Data supporting the publication: An Integrated Smartphone and Physics-Informed Framework for Real-Time Estimation of Rolling Resistance and Tractive Power Losses in Passenger Vehicles

This dataset accompanies the study “An Integrated Smartphone and Physics-Informed Framework for Real-Time Estimation of Rolling Resistance and Tractive Power Losses in Passenger Vehicles” and contains both experimental and derived data used to develop, validate, and deploy a real-time system for estimating rolling resistance and vehicle energy losses. The dataset integrates field measurements, synthetic data augmentation, machine learning outputs, and physics-informed model predictions.<br>The research follows a three-phase methodology. In Phase I, a data-driven Random Forest (RF) model was developed to predict the Rolling Resistance Coefficient (RRC). The base dataset was obtained from field measurements conducted on Dutch highways, where Mean Profile Depth (MPD) was measured using a laser profiling system and RRC was measured using a trailer-based system at controlled conditions (80 km/h, 2.1 bar tire pressure). Due to practical and financial limitations in collecting large-scale experimental data under varying operating conditions, the dataset was expanded using validated relationships from literature to simulate variations in speed, tire temperature, and tire pressure. The final dataset used for model training consisted of 111,180 records with input features including International Roughness Index (IRI), speed, tire pressure, and tire temperature. IRI values were derived from MPD and RRC using the VTI model, enabling compatibility with smartphone-based sensing.<br>In Phase II, a smartphone-based Android application was developed to collect real-time data and deploy the trained RF model. Data acquisition was performed using embedded smartphone sensors (accelerometer, GPS, magnetometer) and an external Tire Pressure Monitoring System (TPMS). The accelerometer data were processed using calibration, projection, filtering (2nd order Butterworth band-pass), and double integration techniques to estimate IRI at 100 m intervals. GPS data were used to compute speed, distance, and road grade, with grade correction applied to improve vertical acceleration accuracy. Tire pressure and temperature were recorded via Bluetooth at 5-second intervals and synchronized with IRI measurements. The RF model was converted into ONNX format for real-time inference within the mobile application. All processed and predicted outputs (IRI, RRC, speed, location, and time) were stored in an SQLite database.<br>In Phase III, a Physics-Informed Neural Network (PINN) was developed to estimate tractive force-related power losses. The model integrates physical laws of vehicle motion with data-driven learning. The governing equations include aerodynamic drag, rolling resistance, and inertial forces. The PINN was trained using a combination of smartphone-derived RRC data and literature-based datasets (e.g., NEDC driving cycle data). The loss function combines data-driven error and physics-based constraints to ensure both predictive accuracy and physical consistency. The model outputs total tractive power loss and isolates the contribution of rolling resistance under real driving conditions.<br>Validation was performed using field experiments on Dutch highways with a passenger vehicle (2018 Seat Ibiza). The smartphone-based IRI measurements were benchmarked against the Roadroid application, showing comparable accuracy. The RF model achieved strong predictive performance (Adjusted R² = 0.83), while the PINN model demonstrated high accuracy in estimating power loss (Adjusted R² = 0.92). Results confirmed that rolling resistance contributes approximately one-third of total tractive power losses under steady driving conditions.<br>No human subjects, personal data, or sensitive information were involved in this study. Data collection was conducted on public road infrastructure under standard driving conditions without intervention or risk. Therefore, formal ethical approval was not required. However, the dataset is associated with collaborative research conducted under the Knowledge-based Pavement Engineering (KPE) program involving Delft University of Technology, TNO, and Rijkswaterstaat. Any use of the dataset must comply with institutional agreements and applicable data governance policies related to infrastructure research.