Automated plan assessment in cognitive manufacturing

Compiling AI Engineering Models for Probabilistic Inference

“…We ran experiments on six different problem instances, five plan assessment instances based on factory models and one diagnosis instance based on a satellite camera model from [4]. The factory models 2 and 3 are variations of the model shown in 1a, factory model 1 is taken from [2]. All models contain one assembly and one or two machining stations.…”

“…The most probable trajectory is then chosen as most probable diagnosis, while the complete list is used as set Θ. While [2] used the Toolbar solver, in this work, we use the more recent, award-winning Toulbar2 solver 1 , which implements many state-of-the-art techniques (including, for example, soft-constraint consistency [12]). …”

“…Recent work developed a model-based diagnosis-inspired solution that first computes the most likely trajectories as the best solutions to a constraint optimization problem (COP) and then computes P (G i | o 0:t ) by summing over goal-achieving trajectories among these, normalizing over all [2]. The work exploits an automatic translation of a PHCA to a COP R = (X, D, C) [4].…”

“…Following the approach of using off-the-shelf tools, [4] provides an automatic translation for constraint optimization, which is common in model-based diagnosis. This work was extended in [2] towards solving plan assessment. However, there is still a gap between high-level modeling on the model-based diagnosis side and the methods and frameworks on the probabilistic reasoning side.…”

“…This is the classical probabilistic reasoning problem of computing posterior marginals. The problem of computing most likely hypotheses and, at the same time, probabilities for certain model states is a variant of the plan assessment problem [2].…”