AI Cake Texture

Founded by VLAIO, this project aims to gain increased insights into the relationship between the composition, the production process, and the quality of baked cake products, and generate insights beyond the expert’s knowledge by using a novel modelling approach for cake. These increased insights, enabled by new AI-based tools, will allow Puratos to stay ahead of the competition when it comes to (i) formulating new and innovative cake mixes, (ii) solving quality issues at producers of patisserie products (Puratos customers), and (iii) facing sourcing issues.

Objective

The general objective of this project is to increase insights into cake properties, and how to steer them, beyond the expert’s current knowledge by using a novel modelling approach for cake. The Smart baking model is an interpretable symbolic regression model that links cake recipes with cake characteristics. By modelling the complex relationship between ingredients, process parameters, and cake properties in a bidirectional way, the resulting model will be used to predict textures and structures based on recipes (forward inference) and derive recipe optimizations for cake improvements based on textural/structural information (backward inference).

 

Partners

Puratos

AI Lab, VUB

 

Impact

  • Efficient new product development: Smarter cake formulation and finished goods creation. The formalization of the food scientists’ knowledge in an interpretable AI model can be reused for training, leading to fewer unknowns, and manageable variability. Concretely, new staff can be trained more effectively, and Puratos’ expert staff can widen its scope through AI assistance during new recipe development. Consequently, new cake recipes that meet evolving consumer expectations will be developed more efficiently, reducing the number of baking trials by up to 50% (less trial & error, less food/ingredients waste).

  • Creative innovation & differentiation: New combinations of raw materials and ingredients (that the ‘human’ food scientist would not think of, but could be generated by the AI model). By formalizing the creation process of (new) cake recipes and consequently employing the AI model, food experts will improve their insight into the multidimensional and multifactorial food design problem beyond what the human brain is capable of (i.e. many ingredient and process combinations).

  • More agile problem solving: Bad outcomes of cake recipe innovation trials can be mitigated more effectively. In case of supply chain issues or other reasons that require the replacement of ingredients, alternative ingredients and combinations thereof to maintain the cake properties, will be identified more efficiently, and Puratos will be able to respond faster to customer inquiries.

  • Cost optimization & reduction: because of higher efficiency in innovation: Reduced raw materials and resources, proposing alternative ingredients while maintaining the cake characteristics.

 

Main Topics

We will look into the following main topics:

  • Domain knowledge formalization
  • Knowledge discovery
  • Image analysis
  • Symbolic regression
  • Hybrid learning systems, hybrid symbolic-subsymbolic representations
  • Forward prediction and backward inference
  • Smart baking

 

Project Info

Start  15/10/2024

End    15/10/2027

Funding VLAIO

Members Prof. Dr. Johan Loeckx, Dr. Leticia Arco García