• Citer

Prediction of cooking-induced food transformations: A survey and outlook towards machine learning

par Ben Jmaa, Ahmed Baha;Hanon, Cécile ;Parfait, Jean-Yves;Debaste, Frédéric
Référence Journal of food engineering, 422, page (113175)
Publication Publié, 2026-05-18

Article révisé par les pairs

• Accès en ligne
• Détails
• Contenu
• Statistiques

Résumé :

Cooking induces coupled physical, chemical, and biological transformations that jointly determine food safety, nutritional value, and sensory quality. Predicting cooking-induced food transformations is therefore a central challenge, receiving growing attention across applications including process optimisation in the food industry, product design, quality control, personalised cooking, and the development of smart appliances and digital twins. Despite extensive progress in food process engineering, robust prediction across foods, geometries, and operating conditions remains difficult due to the multiscale heterogeneity of food matrices, substantial process variability, and experimental data that are often sparse, noisy, or destructive to acquire. To stimulate future research, this survey provides a comprehensive review of modelling approaches for cooking-induced transformations and articulates a structured outlook towards machine learning (ML). We first consolidate the scientific background on food cooking processes. We then introduce a principled taxonomy of conventional models and analyse how their encoded structure can act as inductive bias for data-driven learning. Building on this foundation, we formalise cooking prediction as learning mappings from initial product state and process descriptors to time-resolved or endpoint outcomes, and synthesise data modalities and evaluation protocols required for trustworthy generalisation under distribution shift. Finally, we systematically map existing ML-based predictors, review hybrid and physics-informed integration strategies, and propose a research roadmap towards data-efficient, physically consistent, and uncertainty-aware models that enable robust decision-making and scalable digital twins for cooking processes.