Internet-of-things sensing, storage and calculation integrated 3D intelligent sensing cooking system

Abstract

The invention relates to the technical field of intelligent sensing cooking, in particular to an Internet of Things sensing, storage and calculation integrated 3D intelligent sensing cooking system. The system obtains three-dimensional point cloud data of food materials in real time through a 3D sensing device, solves the shrinkage rate and the form coefficient, and carries out millisecond-level dynamic fine adjustment on the pose and the heating power of an execution mechanism in combination with a physicochemical evolution model. By using a cloud and mist collaborative architecture, the system realizes experience sharing and resource mutual exclusion scheduling among devices in the same area through a real-time cooking feature cache library. And the cloud module realizes continuous evolution and differential issuing of a global model through feature clustering and gradient updating of incremental data. The problems that a traditional cooking robot is single in sensing dimension and poor in multi-machine collaborative conflict and model self-adaption are effectively solved, and the accuracy of automatic cooking and the group intelligence level are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent sensing cooking technology, and in particular to an integrated 3D intelligent sensing cooking system that combines sensing, storage, and computing in the Internet of Things. Background Technology

[0002] With the rapid development of industrialized catering and smart kitchen technology, automated cooking robots have become an important means to solve the problems of catering standardization and reduce labor costs. Current Internet of Things (IoT) cooking systems are gradually evolving from simple single-machine timed operations to multi-machine collaboration and intelligent closed-loop control.

[0003] However, existing cooking systems still have significant shortcomings in practical applications: First, their perception dimensions are limited and lack depth. Traditional solutions often use infrared temperature measurement or 2D vision, which cannot capture in real time the key physicochemical changes of ingredients during high-temperature cooking, such as volume collapse, moisture loss, and morphological disintegration, resulting in low closed-loop control accuracy. Second, their collaborative mechanisms are weak. When multiple devices operate in parallel, downtime often occurs due to spatial overlap or hardware conflicts, and the process experience generated by a single machine is fragmented, making it impossible to achieve knowledge sharing within a local cluster. Finally, their model has poor adaptive capabilities. Due to the lack of an integrated "sensing, storage, and computing" architecture and a cloud-based incremental learning mechanism, the system struggles to cope with the differences in the physicochemical properties of different batches of ingredients, preventing the cooking model from continuously evolving based on real-time feedback. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides an IoT-integrated 3D intelligent sensing cooking system, which aims to perceive the physical and chemical state of ingredients in real time through 3D point cloud, realize multi-machine mutual exclusion scheduling and feature sharing through cloud-fog collaborative architecture, and achieve continuous optimization of the cooking model through incremental learning algorithm.

[0005] This invention provides the following technical solution: an IoT-integrated 3D intelligent sensing cooking system, comprising: The multi-machine collaborative atomization scheduling module is used to receive order information sent by the client and distribute tasks and schedule production according to the current busy / idle status of multiple execution devices; The process matching module is used to retrieve the food processing process sequence based on order information, call the physicochemical evolution model and motion control algorithm, and generate benchmark cooking control parameters. The execution device scheduling module is used to perform mutually exclusive scheduling of environmental resources for multiple execution devices and to establish a real-time cooking feature cache library in a local area. The cooking execution and feedback module is used to send the baseline cooking control parameters to the execution device and use 3D sensing device to perform spatial scanning of the food to be processed, obtain three-dimensional point cloud data to extract the real-time status of the food to be processed. The integrated sensing, storage, and computing process generation module is used to sense real-time visual and thermal field feedback during the cooking process, and to dynamically fine-tune parameters based on the feedback. The adjusted parameters are then stored in the real-time cooking feature cache library to enable data interaction between devices in the same area. The cloud-based enhancement and incremental learning module is used to extract incremental data from the real-time cooking feature cache library, upload it to the cloud server for global learning, and receive global model update packages issued by the cloud. It continuously optimizes the physicochemical evolution model through incremental update algorithms.

[0006] Preferably, the multi-machine collaborative atomization scheduling module, the integrated sensing, storage, and computing process generation module, and the cloud-based enhancement and incremental learning module constitute a cloud-fog collaborative architecture, wherein the integrated sensing, storage, and computing process generation module belongs to the edge node layer, the multi-machine collaborative atomization scheduling module belongs to the fog computing layer, and the cloud-based enhancement and incremental learning module belongs to the cloud layer.

[0007] Preferably, the step of generating baseline cooking control parameters includes: Based on the retrieved food processing sequence, the entire cooking process is decomposed into a baseline timeline model; The physicochemical evolution model is invoked to simulate the expected ripening curve of the food under the reference time axis, and the theoretical shrinkage rate and morphological threshold of the food at each time point are calculated. Match the actuator motion primitives corresponding to the reference time axis to generate the initial pose path of the multi-degree-of-freedom actuator and the initial power curve of the heating unit, which are then issued as reference cooking control parameters.

[0008] Preferably, the step of mutually exclusive scheduling of environmental resources for multiple execution devices includes: Based on the current busy / idle status and location topology of multiple execution devices, the order information is deconstructed into discrete processing time slot sequences and distributed to the corresponding execution devices; Mutual exclusion signals are detected for material feeding points and overlapping areas in the execution space generated by multiple execution devices during the execution process to prevent spatial collisions and hardware conflicts.

[0009] Preferably, the step of extracting the real-time status of the food to be processed includes: The collected raw 3D point cloud data is denoised and smoothed, and the ingredients to be processed are separated by point cloud segmentation. Extract the static initial state of the food to be processed, including the food volume, surface area, aspect ratio, and height information; Extract the dynamic physicochemical state of the ingredients to be processed, including the shrinkage rate and morphology coefficient of the ingredients obtained by temporal comparison point cloud pose calculation during the cooking intervention process.

[0010] Preferably, the dynamic parameter fine-tuning step includes: The dynamic physicochemical state of the ingredients is compared in real time with the theoretical expected values ​​in the benchmark cooking control parameters to calculate the offset of the degree of cooking of the ingredients. Based on the fluctuations in the morphological coefficients, the kinematic parameters of the multi-degree-of-freedom actuator are dynamically adjusted to compensate for the need for physical intervention due to changes in the shape of the food. Based on the rate of change of the shrinkage rate of the ingredients, and combined with real-time thermal field feedback, the heating power of the heating unit is adjusted in real time through the process coefficient.

[0011] Preferably, the data interaction steps between devices in the same area include: When other execution devices in the same area initiate orders for the same dish, the latest process correction record for that dish is obtained by searching the real-time cooking feature cache. The executing device uses the retrieved process correction records as the preset initial values ​​of its own cooking control parameters to achieve rapid convergence and consistent control of cooking processes across devices.

[0012] Preferably, the incremental update algorithm continuously optimizes the physicochemical evolution model by including the following steps: Incremental data is extracted from the real-time cooking feature cache library, and combined with the corresponding 3D point cloud features and energy feedback data. Data cleaning and feature clustering are performed in the cloud to select incremental training samples with statistical significance. The selected incremental training samples are input into the basic physicochemical evolution model in the cloud, and the gradient is updated using the loss function to correct the mapping relationship of the theoretical evolution trajectory of food under different working conditions. The updated model weights are converted into a global model update package and sent to the integrated sensing, storage, and computing process generation module of each execution device. The differential update algorithm replaces the local old model parameters, realizing the global synchronous evolution of cooking intelligence.

[0013] Preferably, the system further includes: a remote monitoring and interactive terminal, used to display the working status and location information of each execution device in real time, and to receive user input instructions to realize the basic configuration and task distribution of the execution devices.

[0014] The present invention has the following beneficial effects: 1. By introducing 3D point cloud dynamic physicochemical analysis technology, the system can calculate the shrinkage rate and morphology coefficient of ingredients in real time during the cooking process. This mechanism shifts the cooking control logic from traditional experience-based parameter-driven to real-time physical feature-driven, effectively compensating for deviations caused by individual differences in ingredients and significantly improving the standardization of dishes and the consistency of output quality.

[0015] 2. In this invention, through the synergistic effect of processing time slot serialization and real-time cooking feature caching library, the system achieves dynamic mutual exclusion of multi-device workspaces at the physical level, avoiding the risk of hardware conflicts; at the logical level, it achieves cross-device process experience pre-alignment, reducing the uncertainty in the system operation process and ensuring the efficient and stable operation of large-scale equipment clusters under complex working conditions.

[0016] 3. This invention combines a cloud-based incremental learning mechanism with continuous clustering and model gradient updates of massive amounts of multidimensional sensory data to achieve dynamic correction of the physicochemical evolution model of cooking. This mechanism endows the system with adaptive learning capabilities for different batches of ingredients and different environmental parameters. Through continuous iteration of the global model, it achieves long-term growth and evolution of the level of cooking intelligence.

[0017] 4. This invention adopts a hierarchical architecture of cloud-fog collaboration, distributing the computing load on demand across the edge node layer, fog computing layer, and cloud layer. The integrated sensing, storage, and computing mode ensures millisecond-level response to local control commands, while remote monitoring and interactive terminals provide visualized management of the global status, effectively balancing the real-time requirements of the underlying execution with the high-efficiency requirements of the upper-layer management. Attached Figure Description

[0018] Figure 1 This is a system structure diagram of the IoT-integrated sensing, storage, and computing 3D intelligent sensing cooking system proposed in this invention. Detailed Implementation

[0019] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] In a first embodiment of the present invention, the present invention provides an IoT-integrated sensing, storage, and computing 3D intelligent sensing cooking system, such as... Figure 1 As shown, it includes: The multi-machine collaborative atomization scheduling module is used to receive order information sent by the client and distribute tasks and schedule production according to the current busy / idle status of multiple execution devices; The process matching module is used to retrieve the food processing process sequence based on order information, call the physicochemical evolution model and motion control algorithm, and generate benchmark cooking control parameters. The execution device scheduling module is used to perform mutually exclusive scheduling of environmental resources for multiple execution devices and to establish a real-time cooking feature cache library in a local area. The cooking execution and feedback module is used to send the baseline cooking control parameters to the execution device and use 3D sensing device to perform spatial scanning of the food to be processed, obtain three-dimensional point cloud data to extract the real-time status of the food to be processed. The integrated sensing, storage, and computing process generation module is used to sense real-time visual and thermal field feedback during the cooking process, and to dynamically fine-tune parameters based on the feedback. The adjusted parameters are then stored in the real-time cooking feature cache library to enable data interaction between devices in the same area. The cloud-based enhancement and incremental learning module is used to extract incremental data from the real-time cooking feature cache library, upload it to the cloud server for global learning, and receive global model update packages issued by the cloud. It continuously optimizes the physicochemical evolution model through incremental update algorithms.

[0021] Preferably, the multi-machine collaborative atomization scheduling module, the integrated sensing, storage, and computing process generation module, and the cloud-based enhancement and incremental learning module constitute a cloud-fog collaborative architecture, wherein the integrated sensing, storage, and computing process generation module belongs to the edge node layer, the multi-machine collaborative atomization scheduling module belongs to the fog computing layer, and the cloud-based enhancement and incremental learning module belongs to the cloud layer.

[0022] Specifically, during the cooking execution phase, the integrated sensing, storage, and computing process generation module, belonging to the edge node layer, collects 3D point cloud data of the ingredients at a preset frequency using built-in 3D sensing devices. The multi-machine collaborative fogging scheduling module, belonging to the fog computing layer, is deployed in a regional gateway or local server and is responsible for managing multiple edge nodes within its region. The cloud-based augmentation and incremental learning module, belonging to the cloud layer, is responsible for long-term, computationally intensive model training.

[0023] Through this implementation process, a hierarchical intelligent system with timely local closed-loop control and global collaborative evolution was constructed. Through the cloud and fog framework, the on-demand allocation of computing load and communication bandwidth was realized. While ensuring the real-time and accurate correction of the single-machine cooking process, it also took into account the orderly concurrent scheduling of multi-machine tasks and the long-term self-evolution of the global model.

[0024] Preferably, the step of generating baseline cooking control parameters includes: Based on the retrieved food processing sequence, the entire cooking process is decomposed into a baseline timeline model; The physicochemical evolution model is invoked to simulate the expected ripening curve of the food under the reference time axis, and the theoretical shrinkage rate and morphological threshold of the food at each time point are calculated. Match the actuator motion primitives corresponding to the reference time axis to generate the initial pose path of the multi-degree-of-freedom actuator and the initial power curve of the heating unit, which are then issued as reference cooking control parameters.

[0025] Specifically, the process matching module first retrieves the standard processing sequence of the target dish, such as preheating, adding ingredients, stir-frying, seasoning, and serving, and deconstructs the entire process into a baseline timeline model with seconds as the smallest unit. For example, for the dish "stir-fried shrimp", the system establishes a time sequence with a total length of 180 seconds and presets action switching points at specific timestamps.

[0026] The system invokes a physicochemical evolution model built upon extensive historical experimental data, inputting the food type and initial weight parameters. The model performs a virtual simulation on a baseline timeline, calculating the physical processes of protein denaturation and moisture evaporation during heating, thereby determining the expected ripening curve for each second. Using this curve, the system can predict the theoretical shrinkage rate and morphological change threshold of the food at different time periods, serving as a benchmark for subsequent real-time sensing.

[0027] Based on the requirements of each process node, the system retrieves motion primitives for the actuators from a pre-set library, such as high-frequency stirring, uniform shaking, and quantitative spraying. Each primitive contains the inverse kinematics of a multi-degree-of-freedom actuator, generating an initial pose path that includes joint angles and end-effector coordinates. Simultaneously, combined with energy requirements, it generates a synchronized initial power curve for the heating unit. Finally, the system integrates the time axis, physicochemical thresholds, pose path, and power curve into a baseline cooking control parameter package and sends it to the actuator.

[0028] This implementation process enables a precise mapping of cooking techniques from experience-based descriptions to digital physical parameters, providing a theoretically predictive control benchmark for the execution equipment.

[0029] Preferably, the step of mutually exclusive scheduling of environmental resources for multiple execution devices includes: Based on the current busy / idle status and location topology of multiple execution devices, the order information is deconstructed into discrete processing time slot sequences and distributed to the corresponding execution devices; Mutual exclusion signals are detected for material feeding points and overlapping areas in the execution space generated by multiple execution devices during the execution process to prevent spatial collisions and hardware conflicts.

[0030] Specifically, the system acquires the location topology and busy / idle status of all executing devices within the current area. When multiple concurrent orders are received, the multi-machine collaborative atomization scheduling module breaks down the process of each order into several discrete processing time slot sequences. For example, the cooking task of device A is broken down into "adding ingredients - heating and stir-frying - removing from the pot", and the task is distributed according to the shortest physical movement path of each device to ensure that the tasks are executed in staggered time scales.

[0031] The system establishes a dynamic virtual spatial envelope model for the multi-degree-of-freedom actuators of each execution device. Before the actuators start, the scheduling module pre-calculates the projection of each device's motion path into the physical space and identifies any overlapping areas in the execution space. If a potential spatial conflict is identified, the system will proactively avoid it by adjusting the start time of the task slots.

[0032] For shared physical resources, such as a single material feeding point or a shared power interface, the system introduces a mutual exclusion semaphore mechanism. When device A prepares to enter the feeding area, it must first request access to the resource from the scheduling module. If the semaphore is already occupied by device B, device A will enter a waiting time slot until the semaphore is released. This dynamic monitoring mechanism can prevent physical collisions between multiple robotic arms in a confined working space in real time and eliminate execution anomalies caused by hardware contention.

[0033] This implementation process enables efficient dynamic control of limited physical resources and workspace. By using dual mutual exclusion detection in both spatiotemporal dimensions, it eliminates the risk of collisions and hardware deadlocks in multi-machine concurrent environments at their source.

[0034] Preferably, the step of extracting the real-time status of the food to be processed includes: The collected raw 3D point cloud data is denoised and smoothed, and the ingredients to be processed are separated by point cloud segmentation. Extract the static initial state of the food to be processed, including the food volume, surface area, aspect ratio, and height information; Extract the dynamic physicochemical state of the ingredients to be processed, including the shrinkage rate and morphology coefficient of the ingredients obtained by temporal comparison point cloud pose calculation during the cooking intervention process.

[0035] Specifically, the system acquires the original point cloud inside the cooking cavity using a 3D sensing device. First, a statistical filter is used to remove outlier noise points, and a bilateral filtering algorithm is employed for feature smoothing to preserve the edge information of the food surface. Subsequently, a random sampling consensus algorithm is used to fit and remove background planes such as the bottom of the pot, and Euclidean clustering is used to completely separate the food to be processed from the environmental background, forming independent food point cloud clusters.

[0036] During the cold state phase before cooking begins, the system establishes axis-aligned bounding boxes or minimum bounding boxes for the food point cloud clusters. The initial volume and surface area of ​​the food are extracted by calculating the closed region enclosed by the point cloud using an integral method. Simultaneously, the aspect ratio and height information of the food are calculated by determining the eigenvalues ​​of the point cloud clusters on the three-dimensional coordinate axes. This data serves as a digital identification tag for this batch of food, used to correct the initial physicochemical evolution presets.

[0037] During cooking intervention, the system continuously captures point cloud changes at a preset frequency. It uses an iterative nearest-point algorithm to align the poses of the food point clouds across consecutive frames. The system compares the current frame's point cloud volume with its initial volume over time, calculating the volume decay percentage in real time, i.e., the food shrinkage rate. It monitors the centroid displacement and normal vector distribution of the food point cloud, extracting the degree of bending, curling, or breakage of the food, defining it as a morphological coefficient. For example, when it detects a drastic change in aspect ratio due to heat-induced shrinkage of meat fibers, the system automatically identifies that the food has entered the high-temperature cooking stage.

[0038] This implementation process enables digital perception of ingredients from geometric space to physicochemical state, which not only accurately identifies individual differences in ingredients but also locks in the real-time evolution characteristics of ingredients during the cooking process through dynamic time-series comparison.

[0039] Preferably, the dynamic parameter fine-tuning step includes: The dynamic physicochemical state of the ingredients is compared in real time with the theoretical expected values ​​in the benchmark cooking control parameters to calculate the offset of the degree of cooking of the ingredients. Based on the fluctuations in the morphological coefficients, the kinematic parameters of the multi-degree-of-freedom actuator are dynamically adjusted to compensate for the need for physical intervention due to changes in the shape of the food. Based on the rate of change of the shrinkage rate of the ingredients, and combined with real-time thermal field feedback, the heating power of the heating unit is adjusted in real time through the process coefficient.

[0040] The system compares the real-time shrinkage rate, morphology coefficient, and other dynamic physicochemical states of the ingredients obtained through 3D point cloud calculations with the baseline physicochemical evolution trajectory generated by the process matching module. Using a preset ripening degree evaluation function, it calculates the degree of ripening deviation between the current state and the standard state. For example, if the measured shrinkage rate is 5% lower than the theoretical value at the current moment, the system determines that the current heat transfer efficiency is lower than expected, and the ingredients are undercooked.

[0041] The system monitors fluctuations in the morphology coefficient in real time. When it detects that the food has curled violently, clumped, or reduced in height due to heat, the original stir-frying trajectory may not be able to effectively reach the center of the food. At this time, the system dynamically adjusts the kinematic parameters of the multi-degree-of-freedom actuator, including correcting the flip angle of the robotic arm end effector, increasing the swing amplitude, or adjusting the stirring speed, to compensate for the blind spots caused by the changes in the physical morphology of the food and ensure uniform physical contact.

[0042] By combining the rate of change of food shrinkage with real-time thermal field feedback collected by sensors inside the pot, a fuzzy control algorithm is invoked. The output power of the heating unit is adjusted in real-time by modifying the process coefficients. If the rate of change of shrinkage is too rapid and the center temperature of the thermal field is too high, the system will automatically reduce the power output or switch to pulse heating mode to prevent the food surface from scorching.

[0043] This implementation process enables a closed-loop transition in the cooking process from timed execution to on-demand intervention, allowing for differentiated compensation based on the varying physicochemical reactions of different ingredients, thus ensuring the effectiveness of physical intervention under complex morphological changes.

[0044] Preferably, the data interaction steps between devices in the same area include: When other execution devices in the same area initiate orders for the same dish, the latest process correction record for that dish is obtained by searching the real-time cooking feature cache. The executing device uses the retrieved process correction records as the preset initial values ​​of its own cooking control parameters to achieve rapid convergence and consistent control of cooking processes across devices.

[0045] Specifically, when a certain execution device in the area is processing a specific dish, its integrated sensing, storage, and computing process generation module fine-tunes parameters based on real-time perceived physicochemical deviations. At the end of cooking or at a specific process node, the system uploads the optimized parameters after this fine-tuning, along with the current batch characteristics of the ingredients, as a process correction record to the real-time cooking feature cache deployed on the fog computing layer in real time. When another execution device in the same area receives an order for the same dish and starts, the execution device scheduling module first performs a high-speed search in the real-time cooking feature cache using the dish identifier to retrieve the most recent and highest-confidence process correction record in that area.

[0046] After receiving the general baseline control parameters from the cloud, Device B does not execute them directly. Instead, it injects the retrieved process correction records into its local controller through weighted fusion as the preset initial values ​​for this cooking action. This means that Device B does not need to perceive differences in ingredients from scratch, but directly inherits the mature experience of Device A, realizing cross-device cold start optimization of the control algorithm.

[0047] Through this implementation process, local cache libraries enable devices in the same area to have collaborative perception and knowledge reuse capabilities, shortening the device's adaptation cycle to environmental and ingredient fluctuations, and ensuring that multiple execution devices can achieve consistent cooking results when processing similar dishes.

[0048] Preferably, the incremental update algorithm continuously optimizes the physicochemical evolution model by including the following steps: Incremental data is extracted from the real-time cooking feature cache library, and combined with the corresponding 3D point cloud features and energy feedback data. Data cleaning and feature clustering are performed in the cloud to select incremental training samples with statistical significance. The selected incremental training samples are input into the basic physicochemical evolution model in the cloud, and the gradient is updated using the loss function to correct the mapping relationship of the theoretical evolution trajectory of food under different working conditions. The updated model weights are converted into a global model update package and sent to the integrated sensing, storage, and computing process generation module of each execution device. The differential update algorithm replaces the local old model parameters, realizing the global synchronous evolution of cooking intelligence.

[0049] Specifically, the cloud server periodically extracts raw incremental data from the real-time cooking feature cache library of fog nodes in various regions. This data includes 3D point cloud temporal features, actuator correction values, and energy feedback data from tens of thousands of cooking processes in different geographical locations and time periods. The system first eliminates abnormal samples caused by hardware failures or human intervention through outlier detection. Then, it uses K-means++ or DBSCAN clustering algorithms to classify the performance of specific ingredients under specific working conditions, thereby selecting an incremental training sample set with high confidence and statistical significance, eliminating the random interference of single-machine experience.

[0050] The system inputs the selected high-value samples into the basic physicochemical evolution model. By constructing a loss function that includes state predictions and actual perceived fine-tuning values, backpropagation training is performed using stochastic gradient descent or the Adam optimizer. This process aims to recalibrate the mathematical descriptions of heat conduction, volume shrinkage, and moisture evaporation in the model, thereby correcting the mapping relationship of the theoretical evolution trajectory of food under different environmental humidity and loads, making the cloud model closer to the real physical world.

[0051] After model training is complete, the cloud does not directly distribute the massive full model file. Instead, it calculates the weight residual between the old and new model weights to generate a lightweight global model update package. This update package is sent to the sensor-memory-computing integrated process generation module of each execution device via the IoT communication link, and uses a differential update algorithm to seamlessly replace the old parameters stored locally during the device's idle period.

[0052] This implementation process established a complete data loop, enabling the cooking robot to gain experience and improving the system's adaptability to various types of ingredients. Based on differential update technology, it ensured the system's efficient operation and maintenance and continuous intelligent upgrades under a massive equipment scale.

[0053] Preferably, the system further includes: a remote monitoring and interactive terminal, used to display the working status and location information of each execution device in real time, and to receive user input instructions to realize the basic configuration and task distribution of the execution devices.

[0054] Specifically, the interactive terminal maintains a persistent connection with the cloud and fog layer via IoT standard protocols. The terminal's homepage retrieves real-time telemetry data from each executing device, displaying the working status, geographical coordinates, and progress bar of the current cooking task for each device through a graphical interface. When a device malfunctions, the terminal provides millisecond-level alerts via push notifications.

[0055] Users interact with the terminal through its management page. When a user sends a task to the terminal, the terminal encapsulates the selected dish ID, target device ID, and desired execution time into an encrypted control command packet. This packet, after cloud authentication, is sent to the corresponding multi-machine collaborative atomization scheduling module and enters the production sequence. Furthermore, users can perform basic configuration settings for the executing device, including adjusting the device's local thermal field preferences, updating the device's physical placement coordinates, or confirming remote OTA upgrades.

[0056] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An IoT-integrated 3D intelligent sensing cooking system, characterized in that: include: The multi-machine collaborative atomization scheduling module is used to receive order information sent by the client and distribute tasks and schedule production according to the current busy / idle status of multiple execution devices; The process matching module is used to retrieve the food processing process sequence based on order information, call the physicochemical evolution model and motion control algorithm, and generate benchmark cooking control parameters. The execution device scheduling module is used to perform mutually exclusive scheduling of environmental resources for multiple execution devices and to establish a real-time cooking feature cache library in a local area. The cooking execution and feedback module is used to send the baseline cooking control parameters to the execution device and use 3D sensing device to perform spatial scanning of the food to be processed, obtain three-dimensional point cloud data to extract the real-time status of the food to be processed. The integrated sensing, storage, and computing process generation module is used to sense real-time visual and thermal field feedback during the cooking process, and to dynamically fine-tune parameters based on the feedback. The adjusted parameters are then stored in the real-time cooking feature cache library to enable data interaction between devices in the same area. The cloud-based enhancement and incremental learning module is used to extract incremental data from the real-time cooking feature cache library, upload it to the cloud server for global learning, and receive global model update packages issued by the cloud. It continuously optimizes the physicochemical evolution model through incremental update algorithms.

2. The IoT-integrated sensing, storage, and computing 3D intelligent sensing cooking system according to claim 1, characterized in that, The multi-machine collaborative atomization scheduling module, the integrated sensing, storage, and computing process generation module, and the cloud-based enhancement and incremental learning module constitute a cloud-fog collaborative architecture. The integrated sensing, storage, and computing process generation module belongs to the edge node layer, the multi-machine collaborative atomization scheduling module belongs to the fog computing layer, and the cloud-based enhancement and incremental learning module belongs to the cloud layer.

3. The IoT-integrated sensing, storage, and computing 3D intelligent sensing cooking system according to claim 1, characterized in that, The steps for generating baseline cooking control parameters include: Based on the retrieved food processing sequence, the entire cooking process is decomposed into a baseline timeline model; The physicochemical evolution model is invoked to simulate the expected ripening curve of the food under the reference time axis, and the theoretical shrinkage rate and morphological threshold of the food at each time point are calculated. Match the actuator motion primitives corresponding to the reference time axis to generate the initial pose path of the multi-degree-of-freedom actuator and the initial power curve of the heating unit, which are then issued as reference cooking control parameters.

4. The IoT-integrated sensing, storage, and computing 3D intelligent sensing cooking system according to claim 1, characterized in that, The steps for mutually exclusive scheduling of environmental resources for multiple execution devices include: Based on the current busy / idle status and location topology of multiple execution devices, the order information is deconstructed into discrete processing time slot sequences and distributed to the corresponding execution devices; Mutual exclusion signals are detected for material feeding points and overlapping areas in the execution space generated by multiple execution devices during the execution process to prevent spatial collisions and hardware conflicts.

5. The IoT-integrated sensing, storage, and computing 3D intelligent sensing cooking system according to claim 1, characterized in that, The steps for extracting the real-time status of the ingredients to be processed include: The collected raw 3D point cloud data is denoised and smoothed, and the ingredients to be processed are separated by point cloud segmentation. Extract the static initial state of the food to be processed, including the food volume, surface area, aspect ratio, and height information; Extract the dynamic physicochemical state of the ingredients to be processed, including the shrinkage rate and morphology coefficient of the ingredients obtained by temporal comparison point cloud pose calculation during the cooking intervention process.

6. The IoT-integrated sensing, storage, and computing 3D intelligent sensing cooking system according to claim 1, characterized in that, The steps for fine-tuning dynamic parameters include: The dynamic physicochemical state of the ingredients is compared in real time with the theoretical expected values ​​in the benchmark cooking control parameters to calculate the offset of the degree of cooking of the ingredients. Based on the fluctuations in the morphological coefficients, the kinematic parameters of the multi-degree-of-freedom actuators are dynamically adjusted to compensate for the need for physical intervention due to changes in the shape of the food. Based on the rate of change of the shrinkage rate of the ingredients, and combined with real-time thermal field feedback, the heating power of the heating unit is adjusted in real time through the process coefficient.

7. The IoT-integrated sensing, storage, and computing 3D intelligent sensing cooking system according to claim 1, characterized in that, The data interaction steps between devices in the same area include: When other execution devices in the same area initiate orders for the same dish, the latest process correction record for that dish is obtained by searching the real-time cooking feature cache. The executing device uses the retrieved process correction records as the preset initial values ​​of its own cooking control parameters to achieve rapid convergence and consistent control of cooking processes across devices.

8. The IoT-integrated sensing, storage, and computing 3D intelligent sensing cooking system according to claim 1, characterized in that, The incremental update algorithm continuously optimizes the physicochemical evolution model through the following steps: Incremental data is extracted from the real-time cooking feature cache library, and combined with the corresponding 3D point cloud features and energy feedback data. Data cleaning and feature clustering are performed in the cloud to select incremental training samples with statistical significance. The selected incremental training samples are input into the basic physicochemical evolution model in the cloud, and the gradient is updated using the loss function to correct the mapping relationship of the theoretical evolution trajectory of food under different working conditions. The updated model weights are converted into a global model update package and sent to the integrated sensing, storage, and computing process generation module of each execution device. The differential update algorithm replaces the local old model parameters, realizing the global synchronous evolution of cooking intelligence.

9. The IoT-integrated sensing, storage, and computing 3D intelligent sensing cooking system according to claim 1, characterized in that, The system also includes a remote monitoring and interactive terminal, which is used to display the working status and location information of each execution device in real time, and to receive user input instructions to realize the basic configuration and task distribution of the execution devices.

Images