The project “Energy‑efficient and resource‑conserving coffee roasting” aimed to provide a precise, holistic monitoring of the roasting process by combining colourimetric and acoustic sensing. The goal was to enable fine‑tuned process control, reduce energy consumption, and improve product quality by ensuring that each batch reaches the desired roast level with minimal waste.

Colourimetric monitoring was implemented through a camera system mounted on the standard viewing window of commercial roasters. A custom illumination rig and image‑processing algorithms were developed to extract the CIE L*a*b* colour coordinates of the bean surface in real time. The system can track the surface colour change during roasting and, by correlating it with the internal state, provide an inline estimate of roast progress. The camera‑based method was tested on a range of drum, fluid‑bed, and rotary roasters. It proved robust against temperature variations in the viewing area, which were mitigated by a small cooling unit. The colour data were used to calibrate the acoustic model and to validate the internal state predictions.

Acoustic monitoring relied on piezoelectric microphones placed on the sampling probe of each roaster. The microphones recorded the impact sound produced when a bean is dropped into the chamber. Fast‑Fourier‑Transform analysis of the recorded waveforms revealed characteristic frequency patterns that change with bean moisture, density, and internal structure. These acoustic signatures were fed into machine‑learning models—Artificial Neural Networks, K‑Nearest‑Neighbour, and Random Forest classifiers—to predict the internal roast state. The Random Forest model achieved the highest accuracy, exceeding 90 % in distinguishing between key roast stages (light, medium, dark). The models were trained on a dataset of 1,200 beans from two species (Peruvian Arabica and Indian Robusta) and validated on independent roasters. The acoustic method was shown to be insensitive to external noise and could be integrated into existing roaster designs with minimal hardware modifications.

Combining the two sensing modalities allowed the system to cross‑validate surface and internal measurements, reducing uncertainty in the roast‑level estimate. The integrated monitoring platform was deployed on a commercial production line, where it enabled a 15 % reduction in energy consumption by shortening the roasting time for each batch without compromising flavour. Additionally, the precise roast‑level control lowered the need for post‑roast blending, thereby improving consistency and reducing waste.

The collaboration brought together PROBAT SE, a leading manufacturer of coffee roasting equipment, and Hochschule Geisenheim University’s Institute for Food Safety and Beverage Technology. PROBAT supplied the roasters, provided engineering support for hardware integration, and facilitated industrial testing. The university conducted the experimental work, developed the image‑processing and acoustic‑analysis algorithms, and performed the machine‑learning modelling. The project was funded by the German Federal Ministry of Education and Research under a three‑year programme (2019‑2022). The partnership ensured that the developed methods were both scientifically rigorous and immediately applicable to commercial coffee production.