Yangguan Chen, Longhan Zhang, Zhehong Ai''\ Yifan Long', Jing Jiang

Research Center for Intelligent Sensing Systems, Zhejiang Lab, Hangzhou, 311121, China;
Hangzhou Institute for Advanced Study, University of Chinese Academy of Sciences,
Hangzhou, Zhejiang, 310024, China 

EXTENDED ABSTRACT: Developing sensor materials involves navigating a complex, high-dimensional variable space. While the widely used one-variable-at-a-time approach has merits, it's limited by only reaching local maxima for specific sensing traits within multi-variable contexts. To optimize sensor material compositions across multiple figures of merit (FoMs), we introduce the Design-Build-Test-MachineLeaming (DBTM) method for targeted and expedited sensor development. For an ammonia colorimetric sensor within the DBTM framework, we create an algorithm-guided autonomous high-throughput platform (AAHP). Utilizing a multi-target algorithm, we concurrently optimize diverse sensing traits. Additionally, we propose a Bayesian algorithm to transform discrete variables into continuous ones for multidimensional exploration. With DBTM, we optimize multiple sensor units in just four iterations within a wide search space of 18 source solutions. Our study showcases optimized sensing units with significant sensitivity and reversibility for ammonia detection across 0.5 ppm to 500 ppm. Leveraging the colorimetric principle, we achieve an ammonia detection limit (LoD) of 50 ppb, verified in thesame batch. Furthermore, for developing an optoelectronic systemcapable of quantifying mixed gases, optimizing the entire sensingarray is essential, a task traditionally challenging. We extend DBTMto colorimetric sensing using a high-throughput autonomous systemthat formulates 96 sensing recipes and evaluates 384 sensing unitsin each iteration. Combining this system with search algorithms, weconstruct a sensor array for mixed binary gas testing, involving threephases: (a) Establishing a pool of sensing units meeting kinetic FoMrequirements (reversibility, response time,sensitivity), respondingindividually to NH3, vapor, and CO, through DBTM iterations; (b)Building a responsivity database for each sensing unit across gases;(c) Optimizing the array using combinatorial methods based on array-relative-error metrics (AREM) and genetic algorithms. Ultimately,a10-unit array with minimal errors is identified, yielding an accuracy of 4.1 ppm for NH3, 0.0098% for CO, and 2.68% for vapor detectionin mixed gases. This underscores the prowess of our DBTM-Searchapproach, harnessing high-throughput data and rapidly exploring therich design space in colorimetric sensor arrays for mixed gas detection.

Keywords: self-driven material research platform; heterogeneousarray optimization; on-demand optimization for materials; gas sensing