Robot-accelerated development of a colorimetric CO2 sensing array with wide ranges and high sensitivity via multi-target Bayesian optimizations

Abstract

The one-variable-at-a-time method for sensor R&D has received extensive research effort, yet it has reached the local maxima for specific sensing characteristics. To achieve the quasi-global maxima for suitable sensors, we developed the Design–Build–Test–Machine learning (DBTM) method for efficiently developing sensors on demand. In addition, the automation of the preparation and characterization processes frees researchers from labor-intensive work, generates adequate high-quality data, and enables researchers to discover valuable information in high-dimensional space. As a proof-of-concept, we built a high-throughput algorithm-driven autonomous system (HAAS) that supports the DBTM approach for developing a CO2 sensor. With the DBTM approach, we can simultaneously optimize multiple sensor units, each for a specific concentration interval. Therefore, such an array can achieve an extensive range and sound sensitivity. Our work demonstrates the superiority of the DBTM method for multi-target and multi-variable sensor development. In contrast to single target optimization in other research areas, multiple characteristics should be improved for sensors. Our multi-target optimization algorithm optimizes four sensor characteristics. Our sensor array could rapidly detect CO2 concentrations from 400 ppm to 30 % with a root mean square error (RMSE) of 0.27 %. The DBTM method is anticipated to be a new paradigm and accelerator of practical application for sensors after the initial proof of the sensing mechanism.