Karin Yu

Last Name

Yu

First Name

Karin

ORCID

0009-0000-0209-9567

Organisational unit

02219 - ETH AI Center / ETH AI Center

Show all metadata

Search Results

Publications 1 - 8 of 8

Experimental and analytical investigation of the crack behaviour of dapped-end beams
Mata Falcón, Jaime; Yu, Karin; Pallarés Rubio, Luis; et al. (2025)
Engineering Structures
Reinforced concrete dapped ends are a widely used connection for precast elements in infrastructure and building structures. However, they are prone to major durability problems due to the formation of an inclined crack at the re-entrant corner of the joint. The crack typically forms at low loads and reaches a significant opening under service load conditions, causing corrosion and stress concentration in the reinforcement, which impairs the safety of the connection. The behaviour of these cracks in a highly disturbed structural region is not well understood. This paper presents the experimental results of 28 dapped-end tests in which the crack behaviour in the re-entrant corner was continuously measured using digital image correlation. The main test variables are the amount of dapped-end reinforcement, the presence of diagonal reinforcement and the ratio of horizontal to vertical dapped-end reinforcement area. The bond behaviour of the dapped-end reinforcement was estimated from the measured crack widths and reinforcement strains, assuming a pull-out mechanism with constant bond shear stresses on both crack sides. The paper proposes a model for evaluating the width of corner cracks at service loads, in which the reinforcement strains are derived from strut-and-tie models. The modelling results agreed satisfactorily well with the experimental observations of this study and 28 additional tests from the literature for dapped-ends with reinforcement diameter below 20 mm.
Distributed Virtual Sensing via Bayesian Filtering for Wind Energy Structures
Yu, Karin (2022)
Nowadays, monitoring systems are used in wind turbine structures as an early warning system to detect structural defects and track the accumulated fatigue damage. This already available data could be further used to sense the climatic conditions such as wave loads or wind pressure in otherwise inaccessible locations like the ocean. The goal of this thesis is to simultaneously estimate spatially distributed loads and full-field vibration responses by applying Bayesian filtering with physics-based models, defining the structural dynamics, and data-driven Gaussian processes. The input varying in time and space is described by a Gaussian process regression model also known as kriging, where the distributed input signals are wave loads. Wind force at the top of the wind turbine tower is added to consider an operational wind turbine and two different types of loading. This spatio-temporal model is then combined with the Kalman filter to deal with the input-state estimation problem. Two approaches are examined: the dual Kalman filter and generalized least squares estimator. Before training the Gaussian process regression model, a suitable kernel and basis functions are selected. The algorithm is first applied to the case of available input measurements and afterwards also to unknown input measurements, which are derived from acceleration measurements. If input measurements are available, the generalized least squares estimator is the optimal predictor for both input and state estimation. Furthermore, fewer covariance matrices need to be tuned compared with the dual Kalman filter. On the contrary, if there are no input measurements, the dual Kalman filter performs better in the identification of the system’s inputs for the full-order model and even if the wind force is considered. However, in the latter case, the large noise from the acceleration measurement for the wind force needs to be smoothed out by an additional filter, the Savitzky-Golay filter. For the modally reduced-order model, the input-state estimation is very poor. To validate this approach, the algorithm should be applied to real-world or experimental data.
A Spatio-Temporal Model for Response and Distributed Wave Load Estimation on Offshore Wind Turbines
Yu, Karin; Tatsis, Konstantinos E.; Dertimanis, Vasilis K.; et al. (2023)
Conference Proceedings of the Society for Experimental Mechanics Series ~ Model Validation and Uncertainty Quantification, Volume 3
Sequential Bayesian inference schemes show tremendous potential for online information extraction from sparsely instrumented, uncertain dynamical systems. Within this context, notable paradigms are the tasks of state, input-state, and joint input-state-parameter estimation. A problem that has been scarcely studied in this context is the concurrent estimation of dynamic states and distributed loads on the basis of output-only (response) measurements. Examples of particular practical interest include the estimation of wind pressure on wind turbine blades, high-rise buildings and bridges, as well as wave loading in offshore structures. In such cases, the sensing of distributed inputs is heavily constrained by the instrumentation cost and the oftentimes limited access for sensor deployment. To tackle this challenge, this contribution investigates the fusion of Gaussian process regression (GPR) models with physics-based system representations for the recursive state and distributed wave load estimation on monopile offshore wind turbines. In particular, the distributed excitation is modeled with a GPR, which enables the implementation of a spatio-temporal filtering for the input process, while the system dynamics are represented by a physics-based model, which is in turn tailored to a recursive Bayesian scheme for the solution of the state estimation problem. The proposed approach is assessed in terms of a simulated case study on the finite element model of an offshore wind turbine.
Grammar-based generation of strut-and-tie models for designing reinforced concrete structures
Yu, Karin; Kraus, Michael Anton; Chatzi, Eleni; et al. (2024)
Computers & Structures
Reinforced concrete structures featuring discontinuity regions are complex to design and often susceptible to errors linked to numerical analysis methods. For such structural design problems, strut-and-tie models offer a simple, intuitive and safe design method based on the lower bound theorem of plasticity. Although intuitive, the derivation of strut-and-tie models requires nonnegligible effort and a certain degree of expertise to navigate the highdimensional design space. The automated generation of strut-and-tie models is nontrivial with existing optimisation-based methods, which struggle with accounting for fabrication aspects or incorporating user adaptations. This paper presents a novel grammar-based approach for generating practical strut-and-tie models by representing them as graphs and constructing a graph grammar. It consists of rules customised to consider engineering judgement, significantly reducing the dimensionality of the design space. The sequential application of such rules allows for human-computer interaction and aids engineers in decision-making, while being kept in the loop. Parsing four common design examples from the literature demonstrates the efficacy of this approach. The developed designs are more practical compared with existing optimisation-based suggestions. This interpretable grammar-based approach closely follows the intuitive decision-making process of practising structural engineers, which could be adapted to support further structural engineering design tasks.
Strut-and-tie models and stress fields: past, present and future
Kaufmann, Walter; Yu, Karin (2025)
fib Symposium Proceedings ~ Proceedings of the fib Symposium 2025
Strut-and-tie models and stress fields are powerful tools for the design of reinforced concrete, providing an intuitive understanding of the force flow and enabling consistent dimensioning including reinforcement detailing. Dating back to the early days of reinforced concrete, their solid theoretical basis was established in the second half of the 20th century by the lower-bound theorem of the theory of plasticity, facilitating their inclusion in modern design codes. Yet, strut-and-tie models are still mainly used by hand, which requires engineering judgement and impairs productivity, particularly if serviceability criteria must be verified and in the assessment of existing structures. Solving these issues is key to their use in industry.
Generative design of reinforced concrete structures incorporating constructability aspects
Yu, Karin; Chatzi, Eleni; Kaufmann, Walter (2025)
fib Symposium Proceedings ~ Proceedings of the fib Symposium 2025
Parametric modelling and generative machine learning (ML) enable examining various design alternatives by systematically exploring the design space. While engineers must take manifold design decisions in a parametric model, including constructability and non-structural aspects, generative ML can produce designs with little intervention once trained on extensive data. Engineering judgement is presumed to be embedded in the data, indicating design preferences. This paper provides an overview of generative design methods, ranging from parametric modelling to generative ML, focusing on integrating expertise such as constructability into reinforced concrete structure design. Based on two case studies, it highlights opportunities and challenges.
A grammar-based framework for strut-and-tie modelling of reinforced concrete structures
Yu, Karin; Kraus, Michael Anton; Chatzi, Eleni; et al. (2024)
Proceedings of IASS Annual Symposia ~ Proceedings of the IASS 2024 Symposium: Redefining the Art of Structural Design
Strut-and-tie models offer a simplified design approach for reinforced concrete structures such as walls or beams and are particularly suitable for static or geometrical discontinuities. They guarantee designs that are safe based on the lower bound theorem of the theory of plasticity. Currently, their manual generation demands significant time and expertise to navigate the solution space for various configu rations with different objectives in mind. Automating the generation of strut-and-tie models has faced several challenges, with previous methods such as discrete layout optimisation or topology optimisation struggling to consider either (i) user adjustments, such as changes of nodal coordinates, or (ii) practical aspects of fabrication and constructability. In response, this work presents a novel grammar-based generative framework that imposes strict and constraining rules, tailored to strut-and-tie models. Unlike previous work, our framework incorporates engineering judgement directly into its rule set, thereby significantly reducing the design space. Fur thermore, the sequential application of rules allows for user intervention and thus human-computer in teraction in the sense of a design co-pilot. We demonstrate the effectiveness of this framework through its application to two use cases: a cantilevered beam with a point load at its end and a dapped-end beam with an opening and two acting loads. The truss structure is represented as a graph and the rules are applied akin to graph grammar. Compared to optimisation-based methods, the developed models are practical, consider the preference towards orthogonal and distributed reinforcement and are typically preferred by professional structural engineers. This marks a first step towards an AI-assisted, grammar based generative design approach for strut-and-tie models. The framework offers interpretability that closely mirrors the intuitive decision-making process employed by human engineers in the selection of suitable strut-and-tie models.
Grammar-based ordinary differential equation discovery
Yu, Karin; Chatzi, Eleni; Kissas, Georgios (2025)
Mechanical Systems and Signal Processing
The understanding and modeling of complex physical phenomena through dynamical systems has historically driven scientific progress, providing essential tools for predicting system behavior under diverse conditions over time. In engineering, the discovery of dynamical systems is indispensable for computational modeling, diagnostics, prognostics, and control of engineered systems. Joining recent efforts that harness the power of symbolic regression in this domain, we propose a novel framework for the end-to-end discovery of ordinary differential equations (ODEs), termed Grammar-based ODE Discovery Engine (GODE). The proposed methodology combines formal grammars with dimensionality reduction and stochastic search for efficiently navigating high-dimensional combinatorial spaces. Grammars serve to inject domain knowledge and provide structure, both constraining and guiding the search for candidate expressions. GODE proves to be more sample- and parameter-efficient than state-of-the-art transformer-based models and to discover more accurate and parsimonious ODE expressions than both genetic programming- and other grammar-based methods, particularly for complex inference tasks, such as the discovery of structural dynamics. Thus, we introduce a tool that could play a catalytic role in dynamics discovery tasks, including modeling, system identification, and monitoring applications.

Publications 1 - 8 of 8

Karin Yu

Last Name

First Name

ORCID

Organisational unit

Filters

Search Results