AI Transfer Learning in Industrial IoT

Projektinformationen

AI-TRAIN

ID Finanzhilfevereinbarung: 957138

DOI

10.3030/957138

Projekt abgeschlossen

EK-Unterschriftsdatum 31 August 2020

Startdatum 1 Oktober 2020

Enddatum 30 September 2021

Finanziert unter

• INDUSTRIAL LEADERSHIP - Innovation In SMEs

Gesamtkosten

Keine Daten

EU-Beitrag

€ 117 000,00

Koordiniert durch

VERTIKAL AI APS
Denmark

Periodic Reporting for period 1 - AI-TRAIN (AI Transfer Learning in Industrial IoT)

Berichtszeitraum: 2020-10-01 bis 2021-09-30

Zusammenfassung vom Kontext und den Gesamtzielen des Projekts

Vertikal AI aims to simplify renewable energy asset management so that less can do more for the planet. Wind energy plays a pivotal role in the transition towards a society powered by renewable energy, but the cost and risk of operating a wind farm must come down.

Vertikal AI’s core product is an AI platform for wind turbine drivetrain condition monitoring, using online data streams from vibration sensor systems. The business of condition monitoring is about delivering early warning on mechanical faults, so that maintenance operations and repair costs can be minimized. Early and robust fault detection also reduces unplanned downtime, and thereby lost power production.

Vibration-based condition monitoring has been the go-to technology for the last two decades in the wind industry, and there are more than 150.000 wind turbines with online sensor systems installed on the gearbox, generator and main bearings. However, the related analysis task is extremely manual and requires highly specialized engineers. Vertikal’s innovative AI technology solves this problem, by automating a large part of the manual parts of condition monitoring, enabling 10x monitoring efficiency while producing significantly earlier warnings than a human can achieve. However, to build an effective AI-based fault detection system, we needed to overcome the lack of training data and a high level of data disparity.

Project AI-train sought to overcome this by researching technology that can enable AI models that work across different sensor system brands, measurement configurations, turbine and component brands. Vertikal AI software is installed on more than 5GW of wind power capacity globally. The research scope included 4 different sensor system brands, 17 different turbine models, and more than 100 unique drivetrains configurations. Building AI for such a diverse fleet would have been extremely difficult without the technology research in project AI-TRAIN. The Innovation Associate has now been hired permanently, and the research output has become part of the core operating model and competitive advantage.

Arbeit, die ab Beginn des Projekts bis zum Ende des durch den Bericht erfassten Berichtszeitraums geleistet wurde, und die wichtigsten bis dahin erzielten Ergebnisse

The project integrated the Innovation Associate into the company by A) Providing in-house training on Vertikal’s proprietary AI technology and B) External certification in vibration diagnostics (CAT-I)

The Innovation Associate built an exploratory data analysis and formulated the experimental design with a purpose to achieve the project goals set forth in the application.

Through several iterations of experiments, the Innovation Associate managed to reach the necessary outcomes, which were implemented in the daily operations of Vertikal AI.

Fortschritte, die über den aktuellen Stand der Technik hinausgehen und voraussichtliche potenzielle Auswirkungen (einschließlich der bis dato erzielten sozioökonomischen Auswirkungen und weiter gefassten gesellschaftlichen Auswirkungen des Projekts)

The key project output was an innovative AI technology for automatic identification principal data components that enable state-of-the-art AI models to be trained and deployed on disparate data and heterogeneous fleets of wind turbines.

The technology research was used to build specific fault detection algorithms of wind turbine drivetrain components; main bearings, gearboxes, and generators. These models have now been deployed on roughly 2.700 wind turbines around the world and have already proved to perform vastly better than previous state-of-the art.

During the project, we also developed better project management practices for data science projects, as well as the technical deployment of AI models at scale.

The expertise on how to train high-capacity AI models on disparate datasets is part of what distinguishes Vertikal AI from all other market players within wind turbine condition monitoring.

publication1.jpg