If the classical approach of physics-based modeling is used to simulate the interaction between cooling and refrigeration circuits and the heat dissipation losses in the powertrain, all subsystems must be resolved with the relevant physical details. To virtually replicate real-world driving conditions, a vehicle model is also required, along with the need to replicate significant parts of the operating strategy and thus the powertrain control. The personnel effort required for such modeling is considerable. Additionally, high-quality measurement data from component test benches (e.g., electric machines, batteries, or combustion engines in hybrid vehicles) and, in some cases, a full-system measurement on a Road2Rig test bench are needed. This significantly increases both time and costs.

The strength of physics-based models lies in their excellent predictive capabilities. Within such a model, components can be resized or completely replaced, and the overall system responds correctly to these changes. This makes this model class particularly well-suited for conceptual studies.

For applications where prototype vehicles are already available or where the cooling/refrigeration circuits of near-series vehicles need to be modeled, an alternative approach appears promising. This approach generates powertrain models directly from data obtained through real-world driving field tests. However, only the (production) sensors installed in the vehicle are available for measurement, and their number and signal quality are often limited. This can be partially compensated by collecting data from a very large number of test drives in field and fleet trials. AI-based approaches are particularly well-suited for modeling such large-scale but lower-quality datasets. These models can accurately describe the behavior of the powertrain, but due to their black-box nature, they do not allow for hardware modifications of individual powertrain components.

This lecture presents an approach that combines physics-based and AI-based models. In the demonstrator project, only field data from production sensors recorded via the OBD interface are used. Subsystems are described using separate neural networks, allowing for the later replacement of individual subsystems with physics-based models. This makes the model suitable for conceptual studies as well.

SmartCorners is an EU-funded project that investigates user-centered comfort using AI methods. Within SmartCorners, AVL Germany is developing a method to train a vehicle climate control algorithm in a user-specific manner with the help of AI. The training environment is represented by a comprehensive vehicle simulation. The aim is to learn the preferences of exemplary virtual occupants and to better respond to the individual needs of the passengers. As the virtual environment is not true to detail and the virtual occupants differ in their preferences from real users, a method is being developed to seamlessly transfer the learning progress to the real vehicle. In addition to improving comfort through individualization, AI is also intended to enhance energy efficiency. Adaptive control of the recirculation flap will be used to find an optimal compromise between the comfort of fresh air and the energy consumption for heating the ambient air instead of the recirculated cabin air.

The global automotive industry is moving towards sustainable mobility through vehicle electrification & autonomous connected vehicles. The vehicle electrification trend is progressing at a good pace with Electric Vehicles (EV) accounting for ~16% of sales worldwide, and expected to increase considerably in the near future. OEMs face many challenges in this relatively new field, starting from manufacturing challenges such as material availability for battery and electric components, as well as availability of charging infrastructure, to vehicle thermal challenges such as battery range and thermal comfort. In extreme weather conditions specifically, the lack of waste heat from the internal combustion engine (ICE), and high resulting energy expenditure of the climate system, lead to a constant trade-off between occupant’s thermal comfort & battery range. Even with the recent proliferation of EVs, range anxiety still exists; in extreme winter weather conditions, this range can reduce by more than 20%, losing power to cabin heaters, seat heaters and defrosters. 

It is important to strike a balance between range and comfort to design an energy efficient Electric Vehicle, and in an environment ripe with competition, and privileging fast development methods, simulation representing realistic usage scenarios can be very beneficial. An advanced virtual Twin can help to size the Heating, Ventilation & Air Conditioning (HVAC) system, and predict battery life, vehicle range and thermal comfort during early design stages. In this study, a virtual co-simulation of the system level & 3D CFD model is used to capture complex interactions across scales, in order to optimize overall vehicle performance. Co-simulation capabilities involving 3D CFD and FEA (Finite Element Analysis) thermal models provide high fidelity predictions of passenger thermal sensation and battery temperature distribution. Further integration with Dymola system behavior models enables the prediction of real-world driving scenarios for all systems in the vehicle. Additional 3D CFD analysis for the underhood airflow is also performed, to characterize the external air received by the front-end heat exchangers of the HVAC system in different vehicle operation conditions. The airflow characteristic is used in the 1D HVAC system analysis to provide realistic boundary condition for the heat exchangers. In addition, this co-simulation approach, the study also presents a methodology for efficient sizing of the HVAC by leveraging reduced order modelling. A set of accurate & detailed cabin comfort thermal results from CFD are aggregated and mapped to define the cabin performance in the system model. With this approach, the vehicle performance can be quickly optimized with multiple quick system model simulations, benefiting from trends captured in detailed 3D CFD response surfaces.