Project type: Bridge Project
Downtime of equipment is costly and a source of safety, security and legal issues. Today, organisations adopt a conservative schedule of preventive maintenance independent of the condition of equipment. This results in unnecessary service costs and occasional interruptions of production due to unexpected failures. Therefore, it is imperative in many domains to transition from regular maintenance to predictive maintenance. A recent report An AI nation: Harnessing the opportunity of artificial intelligence in Denmark estimates that enabling predictive maintenance via AI has a 14-19 billion potential for the Danish private sector.
Relevant domains include medical production (e.g. Novo Nordisk) to introduce condition-based maintenance of the machines that are used in production. The data collected by the equipment manufacturers is often not available in real-time. To address this issue, they need accurate predictive models, based on data collected by sensors under their control. Robot manufactures (e.g. UR) and their integrators want to enable condition-based maintenance of robotic systems. To address this, they need predictive models based on data from robots. In both domains due to reliability and safety requirements it is a prerequisite that data collection and processing are placed in vicinity of the equipment.
The energy sector (e.g. Energinet) wants to incorporate predictive knowledge of equipment performance. They need accurate predictive models, based on available data. E.g. for wind turbines based on local wind conditions as well as the state of the wind turbines. For this case Energinet has to collect the data externally as they do not have access to internal wind turbine data.
The research aim of the project is methods and tools that enable industry to develop new solutions with accurate AI-based maintenance predictions on edge-based software platforms.
The resulting applications must be deployed to collect and process large amounts of data locally. This data feeds high accuracy predictive models deployed at the edge, adapted to changing local conditions, and maintained with minimum intervention from operators.
These models should cover abstractions allowing the understanding of relevant dependencies in the data. The key research problem is to devise architectures and solutions that scale to the entire fleets of equipment with accurate AI predictions.
This also requires that resources in terms of processing power, storage and communication, are optimised in order to obtain low-power and real-time performance, leading to a resource efficiency vs prediction accuracy trade-off. The project will establish a bridge to enable Danish companies to develop and use AI-based predictive maintenance within several domains.
The scientific value of the project is new methods and tools and associated knowledge of their performance and properties in field tests. These are important contributions to provide excellent knowledge to Danish companies and education programs within AI and IoT.
For capacity building the value of the project is to educate 3 PhD students (including 1 Industrial PhD) and 1 Post Doc in close collaboration with industry. The open source availability of general project outcomes and industry collaboration enable several exploitation paths. In addition, for the master-level the projects will offer an industry program to 15 students at 3 universities.
The business and societal value is on a national level estimated to a 14-19 billion potential for the Danish private sector. In the project we target both the medical, robotic, industrial and energy sectors. These are Danish frontrunners in adopting the technology and creating inspiration for wider adoption by the Danish private sector. For the public sector enabling equipment with higher operation efficiency will positively impact the efficiency of the sector.
Within the area of AI the project will research and develop AI models applicable to predictive maintenance. Furthermore, research will consider methods for handling limitations on training data among others due to data ownership restrictions and confidentiality. A final aspect is work on AI models that can adapt to different edge conditions including available processing power and timing deadlines.
Within the area of Cyber-physical systems the project will research and develop software architectures and platforms for edge-based execution of AI models. The outcomes should enable AI models to adapt to changing local conditions, and be maintained with minimum intervention from operators. The key research problem is to devise architectures and solutions that scale to the entire fleets of equipment with accurate AI predictions. This requires methods that optimize resources in terms of processing power, storage and communication in order to obtain low-power and real-time performance, leading to a resource efficiency vs prediction accuracy trade-off.
October 1, 2020 – September 31, 2024 – 3.5 years
Total budget DKK 12.24 million / DIREC investment DKK 6.3 million
University of Southern Denmark
The Maersk Mc-Kinney Moller Institute
E: mbkj@mmmi.sdu.dk
IT University of Copenhagen
Department of Computer Science
Technical University of Denmark
Dept. of Applied Mathematics and Computer Science
Technical University of Denmark
DTU Compute
The Alexandra Institute
The Alexandra Institute
IT University of Copenhagen
Department of Computer Science
IT University of Copenhagen
Department of Computer Science
Universal Robots
University of Southern Denmark
Technical University of Denmark
Universal Robots
Energinet
Octavic
Digital Research Centre Denmark is a unique collaboration between the eight Danish Universities and the Alexandra Institute supported by the Innovation Fund Denmark.