Master Projects

Additive manufacturing and parametric design of Iron-based shape memory couplers

Supervisors: Ali Jafarabadi (, Empa/ETHZ), Prof. Elyas Ghafoori, Dr. Christoph Czaderski, Prof. Dr. Eleni Chatzi
Industry Partner: HILTI

Content
Additive manufacturing technologies have brought engineers a new era by enabling them to design structures to reduce material consumption and, at the same time, realizing additional functionalities. Metal additive manufacturing facilitates the proof of such new design concepts due to fast and cost-competitive prototyping. Combining smart materials such as iron-based shape memory alloys (Fe-SMAs) in harmony with an intelligently architected geometry will result in extraordinary properties. This project aims to study lattice structures in order to develop lightweight mechanical couplers made of Fe-SMAs. The lattice structures will be generated in the Rhino Grasshopper and then after finite el-ement analysis (FEA) of the coupler, it will 3D printed at Empa for experimental evaluation. The stu-dent will be introduced to Rhino Grasshopper and FEA in ABAQUS environment.

Suggested Courses: Method of Finite Elements I

Possible to select as: obligatorishce Projektarbeit, praktische Projektarbeit, forschungsbezogene Projektarbeit

Limitation of offerings: 2 students (group work is possible)

Smart Mesh Refinement with Graph Neural Networks
Supervisors: Dr. Imad Abdallah (), Gregory Duthé (), Prof. Dr. Eleni Chatzi

Content

Most computational methods in engineering require meshes which have been designed by domain experts. Whether it be CFD or FEM, different study objectives will lead to different kinds of meshes, with mesh densities and refinement strategies which vary over the computational domain. To obtain accurate results without over-refinement, designing meshes can often be a tedious and time consuming process. With this project, we aim to investigate the following question: can Graph Neural Networks (GNNs) replace this tiresome but necessary step?

Previous work has started the investigation and built a framework to train a Graph Attention Network to add a small number of extra grid points in the right locations using Genetic Algorithms. The student(s) will build upon this work and will have the following objectives:

• Review the existing method and literature and become familiar with the existing code.
• Define a set of objectives for the optimization process (for CFD as previously started, FEM, etc.)
• Investigate improvements to the training procedure, with more modern Evolutionary Strategies or stochastic gradient descent schemes to speed up the optimization process.
• Validate the developed method with numerical case studies

Suggested Courses: Structural Identification and Health Monitoring

Possible to select as: obligatorishce Projektarbeit, praktische Projektarbeit, forschungsbezogene Projektarbeit

Limitation of offerings: 2 students (group work is possible) 
Language (incl. report, oral presentation and poster): English

Enhancing Dynamic Mobile Sensing Platforms through Shake-Table Testing for Urban Infrastructure Digital Twins

Supervisors: Kiran Bacsa (), Dr. Vasileios Ntertimanis, Pr. Eleni Chatzi

Content

The Singapore-ETH Centre is seeking to implement a “Dynamic Mobile Sensing Platform”, i.e., a new sensing paradigm that leverages the use of satellite data, roving sensors and urban networks to implement an infrastructure’s digital twin.

To put our approach into practice, we seek to build a toy example by mounting mobile sensors and a simple computing unit onto bicycles. Equipped with vibration sensors, our rudimentary roaming sensors measure the road roughness of various roads and relay the diagnosis back to a central server. The end use of the sensing platform will be the Park Connector Network in Singapore, an extensive bicycle lane network that spans the entire island.

The goal of this project is to enhance the capabilities of the proposed "Dynamic Mobile Sensing Platform" by incorporating a shake-table testing methodology. By utilizing a shake-table, the student will subject the instrumented bicycle to controlled and simulated vibrations that mimic real-world road conditions. This testing process will allow for the collection of benchmark data on the performance of the sensing platform under different vibration levels and frequencies, providing insights into its robustness and accuracy. By validating the platform's effectiveness through rigorous shake-table testing, we can ensure its reliability and suitability for real-world applications, ultimately advancing the development of digital twins for urban infrastructure.

Suggested Courses: Structural Identification and Health Monitoring

Suggested Competencies: Comfortable programming & willingness to advance your skillset

Possible to select as: obligatorishce Projektarbeit, praktische Projektarbeit, forschungsbezogene Projektarbeit

Limitation of offerings: 2 students (group work is possible)

Real-time control of MIMO force estimation for Model-Experiment Convergence and Visualization using Augmented Reality

Supervisors: Kostas Vlachas (), Prof. Dr. Fernando Moreu (University of New Mexico), Prof. Dr. Eleni Chatzi

Content

Replicating system response in the lab is done by characterizing the system dynamics with multiple inputs and multiple outputs (MIMO). System ID is obtained in terms of a full order FRF matrix with uncorrelated, random inputs at each input location. The FRF matrix is then inverted, using the Moore-Penrose pseudoinverse process with Tikhonov regularization. The desired response, in terms of a spectral density matrix (SDM), is then used in conjunction with the inverted FRF matrix to find an input. The input is applied experimentally, and the difference between the desired SDM and the experimental SDM is used to iteratively update the input. Filling in the cross terms in the response SDM, inverting the FRF matrix, and quantifying error are ongoing challenges in inverse MIMO.
Real-time control of inputs is seldom done in inverse MIMO, with iterations typically being done in separate experiments and without model updates. Visualization of error between the estimation and the actual response is in general the best assessment for a successful experimental framework after System-ID is achieved.
The experimental system currently available is a 3-story frame structure with up to 3 input locations and up to 6 output locations. Inputs are applied with uniaxial electrodynamic actuators, and outputs are collected with uniaxial accelerometers. Although many inputs and outputs are available, results will be validated on a simple case first, with 2 inputs and 2 outputs.
This project will investigate model update based on real-time experimental results and its projection using Augmented Reality for humanin-the-loop investigation. The updated model will be used to update inputs in real-time as well to better match the response SDM with human visualization of the model. This result will be used to achieve faster error convergence, reduce the number of tests needed, and minimize input forces.

Expected outcomes of project:

1. Theory development: dynamic simulations of MIMO
2. System ID based on multiple inputs and multiple outputs (MIMO)
3. Physics-based, reduced order model based on system ID
4. Real-time model update based on instantaneous difference between reference and real spectrum
5. Augmented Reality programming of dynamics and real-time simulation connection with AR and experiment
6. Error quantification (frequency domain, time domain, etc.) and iterative input update process

Suggested Courses: Structural Identification and Health Monitoring

Suggested Competencies: Python/MATLAB, Strong background on programming with an interest on human-in-the-loop and structural dynamics is recommended.

Possible to select as: obligatorishce Projektarbeit, praktische Projektarbeit, forschungsbezogene Projektarbeit

Limitation of offerings: 2 students (group work is possible)

Further open for:  master students with background of computer science, electrical engineering, mechanical engineering, civil engineering, or related. 

Exploring Mycelium's Electrical Properties

Supervisors: Sophia Ganseboom (), Prof. Dr. Eleni Chatzi

Content

Mycelium is the vegetative part of a fungus, consisting of a network of thread-like structures called hyphae, which grow underground or within a substrate. It plays a critical role in the ecosystem, facilitating nutrient cycling and breaking down organic matter. The research into its potential as a possible biomaterial has gained a lot of momentum recently, while some emerging and very recent efforts have extended the investigation to its electrical properties. The research so far shows that the mycelium is sensitive to external stimuli, which are reflected in changes of its electrical potential. We want to investigate this further and look into its potential as a biological sensor.

In this student project, focused on studying the electrical properties of mycelium, we aim to investigate the mycelium's extracellular electrical potential, which has shown spontaneous spiking activity even in absence of external stimuli. In this project, we will look into how this electrical activity changes when the mycelium is exposed to different external stimuli.
The objective of this project is to analyse these changes and find patterns to quantify them. For this, we will design a series of experiments that allow us to record these changes when the mycelium is under mechanical pressure as well as exposed to light. Additionally, we will explore signal analysis techniques to quantify the recorded responses accurately.
This project offers a unique opportunity to blend biology, engineering, and data analysis for students interested in working on an interdisciplinary theme.

Possible to select as: obligatorishce Projektarbeit, praktische Projektarbeit, forschungsbezogene Projektarbeit

Limitation of offerings: 2 students (group work is possible)

Further open for:  master students with background of material science, computer science, electrical engineering, mechanical engineering, civil engineering, or related. 

Physics-Informed Neural Networks for System Identification: Predicting Acceleration Time Series from Electrical Inputs in End-of-Production-Line Actuator Testing

Supervisors: Dr. Roman Klis (Johnson Electric), Prof. Dr. Eleni Chatzi ()

Content

This master thesis aims to employ physics-informed neural networks (PINNs) to enhance the predictive capabilities of end-of-production-line actuator testing. The testing station currently captures both electrical and vibration time series data during various tests stages. The focus of this research is on utilizing PINNs to predict acceleration time series responses using the available electrical input readings. The study incorporates system identification techniques to improve the accuracy and interpretability of the neural network predictions.

Objective:

The primary objective of this research is to investigate the application of physics-informed neural networks in the context of end-of-production-line actuator testing. Specifically, the study aims to:

• Explore the existing challenges in predicting acceleration responses in the actuator testing process using traditional methods. 
• Develop a physics-informed neural network model capable of predicting acceleration time series data from electrical input readings and observed outputs. 
• Incorporate system identification techniques to enhance the interpretability and understanding of the neural network model. 
• Evaluate the performance of the proposed model in terms of accuracy, efficiency, and robustness under various testing conditions. 
• Provide insights into the practical implementation of physics-informed neural networks for predictive modeling in industrial testing scenarios.

Methodology:

The research will involve a combination of literature review, data preprocessing, model development, and empirical analysis. The study will begin with a thorough review of existing literature on predictive modeling, physics-informed neural networks, and system identification. Subsequently, a physics-informed neural network model will be developed and trained using the available electrical and vibration time series data from the end-of-production-line actuator testing. System identification techniques will be integrated to improve the model's interpretability. The model's performance will be rigorously evaluated and compared against traditional methods.

Johnson Electric (JE) and ETH are working together to develop a proof-of-concept tool that would solve the described problem.

Possible to select as: obligatorishce Projektarbeit, praktische Projektarbeit, forschungsbezogene Projektarbeit

Limitation of offerings: 2 students (group work is possible)

Further open for:  master students with background of material science, computer science, electrical engineering, mechanical engineering, civil engineering, or related.