Overview and details of the sessions of this conference. Please select a date or location to show only sessions at that day or location. Please select a single session for detailed view (with abstracts and downloads if available).
Please note that all times are shown in the time zone of the conference. The current conference time is: 17th May 2024, 12:23:39pm CEST
D321: MACHINE LEARNING IN DESIGN AND PRODUCT DEVELOPMENT
Time:
Wednesday, 22/May/2024:
10:45am - 12:30pm
Session Chair: Gualtiero Fantoni, University of Pisa, Italy
Location:Congress Hall Ragusa
Presentations
Minimizing occupant loads in vehicle crashes through reinforcement learning-based restraint system design: assessing performance and transferability
Janis Mathieu1,2, Parul Gupta3, Michael Di Roberto1, Michael Vielhaber2
1Porsche Engineering Group GmbH, Germany; 2Saarland University, Germany; 3Ilmenau University of Technology, Germany
The optimization of mechanical behavior in safety systems during crash scenarios consistently poses challenges in vehicle development. Hence, a reinforcement learning-based approach for optimizing restraint systems in frontal impacts is proposed. The trained agent, which adjusts five parameters simultaneously, is capable of minimizing loads on a seen and unseen anthropomorphic test device on the co-driver position and is thus able of transferring knowledge. A hundred times higher rate of convergence to reach a similar optimum compared to a global optimization algorithm has been achieved.
A low-cost non-intrusive spatial hand tracking pipeline for product-process interaction
James Gopsill, Aman Kukreja, Christopher Michael Jason Cox, Chris Snider
University of Bristol, United Kingdom
Hands are the sensors and actuators for many design tasks. While several tools exist to capture human interaction and pose, many are expensive and require intrusive measurement devices to be placed on participants and often takes them out of the natural working environment. This paper reports a novel workflow that combines computer vision, several Machine Learning algorithms, and geometric transformations to provide a low-cost non-intrusive means of spatially tracking hands. A ±3mm position accuracy was attained across a series of 3-dimensional follow the path studies.
A conceptual MCDA-based framework for machine learning algorithm selection in the early phase of product development
Sebastian Sonntag, Erik Pohl, Janosch Luttmer, Jutta Geldermann, Arun Nagarajah
University of Duisburg-Essen, Germany
Despite the potential to enhance efficiency and improve quality, AI methods are not widely adopted in the context of product development due to the need for specialized applications. The necessary identification of a suitable machine learning (ML) algorithm requires expert knowledge, often lacking in companies. Therefore, a concept based on a multi-criteria decision analysis is applied, enabling the identification of a suitable ML algorithm for tasks in the early phase of product development. The application and resulting advantages of the concept are presented through a practical example.
Machine learning-based virtual sensors for reduced energy consumption in frost-free refrigerators
Alejandro Alcaraz1, Dennis Ilare1,2, Alessandro Mansutti1, Gaetano Cascini2
1Elettrotecnica ROLD, Italy; 2Politecnico di Milano, Italy
This study explores Machine Learning (ML) integration for household refrigerator efficiency. The ML approach allows to optimize defrost cycles, offering energy savings without complexity or cost escalation. The paper initially presents a State-of-the-Art of ML potential to improve functionality and efficiency of refrigerators. Since frost is the cause of significant energy losses, a ML-based Virtual Sensor was developed to predict frost formation on the evaporator also in low -level refrigerators. The results show the environmental significance of ML in enhancing appliance efficiency.
Automating the assembly planning process to enable design for assembly using reinforcement learning
Rafael Parzeller1,2, Dominik Koziol1, Tizian Dagner1, Detlef Gerhard2
This paper introduces a new concept for the automation of the assembly planning process, to enable Design for Assembly (DfA). The approach involves the application of reinforcement learning (RL) to assembly sequence planning (ASP) based on a 3D-CAD model. The ASP algorithm determines assembly sequences through assembly by disassembly. The assembly sequence is then used for the generation of subassemblies by considering the product contact information. The approach aims to support the creation of the manufacturing bill of materials (MBOM) by automating the assembly planning process.