–
Brake pad wear directly impacts vehicle safety, yet most existing sensors only issue warnings when pads are nearly depleted, offering no estimate of remaining useful life. This thesis investigates a simulation-based machine-learning approach for predictive brake wear modeling. A physics-based digital twin of an automotive braking system is used to generate synthetic data under varying operating and fault conditions. A public real-world EV dataset (EVIoT predictive maintenance) serves as an external reference to assess realism. The digital twin integrates a system-level vehicle and brake model with an Finite Element Analysis -derived wear formulation, where wear rate depends on brake pressure, rotor speed, and accumulated braking events. Time-series signals from each braking event are processed as fixed-length sequences and analyzed using a recurrent neural network to estimate brake wear and classify brake health and fault states.
On another front, sensing technologies in unmanned vehicles are rapidly evolving. In a parallel research effort, we conduct a comprehensive survey of recent advances in sensor technologies for unmanned and autonomous systems, examining applications and enabling techniques such as multi-sensor fusion, calibration, and synchronization to identify current capabilities, limitations, and emerging research directions supporting reliable autonomous operation.