Searcharxiv⌕ Search

arXiv subjects

Dirk Helm

Publications and source records attributed to Dirk Helm.

10 recordsLinked to original sources
arXiv2026

Diversity-Aware Batch-Mode Active Learning for Efficient Sampling in Data-Driven Constitutive Modeling

The constitutive behavior of materials is modeled through relationships between stress, strain, and possibly additional internal variables. This results in relatively high-dimensional feature spaces for machine learning models rendering the efficient generation of informative datasets essential as brute force methods suffer from the curse of dimensionality. This work introduces a diversity-aware batch-mode query-by-committee active-learning strategy to generate datasets of maximum information content at minimum cost. In contrast to existing methods, this novel method selects multiple informative, non-redundant queries per iteration, enabling concurrent generation of informative datasets and reducing the number of machine-learning retraining cycles. A central component of this method is a cosine-similarity-based metric that complements the uncertainty criterion based on committee variance by promoting within-batch diversity. The query selection is guided by committee variance and a diversity-promoting criterion. The approach is benchmarked for efficient stress-space sampling in data-driven constitutive modeling. In this setting, a committee of support vector classifiers approximates the so-called yield surface, which is a manifold dividing the six-dimensional stress space into an elastic and plastic domain. We demonstrate that the method handles different batch sizes robustly, maintains high within batch diversity, and rapidly reduces committee uncertainty. The resulting machine learning yield surfaces achieve predictive accuracy comparable to sequential active learning, while requiring substantially fewer retraining cycles. This makes the proposed approach an efficient strategy for stress space sampling in data driven constitutive modeling and for reducing time to solution via concurrent data collection in each iteration.

physics.comp-ph↗
arXiv2024

Semantic orchestration and exploitation of material data: A dataspace solution demonstrated on steel and copper applications

In materials science and manufacturing, vast amounts of heterogeneous data (e.g., measurement and simulation logs, process data, publications) serve as the bedrock of valuable knowledge for various engineering applications. However, efficiently storing and managing this diverse data poses challenges due to limited standardization and integration across different organizational units. Addressing these challenges is essential to fully unlock the potential of data-driven approaches. This paper introduces novel, comprehensive semantic methodology tailored to materials engineering and realized as a technology stack named Dataspace Management System (DSMS), which powers dataspace solutions that leverage the knowledge encoded in heterogeneous data sources to support data-driven insights and to derive new knowledge. At its core, DSMS offers a distinctive knowledge management approach tuned to meet the specific requirements of the materials science and manufacturing domain, all while adhering to the FAIR principles. DSMS provides functionalities for data integration, linkage, exploration, visualization, processing, data sharing, and services (e.g., consulting) to support engineers in decision-making, design and optimization. We present an architectural overview of DSMS, outlining its core concepts and their technological implementation, as well as demonstrate its applicability to common data-processing tasks through use cases from the StahlDigital and KupferDigital research projects within Germany's MaterialDigital initiative.

cs.DB↗
arXiv2023

Machine learning for structure-guided materials and process design

In recent years, there has been a growing interest in accelerated materials innovation in the context of the process-structure-property chain. In this regard, it is essential to take into account manufacturing processes and tailor materials design approaches to support downstream process design approaches. As a major step into this direction, we present a holistic and generic optimization approach that covers the entire process-structure-property chain in materials engineering. Our approach specifically employs machine learning to address two critical identification problems: a materials design problem, which involves identifying near-optimal material microstructures that exhibit desired properties, and a process design problem that is to find an optimal processing path to manufacture these microstructures. Both identification problems are typically ill-posed, which presents a significant challenge for solution approaches. However, the non-unique nature of these problems offers an important advantage for processing: By having several target microstructures that perform similarly well, processes can be efficiently guided towards manufacturing the best reachable microstructure. The functionality of the approach is demonstrated at manufacturing crystallographic textures with desired properties in a simulated metal forming process.

cond-mat.mtrl-sci↗
arXiv2023

Accurate Distances Measures and Machine Learning of the Texture-Property Relation for Crystallographic Textures Represented by One-Point Statistics

The crystallographic texture of metallic materials is a key microstructural feature that is responsible for the anisotropic behavior, e.g., important in forming operations. In materials science, crystallographic texture is commonly described by the orientation distribution function, which is defined as the probability density function of the orientations of the monocrystal grains conforming a polycrystalline material. For representing the orientation distribution function, there are several approaches such as using generalized spherical harmonics, orientation histograms, and pole figure images . Measuring distances between crystallographic textures is essential for any task that requires assessing texture similarities, e.g. to guide forming processes. Therefore, we introduce novel distance measures based on (i) the Earth Movers Distance that takes into account local distance information encoded in histogram-based texture representations and (ii) a distance measure based on pole figure images. For this purpose, we evaluate and compare existing distance measures for selected use-cases. The present study gives insights into advantages and drawbacks of using certain texture representations and distance measures with emphasis on applications in materials design and optimal process control.

cond-mat.mtrl-sci↗
arXiv2023

Neural Networks for Constitutive Modeling -- From Universal Function Approximators to Advanced Models and the Integration of Physics

Analyzing and modeling the constitutive behavior of materials is a core area in materials sciences and a prerequisite for conducting numerical simulations in which the material behavior plays a central role. Constitutive models have been developed since the beginning of the 19th century and are still under constant development. Besides physics-motivated and phenomenological models, during the last decades, the field of constitutive modeling was enriched by the development of machine learning-based constitutive models, especially by using neural networks. The latter is the focus of the present review, which aims to give an overview of neural networks-based constitutive models from a methodical perspective. The review summarizes and compares numerous conceptually different neural networks-based approaches for constitutive modeling including neural networks used as universal function approximators, advanced neural network models and neural network approaches with integrated physical knowledge. The upcoming of these methods is in-turn closely related to advances in the area of computer sciences, what further adds a chronological aspect to this review. We conclude this review paper with important challenges in the field of learning constitutive relations that need to be tackled in the near future.

cond-mat.mtrl-sci↗
arXiv2022

Machine learning-based sampling of virtual experiments within the full stress state

This paper presents a new machine learning-based approach to investigate anisotropic yield surfaces of sheet metals by means of virtual experiments. The new sampling approach is based on the machine learning technique known as active learning, which has been adapted to efficiently sample virtual experiments with respect to the full stress state in order to identify parameters of anisotropic yield functions. The approach was employed to sample virtual experiments based on the crystal plasticity finite element method (CPFEM) for a DX56D deep drawing steel and compared with two state-of-the-art sampling methods taken from the literature. The resulting points on the initial yield surface for all three sampling methods were used to identify parameters of the anisotropic yield functions Hill48, Yld91, Yld2004-18p and Yld2004-27p. The results show that the new machine learning-based sampling approach has a higher sampling efficiency than the two state-of-the-art sampling methods. Consequently, fewer computationally expensive crystal plasticity simulations are required. By comparing different variants of the Hill48, Yld91, Yld2004-18p and Yld2004-27p yield surfaces, it was also found that identifying parameters of anisotropic yield functions based on virtual experiments sampled within the full stress state can lead to a degraded representation of the in-plane anisotropy. With respect to DX56D deep drawing steel, this degradation was observed for the Yld2004-18p yield function. As a consequence, the representation of the in-plane anisotropy must be carefully reviewed when taking the full stress state into account. In this context, Yld2004-27p was identified as being sufficiently flexible to simultaneously represent the plastic anisotropy of DX56D with respect to the in-plane and out-of-plane behaviour with high accuracy.

cond-mat.mtrl-sci↗
arXiv2021

A multi-task learning-based optimization approach for finding diverse sets of material microstructures with desired properties and its application to texture optimization

The optimization along the chain processing-structure-properties-performance is one of the core objectives in data-driven materials science. In this sense, processes are supposed to manufacture workpieces with targeted material microstructures. These microstructures are defined by the material properties of interest and identifying them is a question of materials design. In the present paper, we addresse this issue and introduce a generic multi-task learning-based optimization approach. The approach enables the identification of sets of highly diverse microstructures for given desired properties and corresponding tolerances. Basically, the approach consists of an optimization algorithm that interacts with a machine learning model that combines multi-task learning with siamese neural networks. The resulting model (1) relates microstructures and properties, (2) estimates the likelihood of a microstructure of being producible, and (3) performs a distance preserving microstructure feature extraction in order to generate a lower dimensional latent feature space to enable efficient optimization. The proposed approach is applied on a crystallographic texture optimization problem for rolled steel sheets given desired properties.

cond-mat.mtrl-sci↗
arXiv2020

Deep Reinforcement Learning Methods for Structure-Guided Processing Path Optimization

A major goal of materials design is to find material structures with desired properties and in a second step to find a processing path to reach one of these structures. In this paper, we propose and investigate a deep reinforcement learning approach for the optimization of processing paths. The goal is to find optimal processing paths in the material structure space that lead to target-structures, which have been identified beforehand to result in desired material properties. There exists a target set containing one or multiple different structures. Our proposed methods can find an optimal path from a start structure to a single target structure, or optimize the processing paths to one of the equivalent target-structures in the set. In the latter case, the algorithm learns during processing to simultaneously identify the best reachable target structure and the optimal path to it. The proposed methods belong to the family of model-free deep reinforcement learning algorithms. They are guided by structure representations as features of the process state and by a reward signal, which is formulated based on a distance function in the structure space. Model-free reinforcement learning algorithms learn through trial and error while interacting with the process. Thereby, they are not restricted to information from a priori sampled processing data and are able to adapt to the specific process. The optimization itself is model-free and does not require any prior knowledge about the process itself. We instantiate and evaluate the proposed methods by optimizing paths of a generic metal forming process. We show the ability of both methods to find processing paths leading close to target structures and the ability of the extended method to identify target-structures that can be reached effectively and efficiently and to focus on these targets for sample efficient processing path optimization.

cs.LG↗
arXiv2020

CupNet -- Pruning a network for geometric data

Using data from a simulated cup drawing process, we demonstrate how the inherent geometrical structure of cup meshes can be used to effectively prune an artificial neural network in a straightforward way.

cs.LG↗