Compact Fluorescence Sensor with Silicon Photomultiplier and Neural Network Enhancement for Real-Time Total Organic Carbon Monitoring in Water.
Journal:
Environmental research
Published Date:
Jan 17, 2026
Abstract
Real-time monitoring of total organic carbon (TOC) is essential for assessing water quality, optimizing treatment processes, and safeguarding aquatic ecosystems. This study reports the development of a compact LED-SiPM fluorescence sensor, integrated with a machine learning model, for accurate and real-time water quality monitoring. The sensor targets tryptophan-like fluorescence (Peak T) as a surrogate for TOC and employs a silicon photomultiplier (SiPM) detector to ensure high sensitivity and reliability. To mitigate interferences from matrix effects such as turbidity and inner-filter phenomena, an additional UV transmittance detector was incorporated. An artificial neural network (ANN) model was trained and optimized using experimental datasets from wastewater effluent, river water, and drinking water. Compared with traditional linear regression, the ANN achieved superior predictive performance, yielding an overall R2 of 0.9448 across all sample types, with the highest accuracy for effluent samples. For effluent measurements, prediction errors ranged from 0.83% to 11.90%, and reproducibility was within 0.20%, confirming excellent repeatability. Field deployment at a municipal wastewater treatment plant further validated the sensor's performance, demonstrating close agreement with conventional laboratory-based TOC analyses. The integration of Internet of Things functionality and cloud-based analytics enables real-time visualization, remote monitoring, and automated alerts. These capabilities position the proposed sensor as a promising platform for intelligent, data-driven water management in both centralized and decentralized treatment systems.
Authors
Keywords
No keywords available for this article.
