Sustainable and Scalable Development of Highly Piezoelectric Thick PVDF Sheets for Hydrophone Applications Using Machine Learning

Abstract The demand for flexible polyvinylidene fluoride (PVDF) sensors in underwater and wearable technologies is increasing due to their ability to maintain stable performance by adapting complex geometries and withstanding mechanical deformation. This study reports the use of a predictive model based on Machine Learning (ML) to develop highly piezoelectric thick PVDF sheets for hydrophone applications. A sustainable and scalable three-stage process comprising extrusion, controlled stretching, and high-voltage contact poling was used for preparing thin piezoelectric PVDF sheets. To identify the structure correlation with the stretching process, these were characterized using Fourier-transform infrared (FTIR) spectroscopy, X-ray diffraction (XRD), scanning electron microscopy (SEM), and quantification of the β-phase was performed using FTIR spectroscopy. In this study, five different machine learning models, viz., Gaussian process regression (GPR), support vector regression (SVR), random forest regression (RFR), artificial neural network (ANN), and genetic algorithm-optimized artificial neural network (GA-ANN) were developed using the outcome from the FTIR analysis of the stretched PVDF sheets. These models were used to predict the optimal stretching conditions to achieve the optimum β-phase content for thick PVDF sheets, which are required for specific applications. The statistical error analysis showed that the GA-ANN model had a higher generalization capability than the other models. The predicted conditions were experimentally validated and employed to develop thick PVDF sheets with high β-phase content. The highly piezoelectric form of the stretched PVDF sheet was obtained by high-voltage contact poling, with a high voltage coefficient (g33) of −300 ± 10 mVm/N. Finally, a prototype PVDF hydrophone was fabricated using the sheets developed with GA-ANN predicted conditions, and its underwater acoustic receiving sensitivity was tested. The hydrophone exhibited a flat frequency response of −204 ± 0.8 dB ref V/μPa in the frequency range of 1–10 kHz. This work demonstrates a promising method for integrating GA-ANN modeling with scalable PVDF processing for tailoring the piezoelectric properties and fabricating flexible underwater acoustic sensing systems.