Manufacturing Informatics and Humanin-the-loop: A case study on Friction Stir Welding

Transcription

1 Manufacturing Informatics and Humanin-the-loop: A case study on Friction Stir Welding 3 rd EPSRC Manufacturing the Future Conference Ali Baraka, A. Adriana Gonzalez-Rodriguez, Dr. George Panoutsos, Dr. Kathryn Beamish and Stephen Cater ambaraka1@sheffield.ac.uk Departement of Automatic Control & Systems Engineering, University of Sheffield, Sheffield, UK TWI Ltd. UK, Granta Park, Cambridge, UK The University of Sheffield 23 rd -24 th Sep 2014, Glasgow

2 HCS Research Manufacturing Informatics and Humanin-the-loop: A case study on Friction Stir Welding Outline Manufacturing Case Study HCS design for manufacturing processes Application to FSW (real-time monitoring and performance evaluation) Summary and Future Work The University of Sheffield 23 rd -24 th Sep rd EPSRC Manufacturing the Future Conference

3 HCS Research HCS for manufacturing processes Advanced manufacturing processes involve generating and storing large amounts of complex data rountinely Human operators are often intrinsic part of the process Not trivial to analyse systematically Why HCS is needed? Develop human-centric engineering/ict systems Human-machine interaction, human-in-the loop, human-like approaches to data mining. The University of Sheffield 23 rd -24 th Sep rd EPSRC Manufacturing the Future Conference

4 HCS Research HCS Applications Example: Real-time process evaluation via natural language Predictive modelling Autonomous systems in manufacturing that are working with and around humans. The University of Sheffield 23 rd -24 th Sep rd EPSRC Manufacturing the Future Conference

5 Friction Stir Welding (FSW) Efficient, versatile, and excellent performance. Initially to weld aluminium alloys. Transfer to weld steel grades. Non-linear, complex and phase change. HCS Objective Manufacturing Case Study To create a real-time process monitoring with Linguistic feedback to the user. The University of Sheffield Fig 1. Schematic diagram of the FSW process 23 rd -24 th Sep rd EPSRC Manufacturing the Future Conference

6 Manufacturing Case Study Process Conditions: Tool Rotation Speed Welding Speed Post-weld properties: Weld Quality Mechanical properties Microstructure Data-driven CI Model Internal Process Variables: Torque Downward Force Traverse Force Temperatures Fig 2. Data-driven CI modelling approaches for FSW Data-driven CI Model The University of Sheffield 23 rd -24 th Sep rd EPSRC Manufacturing the Future Conference

7 Work on aerospace-grade aluminium: Material Grade: AA7xxx Thickness: 6 mm Tool Type: Tri-flute TM MX Number of trials: 34 Work on shipbulding steel plates: Material Grade: DH 36 steel Thickness: 6 mm Tool Type: PcBN Number of trials: 25 Manufacturing Case Study The University of Sheffield 23 rd -24 th Sep rd EPSRC Manufacturing the Future Conference

8 Manufacturing Case Study Raw Data Granular Computing Information Granules Knowledge Extraction Initial NF Structure Parametric Optimisation Final Model Fig 3. Knowledge Extraction From Data and Model Development. The University of Sheffield 23 rd -24 th Sep rd EPSRC Manufacturing the Future Conference

9 Manufacturing Case Study Model prediction Linguistic rule-base Weld quality Mechanical properties Microstructure Data-driven Computational Process Models Indices Fig 4. Novelty Detection Framework. Novelty Detection Confidence of prediction Linguistic-based feedback Example: Sample 4 is a New condition with a system predicted output of Weld Quality = 0. The most relevant rule in the systems relates to Good weld quality, and this prediction is of High confidence The University of Sheffield 23 rd -24 th Sep rd EPSRC Manufacturing the Future Conference

10 Summary HCS for fault detection in real-time process monitoring and performance evaluation Current Work on AA7xxx aluminium allys and DH36 steel Create HCS that provides process understanding and gain insights Can be used for on-line monitoring and prediction in realtime (unlike CFD, FEA) Alert the user when a new system behaviour occurs. 23 rd -24 th Sep 2014 The University The University of Sheffield of Sheffield 3 rd EPSRC Manufacturing the Future Conference

11 future work Introduce real-time decision support systems (DSS) to: Automatically or semi-automatically guide the process into optimum weld conditions by taking into consideration a set of desirable objectives. Assist the process operator in making effective decisions via feedback in natural language. Implement the DSS in real-time (i.e. through-process modelling and optimisation). The modelling framework could be extended to various industrial applications too. The University of Sheffield 23 rd -24 th Sep rd EPSRC Manufacturing the Future Conference