In a cognitive factory setting, product manufacturing is automatically planned and scheduled, exploiting a knowledge base that describes component capabilities and behaviors of the factory. However, because planning and scheduling are computationally hard, they must typically be done offline using a simplified system model, and are thus unaware of online observations and potential component faults. This leads to a problem: Given behavior models and online observations of possibly faulty behavior, how likely is each manufacturing process plan to still succeed? In this work, we first formalize this problem in the context of probabilistic reasoning as plan assessment. Then we contribute a solution which computes plan success probabilities based on most likely system behaviors retrieved from solving a constraint optimization problem. The constraint optimization problem is solved using well-optimized off-the-shelf solvers. Results obtained with a prototype show that our method can guide systems away from plans which rely on suspect components.     «

In a cognitive factory setting, product manufacturing is automatically planned and scheduled, exploiting a knowledge base that describes component capabilities and behaviors of the factory. However, because planning and scheduling are computationally hard, they must typically be done offline using a simplified system model, and are thus unaware of online observations and potential component faults. This leads to a problem: Given behavior models and online observations of possibly faulty behavior...     »