Scaling Agile Mechatronics: An Industrial Case Study

Abstract

The automotive industry is currently in a state of rapid change. The traditional mechanical industry has, forced by electronic revolution and global threats of climate change, transformed into a computerized electromechanical industry. A hybrid or electric car of 2013 can have, in the order of 100 electronic control units, running gigabytes of code, working together in a complex network within the car as well as being connected to networks in the world outside. This exponential increase of software has posed new challenges for the R&D organizations. In many cases the commonly used method of requirement engineering towards external suppliers in a waterfall process has shown to be unmanageable. Part of the solution has been to introduce more in-house software development and the new standardized platform for embedded software, AUTOSAR.

During the past few years, Volvo Cars has focused on techniques and processes for continuous integration of embedded software for active safety, body functions, and motor and hybrid technology. The feedback times for ECU system test have decreased from months to, in the best cases, hours.

Domain-specific languages (DSL), for both software and physical models, have been used to great extent when developing in-house embedded software at Volvo Cars. The main reasons are the close connection with mechatronic systems (motors, powertrain, servos, etc.), the advantage of having domain experts (not necessarily software experts) developing control software, and the facilitated reuse of algorithms. Model-driven engineering also provides a method for agile development and early learning in projects where hardware and mechanics usually are available only late. Model-based testing of the software is performed, both as pure simulation (MIL) and in hardware-in-the-loop (HIL) rigs, before it is deployed in real cars. This testing is currently being automated for several rigs, as part of the continuous integration strategy.

The progress is, however, not without challenges. Details of the work split with Tier 1 suppliers, using the young AUTOSAR standard, and the efficiency of AUTOSAR code are still open problems. Another challenge is to manage the complex model framework required for virtual verification when applied on system level and numerous DSLs have to be executed together.