SINDy + ANN for Magnetoelectric Sensors

Thesis SINDy + ANN Magnetoelectric sensors (Duffing Oscillator) R² SINDy: 0.986 (deriv) R² SINDy: 0.991 (signal) DNNS: 99.85–100%

Designed the excitation and acquisition pipeline (MATLAB + audio interface) for a prototype implantable magnetoelectric sensor, setting excitation frequency and boundary conditions to respect brain-field safety constraints. Applied SINDy to discover the governing equations directly from acquired signals — recovering Duffing-type dynamics with R² ≈ 0.99, yielding a model that predicts both the signal and its derivatives. ANN models (LSTM, MLP) were built in parallel as accuracy benchmarks; SINDy matched or exceeded them while remaining fully interpretable. Validated via energy consistency, stiffness (force-deflection), and damping (phase portrait) analyses.

GitHub

Stiffness force — **SINDy** capturing the EM-sensor’s stiffness force, i.e., non-dimensional force-deflection characteristics.

Damping force — **SINDy** capturing the EM-sensor’s damping behavior (phase portraits, i.e., non-dimensional deflection (y1, U) vs. its derivative (y2, Udot)).