Master Projects

Self-Supervised Learning for Population-Based Condition Monitoring of Industrial Assetsy

Supervisors:  Xudong Jian (), Chara Stoura, Prof. Dr. Eleni Chatzi 

co-Supervisors: Vasileios Vlachou: National Technical University of Athens 2) Prof. Theoklitos Karakatsanis: Democritus University of Thrace

Content
Reliable condition monitoring requires distinguishing fault-related changes from variations caused by normal operating conditions, such as changes in speed and load. This becomes particularly challenging when monitoring a population of machines, where differences between individual assets can further obscure fault signatures and limit the transferability of trained models. This Master’s thesis will investigate self-supervised learning methods for extracting informative representations from measured signals, with the aim of reducing sensitivity to environmental and operational variability while retaining information relevant to fault detection. Building on recent work at the Chair of Structural Mechanics and Monitoring, the project will explore how these methods can support population-based condition monitoring and transfer knowledge between related industrial assets. The study will use selected experimental datasets provided through collaboration with NTUA. The final scope will be defined according to the available measurements and metadata, initially focusing on a related group of machines or components. Public bearing datasets may provide additional validation cases.

Tasks

• Review self-supervised learning and population-based approaches to condition monitoring. 
• Prepare and characterize selected experimental datasets, including available operating-condition and fault labels. 
• Implement and adapt a self-supervised learning approach for fault-sensitive representation learning. 
• Evaluate robustness to changing operating conditions and transferability across related assets, using operating conditions or machines withheld from training. 
• Compare performance with conventional signal features and selected machine-learning baselines.

If time permits: Investigate adaptation to a new asset using limited labeled data and validate the approach on an additional public dataset.

Suggested Courses: Structural Identification and Health Monitoring, Structural Dynamics & Vibrations

Suggested Competences: Python programming, basic knowledge of machine learning and time-series analysis; familiarity with PyTorch would be beneficial.

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

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

Numerical and experimental dynamic characterization of curved elastic beams

Supervisors: Dr. Farhad Zeighami, Dr. Vasileios Ntertimanis, and Prof. Dr. Eleni Chatzi (Chair of Structural Mechanics and Monitoring, ETH Zürich)

Content

Controlling structural vibrations is an important challenge in civil engineering to ensure structural safety, serviceability, and comfort. Conventional vibration control approaches often rely on additional damping elements or resonating components. However, recent research has demonstrated that changes in structural geometry can significantly modify the dynamic response of structures. In specific, geometric nonlinearities, such as buckling and large deformations, can induce coupling between different vibration modes, redistribute resonance peaks, and alter the dynamic forces transmitted to supporting structures. Understanding and exploiting these mechanisms opens new opportunities for developing innovative vibration control systems. This Master's thesis aims to investigate how geometric modifications can be used to tailor the dynamic behavior of slender elastic structures. The student will design and investigate beam-like structures with different geometrical configurations to understand their influence on resonance characteristics, vibration modes, and dynamic force transmission. The research will combine finite element (FE) simulations and experimental investigations, using an electrodynamic shaker and a laser Doppler vibrometer (LDV), to characterize the structural response in both the frequency and time domains. Particular attention will be given to understanding the physical mechanisms governing the observed dynamic behavior and identifying potential applications in civil engineering.

Expected outcomes of project:

1. Review the literature on structural vibrations, geometric effects, buckling analysis, modal coupling, and vibration control. 
2. Design and develop FE models of slender curved beams with different geometrical configurations. 
3. Conduct experimental vibration tests to characterize resonance frequencies, mode shapes, and frequency response functions (FRFs). 
4. Investigate the influence of structural geometry on vibration characteristics and dynamic force transmission through numerical and experimental comparisons. 
5. Explore simplified analytical models and identify potential design strategies for vibration control applications.

Suggested Courses:  Structural Mechanics, Structural Dynamics, Finite Element Methods, Nonlinear Dynamics.

Suggested Competencies: MATLAB or Python programming, familiarity with FE modelling software, and an interest in experimental investigations

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

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

Finite Element Tool for Railway Track-Bridge Interaction

Supervisors: Manuel Zimmermann (1435 Ingenieure AG), Dr. Patrick Braess (ETH Zürich), Prof. Dr. Eleni Chatzi (ETH Zürich, )

Motivation

Where continuous welded rail (CWR) crosses a bridge, track and structure do not act independently. Temperature variations, braking and traction forces and vertical traffic loads cause longitudinal movements of the deck that differ from those of the rail. Because of the interaction, these movements and the thermal expansion of the deck are partly restrained. This restraint leads to additional longitudinal forces in the rail, relative displacements between track and deck and horizontal forces in bearings and piers and can ultimately lead to rail fracture or track buckling. These effects have to be verified according to UIC 774-3, EN 1991-2 (SIA 261) and the Swiss AB-EBV, in particular for long bridges.

With climate change, hot summers are becoming more frequent and more intense. Higher and longer-lasting rail and deck temperatures increase the risk of track buckling and make reliable interaction analyses increasingly important. In Swiss practice, track-bridge interaction analyses have mostly been carried out with a commercial software package that is no longer supported. A simple, transparent and validated tool that can be applied to both standard gauge and metre gauge networks is therefore currently missing.

This thesis aims to develop and validate a finite element tool for longitudinal track-bridge interaction under thermal and traffic loading. To this end, rail and deck can be modelled, for example, as beam elements coupled by nonlinear (bilinear) longitudinal track resistance, with bearings and piers represented as springs (Figure). The simplified linear superposition of load cases should then be compared with a nonlinear analysis and with existing calculations.

Your Task

The objective of this project is to experimentally evaluate the accuracy of tilt estimation, derived from acceleration measurements, by comparing it to direct inclinometer readings. The study involves data collection from both accelerometers and inclinometers positioned in close proximity, followed by tilt computation using a provided estimation algorithm. If necessary, improvements to the algorithm will need to be explored to enhance accuracy and reliability.

Detailed task:

• Review the code requirements (UIC 774-3, EN 1991-2, CEN/TR 17231, AB-EBV) and existing modelling approaches. 
• Implement a 1D (optionally 2D) FE model with nonlinear track resistance for ballasted track and direct rail fastening. 
• Implement the relevant load cases: temperature of deck and rail, braking and traction, vertical traffic loads including moving load positions. 
• Validate the tool against the UIC 774-3 / CEN/TR 17231 benchmark cases and an existing calculation of a Swiss railway bridge. 
• Carry out a parameter study for typical standard and metre gauge bridges (span length, bearing layout, pier stiffness, track type).

Expected project outcomes

1. A documented and validated FEM-tool
2. A validation report comparing the tool with benchmark cases and existing calculations.
3. An assessment of where simplified code checks are sufficient and where a nonlinear analysis is required.
4. Recommendations for the application of the tool in Swiss railway practice.

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

Suggested Background: Structural analysis, finite elementmethod. Interest in railway engineering is an advantage.

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

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

Non-Destructive Testing and AI-Assisted Inspection and Quality Control of Structures

Supervisors: Dr. Yunus Emre Harmanci (, Empa), Prof. Dr. Eleni Chatzi

Content

The safe and economical operation of structures depends on reliable information about their condition, both for existing infrastructure in service and for structural members produced with novel construction and manufacturing methods. Many relevant defects are, however, difficult to detect: some develop gradually and are spread over large surfaces, while others are hidden inside the material or at interfaces and are not visible to the naked eye. Conventional inspection relies largely on visual observation and the accumulated expertise of specialized personnel. This approach is effective but time-consuming and costly, involves many repetitive tasks, and can miss defects that do not appear on the surface. Non-destructive testing (NDT) offers a way to assess structures without impairing their function. Combined with automated data processing and artificial intelligence, it can efficiently identify and locate problematic areas. This allows engineers to prioritize their efforts and make more informed decisions about when detailed investigations or interventions are warranted.

Objectives and Methodology

In this master’s project, NDT and data-driven methods will be applied for the inspection and quality control of structures. Depending on the student’s interests, the work can focus on one of the following directions:

• Computer-vision-based inspection of bridges: Reviewing existing datasets of bridge components with various defect types (e.g., spalling, corrosion/rust, leakage, cracks), implementing deep learning models for automated defect detection, classification, and localisation, and integrating the model into a user-friendly inspection aiding tool.
• Thermography of externally bonded reinforcements: Developing a data processing module for automated analysis of lock-in thermography measurements and an AI-based anomaly detection and segmentation module, validating them using data from previously collected pilot tests, and, if time permits, conducting a field experiment with an operational UAV system.
• Crack detection in wire-arc additively manufactured (WAAM) steel plates: Conducting an experimental campaign on 3D-printed specimens under quasi-static and dynamic loading, recorded with accelerometers and a (high-speed) digital image correlation system. The measurements are compared with an existing analytical solution (a documented Python implementation is provided), and crack length and position are estimated from the measured response.

The student will have the opportunity to study the latest research in the field of non-destructive testing, work with computer vision, UAV technology, and laboratory testing, and collaborate with industry and project partners. This work bridges structural engineering, artificial intelligence, and practical field applications. There will be one interim presentation and a final presentation.

Suggested Courses: Structural Identification and Health Monitoring, Structural Dynamics & VibrationProblems, Non Destructive Evaluation

Suggested Competencies: Python/MATLAB, and interest for computational methods and/or motivation forlaboratory work.

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

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