Problem
Research questions and friction points this paper is trying to address.
BLDC motor modeling
high-fidelity dynamics
real-time control
physics-informed learning
motor state estimation
Residual Physics-Informed Neural Networks for High-Fidelity BLDC Motor Modeling
🤖 AI Summary
Accurate dynamical modeling of brushless DC (BLDC) motors is critical for high-performance robotic joint control, yet conventional approaches often incur prohibitive computational costs that hinder real-time deployment. This work proposes a physics-informed neural network (PINN) based on deep residual architectures to directly predict the six-dimensional state variables from input voltages and excitation parameters in continuous time. The model enforces fidelity to the electromechanical-thermal coupled differential equations through a composite physics-data loss. A novel curriculum scheduling strategy is introduced to progressively activate physical constraints during training, effectively mitigating premature convergence. The resulting surrogate model trains in under two minutes on a CPU and achieves inference latencies of merely 0.1–22 microseconds—yielding a 118× speedup over traditional ODE solvers—while preserving high solution fidelity, thereby enabling ultra-low-latency real-time observation and control.
📝 Abstract
Accurate dynamics modeling of Brushless DC (BLDC) motors is fundamental to high-performance robotic joint control. This paper presents a Physics-Informed Neural Network (PINN) with a deep residual (ResNet) backbone that learns a continuous-time surrogate of the full six-state BLDC motor dynamics. Given simulation time, applied three-phase voltages, and excitation parameters as inputs, the network directly predicts all motor state variables -- rotor angle, angular velocity, three-phase currents, and winding temperature -- while simultaneously satisfying the governing electromechanical and thermal ODEs through a composite physics-data loss. A curriculum scheduling strategy gradually activates the physics penalty to prevent premature convergence. Training runs are completed in under two minutes on a standard CPU. Crucially, once trained, PINN inference achieves latencies of 0.1--22, mu s per query, up to 118x faster than conventional ODE solvers, making it suitable for real-time observer and control applications.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Physics-Informed Neural Networks
Residual Network
BLDC Motor Modeling
Real-time Inference
Curriculum Learning
🔎 Similar Papers
No similar papers found.
