Bayesian Optimization for Customizable Hierarchical Kirigami Piezo-Transmittance Strain Sensor Designs

The significance of IoT systems incorporating green and sustainable energy sources has grown exponentially. Simultaneously, the demand for sensing systems integrated with low-powered and self-powered capabilities has emerged. Piezo-transmittance strain sensors, which operate on the optical transmittance change induced by mechanical deformation of the functional film, have demonstrated remarkable potential for in self-powered soft sensing applications integrated with solar cells. These sensors are notable for their reliability, rapid response, and long-term stability. While soft sensing systems offer advantages because of their compliance and stretchability, their suitability for vertical sensing platforms varies depending on the mechanical properties of the target objects. Consequently, there is a growing need to develop customized sensor characteristic optimization procedures within this field. The utilization of a kirigami structure in piezo-transmittance sensors addresses two benefits. First the repeated unit-cell pattern across the film area ensures sensor uniformity. Additionally, it offers design flexibility, allowing for the accommodation of predicted mechanical deformations within the pre-designed structure. In this research, we optimized a hierarchical kirigami-based structure, focusing on sensor sensitivity within different working ranges, as the functional film for piezo-transmittance strain sensors. First, we utilize the results obtained from finite element simulations as a training dataset to build a neural network surrogate machine learning model. This model accepts geometric variables as input and furnishes data regarding the desired performance metrics, thereby substituting the requirement for more costly physical simulations, facilitating the optimization process. The Bayesian optimization and the cross-entropy method were used to optimize the sensor’s geometric structure, and a comprehensive analysis of the geometric factors was conducted through sensitivity analysis. Through sensitivity analysis, we can identify the major geometric factors that affect the sensor's performance. Additionally, in contrast to auxetic behavior, the sensor's performance is better in a simpler model, specifically at hierarchical level 1. Furthermore, through experimental testing, we can validate the superior performance of the designed kirigami piezo-transmittance strain sensors.