the solution space by deep clustering and logistic regression for multi-objective topology optimization. In the framework, the deep clustering technique classifies the Pareto solution set in design variables space based on structural similarity to reveal representative types, then the logistic regression technique identifies the classification in evaluation criteria space that explains the differences in the types, and then designers arrange these results into design knowledge. This paper discusses its basic validity and possibilities through an application to a simple design problem of the conceptual design of a bridge structure.

Aerodynamic robustness optimization and design exploration of centrifugal compressor impeller under uncertainties

Aerodynamic robustness optimization of centrifugal compressor impeller under multiple uncertainties is an arduous task, due to high dimension, meta modelling workload and trial and error optimization, and an automatic optimization process is often treated as a black box, the underlying mechanism are often not well understood and explored. This paper aims to provide an efficient robustness optimization procedure and explore the underlying physical mechanism for centrifugal compressor impeller under operational and geometric uncertainties. A novel approach is proposed for the multi-objective aerodynamic robustness optimization and design exploration of a centrifugal compressor impeller. A neural-network-based Kriging model is constructed and integrated into the aerodynamic robustness optimization The correlation between the design variables and performance parameters are explored from optimization and directly visualized by self-organizing mapping. Compared to the initial impeller, the average pressure ratio of the optimized impeller increases by 9.3%, the average isentropic efficiency increases by 6.7%, the standard deviations of pressure ratio and efficiency decrease by 7.5% and 15.4% respectively, and the acoustic power level decreases by 11dB. The neural-network-based Kriging model exhibits preferable accuracy for uncertain approximation modeling. The data visualization and interpretation facilitate designers to perform efficient design optimizations. The proposed approach supports design explorations for different applications of turbomachinery.

Verification Method to Optimize Multiple Functions of Valve Train in a Short Time Using Multi-objective Design Exploration

Verification Method to Optimize Multiple Functions of Valve Train in a Short Time Using Multi-objective Design Exploration

Multi-objective design exploration using self-organization map for composite structures made by automated fiber placement

In this study, we apply the sequential approximate multi-objective optimization (SAMOO) to an open-hole laminated plate with partially different lay-up configurations for assuming fabrication by the AFP equipment. The SAMOO is a method to search for the optimal design with a small number of evaluations of the objective function while adding appropriate sample points to the response surface. When the optimization problem deals with many objective functions, it becomes difficult to visualize the design candidates and to understand the relationship between the objective functions. Thus, the self-organizing map (SOM) which displays multi-dimensional data in two dimensions is employed for visualization and clustering of obtained non-dominated solutions. Numerical results showed that the trade-off relationships among the three objectives were clarified and design alternatives could be explored according to the design tastes, and it was confirmed that the SOM was effective for the present optimization problem.

The Effect of Fuel Cell and Battery Size on Efficiency and Cell Lifetime for an L7e Fuel Cell Hybrid Vehicle

The size of the fuel cell and battery of a Fuel Cell Hybrid Electric Vehicle (FCHEV) will heavily affect the overall performance of the vehicle, its fuel economy, driveability, and the rates of fuel cell degradation observed. An undersized fuel cell may experience accelerated ageing of the fuel cell membrane and catalyst due to excessive heat and transient loading. This work describes a multi-objective design exploration exercise of fuel cell size and battery capacity comparing hydrogen fuel consumption, fuel cell lifetime, vehicle mass and running cost. For each system design considered, an individually optimised Energy Management Strategy (EMS) has been generated using Stochastic Dynamic Programming (SDP) in order to prevent bias to the results due to the control strategy. It has been found that the objectives of fuel efficiency, lifetime and running cost are largely complimentary, but degradation and running costs are much more sensitive to design changes than fuel efficiency and therefore should be included in any optimisation. Additionally, due to the expense of the fuel cell, combined with the dominating effect of start/stop cycling degradation, the optimal design from an overall running cost perspective is slightly downsized from one which is optimised purely for high efficiency.

Multi-objective design exploration approach for Ravenscar real-time systems

This article deals with the design exploration and verification of real-time critical systems. Assigning the functions to the tasks of the target real-time operating system is a part of the design process. Finding a suitable design involves many important design decisions that have a strong impact on the system quality criteria. However, with the increasing complexity and scale of today’s systems and the large number of possible design solutions, making design decisions while balancing conflicting quality criteria becomes error-prone and unmanageable for designers. We propose an automated method using a multi-objective evolutionary algorithm guided by an architectural clustering technique. This method allows designers to search the design space for schedulable solutions with respect to multiple competing performance criteria. To assess our method, several evaluations were performed. One of them shows that we were able to produce the exact optimal solution sets for $$55\%$$ of the studied problem instances.

Multiphysics design optimization of RF-MEMS switch using response surface methodology

This paper presents the multi-objective geometric design exploration and optimization of an electrostatic Symmetric Toggle RF-MEMS switch (STS), considering both the electromechanical and RF characteristics simultaneously. The output responses considered for the STS switch optimization are pull-in voltage, switching time, insertion loss in the on-state and isolation in the off-state. Metamodels for the output responses, with respect to geometric design parameters, are developed using Design of Experiments (DOE) based Response Surface Methodology (RSM) and Finite Element Method (FEM) simulations. A single optimization objective function, considering all the four output responses and microfabrication process constraints, is defined and optimized for the design factors using combined desirability function and heuristic search algorithm approach. The predicted values of the output responses are verified through both the electromechanical and electromagnetic FEM simulations. The effect of residual stress, developed in the RF-MEMS switch during the sacrificial layer removal step of the microfabrication process, on both the electromechanical and RF characteristics of the final optimized switch geometry is analyzed using coupled structural-thermal-electric FEM simulations. The proposed DOE and RSM based design optimization technique can be implemented for the design space exploration and optimization of complex MEMS devices which involve coupled multiphysics interactions.

Heat flux reduction mechanism induced by a combinational opposing jet and cavity concept in supersonic flows

The thermal protection on the surface of hypersonic vehicles attracts an increasing attention worldwide, especially when the vehicle enters the atmosphere at high speed. In the current study, the Reynolds-averaged Navier-Stokes (RANS) equations coupled with the Menter׳s shear stress transport (SST) model have been employed to investigate the heat flux reduction mechanism induced by the variations of the cavity configuration, the jet pressure ratio and the injectant molecular weight in the combinational opposing jet and cavity concept. The length of the cavity is set to be 6mm, 8mm and 10mm in order to make sure that the cavity configuration is the “open” cavity, and the jet pressure ratio is set to be 0.4, 0.6 and 0.8 in order to make sure that the flow field is steady. The injectant is set to be nitrogen and helium. The obtained results show that the aft angle of the cavity only has a slight impact on the heat flux reduction, and the heat flux peak decreases with the decrease of the length of the cavity. The design of the thermal protection system for the hypersonic blunt body is a multi-objective design exploration problem, and the heat flux distribution depends on the jet pressure ratio, the aft wall of the cavity and the injectant molecular weight. The heat flux peak decreases with the increase of the jet pressure ratio when the aft angle of the cavity is large enough, and this value is 45°.

多目的設計探査によるフィラー充填ゴムのモルフォロジー設計

多目的設計探査によるフィラー充填ゴムのモルフォロジー設計

材料インフォマティクスによる革新的な材料開発に向けて

Expectation to materials informatics increases toward innovative materials development. Materials informatics is the methodology based on numerical simulation and data mining using machine learning. The objective of this report is to mention an introductive effort for the materials informatics of filled rubber. At first, two data mining techniques i.e. self-organizing map and decision tree which is a kind of machine learning are presented to show their effectiveness with the example of the multi-objective design exploration for the tire shape design. The case for microstructure designs of rubber material becomes the same procedure, too. Next, we mention the other important issues for the materials informatics of materials having microstructure i.e. characterization of morphology and construction of simulation model using the morphological parameters. Morphological measures and random morphological modeling are introduced for their purpose.

Multiobjective Design Exploration of Propeller Airfoils at Low-Reynolds and High-Mach Number Conditions towards Mars Airplane

Multiobjective Design Exploration of Propeller Airfoils at Low-Reynolds and High-Mach Number Conditions towards Mars Airplane

Multi-Objective Aerodynamic Design Exploration of a Static Stable VTOL Rocket

Multi-Objective Aerodynamic Design Exploration of a Static Stable VTOL Rocket

1219 DESTINY軌道設計の多目的設計探査(多目的最適化)

1219 DESTINY軌道設計の多目的設計探査(多目的最適化)

S052034 Multi-Objective Design Exploration of DBD Plasma Actuator for Low Reynolds Number Application

To understand the effects of working parameters of the DBD plasma actuator, investigation with a Multi-Objective Design Exploration(MODE) framework is performed. Experiments with a NACA 0015 airfoil fixed to the stall angle of 12 degrees are conducted at Reynolds number of 63,000. The optimization objectives are to maximize the lift coefficient C_l and minimize the power consumption P. The design variables consist of input power parameters. Despite the small population size, an approximate Pareto-optimal front is found. In the objective function space, a region where there exists a linear relationship between C_l and P is found. After a threshold value, the value of C_l seems to saturate. This paper concentrates on this linear region of the objective function space. We use numerical simulation results with similar parameters to several Pareto-optimal experiment cases to compliment our discussion.

0634 進化計算とLESを用いたロケット射点形状の空力音響多目的設計探査(OS6-7 噴流,後流およびはく離流れ現象の探求と先端的応用,OS6 噴流,後流およびはく離流れ現象の探求と先端的応用)

In this study, multiobjective design exploration for a rocket launch site is conducted using evolutionary computation with large eddy simulation to understand the acoustic characteristics associated with various launch sites. The launch site is described by curved surface. The flat plate inclined with 45 degree is considered as the reference configuration. The objective functions of multiobjective aero-acoustic design optimization are, 1) minimization of averaged sound pressure level near the payload fairing, 2) minimization of maximum pressure on the curved surface of the rocket launch site, and 3) minimization of the change of the curved surface from the reference configuration. Three-dimensional compressible Navier-Stokes equations are solved with the modified weighted compact nonlinear scheme. The total number of evaluation is 2500, and the evaluation of one generation necessitates the use of 6500 nodes using "K" supercomputer. As the results of the flow field analysis of some characteristic non-dominated solutions show the characteristics of acoustic wave and the location of the maximum gradient of the curved surface.

Extraction of design rules from multi-objective design exploration (MODE) using rough set theory

Multi-objective design exploration (MODE) and its application for design rule extraction are presented. MODE reveals the structure of design space from the trade-off information. The self-organizing map (SOM) is incorporated into MODE as a visual data-mining tool for design space. SOM divides the design space into clusters with specific design features. The sufficient conditions for belonging to a cluster of interest are extracted using rough set theory. The resulting MODE was applied to the multidisciplinary wing design problem, which revealed a cluster of good designs, and we extracted the design rules of such designs successfully.

Design Exploration of Helicopter Blades for HSI Noise and Aerodynamic Performance

In this study, the high-speed impulsive (HSI) noise and aerodynamic performance of helicopter blades were improved using multi-objective design exploration (MODE) comprised of multi-objective shape optimization and data mining. To broaden the design space of the problem, geometry definitions with a high degree of freedom were adopted. As a result, remarkable improvements in both HSI noise and aerodynamic performance were achieved. For data mining, analysis of variance (ANOVA) and self-organizing map (SOM) were used to extract design knowledge of the helicopter blades. The results indicated that tip chord length and blade twist are important factors for HSI noise and aerodynamic performance, respectively. Based on the design knowledge obtained from data mining, an additional design variable (twist) was introduced. The solutions obtained from the shape optimization using variable twist showed better blade loading performance than that from shape optimization using fixed twist.

Visualization of Aircraft Design Space

This article discusses visualization of design space by using Self-Organizing Map (SOM) utilized in Multi-Objective Design Exploration (MODE), which has been applied to an actual aircraft design. SOM derived from MODE visualizes performance of design. Visualization and data mining techniques reveal tradeoffs in the design space, sweet-spot in the design space, and key features of the designs in the sweet-spot. Of course, sweet-spot may or may not exist. Many design problems requires detailed analysis among objective functions. However, the visualization of design space will lead to the design knowledge. Visualization is very important in the field of design.