Intelligent Hybrid Neuro-Fuzzy Control System for Industrial Bread Baking Ovens

Abstract

Intelligent control systems can stabilise operational processes and support the development of digital manufacturing in the food industry. This study presents and examines a digital twin of an industrial bread oven with a hybrid neuro-fuzzy controller for temperature and humidity to bake several loaves of bread. Its architecture comprises MLP-Control (4-32-16-2), a discrete Fuzzy-PD corrector and safety rules. The Mamdani controller with 27 rules operates as an independent reference and provides labels for behaviour cloning. MLP-Control accepted those labels in 98.73% of cases for the dryer, 99.10% for the humidifier, and 98.01% for both combined. The Simulation-in-the-Loop evaluation was conducted on 100 pairwise scenarios (seed 42) with a 39 min cycle, 5 s steps and phases at 0–3, 3–6 and 6–39 min. Hybrid + Constraint met the stringent Phase 3 sustainability criterion in 100 out of 100 scenarios, compared with 95 out of 100 for PI. However, PI achieved the lowest final MSE, the shortest mean settling time and the lowest control effort; therefore, it is not claimed that Hybrid + Constraint is superior in all respects. A step-by-step analysis showed that the common fraction within the band was zero in Stages 1–2, indicating the dynamics of the SIL model and its boundary interpretation. In the fair test, MLP-Quality achieved MAE = 0.162318, RMSE = 0.203081 and R2 = 0.686535 on the test set, which was held out with 202,000 samples. Results from the 2025 industry benchmark report are also taken into account and do not indicate how this Hybrid + Constraint configuration performs in a production furnace.