1. Introduction and Background

Data-driven optimisation is central to modern motorsport and automotive engineering. Teams like Mercedes-AMG Petronas Formula One use telemetry, simulation, and predictive algorithms to refine setups, boost reliability, and maximise performance. The rise of high-fidelity simulation tools and open datasets now enables the exploration of similar methods through machine learning.

This dissertation proposes a system that predicts vehicle performance metrics—such as speed, cornering efficiency, and energy consumption—using simulated telemetry data. By analysing relationships between parameters and outcomes, the project aims to identify optimal configurations under varying track and environmental conditions.

Bridging software engineering, data science, and mechanical modelling, the project applies computational intelligence to real-world engineering problems where performance and precision are key.

2. Research Problem and Aim

Problem: Simulation data in vehicle analysis is highly dimensional and non-linear, making it difficult to extract actionable insights using traditional techniques. Aim: To design and evaluate a machine learning model that predicts and optimises vehicle performance metrics using simulated telemetry data.

3. Research Questions and Objectives

Research Questions:

Which machine learning/AI models best predict vehicle performance metrics from simulated telemetry?

How can outputs guide optimisation of vehicle parameters for speed, efficiency, or handling?

What preprocessing or feature engineering methods improve model accuracy and interpretability?

Objectives:

Generate or source a simulated telemetry dataset including throttle, braking, suspension, and tyre data.

Preprocess and normalise data to identify key correlations.

Develop, train, and evaluate predictive models (e.g., Random Forest, Gradient Boosting, Neural Networks).

Implement optimisation methods (e.g., grid or Bayesian search) to find ideal setups.

Assess performance using metrics such as MSE and R², and discuss real-world applicability.

4. Literature Review Summary

Machine learning has proven effective in automotive domains like predictive maintenance, lap time estimation, and EV energy management (Zhao et al., 2021; Kim et al., 2020). Advanced models such as Gradient Boosted Trees and Neural Networks capture complex dependencies better than traditional approaches. While most research relies on proprietary real-world telemetry, simulated data provides a flexible, privacy-safe means of testing and optimisation. Studies (e.g., Singh et al., 2022) indicate increasing simulation realism, making it suitable for developing performance-prediction frameworks.