The development of vehicles has always been a discipline of mechanical engineering. After all, cars were always about the engine, the power, the efficiency. Traditionally, the development of the complex overall automotive system has always been carried out in accordance with classical principles of mechanical engineering, for example by using development models like the V-model. Although the proportion of electronics in automobiles has increased sharply over the past few decades, relatively little has changed here. However, this constant in automotive development is now being challenged by many OEMs. Due to many teams working remotely, driven by the Coronavirus pandemic, as well as worldwide supply bottlenecks for raw materials and semiconductors, a radical rethinking is now taking place. Model-driven engineering, i.e. virtualized prototyping, virtual testing and virtual approval, are pointing the way to the future of automotive development.
With the shift toward electromobility and the availability of high-performance microelectronics, true digitization in the automotive sector has begun in recent years. Today, electronics and software are required by every function in a vehicle. Where hardware functions used to dominate in the past, all networking, automation, and personalization functions are now implemented in software.
Today, we even speak of a software-defined vehicle: Software not only defines the customer experience, it also determines which hardware must be used in the vehicle’s electronic control units.
Current trends towards the fully electrified vehicle, autonomous driving, and new mobility services such as car sharing or teleoperated driving are enabled in particular by software and microelectronics and are leading to a disruptive change in the entire value chain.
These trends not only impact usage, but also require new development methods and forms of collaboration between partners along the entire value chain – from microelectronics manufacturers to mobility providers and transport infrastructure operators.
As far as safety and the driving experience are concerned, sensors rather than mechanics will play the key role in the vehicle in the future. For autonomous driving, a vehicle must exceed the driver’s sensory and cognitive abilities with powerful sensor technology. The task is to select from the numerous options of sensors, architectures, and data fusion approaches the one offering the best possible performance, security, and price. However, completely new development methods are now required. Similar to the development of human senses, which could also only develop in interaction with each other and the environment, the development of automotive sensor systems cannot take place in isolation from the overall system. The influences of the hardware and the environment must be evaluated and tested under all possible driving situations.
If this were done on the road, it would add up to many millions of road kilometers before a new vehicle model receives approval. Instead, agile principles must be applied, as they have been successfully used in software development for a long time: It must be possible to implement changes to the system quickly and their effects must be immediately visible to the developer. However, this type of agility is more difficult than with pure software products due to the hardware dependency. However, this can be solved by consistently using a virtual development methodology and virtualized hardware. Similar to software development, implementation details can thus be abstracted and the development of the overall system can be developed from the functionality side. This is the only way in which it will be possible in the future to achieve sufficient test coverage for the validation of AI-based ADAS/AD systems. In this respect, virtual development and virtual testing are the most important tools in future automotive development.
However, to implement the concept of virtual development, all partners within the automotive value chain must work closely on a joint solution. Among other things, large gaps currently exist in the following areas:
- Physics-based and validated sensor models,
- Tool-agnostic virtual ECU models,
- Compatible tools that allow quick and easy configuration,
- Databases with standardized driving scenarios.
Many of the aspects mentioned still require extensive research. Once the above-described requirements for virtual development have been met, the question of a legal road-approval still arises, however. While it is primarily the hardware that is validated in conventional development, in the virtual development process the simulation tools and the models have to be validated instead so that a virtual approval can take place. It must be factually proven that the virtually driven road kilometers are equivalent to real road kilometers.
In summary, the standard use of virtual sensors and virtual ECUs are the key to testing and validating future autonomous driving functions. However, there are still a number of challenges to be addressed, particularly in the area of collaboration among the partners of the value chain, before such a virtual development methodology can be used across the board.
The post Virtual Testing Of Automotive Sensor Systems appeared first on Semiconductor Engineering.
