Machine learning-driven fiber steering of variable-stiffness piezoelectric composites for vibration energy harvesting

Autonomous wireless sensing for structural-health monitoring and Internet-of-Things applications increasingly demands battery-free power sources, motivating the conversion of ambient mechanical vibration into usable electrical energy. This study introduces an interpretable machine-learning framework for designing variable-stiffness piezoelectric composite vibration energy harvesters, in which curvilinear (tow-steered) fibers provide a continuous means of tailoring both the resonant frequency and in conjunction with the electrode layout, the harvested power. A coupled electromechanical first-order shear-deformation finite-element model, using the MITC4 formulation to suppress shear locking, is developed and verified against analytical plate solutions and an established distributed-parameter harvester benchmark. A 1200-sample Latin-hypercube dataset spanning the fiber path, plate geometry and electrical load is used to train four surrogates Gaussian process regression, random forest, gradient boosting and a multilayer perceptron for the fundamental frequency, peak power and volumetric power density. Gaussian process regression is the most accurate (R 2 ≈ 0.998). A SHAP analysis, cross-validated by permutation importance, identifies the governing variables, and a surrogate-driven Pareto optimization balances power density against resonance matching. Through a fully controlled comparison at fixed geometry, mass and resonance, fiber steering is shown to provide essentially no benefit with a full-coverage electrode but to increase peak harvested power by up to about 10% once the electrode is localized near the clamped root, by concentrating the modal strain where the charge is collected; all designs are revalidated by direct finite-element analysis to within 3%. Fiber steering and electrode geometry must therefore be co-designed—an interpretable guideline for variable-stiffness piezoelectric-based vibration energy harvesting.