Embedded learning and optimization for the next generation of smart industrial control systems

Thanks to the increasing capabilities of digital technologies, the next generation of industrial control systems are expected to learn from streams of data and to take optimal decisions in real-time, leading to increased performance, safety, energy efficiency, and ultimately value creation.Numerical optimization is at the very core of both learning and decision-making, and machine learning algorithms and artificial intelligence raise huge worldwide research interest, often using cloud computing and large data centers for their optimization computations.However, in order to bring learning- and optimization-based automated decision-making into smart industrial control systems (SICS), two important bottlenecks have to be overcome: (1) computational resources on industrial control systems are locally embedded and limited, and (2) industrial control applications require reliable algorithms, with interpretable and verifiable behavior.

Both requirements partially stem from safety aspects, which are crucial in applications where a single computation error can cause high economic and environmental cost or even damage to people.Pushing the performance boundary of SICS to leverage advanced digital technologies will therefore involve both fundamental new research questions and technological solutions, calling for a new set of advanced methods for embedded learning- and optimization-based control algorithms.

Through its 15 PhD students hosted and seconded at 11 top European research centers (6 academic, 5 industrial) and 4 partner organizations in the US, Japan and China, ELO-X will address the timely and pressing need for highly qualified and competent researchers who will develop embedded learning- and optimization-based control methodologies for SICS, thus enabling new and possibly game-changing digital technologies for important EU industries.