STI Industry Day 2023

Engineering Industry Day

Wednesday March 8, 2023 – SwissTech Convention Center

Come and meet CIS during the event

Innovation is one of the basic missions of EPFL and of primary importance to the School of Engineering (STI). After 3 successful editions and growing number of attendees (over 600 in 2019), EPFL is organising the 5th edition on Wednesday March 8, 2023. The event is a great opportunity for companies, research laboratories, start-ups and researchers to connect and establish new collaborations.

CIS will be attending the event and hosting a panel discussion on AI for engineering. We will also showcase the latest research in machine learning and robotics. Join us to experience demonstrations from the CIS integrative research pillars on Intelligent Assistive Robotics, AI4Medicine, EdgeAI, and DigitalTwin. Be sure to stop by our booth to meet us and learn about our activities.

View the full list of talks, roundtables and exhibitors on the official website of the event. 

Come and listen to CIS Academic Director Prof. Pierre Vandergheynst as he shares his view on how intelligent systems will shape our future. 

When: 15:50

The CIS will be hosting a discussion panel on the topic of AI for Engineering. Gathering experts on AI, Machine Learning and robotics, we will explore the impact that AI can have on engineering, whether to optimize processes, reduce costs, or increase efficiency. 

Moderator: Dr. Jan Kerschgens, CIS
Panel experts: Prof. Michele Cerriotti, Prof. Adrian Ionescu and Prof. Olga Fink.

Time: 16:50

Come by our booth during breaks to meet our team and learn about our activities.

Abstract: In collaboration with the Swiss Federal Institute of Technol- ogy Lausanne (EPFL) and the Engineering School of Geneva (HEPIA), Onescope has created the first 3-in-1 medical de- vice – Pneumoscope – combining a digital stethoscope, an oximeter and a thermometer. Lung sounds, oxygen blood saturation, heart rate, and temperature are simultaneously transmitted to an intuitive mobile app for real-time analy- sis and recognition of respiratory acoustic signatures by AI algorithms. The app not only informs the user on the likely diagnosis, but also gives the severity of the disease and sug- gests the best care pathway.

Presenter: Annie Hartley

Abstract: Instead of improving the mobility of human body via wearable device, rendering the indoor infrastructures mobile can also ease the life of patients with limited mobility. Such scenes can be in the elderly care facilities or hospitals, where mobile chairs automatically move around to provide help or avoid collision, and the operating table approaches a person in bed.

Presenter: Anastasia Bolotnikova 

Abstract: Emotional stress has been classified as a major contributor to several social problems related to crime, healthcare, and quality of life. While blood cortisol tests, electroencephalography (EEG) and physiological parameter measurements are gold standards for assessing stress, they are invasive or inconvenient and thus unsuitable for wearable real-time stress monitoring. Human sweat has started to be an attractive area of research for measuring cortisol for stress detection. The objectives of this research are a) to detect long-term stress by monitoring EEG signals and cortisol level in human sweat; 2) to create a specific dynamical digital fingerprint of unique EEG and cortisol responses that can reduce the stress level. The collaboration between ESL and NANOLAB targets the development of a wearable device to read data from EEG and cortisol sensors to quantify the level of stress in real time by using different embedded machine learning (ML) algorithms. The prototype is developed for real-time EEG-based subject monitoring. The next steps include the integration of the cortisol monitor system into the existing prototype and studying the correlation between EEG and cortisol data for stress detection purposes.

Presenter: Nataliya YAKYMETS

Abstract: Respiratory rate (RR) is a valuable but frequently misinterpreted vital sign. Manual estimation is time consuming and error prone; and even when correct, requires a challenging feat of memory for comparing to age-adjusted norms. We use 2,875 digital 15 second lung auscultation recordings from 126 pediatric outpatients to develop DeepRR: a deep learning model that predicts the respiratory rate.

Presenter: Annie Hartley

Abstract: “DISCO (Distributed Collaborative Learning) is an open-source, open-access, web-based platform that allows several collaborators to build AI models together, without sharing any data.
It provides an easy-to-use, code-free interface to access federated and decentralized learning and its Javascript codebase can also be integrated into mobile applications.
Instead of copy-pasting data to a central server, DISCO distributes the model to the source of the data. The training thus takes place locally on each user’s device, and the model updates are then aggregated into a global model that has learned from all collaborator’s data, without compromising their privacy.
Distributed collaborative learning has the potential to transform access to critical data and improve predictive tasks from medicine to climate change. Join us.”

Presenter: Annie Hartley

Abstract: Gas bearings use pressurized gas as a lubricant to support and guide rotating machinery. These bearings have a number of advantages over traditional lubricated bearings, including higher efficiency in a variety of applications and reduced maintenance requirements. However, they are more complex to operate and exhibit nonlinear behaviors. Real-time simulation of these systems could force a breakthrough in smart monitoring with physically interpretable anomaly detection techniques. In contrast to purely data-driven predictive maintenance frameworks, the goal of this work is to monitor such a complex dynamical system with an accurate physics-based surrogate model, empowered with data assimilation techniques only where needed. This is the so-called hybrid twin paradigm gaining more and more importance in a variety of engineering applications. We focus herein on the first step toward a hybrid twin, that is on the development of an efficient parametric surrogate model based on the governing equations of the system. A major scientific challenge is the real-time constraint since gas bearings supported rotors operate at extremely high speeds to maximize their performances.

Presenter: Dimitri Goutaudier


Contact

Any questions regarding the event? Please contact us:
[email protected]