srMO-BO-3GP: A sequential regularized multi-objective constrained Bayesian optimization for design applications.
Abstract not provided.
SEARCH · Engineering Papers
Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.
Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.
Abstract not provided.
Abstract not provided.
The paper presents a novel approach to applying Bayesian Optimization (BO) in predicting an unknown constraint boundary, also representing the discontinuity of an unknown function, for a feasibility check on the design space, thereby representing a classification tool to discern between a feasible and infeasible region. Bayesian optimization is a low-cost black-box global optimization tool in the Sequential Design Methods where one learns and updates knowledge from prior evaluated designs, and proceeds to the selection of new designs for future evaluation. However, BO is best suited to problems with the assumption of a continuous objective function and does not guarantee true convergence when having a discontinuous design space. This is because of the insufficient knowledge of the BO about the nature of the discontinuity of the unknown true function. In this paper, we have proposed to predict the location of the discontinuity using a BO algorithm on an artificially projected continuous design space from the original discontinuous design space. The proposed approach has been implemented in a thin tube design with the risk of creep-fatigue failure under constant loading of temperature and pressure. The stated risk depends on the location of the designs in terms of safe and unsafe regions, where the discontinuities lie at the transition between those regions; therefore, the discontinuity has also been treated as an unknown creep-fatigue failure constraint. The proposed BO algorithm has been trained to maximize sampling toward the unknown transition region, to act as a high accuracy classifier between safe and unsafe designs with minimal training cost. The converged solution has been validated for different design parameters with classification error rate and function evaluations at an average of <1% and ~150, respectively. Finally, the performance of our proposed approach in terms of training cost and classification accuracy of thin tube design is shown to be better than the existing machine learning (ML) algorithms such as Support Vector Machine (SVM), Random Forest (RF), and Boosting.
This presentation provides an overview of emulation, calibration and experiment design for computer experiments. Emulation refers to building a statistical surrogate from a carefully selected and limited set of model runs to predict unsampled outputs. The standard kriging approach to emulation of complex computer models is presented. Calibration refers to the process of probabilistically constraining uncertain physics/engineering model inputs to be consistent with observed experimental data. An initial probability distribution for these parameters is updated using the experimental information. Markov chain Monte Carlo (MCMC) algorithms are often used to sample the calibrated parameter distribution. Several MCMC algorithms commonly employed in practice are presented, along with a popular diagnostic for evaluating chain behavior. Space-filling approaches to experiment design for selecting model runs to build effective emulators are discussed, including Latin Hypercube Design and extensions based on orthogonal array skeleton designs and imposed symmetry requirements. Optimization criteria that further enforce space-filling, possibly in projections of the input space, are mentioned. Designs to screen for important input variations are summarized and used for variable selection in a nuclear fuels performance application. This is followed by illustration of sequential experiment design strategies for optimization, global prediction, and rare event inference.
A model of a linear aerospike rocket nozzle that consists of coupled aerodynamic and structural analyses has been developed. A nonlinear computational fluid dynamics code is used to calculate the aerodynamic thrust, and a three-dimensional fink-element model is used to determine the structural response and weight. The model will be used to demonstrate multidisciplinary design optimization (MDO) capabilities for relevant engine concepts, assess performance of various MDO approaches, and provide a guide for future application development. In this study, the MDO problem is formulated using the multidisciplinary feasible (MDF) strategy. The results for the MDF formulation are presented with comparisons against sequential aerodynamic and structural optimized designs. Significant improvements are demonstrated by using a multidisciplinary approach in comparison with the single- discipline design strategy.
Most grant goals were accomplished over the course of the grant, albeit only partial success was achieved in applying ASDL s existing advanced design methods to HAVOC. The limited success in applying ASDL's advanced design methods occurred due to: fundamental 20 of 197 limitations in the RSM metamodeling process used; the highly computationally intensive nature of HAVOC as compared to many other advanced system synthesis and sizing codes; and the highly curved nature of low I(sub sp) rocket system spaces. These challenges encountered during the grant led to ongoing efforts in more efficient and robust sequential metamodeling/surrogate optimization based advanced design methods which are expected to overcome all of the challenges encountered with RSM during the grant. Other achievements over the course of the grant included the documentation of the existing HAVOC cost module and formulation of new cost estimation relationships in support of APS analysis of 3rd Gen RLV. Further achievements beyond the scope of the grant as originally defined were also achieved such as HAVOC code and portability improvements, historical propulsion performance and reliability regression relationships, and new visualization routines to vacilitate visualization of HAVOC DOE case file results.
We present a control co-design software framework that can be used to optimize floating wind turbines and their controllers. Because this framework has many options for design variables, constraints, and merit figures, along with modeling fidelity levels, we seek to demonstrate best practices for using the tool while designing a floating platform for the new 22 MW offshore reference wind turbine developed within the International Energy Agency Wind Technology Commercialization Programme 55 on Reference Wind Turbines and Farms. During these studies, we evaluate the use of different simulation fidelity levels, the effect of using different load cases for controller tuning, and the difference between sequential and simultaneous control co-design solutions. Based on these efforts, we suggest using an algorithm that performs an initial search of the design space before optimization. We find that solving smaller optimization problems, in a sequential manner, leads to more reliable outcomes in fewer iterations than larger, simultaneous control co-design solutions. However a simultaneous CCD solution produces a platform with a 2% lower mass than the sequential CCD outcome.
This study proposes a novel approach to explore the solution space of a wind turbine rotor design process. The goal is to offer to blade designers the possibility to select an optimal rotor for given market conditions, assessing trade-offs and alternatives in a matter of minutes. The design process consists of sequential aerodynamic-structural optimizations, where the two loops are linked via a blade-loading parameter that can be varied by the user. A design study is performed starting from the IEA Wind Task 37 land-based reference wind turbine. The solution space is characterized for rotor diameters in the range of 130–160 m, rated generator power values in the range of 3.0–6.0 MW, and tip-speed ratios in the range of 7.5–12.5. Results are discussed highlighting optimal design decisions and trade-offs.
This paper considers the problem of optimal regulator design of linear multivariable systems with prescribed pole locations and/or poles corresponding to specified relative stability. A sequential method based on the frequency-domain optimality condition is proposed for achieving the desired pole assignment and determination of the corresponding quadratic performance index. This design method enables the retention of some stable open-loop poles and the associated eigenvectors in the closed-loop system. An illustrative example is provided to demonstrate the effectiveness of the proposed method.
This paper presents a control co-design (CCD) framework tailored to the hydraulic and mechanical subsystems of hydropower systems, aimed at enhancing their dynamic performance and responsiveness to grid demands. Conventional sequential design approaches often fall short in capturing the coupled interactions between plant dynamics and control objectives. The proposed CCD methodology enables simultaneous optimization of key hydro plant parameters - such as governor settings, penstock characteristics, and valve control - with the control system architecture. Results indicate that both CCD and sequential design approaches achieve comparable transient and steady-state performance, with minor discrepancies arising from differences in their optimized parameters. However, CCD demonstrates a distinct advantage by tripling the water time constant (Tw) while maintaining similar dynamic performance to the sequential approach. This increase in Tw enables optimization of penstock dimensions, as it is directly influenced by penstock design, thereby contributing to overall cost optimization.
A subcritical facility driven by an electron accelerator is planned at the Kharkov Institute of Physics and Technology (KIPT) in Ukraine for medical isotope production, materials research, training, and education. The conceptual design of the facility is being pursued through collaborations between ANL and KIPT. As part of the design effort, the high-fidelity analyses of various target options are performed with formulations to reflect the realistic configuration and the three dimensional geometry of each design. This report summarizes the results of target design optimization studies for electron beams with two different beam profiles. The target design optimization is performed via the sequential neutronic, thermal-hydraulic, and structural analyses for a comprehensive assessment of each configuration. First, a target CAD model is developed with proper emphasis on manufacturability to provide a basis for separate but consistent models for subsequent neutronic, thermal-hydraulic, and structural analyses. The optimizations are pursued for maximizing the neutron yield, streamlining the flow field to avoid hotspots, and minimizing the thermal stresses to increase the durability. In addition to general geometric modifications, the inlet/outlet channel configurations, target plate partitioning schemes, flow manipulations and rates, electron beam diameter/width options, and cladding material choices are included in the design optimizations. The electron beam interactions with the target assembly and the neutronic response of the subcritical facility are evaluated using the MCNPX code. the results for the electron beam energy deposition, neutron generation, and utilization in the subcritical pile are then used to characterize the axisymmetric heat generation profiles in the target assembly with explicit simulations of the beam tube, the coolant, the clad, and the target materials. Both tungsten and uranium are considered as target materials. Neutron spectra from tungsten and uranium are very similar allowing the use of either material in the subcritical assembly without changing its characteristics. However, the uranium target has a higher neutron yield, which increases the neutron flux of the subcritical assembly. Based on the considered dimensions and heat generation profiles, the commercial CFD software Star-CD is used for the thermal-hydraulic analysis of each target design to satisfy a set of thermal criteria, the most limiting of which being to maintain the water temperature 50 below the boiling point. It is found that the turbulence in the inlet channels dissipates quickly in narrow gaps between the target plates and, as a result, the heat transfer is limited by the laminar flow conditions. On average, 3-D CFD analyses of target assemblies agree well with 1-D calculations using RELAP (performed by KIPT). However, the recirculation and stagnation zones predicted with the CFD models prove the importance of a 3-D analysis to avoid the resulting hotspots. The calculated temperatures are subsequently used for the structural analysis of each target configuration to satisfy the other engineering design requirements. The thermo-structural calculations are performed mostly with NASTRAN and the results occasionally compared with the results from MARC. Both, NASTRAN and MARC are commercially available structural-mechanics analysis software. Although, a significant thermal gradient forms in target elements along the beam direction, the high thermal stresses are generally observed peripherally around the edge of thin target disks/plates. Due to its high thermal conductivity, temperatures and thermal stresses in tungsten target are estimated to be significantly lower than in uranium target. The deformations of the target disks/plates are found to be insignificant, which eliminate concerns for flow blockages in narrow coolant channels. Consistent with the specifications of the KIPT accelerator to be used in this facility, the electron beam power is 100-kW with electron energy in the range of 100 to 200 MeV. As expected, the 100 MeV electrons deposit their energy faster while the 200-MeV electrons spread their energy deposition further along the beam direction. However in that electron energy range, the energy deposition profiles near the beam window require very thin target plates/disks to limit the temperatures and thermal stresses.
Minimum-weight design of simply-supported, symmetrically laminated, thin, rectangular, specially orthotropic laminated plates for buckling and postbuckling strength is investigated. The postbuckling analysis is based on an Marguerre-type energy method. The failure load of laminates is calculated by the maximum strain failure criterion based on the in-plane strains. Design variables are individual layer thicknesses with specified fiber orientations. Optimization with discrete valued design is achieved by introducing additional penalty terms to the regular pseudoobjective function of sequential unconstrained minimization technique. The proposed optimization technique is applied to the design of rectangular laminates with various aspect ratios loaded by axial compressive loads.
The optical design of the Long Trace Profiler optical system is explored with a commercial raytrace program, Zemax OpticStudio™ (ZOS)1, with the intent of finding and correcting sources of systematic error. ZOS provides both geometric raytracing tools and physical optics Gaussian beam propagation and diffraction image calculation tools, and two design modes, sequential component (SC) and non-sequential component (NSC) that are optimized for different aspects of the design process. The original LTP-II system employs a singlet lens with a 1250 mm focal length. It is optimized to provide minimum distortion over a surface slope angle range of ±5 mrad. Using the ZOS tools, we are able to simulate ghost rays that produce distortion in the beam spot image and can minimize the distortion by deliberate misalignment of the beamsplitter (BS) components. Unfortunately, the reference beam is compromised because of the component tilts. The most recent LTP500 system design simplifies the optical system and makes the reference beam usable again, even with misalignment of the polarizing beamsplitter (PBS). Two lenses are designed for the LTP500 - a cemented doublet that has been fabricated, and a singlet with one aspheric surface. Both have focal distances of 500 mm with an expanded angular measurement range of ±10 mrad. The aspheric singlet provides superior performance. ZOS allows the import of wavefront measurement data produced by commercial interferometer software. We apply the wavefront error measurement from the cemented doublet to the model to show that the 19nm RMS wavefront error needs to be improved by at least a factor of ten in order to reduce the systematic error to a level that will allow the LTP to approach its design limit of a few tens of nanoradians.
The design and analysis methods for offshore wind turbines must consider the aerodynamic and hydrodynamic loads and response of the entire system (turbine, tower, substructure, and foundation) coupled to the turbine control system dynamics. Whereas a fully coupled (turbine and support structure) modeling approach is more rigorous, intellectual property concerns can preclude this approach. In fact, turbine control system algorithms and turbine properties are strictly guarded and often not shared. In many cases, a partially coupled analysis using separate tools and an exchange of reduced sets of data via sequential coupling may be necessary. In the sequentially coupled approach, the turbine and substructure designers will independently determine and exchange an abridged model of their respective subsystems to be used in their partners' dynamic simulations. Although the ability to achieve design optimization is sacrificed to some degree with a sequentially coupled analysis method, the central question here is whether this approach can deliver the required safety and how the differences in the results from the fully coupled method could affect the design. This work summarizes the scope and preliminary results of a study conducted for the Bureau of Safety and Environmental Enforcement aimed at quantifying differences between these approaches through aero-hydro-servo-elastic simulations of two offshore wind turbines on a monopile and jacket substructure.
Power converters are pervasive in modern electronic component design. They can be found in all electronic devices from household appliances and cellphone chargers to vehicles. Currently, designing new circuit topologies is hard because it requires human expertise based on experience and is difficult to automate. However, artificial-intelligence-assisted design can significantly facilitate the development of new power converters and/or improve the final result. Intelligently designed highly efficient power converters can have a significant effect on many important attributes, such as power efficiency, layout size, cost, heat dissemination, energy requirements, etc. We propose Autonomous Topology Generator (AutoTG), a reinforcement-learning-based framework that generates power converter topology candidates based on user specifications, optimized for user preferences. By modeling power converter design as a symbolic optimization problem, we sequentially sample components in an autoregressive manner until new topologies are formed, providing both the topology specification and the sizing (magnitude of each component parameter) of the proposed power converter. Here, we provide an empirical evaluation and show that AutoTG is able to generate varied high-efficiency topologies within component restrictions based on user input and show that previously unknown topologies can be found for further evaluation.
Explosives and explosive devices such as detonators are typically tested by applying a range of stimuli such as voltage or mechanical shock, and recording binary “detonated/did not detonate” responses. These are analyzed using maximum likelihood or generalized linear models to provide estimates of quantities such as the all-fire and no-fire points. Given that the true threshold for detonation is unknown a priori , sequential design methods are typically used to optimize the set of test points. One popular method, implemented in commercial software, is Neyer’s algorithm. To support simulation and experimental design, we have developed code in the R programming language to duplicate the functions of the Neyer software. We provide code for the simulator along with a description and examples of usage.
The currently common sequential design process for engineering systems is likely to lead to suboptimal designs. Recently developed decomposition methods offer an alternative for coming closer to optimum by breaking the large task of system optimization into smaller, concurrently executed and, yet, coupled tasks, identified with engineering disciplines or subsystems. The hierarchic and non-hierarchic decompositions are discussed and illustrated by examples. An organization of a design process centered on the non-hierarchic decomposition is proposed.
The currently common sequential design process for engineering systems is likely to lead to suboptimal designs. Recently developed decomposition methods offer an alternative for coming closer to optimum by breaking the large task of system optimization into smaller, concurrently executed and, yet, coupled tasks, identified with engineering disciplines or subsystems. The hierarchic and non-hierarchic decompositions are discussed and illustrated by examples. An organization of a design process centered on the non-hierarchic decomposition is proposed.