AI-supported Digital Twin Synthesis supporting vehicle development and testing for novel propulsion systems
The AVL use case consists of several main tasks along with AI-augmented methods for modelling (vehicle design phase), vehicle development and test (including verifying testing equipment and testing methods) and vehicle operation (including data acquisition). DevOps loops are essential and will be performed with a higher frequency between development/test and operation and a somewhat lower frequency between operation and design.
Task 1: AI-augmented Digital Twin Synthesis
Digital Twins for vehicle prototypes and vehicles in operation as well as for their testing environments will dominate future vehicle development. The degree of automation of the synthesis of such digital twins will determine their application efficiency. AI-aided methods will be applied to create and/or parameterize Digital Twins. Furthermore, the underlying data quality (amount, data context, corner cases, etc.) needed to apply AI-aided methods is considered essential. A major focus will take place in the development of (again: AI-aided) methods to generate high-quality test cases, which produce the aimed data quality.
In the case of pure operational data, requirements for data creation will be defined, which go beyond existing logging mechanisms or sensor facilities.
Task 2: AI-augmented Test Reduction
Vehicle tests are expensive, regardless of whether they are carried out on a test bench or as in-vehicle tests on the road. In the past, simulation and parameterization models supported the reduction of hardware-based test execution. Due to new requirements resulting from new types of propulsion systems or environmental regulations (e.g. the upcoming RDE standards), the already multi-dimensional test space is being multiplied with the introduction of new dimensions, such as a complex operating strategy for battery and hybrid vehicles or the right choice of test routes for RDE. To cope with such complexity, AI-supported methods will be used to reduce the number of required test scenarios without risking disadvantages such as a reduction in test quality as such. This can be done, for example, by distinguishing tests that are relevant for a specific optimization goal from less relevant or redundant tests using AI-aided methods. Monitoring operational data from existing vehicle fleets will add value to this, as AI-based methods will be used to extract virtual test cases that replace the need for traditional test case execution.
Task 3: AI-augmented Maturity Assessment
As a direct consequence of Task 1 and 2, the extensive use of models (e.g. in the form of digital twins), and the application of AI-supported methods for their synthesis or for reducing the test space, the validity of the overall result must be evaluated and measured. If Digital Twins are used e.g. to predict a certain vehicle characteristic at an earlier stage of development or even design, a maturity assessment of this prediction allows a vehicle engineer to focus even more on simulation frontloading or to enable him to drive DevOps cycles with even higher frequency. Furthermore, and especially with DevOps-based methods, a reduced test space carries the risk of incomplete tests with fatal consequences when updates are rolled out on an existing vehicle fleet. AI-based methods for estimating the test coverage would be an important contribution here.