The project, funded under the German Federal Ministry of Education and Research (FKZ 03EE5038) and carried out from 1 June 2020 to 30 June 2023, aimed to improve the operation of steam generators in waste‑incineration plants through digital methods. The research was conducted by a consortium of the Leibniz University Hannover (Institute of Process Engineering, IKW), OFFIS – German Research Center for Environmental, Safety and Energy Technology, and EEW Energy from Waste GmbH, which supplied operational data and facilitated field experiments at a reference plant.
The technical work focused on characterising the highly variable waste feed that drives the steam generator. In the reference plant, two combustion lines feed waste from a bunker into a feed system that delivers the material to a four‑zone grate. The waste passes through a chute, is dosed by a hammer, and slides onto the grate. Because the waste composition is unknown before combustion, the project measured residence times of the material in the feed system using tracer techniques. Visual tracers and a radioisotope tracer were deployed, and the resulting residence‑time distributions were correlated with key operating parameters such as feed rate, grate temperature and pressure drop. The analysis revealed that residence time is a strong predictor of subsequent grate fouling and emission behaviour.
Building on the residence‑time data, the team developed a machine‑learning framework to estimate the heating value of the waste at the moment of feed. Video recordings of the waste being loaded onto the chute were processed with image‑recognition algorithms that extracted colour and texture features. Because the dataset was small and unevenly distributed, a Random‑Forest classifier was chosen and achieved the highest predictive accuracy among the tested algorithms. The model was used to classify waste into heating‑value categories, enabling the plant operators to identify critical batches that could lead to emission spikes. In addition, the Random‑Forest model was trained on meteorological data to predict heating values when video data were unavailable, further extending its applicability.
The project also produced a comprehensive information tool that integrates the developed methods with existing plant data. The tool displays real‑time measurements, predicts pollution levels based on pressure‑loss calculations, and overlays processed video material. A demonstration at the reference plant showed that the tool could present a clear, actionable overview of plant status, thereby supporting operators in making informed decisions.
Collaboration was structured around eight work packages. WP 1 established continuous data exchange and operational support from EEW. WP 2 produced an initial concept for boiler monitoring, including a catalogue of operating states and control actions. WP 3 and WP 4 carried out the residence‑time measurements and correlation analysis. WP 5 focused on machine‑learning prediction of heating values, while WP 6 developed the image‑recognition pipeline. WP 7 integrated all results into the information tool, and WP 8 demonstrated the complete system in the reference plant. Regular weekly meetings among the core team (IKW, OFFIS, EEW) ensured alignment of objectives and timely progress.
Overall, the project delivered a validated methodology for measuring waste residence times, a machine‑learning model for heating‑value estimation, and an integrated information system that together enable more reliable and efficient operation of steam generators in waste‑incineration plants.