A machine learning-guided design and manufacturing of wearable nanofibrous acoustic energy harvesters

Nanofibrous acoustic energy harvesters (NAEHs) have emerged as promising wearable platforms for efficient noise-to-electricity conversion in distributed power energy systems and wearable sound amplifiers for assistive listening devices. However, their real-life efficacy is hampered by low power output, particularly in the low-frequency range (< 1 kHz). This study introduces a novel approach to enhance the performance of NAEHs by applying machine learning (ML) techniques to guide the synthesis of electrospun polyvinylidene fluoride (PVDF)/polyurethane (PU) nanofibers, optimizing their application in wearable NAEHs. We use a feed-forward neural network along with solving an optimization problem to find the optimal input values of the electrospinning (applied voltage, nozzle-collector distance, electrospinning time, and drum rotation speed) to generate maximum output performance (acoustic-to-electricity conversion efficiency). We first prepared a dataset to train the network to predict the output power given the input variables with high accuracy. Upon introducing the neural network, we fix the network and then solve an optimization problem using a genetic algorithm to search for the input values that lead to the maximum energy harvesting efficiency. Our ML-guided wearable PVDF/PU NAEH platform can deliver a maximal acoustoelectric power density output of 829 µW/cm3 within the surrounding noise levels. In addition, our system can function stably in a broad frequency (0.1–2 kHz) with a high energy conversion efficiency of 66%. Sound recognition analysis reveals a robust correlation exceeding 0.85 among lexically akin terms with varying sound intensities, contrasting with a diminished correlation below 0.27 for words with disparate semantic connotations. Overall, this work provides a previously unexplored route to utilize ML in advancing wearable NAEHs with excellent practicability.