Methodology for building an automated system for operational monitoring of individual biogas production parameters [in Ukrainian]
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
A methodology for building an automated system for real-time monitoring of selected biogas production parameters – temperature regime and visual-statistical features of the substrate (mean pixel value, standard deviation, skewness, kurtosis) - is proposed. The system integrates four functional subsystems: mobile IoT-based environmental monitoring using an ESP32 microcontroller with data transmission via the MQTT protocol, computer vision technologies for rapid biomass quality assessment using the Otsu method with prior application of CLAHE, an ensemble of machine learning algorithms (XGBoost, LSTM, ResNet-50) for biogas yield prediction, and the Harrington generalized desirability function for generating operator recommendations. The mathematical framework of the system is systematized, including formulas for computing image statistical features, a quality assessment method based on normalized distance from reference values, a double exponential desirability function with linear normalization, and regression analysis metrics. The system's responsiveness is ensured by data transmission from the IoT device to the Amazon Web Services cloud infrastructure in under 5 seconds and a complete processing cycle - from image acquisition to operator recommendation – in 3.2 minutes, which is substantially less than the characteristic time constants of the anaerobic digestion process. The architecture follows an open-loop scheme, where the system output is not a direct control action but rather operator recommendations with biomass condition classification across five quality levels. A comparative analysis of five segmentation methods on a set of 26 images confirmed the advantage of the CLAHE + Otsu method with the lowest coefficient of variation of 3.63 %. Feature importance analysis using Random Forest showed that the visual quality score is the most influential predictor with a relative importance of 0.301. The results of 6-month pilot plant testing confirmed the effectiveness of the approach: 98.6 % of scheduled measurements collected, predictive model coefficient of determination R² = 0.91 after a retraining cycle on 100 validated records, and RMSE reduction of 42.6 % – from 0.61 to 0.35 m³/day.
Fields of Study & Subjects
Anaerobic Digestion and Biogas Production
Smart Agriculture and AI
Arduino and IoT Applications
Publication details
- Journal / source
- Proceedings of the NTUU “Igor Sikorsky KPI” Series Chemical engineering ecology and resource saving
- Issue
- 1
- Pages
- 53–61
- Publication date
- 30 March 2026
- Language
- EN