“The goal is to turn data into information, and information into insight.”

Carly Fiorina, Former CEO of HP


Predictive maintenance

Predictive maintenance helps to reduce downtimes and equipment failure with more on-point settings and thus significantly reduce costs. Artificial intelligence based insights and recommendations do not only make maintenance more efficient for personnel, but they can also investigate and understand what leads to downtimes and how they can prevent them in the first place.


District heating network

Vienna has one of the biggest district heating networks in Europe, consisting of thousands of converter stations cooling down the water before it is delivered to the end customer.

ruag logo
Downtimes occur on a daily basis, personnel has to stop nearly every other scheduled maintenance work in order to repair a broken machine, which makes it a highly inefficient process.
Using historical data our machine learning algorithm predicts seven days in advance potential downtimes. Moreover, the solution indicates unknown correlations between certain data sets and downtimes helping to understand what caused those downtimes.
icon digitalisierung result
Based on these insights, more efficient maintenance schedules can be set in order to avoid those downtimes. The knowledge of root causes can be used in order to have more on-point settings, first as a recommendation system and later a gradual transfer to an automated setting of parameters can be done.


The process

In a first step, the team of the utility company and craftworks were sitting together in order to find out which problems should be tackled and what the goals are. In this case, it was the inefficiency in the maintenance work that should be addressed. After the use case was defined, the teams from both companies checked if there was enough quality data and got the first understanding of processes, parameters, and their values. In certain cases, it might be necessary to use additional sensors from f.e. trusted craftworks partner companies. In a next step, related data and patterns in sensor-, machine, and process data were recognized, before a machine learning model and architecture was developed. At the end of the first phase, a simple software prototype was reviewed verifying the recommendations and identifying enhancements.

Our clients