Towards Inferring Environment Models for Control Functions from Recorded Signal Data

In the automotive domain, control functions (e.g., ACC or brake booster) are mainly validated through road tests by means of performing specific driving maneuvers. In many cases, however, there is only an indirect connection between the inputs at the system level (e.g., position of the brake pedal) and the inputs to a tested component (e.g., negative pressure of a brake booster). In order to validate that a software component was tested sufficiently, engineers have to analyze recorded data after road tests. We present an approach for inferring automata models from such data. These (small) automata are easier to analyze than hours of raw signal data: they exhibit specific states and transitions for different test scenarios, which allow engineers to understand how a function was exercised during a road test. Technically, we generate models in three steps: we (1) identify segments of consistent behavior, (2) classify these segments, and (3) generate automata models from sequences of classified segments. We evaluate the presented approach on speed and acceleration data from a small number of road tests.