We assess technical and commercial risks and opportunities of starting a company that applies research conducted in the field of Deep Reinforcement Learning for practical logistics and supply chain. In particular, we examine how the developed knowledge can be used to develop software to automate operational decision making at companies. The project includes the development of a mock-up prototype solution that enables us to more deeply assess technical feasibility while showcasing our capabilities to potential interested parties.

This research will contribute to several ongoing projects and programmes DATAS for multi-channel, multi-company collaboration (NWO file number: 438-15-507) DATA2MOVE (Research Program in Data Science Center Eindhoven and submitted proposal (NWO file number: BDL.16C2.002) PERFORM (ITN EUROPEAN TRAINING NETWORK)

The project developed optimization models to increase the availability and accessibility of new treatments for rare diseases. First, the project analyzed the impact of subsidies on the development on new drugs and social welfare. Inspired by recent initiatives in Europe, the project examined novel pricing and payment schemes, such as exogenous pricing and outcome-based payment. Second, the project developed optimization models to entice drug repurposing for unmet clinical needs. Third, the project analyzed adaptive drug approval programs to enable faster access to medicines. Lastly, the project analyzed a portfolio of operational problems for pharmaceutical (bio)manufacturers to reduce drug manufacturing costs.

More and more people are being threatened and in some cases even murdered in the Netherlands, while capacity for individual protection (e.g., bodyguards) remains scarce. In this project, mathematical models developed for optimal deployment of protection will be tested at the Dutch National Police, using serious gaming. A workshop with policy makers will also be organised to discuss options for implementing these models. This project will contribute to a more effective deployment of individual protection capacity and as such improve national security.

System failures of capital goods such as aircraft, MRI-scanners, and lithography machines have severe consequences in terms of costs and sometimes human safety. For example, failures in production systems can lead to the standstill of factories and penalties for delayed orders. Failures of medical equipment, bridges, or aircraft can lead to safety issues and human fatalities. Failures are usually the result of degradation exceeding a critical level. Degradation can be measured by monitoring health indicators such as temperature, and vibration. Nevertheless, accurate prediction of failure times is difficult because the event we want to predict, a failure, is rarely observed and degradation processes have uncertain evolution. The traditional approach deals with this uncertainty by fitting statistical degradation models on measurements made on several systems over time in order to predict failures. These statistical models are then used as input to decision models that optimize maintenance decisions for all systems collectively. The advent of modern inexpensive sensor technology and their connection to the so-called Internet-of-Things, makes it possible to generate a real-time stream of degradation data for each individual system that has installed sensors. This offers new opportunities to learn the degradation behavior of each system individually and to integrate this directly with the decision making for individual machines and their logistic support system. The objective of this research is therefore twofold: To (i) make tractable models to learn the degradation behavior for both system populations as well as for individual systems based on real-time degradation data and (ii) integrate this directly with decision making about maintenance and surrounding logistics such as service engineers and parts. Leveraging this new data availability in this way has the promise of better failure predictions (fewer false positives and negatives) and smarter maintenance leading to decreased monetary and societal losses.