Machine‐Learning‐Driven Inverse Design of NURBS‐Parameterized Auxetic Piezoelectric Energy Harvesters for Self‐Powered Wireless Sensing
Year: 2026
Type: article
Abstract: Low‐frequency vibration energy is abundant in infrastructure, transportation, and industrial equipment, but piezoelectric harvesters face a power–stress‐concentration trade‐off: the large strain required for high output can intensify localized peak stress. In this paper, we develop a machine‐learning‐assisted inverse‐design framework for NURBS‐parameterized auxetic piezoelectric energy harvesters. NURBS control variables define a compact and continuous design space for curved re‐entrant lattices, expanding deformation and load‐transfer pathways beyond straight‐wall geometries. A finite‐element dataset generated by Latin hypercube sampling trains a neural‐network surrogate to predict maximum average output power and peak von Mises stress from geometric parameters. Coupling this surrogate with NSGA‐II enables co‐optimization of electrical output and stress concentration. The optimized geometry increases effective biaxial stress/strain in the piezoelectric layer while redistributing localized stress. Under 0.5 g excitation at resonance, the optimized harvester delivers maximum average output power and a normalized power density of . Integrated with a self‐powered LoRa sensing node, the device transmits temperature data over . This work offers a surrogate‐assisted route to high‐power, stress‐aware energy harvesters and energy‐autonomous IoT sensing.
Source: Advanced Science
DOI: 10.1002/advs.78735
Language: English
Subfield: Mechanical Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $6,730
OpenAlex ID: W7236701343
