Field Results of New Sodar Transducer Horn DEWEK 2015 Proceedings Andrew Hastings-Black Introduction Many thanks to DEWEK for accepting this presentation. Education B.A., Physics, Tufts University M.Sc. Candidate, Applied Physics, Columbia University Industry Second Wind, Inc., Triton R&D, 2008 – 2014 Vaisala Oyj, Triton R&D, 2014 – present day Page 2 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Triton Sonic Wind Profiler Page 3 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Triton Sonic Wind Profiler Commercialized in 2008 Monostatic, phased array sodar, 7W power consumption 500+ systems delivered, 6 continents 13 million operational hours with 98.5% uptime Heated mirror option and supplementary fuel cell option Used both in WRA and wind farm operations Excellent reliability and performance, widely adopted by largest developers Bankable data – accepted by key financiers and independent engineers Page 4 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Triton Sonic Wind Profiler Triton cross-section Page 5 © Vaisala 11/10/2023 Speaker array Internal use DEWEK 2015 Proceedings Individual speaker Triton Sonic Wind Profiler Wide deployment in diverse conditions shows weaknesses Atmospheric absorption of sound, geographical+seasonal 𝐶𝑇2 stability variations, diurnal Speaker efficiency temperature-dependence, seasonal Turbine hub heights increasing Data recovery financing requirements ~90% Clear need for higher data recovery at higher heights Redesign speaker horn Page 6 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Wave front path Page 7 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings New Horn Design Key Characteristics Horn length creates 4.5 kHz resonant peak Increases speaker efficiency in SoDAR bandwidth Increases signal strength and microphone sensitivity Improves SNR Central waveguide Increases speaker Directivity Index Reduces phased array sidelobe intensity Reduces fixed echoes Enables operation at multiple wavelengths Page 8 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Frequency Response Page 9 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Polar Response Page 10 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Directivity Index Page 11 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Hardware and Operating Strategy KSN 1005A Speakers 𝜆 = 76 𝑚𝑚 𝑓𝑡𝑟𝑎𝑛𝑠𝑚𝑖𝑡 ∝ 𝑇𝑒𝑚𝑝𝑒𝑟𝑎𝑡𝑢𝑟𝑒 42 Vpp 70 ms pulse width G2 Speakers 𝜆 ∈ 70, 72, 74, 76 , 78, 80 mm 𝑓𝑡𝑟𝑎𝑛𝑠𝑚𝑖𝑡 ∝ 𝑇𝑒𝑚𝑝𝑒𝑟𝑎𝑡𝑢𝑟𝑒, 𝜆 42 Vpp 70 ms pulse width NB: Only changes to Triton configuration are (1) speaker horn, and (2) operating strategy to stay near resonant peak Page 12 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Comparison with Legacy Triton G2 Speaker Triton Performance Upgrade (“TPU”) 160 TPUs deployed world wide 23 TPU-Legacy pairs within 50 km and in similar terrain Total dataset length: 20 years Average measurement campaign: 10 months Data filtered using standard Triton filtering guidelines QF > 90% |VWS| < 1.5 m/s Data recovery = Page 13 © Vaisala 11/10/2023 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑠𝑎𝑚𝑝𝑙𝑒𝑠 𝑎𝑓𝑡𝑒𝑟 𝑓𝑖𝑙𝑡𝑒𝑟𝑖𝑛𝑔 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑜𝑡𝑒𝑛𝑡𝑖𝑎𝑙 𝑠𝑎𝑚𝑝𝑙𝑒𝑠 𝑖𝑛 𝑚𝑒𝑎𝑠𝑢𝑟𝑒𝑚𝑒𝑛𝑡 𝑝𝑒𝑟𝑖𝑜𝑑 Internal use DEWEK 2015 Proceedings Seasonal Distribution of Data Page 14 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Data Recovery Plots 92% 80 m TPU Legacy Uptime Page 15 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Page 16 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Page 17 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Data Recovery Improvement Improvement at nearly all sites: +3% @ 80m, +7% @ 100m, +10% @120m Normalizing for uptime: 20/23 TPU sensors get 90% data recovery at 80m 23/23 TPU sensors get 85% 14/23 Legacy sensors get 90% data recovery at 80m 20/23 Legacy sensors get 85% Biggest improvement in winter months +7% @ 80m, +11% @ 100m, +16% @120m Page 18 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Data Accuracy Resonant design difficult to manufacture. Very important that speakers be “well-matched” Poor matching causes errors in wind speed measurement 𝜎𝑠𝑝𝑒𝑎𝑘𝑒𝑟 → 𝜎𝑏𝑒𝑎𝑚 𝑎𝑛𝑔𝑙𝑒 → 𝜎𝑤𝑖𝑛𝑑 𝑠𝑝𝑒𝑒𝑑 If we quantify 𝜎𝑤𝑖𝑛𝑑 𝑠𝑝𝑒𝑒𝑑 as a function of 𝜎𝑠𝑝𝑒𝑎𝑘𝑒𝑟 we can control speaker matching to ensure that every array has consistent beam pattern Simulate phased array, compare with field data Page 19 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Monte Carlo Simulation Detail R (v. 2.14) RStudio (v. 0.97) Linux Fedora release 14 (Laughlin) ‘seque’ package by J.D. Long Calls virtual clusters using Amazon EC2 instances 𝑚𝑎𝑡𝑐ℎ𝑖𝑛𝑔 𝑐𝑜𝑒𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑡 = 𝑟 = 𝑀𝑜𝑑(𝐴𝑛𝑜𝑟𝑚 𝑒 𝑖𝜃𝑛𝑜𝑟𝑚 − 𝑒 𝑖∗0 ) 6 wavelengths (70 mm – 80 mm) 100 random permutations, per-grouping, per-wavelength 50850 data points per simulation 36 speakers per array 7.7 billion calculations Page 20 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Data Accuracy Page 21 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Simulation Page 22 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Simulation Page 23 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Simulation Page 24 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Data Accuracy Grouping 1 Page 25 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Data Accuracy Grouping 2 Page 26 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Data Accuracy Grouping 3 Page 27 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Data Accuracy Grouping 4 Page 28 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Data Accuracy Grouping 5 Page 29 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Data Accuracy Triton Matching Passband Page 30 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Data Accuracy No significant difference in uncertainty between wavelengths Final passband between Grouping 1 and Grouping 2 3 standard deviations ~ 1.25% total wind speed error How does the model agree with field data? Page 31 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Data Accuracy Page 32 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings Further Research Upcoming accuracy evaluation from 3Tier using tower datasets from resource assessment campaigns ~1% sensor uncertainty for Legacy and TPU Tritons Monte Carlo Simulation of enclosure effects Include inhomogeneous atmosphere TPU development partners: Thanks for listening! Page 33 © Vaisala 11/10/2023 Internal use DEWEK 2015 Proceedings