Chef machine intelligent recipe recommendation and operation control method

Abstract

The invention relates to the technical field of intelligent cook machine recommendation, and discloses an intelligent cook machine recipe recommendation and operation control method, which comprises a user input module used for receiving user instructions including food material types, cooking targets, diet restrictions and equipment models; the recognition module is used for collecting food material weight and state data; the data processing module adopts a multi-modal fusion algorithm; the recipe recommendation module is based on an improved collaborative filtering algorithm and a knowledge graph; and the output module is used for controlling the instruction generation unit to output a sectional operation instruction set according to the selected recipe. By integrating the image recognition unit and the weight sensor, the type, weight and state data of food materials placed in the chef machine by a user can be collected in real time, and correction processing is carried out in combination with the environment temperature and humidity sensor; therefore, the existing food material inventory of the user can be accurately considered during recipe recommendation, and the problem that the existing food materials of the user cannot be fully utilized during recipe recommendation by an existing system is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent recommendation for a chef machine, and in particular to an intelligent recipe recommendation and operation control method for a chef machine. Background Art

[0002] With the continuous development of science and technology, intelligent technology has gradually penetrated into all areas of our lives, and kitchen appliances are no exception. Traditional chef machines mainly rely on manual operation, and users need to manually set cooking parameters such as temperature, time and stirring speed according to their own experience and recipes. In recent years, with the rapid development of artificial intelligence and Internet of Things technologies, intelligent chef machines have begun to emerge. These chef machines can automatically adjust cooking parameters through built-in intelligent algorithms and sensors to improve cooking accuracy and efficiency.

[0003] After searching, the Chinese patent number CN111965987B discloses a recipe recommendation method, device, storage medium and smart home system, the method comprising: if a recipe recommendation request is monitored, outputting inquiry information for determining the dietary needs of each dining member; receiving feedback information corresponding to the inquiry information; determining the dietary needs of each dining member according to the feedback information; judging whether the recipes matching the existing ingredients in the food library are suitable for the dietary needs of each dining member, and obtaining a judgment result; according to the judgment result, adding descriptive information of the dining members suitable for the recipe and / or descriptive information of the dining members not suitable for the recipe to the recipe, obtaining and outputting a recommended recipe, thereby realizing the recommendation of recipes according to the dietary needs of diners.

[0004] The above recipe recommendation method can recommend recipes according to the needs of diners, but the existing system often fails to fully consider the user's existing food inventory when recommending recipes, resulting in the recommended recipes cannot be directly implemented; there is a lack of linkage between traditional cooking equipment and recipe recommendation systems, and cooking parameters cannot be automatically adjusted according to the user's dietary needs, affecting cooking efficiency and user experience. Based on this, the present invention designs a smart recipe recommendation and operation control method for a chef machine to solve the above problems. Summary of the invention

[0005] The purpose of the present invention is to provide a method for intelligent recipe recommendation and operation control of a chef machine, which solves the problem in the background technology that the user's existing food inventory cannot be considered.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] A method for intelligent recipe recommendation and operation control of a chef machine, comprising:

[0008] A user input module for receiving user instructions including ingredient type, cooking goal, dietary restrictions and equipment model;

[0009] The recognition module includes an image recognition unit and a weight sensor, which is used to collect food weight and status data and correct the processing logic in combination with the ambient temperature and humidity sensor;

[0010] The data processing module uses a multimodal fusion algorithm to map user input and sensor data to a high-dimensional feature space and generate a feature vector V = (v 1 ,v 2 ,v n ), the dimensions include food moisture content, hardness, heat capacity and equipment load factor;

[0011] The recipe recommendation module recommends candidate recipes from the recipe library based on the improved collaborative filtering algorithm and knowledge graph, and calculates the fitness score S = α·user preference matching degree + β·device compatibility + γ·nutritional balance coefficient (α+β+γ=1);

[0012] The output module controls the instruction generation unit to output a segmented operation instruction set according to the selected recipe.

[0013] Preferably, the user input module supports voice recognition and touch input, and integrates an allergen database covering FDA's eight major categories of allergens for real-time filtering.

[0014] From the above technical solution, it can be seen that the user first provides key cooking information through the voice recognition or touch input module, including ingredient type, cooking goals (such as cake, bread, etc.), dietary restrictions (such as gluten-free, low sugar, etc.) and the model of the chef machine used. This information is processed instantly and passed to the data processing module; at the same time, the allergen database integrated in the user input module filters out ingredient options that do not meet the user's health needs in real time, and performs precise matching based on the eight major categories of allergens specified by the FDA.

[0015] Preferably, in the recognition module, the image recognition unit adopts an improved YOLOv5 model, the training data set contains ≥ 500,000 images of food under multiple lighting conditions, and a generative adversarial network is introduced to enhance data diversity.

[0016] Through the above technical solution, it can be seen that the recognition module is immediately started, and the image recognition unit uses the improved YOLOv5 model to quickly and accurately recognize the images of the ingredients placed in the food processor. At the same time, the weight sensor accurately measures the weight of the ingredients, and the ambient temperature and humidity sensor monitors the current cooking environment. These data are used together to correct the processing logic of the ingredient status to be closer to the actual cooking conditions.

[0017] Preferably, in the data processing module, the multimodal fusion algorithm is implemented using the deep learning framework TensorFlow, and the time series data is modeled using a long short-term memory network to capture the dynamic characteristics of food status changing over time. The number of LSTM layers is 3, the number of hidden units is 256, the optimizer is Adam, the learning rate is set to 0.001, and the number of training iterations is 100,000 to ensure the accuracy and robustness of the feature vector V.

[0018] From the above technical solution, it can be seen that the data processing module receives the above-mentioned multi-source data, and uses the multimodal fusion algorithm under the TensorFlow framework, especially the long short-term memory network (LSTM), to perform deep modeling on the time series data to capture the subtle characteristics of the food status changing over time. This process not only takes into account the static properties of the food (such as water content and hardness), but also incorporates dynamic change factors to generate a high-dimensional feature vector V, providing a rich data foundation for subsequent recipe recommendations.

[0019] Preferably, in the recipe recommendation module, the collaborative filtering algorithm adopts a hybrid model, including matrix decomposition based on user behavior with potential factor number k=50 and a graph neural network based on the knowledge graph. The recipe recommendation module also includes a real-time feedback reinforcement learning module with a Q-learning update frequency Δt=10min. In the recipe recommendation module, the knowledge graph is constructed based on the Neo4j graph database, including ingredients, recipes, cooking skills and nutritional information. The graph information is embedded into a low-dimensional vector space through graph embedding technology to enhance the semantic understanding and reasoning ability of the recommendation system.

[0020] Through the above technical solutions, we can know that the recipe recommendation module based on the improved collaborative filtering algorithm and knowledge graph can select the most suitable candidate recipes from the huge recipe library by combining user preferences, device compatibility and nutritional balance coefficient. Among them, the graph neural network enhances the semantic understanding and reasoning ability based on the knowledge graph, making the recommendation more intelligent and personalized.

[0021] Preferably, in the real-time feedback reinforcement learning module, the Q-learning algorithm uses an ε-greedy strategy to balance exploration and utilization, and the initial value of ε is set to 0.5. As the number of training iterations increases, the ε value decays exponentially to 0.1 to optimize long-term benefits. The real-time feedback reinforcement learning module continuously optimizes the recommendation strategy through the Q-learning algorithm to ensure the accuracy of long-term recommendations and user satisfaction.

[0022] Preferably, in the output module, the instruction granularity includes a speed control curve, a temperature PID control parameter and a safety monitoring module. The time resolution of the speed control curve is ≤100ms and the dynamic adjustment error is ±5rpm. Among the temperature PID control parameters, Kp=2.5, Ki=0.1, Kd=0.05. The safety monitoring module is used to detect current fluctuations and vibration amplitudes in real time, with a threshold of ±15% of the rated value and an alarm threshold ≥3mm.

[0023] Through the above technical solution, it can be seen that after selecting a recipe, the control instruction generation unit of the output module generates a segmented operation instruction set based on the recipe details and the characteristics of the chef machine. These instructions cover key cooking stages such as preheating, stirring, heating and keeping warm. Each stage is equipped with precise operating parameters (such as speed control curve, PID temperature control parameters) and time nodes. In the stirring stage, the system will dynamically adjust the stirring mode according to the characteristics of the ingredients, while the heating stage uses the PID temperature control algorithm combined with fuzzy logic control to achieve fine temperature adjustment to ensure the stability and efficiency of the cooking process.

[0024] Preferably, in the output module, the segmented operation instruction set includes a preheating stage, a stirring stage, a heating stage and a heat preservation stage, each stage includes detailed operation parameters and time nodes, a linear heating strategy is adopted in the preheating stage, and the heating rate does not exceed 5°C / min, the stirring mode is dynamically adjusted according to the characteristics of the food in the stirring stage, and the PID temperature control algorithm is adopted in the heating stage, combined with fuzzy logic control to fine-tune the temperature to improve the temperature control accuracy and response speed.

[0025] Preferably, an exception handling mechanism is also included, which automatically switches to a safety mode and sends a Bluetooth alarm when it is detected that the temperature gradient change rate is ≥5°C / s or the torque fluctuation exceeds the mean ±20% for 2s.

[0026] Preferably, in the exception handling mechanism, when entering the safety mode, the food processor automatically adjusts to the lowest power state and starts a fault diagnosis program, which analyzes sensor data in real time based on a machine learning model to identify the fault type of overheating, overload or mechanical failure. The diagnosis results are sent to the user's mobile phone App via Wi-Fi or Bluetooth.

[0027] Through the above technical solutions, it can be seen that during the entire cooking process, the safety monitoring module continuously monitors key indicators such as current fluctuations, vibration amplitude, and temperature gradient change rate. Once an abnormal situation is detected (such as excessive temperature or abnormal torque), the system will immediately trigger the abnormality handling mechanism, automatically switch to safety mode, reduce power and start the fault diagnosis program. The program quickly identifies the type of fault based on the machine learning model, and instantly notifies the user via Wi-Fi or Bluetooth, while providing preliminary fault solutions or suggesting to contact maintenance services.

[0028] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0029] 1. The present invention, by integrating an image recognition unit and a weight sensor, can collect data on the type, weight and status of ingredients placed in the food processor by the user in real time, and perform correction processing in combination with the ambient temperature and humidity sensor, thereby ensuring that the user's existing food inventory can be accurately considered when recommending recipes. This solves the problem that the existing system cannot fully utilize the user's existing food when recommending recipes, greatly improves the practicality of recipe recommendations and the user's cooking efficiency, and makes the cooking process smoother and more efficient.

[0030] 2. The present invention, through the high-dimensional feature vector generated by the data processing module, and the intelligent recommendation of the recipe recommendation module based on the improved collaborative filtering algorithm and the knowledge graph, not only takes into account user preferences and equipment compatibility, but also incorporates the nutritional balance coefficient, thereby realizing the deep linkage between the cooking equipment and the recipe recommendation system. The output module generates a segmented operation instruction set according to the selected recipe, covering key cooking stages such as preheating, stirring, heating and insulation, and is equipped with precise operation parameters and time nodes.

[0031] 3. The present invention can timely detect and handle abnormal conditions in the cooking process by real-time monitoring of the temperature gradient change rate and torque fluctuation. Once abnormal temperature changes or torque fluctuations beyond the set range are detected, the system will automatically switch to the safety mode and start the fault diagnosis program for real-time analysis. This improvement greatly enhances the safety of the cooking process and effectively avoids safety accidents caused by equipment failure or improper operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is an overall flow chart of the intelligent recipe recommendation and operation control method of the chef machine of the present invention;

[0033] Figure 2 It is a working flow chart of the user input module of the present invention;

[0034] Figure 3 is a work flow chart of the identification module of the present invention;

[0035] Figure 4 A workflow diagram of the recipe recommendation module of the present invention;

[0036] Figure 5 It is a flow chart of the segmented operation instruction set of the output module of the present invention. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] See also Figure 1-Figure 5 In an embodiment of the present invention, a method for intelligent recipe recommendation and operation control of a chef machine includes:

[0039] A user input module for receiving user instructions including ingredient type, cooking goal, dietary restrictions and equipment model;

[0040] The recognition module includes an image recognition unit and a weight sensor, which is used to collect food weight and status data and correct the processing logic in combination with the ambient temperature and humidity sensor;

[0041] The data processing module uses a multimodal fusion algorithm to map user input and sensor data to a high-dimensional feature space and generate a feature vector V = (v 1 ,v 2 ,v n ), the dimensions include food moisture content, hardness, heat capacity and equipment load factor;

[0042] The recipe recommendation module recommends candidate recipes from the recipe library based on the improved collaborative filtering algorithm and knowledge graph, and calculates the fitness score S = α·user preference matching degree + β·device compatibility + γ·nutritional balance coefficient (α+β+γ=1);

[0043] The output module controls the instruction generation unit to output a segmented operation instruction set according to the selected recipe.

[0044] The working principle of the embodiment of the present invention is: the user first provides key cooking information through the voice recognition or touch input module, including the type of ingredients, cooking goals (such as cake, bread, etc.), dietary restrictions (such as gluten-free, low sugar, etc.) and the model of the chef machine used. This information is processed in real time and passed to the data processing module; at the same time, the allergen database integrated in the user input module filters out the ingredient options that do not meet the user's health needs in real time, and performs precise matching based on the 8 major categories of allergens specified by the FDA.

[0045] The recognition module is immediately started, and the image recognition unit uses the improved YOLOv5 model to quickly and accurately recognize the images of the ingredients placed in the food processor. At the same time, the weight sensor accurately measures the weight of the ingredients, and the ambient temperature and humidity sensor monitors the current cooking environment. These data are used together to correct the processing logic of the ingredient status to be closer to the actual cooking conditions.

[0046] The data processing module receives the above-mentioned multi-source data and uses the multimodal fusion algorithm under the TensorFlow framework, especially the long short-term memory network (LSTM), to perform deep modeling on the time series data to capture the subtle characteristics of the food status changing over time. This process not only takes into account the static properties of the food (such as water content and hardness), but also incorporates dynamic change factors to generate a high-dimensional feature vector V.

[0047] The recipe recommendation module based on the improved collaborative filtering algorithm and knowledge graph combines user preferences, device compatibility and nutritional balance coefficient to select the most suitable candidate recipes from the huge recipe library. Among them, the graph neural network enhances the semantic understanding and reasoning ability based on the knowledge graph, making the recommendation more intelligent and personalized.

[0048] After selecting a recipe, the control instruction generation unit of the output module generates a segmented set of operation instructions based on the recipe details and the characteristics of the chef machine. These instructions cover key cooking stages such as preheating, stirring, heating and keeping warm. Each stage is equipped with precise operating parameters (such as speed control curve, PID temperature control parameters) and time nodes. In the stirring stage, the system will dynamically adjust the stirring mode according to the characteristics of the ingredients, while the heating stage uses the PID temperature control algorithm combined with fuzzy logic control to achieve fine temperature adjustment.

[0049] During the entire cooking process, the safety monitoring module continuously monitors key indicators such as current fluctuations, vibration amplitude, and temperature gradient change rate. Once an abnormal situation is detected (such as excessive temperature or abnormal torque), the system will immediately trigger the abnormality handling mechanism, automatically switch to safety mode, reduce power and start the fault diagnosis program. The program quickly identifies the type of fault based on the machine learning model and notifies the user instantly via Wi-Fi or Bluetooth, while providing preliminary fault solutions or suggesting to contact maintenance service.

[0050] See also Figure 1-Figure 5 In the embodiment of the present invention, the user input module supports voice recognition and touch input, and integrates the allergen database and covers FDA's 8 major categories of allergens for real-time filtering.

[0051] In the recognition module, the image recognition unit uses an improved YOLOv5 model. The training data set contains ≥ 500,000 images of food under multiple lighting conditions, and an adversarial generative network is introduced to enhance data diversity.

[0052] In the data processing module, the multimodal fusion algorithm is implemented using the deep learning framework TensorFlow. The long short-term memory network is used to model the time series data to capture the dynamic characteristics of the food status changing over time. The number of LSTM layers is 3, the number of hidden units is 256, the optimizer is Adam, the learning rate is set to 0.001, and the number of training iterations is 100,000 to ensure the accuracy and robustness of the feature vector V.

[0053] In the recipe recommendation module, the collaborative filtering algorithm adopts a hybrid model, including matrix decomposition based on user behavior with potential factors k=50 and a graph neural network based on the knowledge graph. The recipe recommendation module also includes a real-time feedback reinforcement learning module with a Q-learning update frequency of Δt=10min. In the recipe recommendation module, the knowledge graph is constructed based on the Neo4j graph database, which includes ingredients, recipes, cooking skills and nutritional information. The graph information is embedded into a low-dimensional vector space through graph embedding technology to enhance the semantic understanding and reasoning capabilities of the recommendation system.

[0054] In the real-time feedback reinforcement learning module, the Q-learning algorithm adopts the ε-greedy strategy to balance exploration and utilization. The initial value of ε is set to 0.5. As the number of training iterations increases, the ε value decays exponentially to 0.1 to optimize long-term benefits.

[0055] In the output module, the instruction granularity includes the speed control curve, temperature PID control parameters and safety monitoring module. The time resolution of the speed control curve is ≤100ms and the dynamic adjustment error is ±5rpm. The temperature PID control parameters are Kp=2.5, Ki=0.1, Kd=0.05. The safety monitoring module is used to detect current fluctuations and vibration amplitudes in real time. The threshold is ±15% of the rated value and the alarm threshold is ≥3mm.

[0056] In the output module, the segmented operation instruction set includes preheating stage, stirring stage, heating stage and insulation stage. Each stage contains detailed operation parameters and time nodes. The preheating stage adopts a linear heating strategy, and the heating rate does not exceed 5℃ / min. The stirring stage dynamically adjusts the stirring mode according to the characteristics of the ingredients. The heating stage adopts a PID temperature control algorithm, combined with fuzzy logic control to fine-tune the temperature to improve the temperature control accuracy and response speed.

[0057] The working principle of the embodiment of the present invention is as follows: Taking the production of whole wheat bread as an example, when the user selects the "whole wheat bread" target and puts in 500g of high-gluten flour, the image recognition unit analyzes the uniformity of flour particle distribution in real time by improving the attention mechanism module (SE Block) of YOLOv5, and the weight sensor synchronously collects flour weight data to generate a 498±2g measurement value. When the ambient temperature and humidity sensor detects 25℃ / 65%RH, the data processing module starts the LSTM time series prediction model, calculates the expected water absorption change according to the formula ΔW=0.023×(RH-60%)×t, and dynamically adjusts the liquid addition amount of the recommended recipe. The multimodal feature vector V is generated in this process, and the heat capacity parameter is calculated by the DSC (differential scanning calorimetry) model.

[0058] When the recipe recommendation module is started, the Neo4j graph database executes the Cypher query statement:

[0059] MATCH(u:User)-[PREFERS]->(i:Ingredient)<-[CONTAINS]-(r:Recipe)

[0060] WHEREr.cuisineType='European'AND r.cookingTime<120

[0061] WITH r,COUNT(i)AS matchScore

[0062] ORDER BY matchScore DESC LIMIT 10

[0063] The graph neural network then uses the GraphSAGE algorithm to perform neighborhood aggregation, and combined with the user's historical operation records that 75% prefer medium-speed mixing, generates the device compatibility parameter β = 0.32. The reinforcement learning module dynamically adjusts the recommendation weight through the Q-value update formula Q(s,a)←Q(s,a)+0.1[γmaxQ(s',a')-Q(s,a)], and finally selects the whole wheat bread recipe with a fitness score of S = 0.87, which is higher than the threshold of 0.8 and enters the candidate list.

[0064] The segmentation instructions generated by the output module then include:

[0065] Preheating stage: PID algorithm is used to control the power of the heating tube according to the formula u(t)=2.5e(t)+0.1∫e(t)dt+0.05de(t) / dt, and the temperature control accuracy of ±1.5℃ is achieved by combining infrared thermal imaging feedback. When the temperature difference between the bottom and top of the cavity is detected to be >8℃, the centrifugal fan is activated to balance the thermal field;

[0066] Mixing stage: The dough shear stress is calculated based on the Hertz contact theory, and the planetary mixing trajectory is adjusted dynamically. When the torque sensor detects a peak value of 3.2N·m, it automatically switches to the pulse mixing mode (200rpm / 5s interval) to prevent excessive gluten development;

[0067] Abnormal processing: When the Hall sensor detects that the motor current harmonic distortion rate is >12%, it triggers the fault feature extraction based on wavelet packet decomposition, identifies the bearing wear type through the SVM classifier, and starts the backup capacitor group for instantaneous power compensation to ensure that the speed fluctuation is <±3rpm.

[0068] By aligning cross-modal features, closed-loop control from micro-ingredient characteristics to macro-equipment operation is achieved. Experimental data show that compared with traditional methods, this system reduces the dough formation time prediction error by 42% and improves energy efficiency by 28%.

[0069] See also Figure 1-Figure 5 In an embodiment of the present invention, an exception handling mechanism is also included. When it is detected that the temperature gradient change rate is ≥5°C / s or the torque fluctuation exceeds the mean ±20% for 2s, it automatically switches to the safety mode and sends a Bluetooth alarm.

[0070] In the exception handling mechanism, when entering the safety mode, the food processor automatically adjusts to the lowest power state and starts the fault diagnosis program. The program analyzes the sensor data in real time based on the machine learning model to identify the fault type of overheating, overload or mechanical failure. The diagnosis results are sent to the user's mobile phone app via Wi-Fi or Bluetooth.

[0071] The working principle of the embodiment of the present invention is: when the food processor is in operation, the system will monitor the temperature gradient change rate and torque fluctuation in real time. The temperature gradient change rate is monitored by high-frequency sampling of the internal temperature of the device through a temperature sensor, and the ratio of the temperature difference between adjacent sampling points to the time interval is calculated. If the ratio is ≥5°C / s, it indicates an abnormal temperature change. Torque fluctuation monitoring is to continuously obtain the device operation torque data through a torque sensor and compare it with the set torque mean. If the fluctuation exceeds the mean ±20% and lasts for 2s, it means that the device is running with abnormal load or mechanical failure.

[0072] Once the above abnormal conditions are detected, the system will automatically switch to safe mode. In safe mode, the food processor will first automatically adjust to the lowest power state. At this time, the device only maintains the power supply of the basic control circuit and monitoring circuit, and stops all major high-power operations such as heating and stirring to reduce energy consumption and potential risks. At the same time, the fault diagnosis program is quickly started. The program is based on a pre-trained machine learning model and is trained through feature extraction and classification algorithms. The program analyzes the current real-time collected sensor data, extracts key features, such as temperature change curve features, torque fluctuation frequency features, etc., and compares and matches them with the fault feature library in the model to quickly identify the fault type of overheating, overload or mechanical failure.

[0073] The diagnostic results are sent to the user's mobile phone app via Wi-Fi or Bluetooth module, informing the user in detail of the fault type, possible causes and recommended treatment measures, such as whether to pause cooking, check the amount of ingredients or contact after-sales maintenance, etc., to ensure that the user can understand the equipment status in time and take corresponding measures to ensure the safety and smooth progress of the cooking process.

[0074] Working principle: The user first inputs key cooking information, such as ingredient type, cooking goal, dietary restrictions and equipment model, through the voice recognition or touch input module. At the same time, the allergen database filters out the ingredient options that do not meet health needs in real time. Then, the recognition module is started, the image recognition unit uses the improved YOLOv5 model to accurately identify the ingredients, the weight sensor measures the weight of the ingredients, and the ambient temperature and humidity sensor monitors the cooking environment. These data jointly correct the ingredient state processing logic. The data processing module receives multi-source data, and uses the multimodal fusion algorithm under the TensorFlow framework, especially the LSTM network, to deeply model the time series data and generate a high-dimensional feature vector V containing dimensions such as moisture content and hardness of the ingredients. The recipe recommendation module is based on the improved collaborative filtering algorithm and knowledge graph, combined with user preferences, equipment compatibility and nutritional balance coefficient, to select the most suitable candidate recipes from the recipe library. After selecting the recipe, the control instruction generation unit of the output module generates a segmented operation instruction set covering preheating, stirring, heating and insulation stages based on the recipe details and the characteristics of the chef machine. Each stage is equipped with precise operation parameters and time nodes. During the cooking process, the safety monitoring module continuously monitors key indicators. Once an abnormal situation is detected, the system triggers the exception handling mechanism, switches to safety mode, reduces power and starts the fault diagnosis program, notifies the user via Wi-Fi or Bluetooth and provides solutions.

[0075] Taking whole wheat bread making as an example, after the user selects the target and puts in high-gluten flour, the image recognition unit analyzes the uniformity of flour particle distribution, the weight sensor collects weight data, the ambient temperature and humidity sensor detects environmental parameters, and the data processing module starts the LSTM time series prediction model, dynamically adjusts the amount of liquid added to the recipe, and generates a multimodal feature vector V. The recipe recommendation module generates equipment compatibility parameters through Neo4j graph database query and graph neural network neighborhood aggregation, combined with user historical operation records, and the reinforcement learning module dynamically adjusts the recommendation weight, and finally selects the whole wheat bread recipe with a fitness score higher than the threshold. The segmented instructions generated by the output module include using the Smith estimate compensation PID algorithm to control the power of the heating tube, and dynamically adjusting the stirring trajectory by calculating the dough shear stress based on the Hertz contact theory. When the abnormal temperature gradient change rate or torque fluctuation is detected, the exception handling mechanism switches to safe mode, starts the fault diagnosis program, identifies the fault type through the machine learning model, and notifies the user to ensure a safe and smooth cooking process.

[0076] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and alterations may be made to the embodiments without departing from the principles and spirit thereof, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent recipe recommendation and operation control of a chef machine, characterized in that: include: A user input module for receiving user instructions including ingredient type, cooking goal, dietary restrictions and equipment model; The recognition module includes an image recognition unit and a weight sensor, which is used to collect food weight and status data and correct the processing logic in combination with the ambient temperature and humidity sensor; The data processing module uses a multimodal fusion algorithm to map user input and sensor data to a high-dimensional feature space and generate a feature vector V = (v1, v2, v n ), the dimensions include food moisture content, hardness, heat capacity and equipment load factor; The recipe recommendation module recommends candidate recipes from the recipe library based on the improved collaborative filtering algorithm and knowledge graph, and calculates the fitness score S = α·user preference matching degree + β·device compatibility + γ·nutritional balance coefficient (α+β+γ=1); The output module controls the instruction generation unit to output a segmented operation instruction set according to the selected recipe.

2. The method for intelligent recipe recommendation and operation control of a chef machine according to claim 1, characterized in that: The user input module supports voice recognition and touch input, and integrates an allergen database covering FDA's eight major categories of allergens for real-time filtering.

3. The method for intelligent recipe recommendation and operation control of a chef machine according to claim 1, characterized in that: In the recognition module, the image recognition unit adopts an improved YOLOv5 model, the training data set contains ≥500,000 food images under multiple lighting conditions, and an adversarial generative network is introduced to enhance data diversity.

4. The method for intelligent recipe recommendation and operation control of a chef machine according to claim 1, characterized in that: In the data processing module, the multimodal fusion algorithm is implemented using the deep learning framework TensorFlow, and the time series data is modeled using a long short-term memory network to capture the dynamic characteristics of food status changing over time. The number of LSTM layers is 3, the number of hidden units is 256, the optimizer is Adam, the learning rate is set to 0.001, and the number of training iterations is 100,000 to ensure the accuracy of the feature vector V.

5. The method for intelligent recipe recommendation and operation control of a chef machine according to claim 1, characterized in that: In the recipe recommendation module, the collaborative filtering algorithm adopts a hybrid model, including matrix decomposition based on user behavior with potential factors k=50 and a graph neural network based on the knowledge graph. The recipe recommendation module also includes a real-time feedback reinforcement learning module with a Q-learning update frequency of Δt=10min. In the recipe recommendation module, the knowledge graph is constructed based on the Neo4j graph database, which includes ingredients, recipes, cooking skills and nutritional information. The graph information is embedded into a low-dimensional vector space through graph embedding technology to enhance the semantic understanding and reasoning capabilities of the recommendation system.

6. The method for intelligent recipe recommendation and operation control of a chef machine according to claim 5, characterized in that: In the real-time feedback reinforcement learning module, the Q-learning algorithm adopts the ε-greedy strategy to balance exploration and utilization. The initial value of ε is set to 0.

5. As the number of training iterations increases, the ε value decays exponentially to 0.1 to optimize long-term benefits.

7. The method for intelligent recipe recommendation and operation control of a chef machine according to claim 1, characterized in that: In the output module, the instruction granularity includes a speed control curve, a temperature PID control parameter and a safety monitoring module. The time resolution of the speed control curve is ≤100ms and the dynamic adjustment error is ±5rpm. Among the temperature PID control parameters, Kp=2.5, Ki=0.1, and Kd=0.

05. The safety monitoring module is used to detect current fluctuations and vibration amplitudes in real time, with a threshold of ±15% of the rated value and an alarm threshold of ≥3mm.

8. The method for intelligent recipe recommendation and operation control of a chef machine according to claim 1, characterized in that: In the output module, the segmented operation instruction set includes a preheating stage, a stirring stage, a heating stage and a heat preservation stage. Each stage includes detailed operation parameters and time nodes. The preheating stage adopts a linear temperature rise strategy, and the heating rate does not exceed 5°C / min. The stirring stage dynamically adjusts the stirring mode according to the characteristics of the food. The heating stage adopts a PID temperature control algorithm, combined with fuzzy logic control to fine-tune the temperature to improve the temperature control accuracy and response speed.

9. The method for intelligent recipe recommendation and operation control of a chef machine according to claim 1, characterized in that: It also includes an exception handling mechanism. When it detects that the temperature gradient change rate is ≥5℃ / s or the torque fluctuation exceeds the mean ±20% for 2s, it automatically switches to safety mode and sends a Bluetooth alarm.

10. The method for intelligent recipe recommendation and operation control of a chef machine according to claim 9, characterized in that: In the exception handling mechanism, when entering the safety mode, the food processor automatically adjusts to the lowest power state and starts the fault diagnosis program. The program analyzes the sensor data in real time based on the machine learning model to identify the fault type of overheating, overload or mechanical failure. The diagnosis results are sent to the user's mobile phone app via Wi-Fi or Bluetooth.

Images