DMCA
Applying Automata Learning to Embedded Control Software
| Citations: | 1 - 0 self |
BibTeX
@MISC{Smeenk_applyingautomata,
author = {Wouter Smeenk and Frits W. Vaandrager and David N. Jansen},
title = {Applying Automata Learning to Embedded Control Software },
year = {}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
We explore whether LearnLib, a state-of-the-art automata learning tool, is able to learn a model of the Engine Status Manager (ESM), a piece of control software that is used in many printers and copiers of Océ. Finding counterexamples for incorrect hypotheses for the ESM turns out to be challenging due to the large number of inputs. A first contribution of this paper consists of two novel heuristics for test selection, implemented on top of LearnLib, that allowed us to learn an almost complete model of the ESM. The first heuristic is based on the intuition that for many systems it is possible to reach the majority of the states using only a small number of different inputs. The idea is to divide the input alphabet into a number of subalphabets, and to divide the test phase into subphases for each subalphabet. The second heuristic is a randomized variant of the W-method of Chow and Vasilevskii. To the best of our knowledge, this is the first paper in which active automata learning is applied to industrial embedded control software. Using our new test selection heuristics, we succeeded to learn a model of the ESM with 3.326 states. More than 263 million test queries were not enough to find a counterexample for this model. We also constructed another model directly from the ESM software with 3.410 states, indicating that we succeeded to learn 97,5 % of the model using automata learning. Our work provides a challenging benchmark for the automata learning and model-based testing communities.
Keyphrases
automaton learning embedded control software control software complete model different input test query large number test phase randomized variant novel heuristic esm software state-of-the-art automaton learning tool new test selection heuristic first paper second heuristic test selection active automaton learning small number many printer first contribution challenging benchmark many system first heuristic engine status manager incorrect hypothesis model-based testing community
