Experience report on the application of genetic algorithms to reduce costs of the software validation process in the automotive sector during an engine control unit project

The number of electronic control units (ECU) installed in vehicles is increasingly high. Manufacturers must improve the software quality and reduce cost by proposing innovative techniques. This research proposes a technique being able to generate not only test-cases in real time but to decide the best means to run them (hardware-in-the-loop simulations or prototype vehicles) to reduce the cost and software testing time. It is focused on the engine ECU software which is one of the most complex software installed in vehicles. This software is coded by using Simulink® models. Two genetic algorithms (GAs) were coded. The first one is in charge of choosing which parts of the Simulink® models should be validated by using hardware-in-the-loop (HIL) simulations and by using prototype vehicles. The second one tunes the inputs of the software module (SM) under validation to cover these parts of the Simulink® models. The usage of dynamic-linked libraries (dlls) is described to deal with the issues linked to SM interactions when running HIL simulations. GAs found at least 7 more bugs than traditional techniques and improved the functional and code coverage by between 3 and 11% for functional coverage and by between 1.4 and 7% for code coverage depending on the SM complexity. The validation time is reduced by 11.9% compared to traditional techniques. GAs perform better than traditional techniques improving software quality and reducing costs and validation time. The usage of dlls allows testing the software in real time as described in this study.