Machine Learning in the Industrial Sector, with SMS group at Messe Düsseldorf

Thomas Wiecki @thomaswiecki

Am Staad - Düsseldorf
Am Staad

Fri, Jun 28, 2019 at 4:30 PM - 6:30 PM

View event

View event


Share with friends


This Meetup will be a bit different. It will take place at the Messe Düsseldorf. You don't have to buy a ticket for the Exhibition to come. We will hear from SMS group about using Machine Learning in an industrial setting and see their function live in action. Logistics: - Adress: GPS coordinates: 51.269011, 6.727094 Address for car navigation system: D-40474 Düsseldorf, Am Staad (Stockumer Höfe) - We recommend you to come by train: To get to the exhibition centre, take tram no. U78 MERKUR SPIEL-ARENA/Messe Nord or U79 Messe Ost (Exit at MesseOst/Stockumer Kirchstr.) or bus no. 722 (Exit at Messe Ost or Messe Süd/CCD). There are two major transport hubs where you are likely to change: Hauptbahnhof (central station) and Heinrich-Heine-Allee. Nearly all destinations in and around Düsseldorf can be reached from these stations. - All listed Meetup participants are asked to register at the Info Counter at the North Entrance and receive a service ticket there. This service ticket entitles to admission from 6 p.m. onwards. We will meet you there and lead you to the Meetup space. There will be two short talks: Azure Machine Learning Studio on a continuous caster, Sergey Gorainov The Azure Machine Learning Studio is a hands-on session. The tool is free to use and is available on the Internet at https://studio.azureml.net/. Sergey shows how non programming specialists are able to design a machine learning approach using the drag and drop function. Options for inserting existing code are shown for advanced users. The path from raw data to finished solutions is also shown. Novelty detection for condition monitoring, Stefan Klanke Monitoring the condition of critical production equipment is a key factor in reducing unplanned downtimes. Still, for many types of faults, optimally adjusting alarm thresholds remains an unresolved problem: A system that is too sensitive quickly loses its credibility due to false alarms. At the other extreme, faults are detected too late, if at all. We use novelty detection algorithms to model the joint distribution of process variables measurements, and statistics derived from measurements. Such models are trained separately for each machine, based on data taken in a healthy condition.