FLACS-based gas detection layout optimisation and the roll-out of the FLACS-HPC service Lars Pesch Gexcon
Outline The Fortissimo II project & collaboration with EPCC + Micropack How to support CFD-based gas detector layouting HPC: product offering, status & outlook
Fortissimo II Project funded by the EU Horizon 2020 research and innovation programme Full title: Factories of the Future: Resources, Technology, Infrastructure and Services for Simulation and Modelling Topic: ICT Innovation for Manufacturing SMEs (part of the initiative) Action duration: 36 months Action start: 1 November 2015
Partners Consortium of 38 beneficiaries 13 core partners: universities/research HPC providers commercial HPC hw/service providers technology commercialization experts and market analysts 25 experimenters e.g. Gexcon
Experiments each experiment is set up as a mini project contractually ring-fenced, no liability across the whole consortium detailed work plan based on work package description exploitation plan and/or business scenario
Gexcon s experiment Topic: Build an HPC-Cloud service for optimising gas and flame detector layouts in hazardous manufacturing and production plants. Ca. 1.5 py (of which ca. 1 py at Gexcon) Partners: 1. UEDIN/ (Edinburgh Parallel Computing Centre) 2., Aberdeen
Partners: EPCC
Partners: Micropack
Gas detector layout optimisation Questions to answer: How many detectors? Where should they be located? Are lines better than points? Voting: what and how? Performance? What is gained by improved detection?
Benefits of CFD-based gas detection system design Repeatable base Build understanding of the flow Based on actual cloud/flow behaviour Improved layout design: reduction in detection times and total risk Cost-effective design: lower CAPEX and OPEX (fewer detectors @ same/better performance)
Challenges Create, organise and handle many scenarios. Run many simulations. No direct support for detectors and detection systems in FLACS. Lots of data to sift through Need different plot types than what is currently in Flowvis.
Solutions Create, organise and handle many scenarios. Run many simulations. FLACS-Risk FLACS-Risk No direct support for detectors and detection systems in FLACS. Lots of data to sift through Need different plot types than what is currently in Flowvis.
FLACS-Risk: set up detectors
FLACS-Risk: define layouts
Layout optimisation Steps towards CFD-based optimisation: Evaluate layouts to find the best one from a set of predefined layouts to minimise (e.g.) detection times/ cloud sizes at detection over a set of scenarios. verification functionality Based on possible detector positions (i.e. without given layouts) arrive at a layout that balances detection criteria (e.g. as many leaks possible as early as possible) with a minimal number of detectors. combinatorial optimization/milp
Optimisation problems several formulations, e.g. p-median: Locate P sensors for a given set of scenarios (given by damage coefficients ) such that the sum of the (weighted) damage coefficients is minimized. p-center: minimize the maximum damage coefficient. different solvers
Optimisation problem (p-median) with
Example setup 8 wind directions, 3 wind speeds 10 leak locations/ca. 5 directions 3 leak rates 3444 scenarios 1097+6 detectors
System performance results (statistics)
Result variation (7 detectors) 20 experiments leaving away 10% of the scenarios: 00: [ 299,593,618,639,723,740, 806] 01: [298, 593,618,639, 740,790,806] 20 experiments leaving away 20% of the scenarios: 00: [299,593,618,639,723,740,806] 02: [298,593,618,639,723,740,806]... 16: [298,593,639,723,740,791,806] 18: [298,593,618,639,723,740,806] xx: [298,299,593,618,639,723,740,790,791,806] modest variation in resulting layouts
Micropack s assessment Examples of engineering considerations: blocking walkways? typical personnel activity?
Micropack s assessment
Seeking pilot projects Gexcon & Micropack are interested in studying real geometries for the purpose of methodology development and validation. Compare detector placement results acquired by traditional/volumetric placement and by CFD simulation/optimised placement. Prerequisite: share the facility geometry with both Gexcon and Micropack. The model will be loaded into Micropack s Hazmap3D modelling package, and FLACS, and a range of detector placement layouts will be tested for comparison. The aim of the project is the develop industry-leading research into future detection system design methodology. Interested? Talk to Lars / Lars.Pesch@gexcon.com
Timeline autumn 2017: verification tool late 2017/early 2018: optimization functionality within FLACS-Risk
FLACS-HPC
FLACS-HPC Many FLACS-simulations run well in massively parallel HPC environments. Gexcon & EPCC have collaboratively offered a FLACS- HPC service for FLACS license holders. Tech: Linux cluster, remote login & file transfer. Reduce the «pain» by providing a frontend for controlling many simulations on the HPC backend from the FLACS-Risk GUI. Service, licensing etc.
FLACS-Risk & HPC
Timeline summer 2017: v1.1 with HPC access
Summary Fortissimo II project CFD/FLACS-based gas detection system optimization FLACS-HPC Interested in pilot projects? Talk to Lars!
Questions?