Digital recipe learning system for commercial kitchen environments

Abstract

This disclosure presents a digital recipe learning system for commercial kitchen environments. It includes a processing unit that performs a fusion process on multimodal recipe data to generate a structured timeline; a dispatch unit that dynamically assigns multiple orders based on load status and distributes the structured timeline to workstations; a display unit that outputs preparation instructions; a sensing unit that captures real-time sensing events; and a dynamic progression unit that updates the execution progress based on real-time sensing events. Through these methods, this disclosure converts implicit cooking operations into machine-readable parameters, ensuring consistency in flavor across branches and effectively shortening the training cycle for new chefs.

Claims

1. A digital recipe learning system for use in a commercial kitchen environment, comprising: a processing unit for performing a fusion process on multimodal recipe data to generate a structured timeline; a dispatching unit, signal-connected to the processing unit, for performing dynamic dispatching of multiple orders based on load status, assigning the structured timeline to at least one workstation; a display unit, signal-connected to the dispatching unit, for outputting at least one preparation guide corresponding to the structured timeline in the at least one workstation; a sensing unit, signal-connected to the processing unit, for capturing a real-time sensing event of the at least one workstation; and a dynamic advancement unit, signal-connected to the sensing unit and the processing unit, for updating the execution progress of the structured timeline based on the real-time sensing event.

2. The digital recipe learning system as described in claim 1, wherein, The at least one workstation is one of a human chef workstation, a robotic arm workstation, or a combination thereof. When the at least one workstation is such a combination, the dispatching unit is used to synchronously assign the structured timeline to the human chef workstation and the robotic arm workstation to establish dual-track collaboration.

3. The digital recipe learning system as described in claim 1, wherein, The multimodal recipe data includes a voice input, a visual input, and a text input. The processing unit uses the voice input as a time anchor, the visual input as an event verification, and the text input to perform semantic constraints in order to complete the fusion process.

4. The digital recipe learning system as described in claim 1, wherein, It also includes an edge learning unit that is signal-connected to the dynamic advancement unit to record a deviation log when the execution progress is changed, and to perform incremental training at an edge node based on the deviation log.

5. The digital recipe learning system as described in claim 4, wherein, The dynamic advancement unit is used to calculate a completion score corresponding to the real-time perceived event, and when the completion score meets a threshold for a consecutive predetermined time, it triggers an update of the execution progress, and provides a reversible buffer time to receive an overwrite operation, without including the overwrite operation in the deviation log.

6. The digital recipe learning system as described in claim 4, wherein, The edge learning unit includes a perceptual model with a general model parameter and a low-rank adapter. The edge learning unit freezes the general model parameter during the incremental training and trains the low-rank adapter based on the bias log and a knowledge distillation loss function.

7. The digital recipe learning system as described in claim 1 further includes an overlay unit connected to the processing unit. The overlay unit is used to receive target specification parameters, scale the corresponding main ingredients in the structured timeline according to a linear overlay operation, and scale the corresponding sensitive variables in the structured timeline according to a nonlinear overlay operation.

8. The digital recipe learning system as described in claim 1 further includes a flavor optimization unit, which is signal-connected to the processing unit. The flavor optimization unit is used to receive customer feedback and generate an optimization suggestion parameter. The flavor optimization unit has a flavor protection mechanism, which includes an adjustable whitelist, a maximum adjustment range limit, and a cosine similarity comparison to ensure that the similarity between the flavor vector of the new recipe after adopting the optimization suggestion parameter and the flavor vector of the original recipe is higher than a threshold.

Images