Machine-Learning Mapping of Fabric Knit Parameters for Optimal Biodegradable Poly(lactic Acid) Textile Triboelectric Nanogenerators

Abstract Self-powered wearable systems require energy harvesters that are mechanically compliant, comfortable, and environmentally sustainable. Here, we report a bio-based, biodegradable textile triboelectric nanogenerator fabricated from knitted poly(lactic acid) (PLA) and use machine learning to elucidate how macroscopic knit architecture influences triboelectric performance. Interlock structures were varied in gauge, stitch length, thread number, and fabric area to study combination−performance relationships. Bayesian optimization with a discrete Gaussian process regression model was employed to identify the fabrication parameters (stitch length, gauge, and fabric area) that maximized the voltage output. Gauge 8 showed the lowest response variability and the most stable behaviour during optimization. In subsequent characterization with a differential probe, the selected PLA fabric paired with PTFE film yielded VRMS = 1.11 V at 40 N and 1.5 Hz. PLA formed effective triboelectric pairs with polytetrafluoroethylene (PTFE), which produced the highest output, and styrene-ethylene/butylene-styrene (SEBS), which offered comparable performance and improved textile compatibility. The device was able to power an LED under high-load and high-frequency actuation, maintaining stable performance over 10,000 cycles. These results highlight knit architecture as a key design lever for sustainable, fabric-integrated triboelectric nanogenerators (TENGs) and demonstrate the value of small-data Bayesian optimization in wearable energy harvesting.