Integrated Computational Materials, Process and Product Engineering – IC-MPPE 2022

Condition Monitoring (i.e. estimating a physical system's current fault state from accessible observations (sensor readings)) can sigificantly lower a systems' operating cost by facilitating longer service intervals and shorter service interventions. Model-Based Condition Monitoring (MBCM) uses inverse models to infer the current fault state. Constructing accurate, robust and compact inverse models is a daunting task, though. Hybrid semiparametric modelling of physical systems (eg combining knowledge and observation) is a viable option to deal with these conflicting design goals.
This talk will look back at findings from past research activities as well as forward to the new prospects of probabilistic hybrid models.
Specifically, I will detail MCL's MBCM workflow collecting insights across multiple projects and industries including software tools developed and further ones planned. I will contrast classical block-based hybrid models with more flexible approaches arising from the rich field of physics-informed machine learning. Finally I will motivate the future focus on probabilistic hybrid models.