AI-Based Generative Engineering for Hybrid Vehicle Design

The automotive and aerospace industries face increasing pressure from new environmental regulations, necessitating high-quality and cost-effective products that cater to the market’s demands. The future of transportation is shifting toward hybrid and electric vehicles (xEV), making the development of powertrain systems more intricate. Meeting stringent development timelines adds to the challenges faced by the industry.

Currently, designing xEV systems relies on the experience of designers to achieve optimal performance while satisfying requirements and constraints. The process involves iterative engineering, modifying designs based on simulation results during collaborative and individual phases. However, time constraints prevent convergence to a globally optimal design due to limited iterations. The conceptual design step is crucial in determining the principal solution, significantly impacting the total product cost.

To efficiently select candidates from various alternatives, a systemic approach is essential. This approach assists engineers in the conceptual design research phase by automatically exploring architectures and concepts that fulfill requirements, avoiding fixation on a single design and performing parameter sensitivity analysis. Such approach reduces the number of iterations by utilizing computer tools to automate tasks and conduct detailed design phases for promising candidates.

A New Approach to System Modeling

Generative engineering is a new approach to system modeling using object-oriented modeling coupled with artificial intelligence and 3D generative algorithms; it allows to exhaustively explore all possible combinations, to manage uncertainty and variability and to explore quickly all the functional architectural and 3D integration solutions meeting the requirements.

With this object-oriented system modeling based on the Python programming language, knowledge about a system is encapsulated in objects which are defined by a set of attributes and methods. The application of algorithms to these objects allows the execution of automated tasks in the system design process, such as architecture research.

The system modeling integrates the product requirements, the objectives to be minimized, the technical knowledge (e.g., design rules) and the functional (or physical) description and/or the 3D description of the system.

The advantages of generative engineering approach are:

explore all possible configurations to avoid late questions
identify the optimal global solution with respect to the system requirements to be met
make informed decisions upstream on structuring choices
achieve rapid design iteration loops.

Application to xEV Systems Architectures

Figure 1. xEV design process based on an engineering system approach. (Image: Dessia)

The vehicle system is broken down into various systems and subsystems, such as the powertrain system and the cooling subsystem (Figure 1). The design process commences with the definition of the vehicle’s requirements, encompassing performance and profile specifications, alongside design constraints like packaging limitations. These requirements are then applied to the powertrain system level, where design engineers face decisions concerning powertrain architecture, component technology, and sizing to fulfill the powertrain requirements.

These choices made at the powertrain level subsequently establish the prerequisites for the cooling subsystem. At the cooling subsystem level, decisions regarding architecture and components must be taken. The choices made at different system levels, both for architecture and components, significantly influence overall system objectives, such as energy consumption and costs.

Notably, the impact of the system architecture on system performance may outweigh that of component sizing and control. The methodology facilitates architects and engineers in making optimal choices for architecture and components at distinct system levels: powertrain system and cooling subsystem.

Design exploration at powertrain system level (HEV)

Figure 2. HEV powertrain system modeling. (Image: Dessia)

The powertrain system is modeled using the approach presented in the previous section. It is decomposed into the following components, as shown in Figure 2: internal combustion engine (ICE), battery, electrical machine (s), and gearbox(s). At this stage, designers consider the powertrain components at a high level of abstraction.

For instance, the ICE is modeled by an object that provides a torque and rotational speed with a single output mechanical connection. A gearbox is modeled by an object with two mechanical connections (input and output) capable of providing a torque and speed ratio.

Figure 3. Examples of possible HEV powertrain architectures. (Image: Dessia)

The proposed powertrain system model, coupled with a decision-tree generative algorithm, allows for the generation of all possible solutions for connecting the components of the powertrain system. Figure 3 presents some examples of powertrain architecture generated, with each architecture represented by a graph.

To meet the powertrain requirements, each architecture requires specific component specifications. For example, to fulfill the same powertrain requirements (for wheel torque and speed), a series and parallel HEV powertrain architecture will need different ICE torque and speed, resulting in different ICE sizing.

By generating all powertrain architecture and associated component sizes, engineers can decide on the optimal solution. The choice of optimal solution defines the cooling subsystem requirements.

Design exploration at cooling subsystem level

The cooling subsystem is modeled using a similar approach to that used for the powertrain system. The thermal needs are defined by the powertrain components, while the pump and radiator components enable the cooling subsystem to evacuate and fulfill the cooling requirements.

Figure 4. Examples of possible cooling architectures. (Image: Dessia)

Similar to powertrain analysis, a generative algorithm is used to generate all possible cooling subsystem architectures. Figure 4 provides examples of cooling solutions represented by graphs. At this level of abstraction, each component has two hydraulic ports: an input and an output. The generative algorithm used here aims to search for all possible combinations of component ports, where each combination represents a cooling graph architecture.

Cooling components must be sized to satisfy the cooling requirements for each architecture. For a given set of thermal losses from powertrain components, pumps will operate at different operating points for an architecture in which all components are placed in series compared to one in which they are placed in parallel. Optimizing the pump’s operating point can significantly affect the pump’s energy consumption.

Generative Engineering at Renault

In order to make its engineering processes more efficient and to increase the performance of the automotive product, Renault has developed various generative engineering applications using Dessia framework for the generation of electrical harness architectures, but also for the technical definition of the thermal system architecture of a hybrid powertrain vehicle platform.

The main ambition for generative engineering was to associate very early functional specifications and 3D architecture drawing and to multiply the number of iterations of technical definition via generative AI algorithms in order to explore as soon as possible the field of possible solutions.

It became possible to go through the design space very widely. In concrete terms, it was possible to carry out studies of the sensitivity of the requirements fields, analyses of the incompatibility of inter-system choices or even analyses of the consequences of local choices on the overall structure of the product.

Figure 5. Generative engineering with functional modeling and 3D integration. (Image: Renault Group)

Thanks to these applications, a hundred eligible solutions were produced with a time saving of more than 80 percent, using at least three times less resources.

Regarding the architecture of the thermal system (Figure 5), the objective was to optimize the energy required for cooling, the layout of the main components, the length of the hoses and the other networks, both electrical and air conditioning.

For this, the first step consisted in a functional approach as described in the previous paragraph. The objective is to generate and optimize a panel of functional water circuit architectures based on the cooling system requirements. The hydraulic permeability of each solution is optimized on cycle in order to minimize the energy expenditure to meet the cooling requirements.

On the HEV vehicle studied, the study allowed to make impact analyses of several modifications of the water circuit architecture, such as to evaluate the impact of cooling the battery either by the air conditioning rather than by the water circuit. In a few minutes without having to trace any hoses manually, a reduction of 40 percent in the cost of water circuit components could be achieved.

Dessia AI-based generative engineering has a great potential, for example:

At the time of project scoping: The increasing complexity linked to the need for performance and the availability of more and more usage data pushes toward processes capable of apprehending all these dimensions, systematically exploring the field of possibilities and determining the strengths and weaknesses of solutions. The earlier these assessments are carried out, the better the engineering performance, avoiding design times penalized by late iterations while using scarce expert resources much more efficiently.

In RFQ phase: Where it is necessary to be able to quickly configure the most performing solution to fulfill the customer’s requirements by re-using COTS or by looking for an innovative solution; a generative engineering application integrating data and knowledge allows to generate in a few hours all the possible solutions, to rank them according to attributes like performance, cost, commonality rate, and to take an informed decision on the solution to propose to the customer.

This article was written by PierreEmmanuel, CEO of Dessia (Paris, France). For more information visit here .