Risk-Priced Radio Resources for In-Line Spectroscopy and Reaction Monitoring in Chemical Plants
Year: 2026
Type: preprint
Abstract: In-line spectroscopy now carries the chemical plant’s quality decisions over shared industrial wireless links, yet the schedulers distributing those links optimize throughput, fairness or delay and are blind to chemistry: an MIR quantification stream survives compression, while a runaway precursor confined to a narrow wavenumber window disappears with it. We present ChemRAN, a chemistry-aware orchestration framework for wireless process analytical technology networks. ChemQoI predicts how an allocation changes a stream’s analytical error, detection sensitivity and measurement-failure probability, and converts them into an expected chemical risk reduction using a sparse spectral sketch and a process cost matrix. SCARF allocates each window by marginal risk reduction under a hard safety floor for hazard streams, and RECAP couples allocation with offloading through a shared adaptation tensor and asynchronous updates conditioned on chemistry descriptors. On the Tennessee Eastman process, Buchwald–Hartwig and Suzuki–Miyaura screens and in-line MIR soft sensing, ChemRAN attains 0.9755 F1, 0.966 yield 𝑅2 and 1.08 g L−1 𝜅-casein RMSECV, cuts the risk-weighted chemical loss by 19% and the hazard miss rate from 2.31% to 0.88%, and retains over 90% of its performance on unseen plants and campaigns.
Source: ChemRxiv
Institution: Zhongnan University of Economics and Law
Language: English
Subfield: Analytical Chemistry
Field: Chemistry
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7223178909