Generic Building Blocks for Understandable Deep Learning Based Outage Predictions
The goal of COGNITUS is to provide a Deep Learning pipeline equipped with a set of generic algorithmic building blocks for predicting outages of machineries based on sensor data streams.
Innovation in COGNITUS is driven by its interdisciplinary nature, which enables the combination of expertise drawn from the fields of data science and maintenance planning. This yields novel integrated data science and maintenance planning methods that tackle several open scientific and technical challenges, i.e.: training and deploying Deep Learning models on-top of horizontally scalable data streaming frameworks; exploring and visualizing streaming data; and investigating probing techniques for studying the inner structure of trained Deep Learning models. COGNITUS will also implement prototypical dashboards and investigate how algorithmic predictions can be communicated to the end user while outperforming existing maintenance software solutions in terms of user experience.
The expected results of COGNITUS are:
- Data-driven maintenance and decision support models considering algorithmic predictions in decision making.
- Data analytics and visualization algorithms operating on-top of data streams.
- Software libraries supporting training and evaluation of Deep Learning based outage prediction models.
- Two pilot demonstrators in the fields of production and logistics to assess and quantify the benefits on operational and business level.
Partners: SPAR, Swarovski, LineMetrics, Fraunhofer Austria
Facts:
- Projektbeginn: October 2019
- Projektdauer: 36 Monate
- Budget: ca. 700k EUR
- Förderung: FFG
Webpage: https://cognitus-project.info/