Design of Virtual Torque Sensor for Class 8 Heavy-duty Trucks using AI/ML Methods and Testing

Date of Award

6-2025

Degree Name

Doctor of Philosophy

Department

Mechanical and Aerospace Engineering

First Advisor

Muralidhar Ghantasala, Ph.D.

Second Advisor

Richard Meyer, Ph.D.

Third Advisor

Bade Shrestha, Ph.D.

Fourth Advisor

Johnson Asumadu, Ph.D.

Keywords

AI/ML, Class 8, micro-controller, torque, trucks, virtual sensor

Abstract

Accurate engine torque estimation is essential for gear shifting, fuel economy, and engine health monitoring in Class 8 heavy-duty trucks, yet existing torque estimation methods remain largely inaccurate, leading to increased fuel consumption, clutch wear, and limited prognostics. To address this, VTS v1 was developed using the vehicle’s existing flywheel speed sensor to estimate torque from n-th harmonic order linked to flywheel torque. Testing on two Class 8 engines revealed that reciprocating engine masses affected accuracy across the entire engine speed and load conditions. A compensation strategy based on instantaneous phase measurement from engine speed signal, reduced the Mean Absolute Percentage Error (MAPE) from 18% to 8%. Further validation using a Cummins ISX 15 engine model in AMEsim confirmed the impact of reciprocating mass and identified additional factors. To improve accuracy, two AI/ML-based models, a feedforward neural network and an LSTM, were trained on on-road vehicle data, with the LSTM achieving the best results (R² = 0.97, MAPE less than 3% in most gears). This work demonstrates a practical, low-cost approach for accurate, real-time engine torque estimation in Class 8 trucks, paving the way for improved fuel efficiency, and engine health management for commercial fleets.

Access Setting

Dissertation-Abstract Only

Restricted to Campus until

6-1-2035

This document is currently not available here.

Share

COinS