Metrology for Factory of Future (MET4FoF):
Most studies in machine learning (ML) have focused on maximizing algorithmic accuracy without taking into account the uncertainties. This leads to overconfident predictions and increased risk for adopters especially with sensor degradation and dynamic and unpredictable dynamic environment. Thus, in Metrology for Factory of Future project, we explore the use of software agents to reason and autonomously act on the uncertainties of the sensors and ML models. For example, when a sensor degrades and increases the prediction’s uncertainty, the agents can exclude the particular sensor from its computation using the other healthy sensors. As data is being collected as time progresses, the agents retrain the ML model to decrease its overall uncertainty. We are testing our approach using two testbeds: condition monitoring of hydraulic system at ZEMA, Germany and quality prediction of radial forging at Advanced Forging Research Centre, Glasgow.
Find out more at https://www.ptb.de/empir2018/met4fof/home/
Funded by: H2020, EUROMET
Researcher: Xiang Bang Yong