Reducing the diagnostic uncertainty of a paper input module by active testing

Model-Based Diagnosis (MBD) algorithms often yield a large number of diagnoses, severely limiting their practical utility. A practical way to deal with such diagnostic uncertainty is to use the active testing approach implemented by Fractal (FRamework for ACtive Testing ALgorithms). Fractal computes a sequence of control settings for reducing the number of diagnoses given a system description and an initial observation. In this report we demonstrate the applicability of Fractal to real-world applications, such as the diagnosis of a Paper Input Module (PIM), part of a heavy-duty printer. In this report we show that (1) Fractal computes intuitive results given the PIM model and an initial observation, (2) Fractal scales to large systems and is computationally efficient, (3) and Fractal leads to a steep reduction in the number of diagnoses even with very coarse-grained models and in the lack of prior failure probabilities. As a result, this report validates the use of Fractal for a range of on-board and off-line diagnostic reasoning tasks.