Learning the characteristics of engineering optimization problems with applications in automotive crash

Long, Fu Xing; Stein, Bas van; Frenzel, Moritz; Krause, Peter; Gitterle, Markus; Bäck, Thomas

doi:10.1145/3512290.3528712

Cited by 20 publications

(5 citation statements)

References 45 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This required developing appropriate optimizers to search in the latent space of the generator. Developing closed-loop optimizers for visual cortex neurons is costly, so we first simulated the process with a more tractable synthetic problem, a common strategy in the field of evolutionary computing [15, 9]. As image-computable models, convolutional neural networks (CNNs) serve as good approximations of the primate visual system[16].…”

Section: Resultsmentioning

confidence: 99%

Neural Dynamics of Object Manifold Alignment in the Ventral Stream

Wang,

Ponce

2024

Preprint

View full text Add to dashboard Cite

Visual neurons respond across a vast landscape of images, comprising objects, textures, and places. Natural images can be parameterized using deep generative networks, raising the question of whether latent factors learned by some networks control images in ways that better align with visual neurons. We studied neurons in areas V1, V4 and posterior IT, optimizing images using a closed-loop evolutionary algorithm. We used two generative image models: (1) DeePSim, which parameterizes local image patterns, and (2) BigGAN which parameterizes object identity and nuisance variables. We found that neurons could guide image optimization on both pattern- and object-based image manifolds across areas; V1 aligned best with the DeePSim image space, whereas PIT aligned well with both DeePSim and BigGAN spaces. While initially PIT neurons responded well to the textural manifold, their responses to objects also emerged over time, suggesting that object-like responses required further processing. We identified similar local features common to both textural and object images, but not optimal global configuration. We conclude that visual cortex neurons are aligned to a representational space not yet captured by current artificial model of the visual system.

show abstract

Section: Resultsmentioning

confidence: 99%

Neural Dynamics of Object Manifold Alignment in the Ventral Stream

Wang,

Ponce

2024

Preprint

View full text Add to dashboard Cite

show abstract

“…Regarding the optimization itself based on metamodels, the selection and hyperparameter tuning and optimization represent current research activity as well. Particularly in the domain of car crash, we refer to 66,67 for recent developments. Including further machine learning based prediction into a multifidelity type of optimization is a promising and active field of research.…”

Section: Discussionmentioning

confidence: 99%

Multidisciplinary optimization of automotive mega-castings merging classical structural optimization with response-surface-based optimization enhanced by machine learning

Triller,

Lopez,

Nossek

et al. 2023

Sci Rep

Self Cite

View full text Add to dashboard Cite

Large high pressure die castings (HPDC), recently referred to as mega-castings, can replace plenty of steel metal sheets usually employed for body-in-white (BIW) structures. They can save manufacturing expense and unleash additional lightweight potential thanks to additional design freedom and material properties. The BIW plays a major role in automotive design since it must fulfill numerous structural targets ranging from stiffness for vehicle dynamics, dynamic responses for NVH (noise, vibration, harshness), driving comfort standards and several passive safety requirements. The use of mega-casting structures leads to additional requirements with respect to castability and material quality. Achieving a lightweight design considering requirements related to crash or castability is a challenge on its own, due to the high computational cost of related simulation techniques. Considering multiple requirements simultaneously, therefore often leads to non-weight-optimal structures. To exploit the full lightweight potential, we present a generative multidisciplinary optimization pipeline for the structural design of automotive mega-casting parts in this paper. The approach combines established methods in automotive industry such as topology optimization and response-surface-based (RSM) optimization and enhances the latter by machine learning (ML) based clustering and classification. In a first step topology optimization is employed to derive optimal load-paths for multidisciplinary loading conditions. For this purpose, casting manufacturing constraints as well as more than hundred linearized loads are used to incorporate NVH and passive safety requirements. In a next step the optimal thickness distribution and rib orientation of the structure is achieved using RSM optimization algorithms for the computationally expensive nonlinear crash and casting simulations. Performance indicators are treated by unsupervised learning based on clustering. This enables classification constraints based on simulation field results from hundreds of samples to be included into RSM optimization. It resolves a typical risk of pure scalar, regression-type targets, where supposed optimal results fail when domain experts examine the full field result of the corresponding simulation. It is shown how this approach is superior in achieving a weight-optimal design and turnaround time compared to a design workflow classically used for BIW structures.

show abstract

“…Based on the landscape characteristics (ELA features) of the given problem instance (Step 2), which are then computed on the sample (X , F), the objective of the pipeline is to identify representative, fast-to-evaluate functions with similar characteristics (Step 3). At the moment, this is done by using a random function generator by which we generate thousands of random functions from which we select one that minimizes the Euclidean distance between its ELA feature vector and the target feature vector [12]. These fast-to-evaluate functions are used to select and configure the optimal optimization algorithm Solving optimization problem (using best algorithm and hyperparmeter)…”

Section: Proposed Pipelinementioning

confidence: 99%

“…The pipeline uses cheap to evaluate proxy functions for the algorithm selection and hyper-parameter tuning. These proxy functions are found using the methodology proposed by Long et al [12]. In this approach, Exploratory Landscape Analysis (ELA) [13] features are computed and utilized to select appropriate generated artificial functions such that these functions closely resemble the optimization landscape.…”

Section: Introductionmentioning

confidence: 99%

AI for Expensive Optimization Problems in Industry

Stein

Winter

Bäck

et al. 2023

2023 IEEE Conference on Artificial Intelligence (CAI)

Self Cite

View full text Add to dashboard Cite

The optimization of real-world engineering problems can be a challenging task, owing to the limited understanding of problem characteristics and the high cost of evaluating objectives and constraints in terms of computing time or licenses. This study proposes an AI-assisted optimization pipeline that addresses these challenges by using proxy functions in order to select and optimize the optimization algorithm and its hyperparameters, thereby significantly accelerating the optimization process on the real (expensive) problem. These proxy functions are inexpensive to evaluate and are selected to exhibit similar landscape characteristics as the original problem. To obtain such proxy functions, we adopt an approach, which involves computing Exploratory Landscape Analysis (ELA) features to characterize the problem's landscape. The ELA features are then used to identify an artificial function that replicates the original problem's properties, which can then be employed as a lowcost proxy function for the hyper-parameter optimization of our pipeline. Several real-world industrial applications are discussed as use-case of our proposed approach.

show abstract

Learning the characteristics of engineering optimization problems with applications in automotive crash

Cited by 20 publications

References 45 publications

Neural Dynamics of Object Manifold Alignment in the Ventral Stream

Neural Dynamics of Object Manifold Alignment in the Ventral Stream

Multidisciplinary optimization of automotive mega-castings merging classical structural optimization with response-surface-based optimization enhanced by machine learning

AI for Expensive Optimization Problems in Industry

Contact Info

Product

Resources

About