© 2017 - Mike Ashcroft
2014 - ongoing
I am currently working with SKF to design a root cause diagnostics system for use in their manufacturing processes. When defects are observed the system is able to offer ranked suggested root causes to fix the observed problems.
We are using a Bayesian network model built from multinomial-multinomial noisy-or nodes. The model will initially be constructed from expert knowledge, but will then adapt to incoming data and user feedback.
The software behind this process is to be made available as the ABN (Adaptive Bayesian Network) R package. More on this as we proceed.