A method, apparatus, soy milk maker, and storage medium for controlling the soy milk production process.

By combining multimodal perception and cooking state recognition models, intelligent closed-loop control of the soy milk maker's soy milk making process is achieved, solving the problem of low intelligence and adaptability of the soy milk maker, and improving the soy milk maker's safety, stability and personalized taste matching ability.

CN123113448APending Publication Date: 2026-11-17GREE TOSOT (SUQIAN) HOME APPLIANCES CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611164224.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-03
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing household soy milk makers have low levels of intelligence and adaptability, making them unable to adapt to the physical properties of different ingredients and the diverse taste needs of users. This results in a monotonous taste in the finished soy milk, insufficient release of nutrients, and safety and stability issues such as overflowing and burning.

Method used

A multi-modal sensing module is used to collect multi-dimensional feature vectors of the slurry in real time. A pre-built cooking state recognition model is used to predict trends and generate predictive anti-overflow and anti-burning commands. The heating power and stirring speed are dynamically adjusted to control the slurry making process and build a closed-loop control system.

Benefits of technology

It enables proactive identification and intervention of risks of overflow and scorching, improves the stability and reliability of the soy milk making process, accurately matches users' personalized taste needs, and improves the efficiency of nutrient extraction and taste quality of the finished soy milk.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123113448A_ABST
    Figure CN123113448A_ABST
Patent Text Reader

Abstract

This invention discloses a method, device, soymilk maker, and storage medium for controlling the soymilk making process, belonging to the field of food processing technology. The method includes: after the soymilk maker starts its soymilk-making operation, acquiring multi-dimensional feature vectors of the soymilk liquid in the soymilk maker through a multi-modal sensing module; wherein the multi-dimensional feature vectors include the concentration gradient of the soymilk liquid, the foam height, and the bottom temperature of the soymilk maker; based on the multi-dimensional feature vectors, using a cooking state recognition model to predict the trend of the soymilk liquid, obtaining a predicted trend; judging whether there is a risk of overflow and whether there is a risk of scorching at the bottom of the soymilk maker based on the predicted trend; when an overflow risk is determined, generating a predictive anti-overflow command; and when a scorching risk is determined, generating a predictive anti-scorching command; and controlling the soymilk-making operation based on the predictive anti-overflow command and / or the predictive anti-scorching command. This invention can improve the intelligent control effect of the soymilk maker.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of food processing equipment technology, and in particular to a method, apparatus, soy milk maker, and storage medium for controlling the soy milk production process. Background Technology

[0002] Currently, most mainstream household soy milk makers on the market adopt a fixed time sequence control scheme. The heating, stirring, and cooking process is preset before the equipment leaves the factory, and the process parameters cannot be dynamically adjusted during operation. Due to the significant differences in the physical properties of ingredients such as soybeans, black beans, and various grains, different ingredients have different characteristics in terms of particle size, cooking viscosity, and high-temperature foam generation rate. At the same time, users have different personalized needs for the consistency and smoothness of soy milk. The traditional fixed soy milk making process cannot adapt to the physical characteristics of various ingredients and the differentiated taste needs of users. In the end, it is very easy to make soy milk with a monotonous taste and serious homogenization. Moreover, the nutrients such as protein and trace elements inside the ingredients cannot be fully extracted and released, resulting in poor soy milk quality.

[0003] In terms of equipment safety control and operating condition adjustment, traditional soymilk makers rely primarily on temperature sensors or anti-overflow electrodes for passive detection to prevent overflow and smearing. Their control logic and trigger thresholds are fixed and rigid, resulting in extremely poor adaptive adjustment capabilities. During actual soymilk making, factors such as fluctuations in water volume, differences in altitude and air pressure in different regions, and variations in the types of ingredients can cause dynamic changes in the foam generation rate and accumulation height during the boiling process. Fixed detection thresholds and control logic cannot adapt to these complex and changing conditions, leading to frequent problems such as overflow, smearing at the bottom of the pot, and uneven heating of the soymilk, severely impacting the stability of the equipment and the soymilk making effect.

[0004] While some intelligent soymilk makers have added auxiliary functions such as remote control and fixed menu switching, these are only basic functional expansions. The core soymilk making system is still in an open-loop control mode and has not established a complete working condition perception and feedback adjustment mechanism. It is unable to collect key operating status parameters such as soymilk concentration, real-time foam height, and temperature gradient inside the pot in real time. It is also difficult to dynamically optimize core process parameters such as heating power, stirring speed, and cooking time based on real-time working conditions. The intelligent control capability has obvious shortcomings.

[0005] While existing technologies include solutions for detecting scorching at the bottom using turbidity sensors and analyzing material composition based on transmission spectroscopy, these technologies have limited sensing dimensions and can only detect single operating parameters. They cannot simultaneously and accurately quantify multi-dimensional core data such as particle size distribution, foam dynamics, and temperature field distribution within the pot during the pulping process. Furthermore, they only have state detection capabilities and have not built a predictive closed-loop control model based on multi-source sensing data. Consequently, they cannot anticipate and proactively intervene in issues such as overflow and scorching at the bottom, and still suffer from control lag.

[0006] In addition, existing home soy milk makers lack a user taste learning system, making it impossible to collect, store, and analyze user feedback on the taste and preferences of the finished soy milk. They also cannot iteratively optimize subsequent soy milk making strategies based on user feedback and lack personalized adaptive optimization capabilities. Overall, existing soy milk makers have low levels of intelligence and adaptability, failing to meet users' needs for high-quality, personalized soy milk making, which significantly hinders the improvement of user experience and core market competitiveness. Summary of the Invention

[0007] This invention provides a method, apparatus, soymilk maker, and storage medium for controlling the soymilk production process, aiming to improve the intelligent control effect of the soymilk maker and enhance the user experience.

[0008] In a first aspect, embodiments of the present invention provide a method for controlling the soy milk production process in a soy milk maker, comprising: After the soymilk maker starts the soymilk making operation, a multi-dimensional feature vector of the soymilk liquid in the soymilk maker is collected through a multi-modal sensing module; wherein, the multi-dimensional feature vector includes the concentration gradient of the soymilk liquid, the foam height and the bottom temperature of the soymilk maker. Based on the multidimensional feature vector, the trend of the slurry is predicted using a pre-built cooking state recognition model to obtain the predicted trend of the slurry. Based on the predicted trend of the slurry, determine whether there is a risk of the slurry overflowing, and determine whether there is a risk of the soy milk maker burning at the bottom; When it is determined that there is a risk of overflow of the slurry, a predictive anti-overflow command is generated; and when it is determined that there is a risk of scorching at the bottom of the soy milk maker, a predictive anti-scorching command is generated. The soymilk making operation of the soymilk maker is controlled based on the predictive anti-overflow command and / or the predictive anti-smear command.

[0009] Secondly, embodiments of the present invention provide a soy milk maker's pulping control device, comprising: The feature acquisition unit is used to acquire multi-dimensional feature vectors of the soy milk maker through a multi-modal sensing module after the soy milk maker starts the soy milk making operation; wherein, the multi-dimensional feature vectors include the concentration gradient of the soy milk maker, the foam height and the bottom temperature of the soy milk maker. The trend prediction unit is used to predict the trend of the slurry based on the multidimensional feature vector and using a pre-built cooking state recognition model to obtain the predicted trend of the slurry. The risk assessment unit is used to determine whether there is a risk of overflow of the slurry based on the predicted trend of the slurry, and to determine whether there is a risk of the soy milk maker burning at the bottom. The instruction generation unit is used to generate a predictive anti-overflow instruction when it is determined that there is a risk of the slurry overflowing; and to generate a predictive anti-sticking instruction when it is determined that there is a risk of the soy milk maker sticking to the bottom. An operation control unit is used to control the soymilk making operation of the soymilk maker based on the predictive anti-overflow command and / or the predictive anti-smear command.

[0010] Thirdly, embodiments of the present invention provide a soymilk maker, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the soymilk maker soymilk making control method as described in the first aspect.

[0011] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the soymilk maker's soymilk-making control method as described in the first aspect.

[0012] This invention provides a method, apparatus, and storage medium for controlling the soy milk maker's soy milk production. The method includes: after the soy milk maker starts its soy milk production operation, acquiring a multi-dimensional feature vector of the soy milk in the soy milk maker through a multi-modal sensing module; wherein the multi-dimensional feature vector includes the concentration gradient of the soy milk, the foam height, and the bottom temperature of the soy milk maker; based on the multi-dimensional feature vector, using a pre-built cooking state recognition model to predict the trend of the soy milk, obtaining a predicted trend of the soy milk; determining whether there is a risk of overflow of the soy milk and whether there is a risk of scorching at the bottom of the soy milk maker based on the predicted trend of the soy milk; when it is determined that there is a risk of overflow of the soy milk, generating a predictive anti-overflow command; and when it is determined that there is a risk of scorching at the bottom of the soy milk maker, generating a predictive anti-scorching command; and controlling the soy milk production operation of the soy milk maker based on the predictive anti-overflow command and / or the predictive anti-scorching command. This invention utilizes multimodal sensing technology to simultaneously collect core operating parameters across multiple dimensions, including slurry concentration gradient, foam height, and pot bottom temperature. Based on a pre-trained cooking state recognition model, it predicts the trend of the slurry-making process and provides early warnings of overflow and scorching risks. Compared to traditional soymilk makers employing single-point passive detection, fixed threshold control, and open-loop regulation, this invention enables proactive identification and intervention of abnormal operating conditions, completely resolving the lag issues inherent in traditional control methods. It effectively avoids various malfunctions during slurry making, such as overflow, scorching, and uneven heating, significantly improving the operational stability and reliability of the entire slurry-making process. Furthermore, this invention constructs a complete closed-loop control system for slurry making based on multi-source real-time sensing data. This system dynamically adapts and optimizes slurry-making process parameters according to the physical characteristics of different ingredients, accurately matching diverse and personalized taste preferences. It effectively improves the efficiency of nutrient extraction and the taste quality of the finished soymilk, fundamentally enhancing the adaptive control capability and intelligence level of the soymilk maker, significantly optimizing the user experience, and improving the product's market competitiveness. Attached Figure Description

[0013] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 A schematic flowchart of a soybean milk making and pulp control method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the first sub-process of a soybean milk making and pulp control method provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the second sub-process of a soybean milk making and pulping control method provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the third sub-process of a soybean milk making and pulp control method provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the fourth sub-process of a soy milk making and pulp control method provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the fifth sub-process of a soy milk making and pulp control method provided in an embodiment of the present invention; Figure 7 A schematic diagram of the workflow of a soybean milk making and pulp control method provided in an embodiment of the present invention; Figure 8 This is a model architecture diagram of a soybean milk maker's pulp control method provided in an embodiment of the present invention; Figure 9 A system architecture diagram of a soybean milk maker's pulp control method provided in an embodiment of the present invention; Figure 10 This is a collaborative architecture diagram of a soybean milk maker's pulp control method provided in an embodiment of the present invention; Figure 11 A schematic block diagram of a soybean milk maker's pulping control device provided in an embodiment of the present invention; Figure 12 This is a first schematic block diagram of a soybean milk maker's pulping control device provided in an embodiment of the present invention; Figure 13 This is a second schematic block diagram of a soybean milk maker's pulping control device provided in an embodiment of the present invention; Figure 14 This is a schematic block diagram of the third sub-section of a soybean milk maker's pulping control device provided in an embodiment of the present invention; Figure 15 This is a fourth schematic block diagram of a soybean milk maker's pulping control device provided in an embodiment of the present invention; Figure 16 This is a fifth schematic block diagram of a soybean milk maker's pulping control device provided in an embodiment of the present invention; Figure 17 This is a schematic block diagram of a soymilk maker provided in an embodiment of the present invention. Detailed Implementation

[0015] The technical solutions of 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, not all, of the embodiments of the present invention. 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.

[0016] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0017] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0018] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0019] Please see below. Figure 1 This invention provides a method for controlling the soy milk production process in a soy milk maker, specifically including steps S101 to S105.

[0020] Step S101: After the soymilk maker starts the soymilk making operation, the multi-dimensional feature vector of the soymilk liquid in the soymilk maker is collected through the multi-modal sensing module; wherein, the multi-dimensional feature vector includes the concentration gradient of the soymilk liquid, the foam height and the bottom temperature of the soymilk maker. Step S102: Based on the multidimensional feature vector, use the pre-built cooking state recognition model to predict the trend of the slurry and obtain the predicted trend of the slurry. Step S103: Determine whether there is a risk of overflow of the slurry based on the predicted trend of the slurry, and determine whether there is a risk of the soy milk maker burning at the bottom; Step S104: When it is determined that there is a risk of overflow of the slurry, a predictive anti-overflow command is generated; and when it is determined that there is a risk of scorching at the bottom of the soy milk maker, a predictive anti-scorching command is generated. Step S105: Control the soymilk making operation of the soymilk maker based on the predictive anti-overflow command and / or the predictive anti-smear command.

[0021] In this embodiment, combined with Figure 7After the soymilk maker starts making soymilk, a multi-modal sensing module collects multi-dimensional feature vectors of the soymilk, including the soymilk concentration gradient, foam height, and the temperature of the bottom of the soymilk pot. Then, based on the multi-dimensional feature vectors, a cooking state recognition model predicts the changing trend of the soymilk. Based on the predicted trend, it determines whether there is a risk of overflow or a risk of the soymilk maker burning the bottom. For each prediction, if there is a risk of overflow, a predictive anti-overflow command is generated; if there is a risk of burning the bottom, a predictive anti-burn command is generated. The soymilk maker's operation is then adjusted according to the corresponding commands. This embodiment uses multi-modal sensing technology to simultaneously collect multi-dimensional core operating parameters such as soymilk concentration gradient, foam height, and bottom temperature, and uses a pre-trained cooking state recognition model to predict the soymilk making trend and provide early warnings of overflow and burning risks. Compared to traditional soymilk makers that employ single-point passive detection, fixed threshold control, and open-loop regulation, this embodiment enables proactive identification and intervention of abnormal operating conditions. This completely resolves the lag issues inherent in traditional control methods, effectively avoiding various malfunctions during the soymilk-making process such as overflow, scorching, and uneven heating, significantly improving the operational stability and reliability of the entire process. Furthermore, this embodiment constructs a complete closed-loop regulation system based on multi-source real-time sensing data. It can dynamically adapt and optimize soymilk-making process parameters according to the physical characteristics of different ingredients, accurately matching diverse and personalized taste preferences. This effectively improves the efficiency of nutrient extraction and the taste quality of the finished soymilk, fundamentally enhancing the soymilk maker's adaptive control capabilities and intelligence level, significantly optimizing the user experience, and increasing the product's market competitiveness.

[0022] In one embodiment, such as Figure 2 As shown, the soymilk maker is equipped with a heating pot, and the multimodal sensing module includes an optical turbidity sensor, an image acquisition module and a temperature and pressure sensor. Step S101 includes steps S201 to S203.

[0023] Step S201: Collect the transmitted light intensity of the slurry using an optical turbidity sensor, obtain the corresponding turbidity based on the transmitted light intensity, and obtain the concentration gradient of the slurry based on a pre-established average particle size-turbidity mapping database. Step S202: Acquire an image of the foam area of ​​the slurry using the image acquisition module, and segment the foam area image using a lightweight semantic segmentation model to extract the foam height of the foam area image; Step S203: Collect the bottom temperature of the heating vessel using a temperature and pressure sensor.

[0024] This embodiment adopts a multi-source data fusion sensing scheme that combines optical sensing, machine vision, and high-precision temperature and pressure acquisition. It can simultaneously collect multiple core operating parameters of the slurry in real time, effectively making up for the shortcomings of traditional soy milk makers, such as limited detection dimensions of a single sensor, large data deviation, and blind spots in operating condition perception. It can capture the dynamic change characteristics of the slurry in all aspects during the slurry making process, and provide comprehensive, accurate, and reliable input data for subsequent cooking status identification and risk trend prediction.

[0025] The hardware component of this embodiment consists of an optical turbidity sensor, a miniature image acquisition module, and a high-precision temperature and pressure sensor group working together. The optical turbidity sensor, specifically the TSL2561 model, has a center wavelength of 550nm±10nm and is vertically mounted on the side wall of the heating vessel. It can collect the intensity of transmitted light in the slurry in real time through a built-in transmit-receive optical path under continuous slurry flow conditions. In the preliminary stages of this embodiment, numerous optical calibration experiments were conducted on soybean milk samples with different ingredient types, soaking times, and degrees of grinding, constructing a comprehensive average particle size-turbidity mapping database. For example, after thorough grinding, the average particle size of soybeans is less than 10μm, corresponding to a turbidity range of 0-30 NTU; after coarse grinding of grains, the average particle size is greater than 30μm, corresponding to a turbidity higher than 70 NTU. This embodiment can accurately obtain the current concentration gradient and average particle size of the slurry based on the real-time collected transmitted light intensity and the calibration database through dynamic lookup operations, enabling effective inference of the particle size distribution of the slurry within the range of 2-50μm.

[0026] The miniature image acquisition module features 1080P high-definition resolution, with the lens centrally mounted downwards on the top of the heating vessel, providing a complete field of view covering the entire slurry surface area inside the vessel. This module utilizes a lightweight semantic segmentation model optimized based on a YOLOv5 pixel-level segmentation head, enabling real-time image segmentation of the foam area on the liquid surface. It accurately extracts bubble edge contours and fits the overall foam morphology, combining this with pre-stored heating vessel inner diameter calibration parameters to calculate real-time foam height with high precision. Simultaneously, a 3-second sliding window algorithm is introduced to smooth and filter the foam height time-series data, effectively suppressing image noise and data fluctuations caused by instantaneous boiling of the slurry, accurately extracting the real-time foam rise rate, and ensuring the stability and accuracy of foam state detection.

[0027] A high-precision temperature and pressure sensor is integrated and installed at the center of the bottom of the heating vessel. It can simultaneously collect the bottom temperature and internal pressure parameters. The temperature detection accuracy can reach ±0.1℃, and the pressure detection accuracy can reach ±0.01kPa. It can accurately reflect the dynamic changes of the temperature and pressure fields during the pulping process, and provide high-precision basic data support for predicting and judging the risks of scorching and overflow.

[0028] In one embodiment, such as Figure 3As shown, step S102 includes steps S301 to S304.

[0029] Step S301: Input the multidimensional feature vector into the cooking state recognition model; wherein, the cooking state recognition model is a CNN-LSTM fusion model; Step S302: Extract local spatial variation patterns from the multidimensional feature vectors using the CNN layer in the CNN-LSTM fusion model; Step S303: Use the LSTM layer in the CNN-LSTM fusion model to establish a temporal linkage relationship for the multidimensional feature vectors and learn the early warning patterns of overflow and burnt bottom accidents; Step S304: Combine the local spatial change mode and the precursor mode to output the foam rising rate and the probability value of bottom sludge clogging, and use the foam rising rate and the probability value of bottom sludge clogging as the predicted trend of the slurry; wherein, the probability value of bottom sludge clogging is obtained based on the concentration gradient of the slurry and the rising rate of the bottom temperature.

[0030] Combination Figure 8 This embodiment employs a CNN-LSTM fusion network to construct a temporal fusion-based cooking state recognition model. Leveraging the advantages of the dual-network structure, it extracts spatial features and mines temporal patterns related to the soy milk preparation process. The model can be pre-trained using a dataset of 2000 sets of soybean milk preparation experiments covering different ingredient types, water amounts, altitudes, and atmospheric pressures, demonstrating excellent generalization and recognition capabilities for various conditions. The model takes the three-dimensional feature vectors of concentration gradient, pot bottom temperature, and foam height output by the sensing module as input. First, the CNN layer extracts local spatial features of the multi-dimensional features, accurately capturing local changes in the characteristics of a single moment, such as recognizing the instantaneous accelerated rise in foam height and the local abrupt changes in soy milk concentration and pot bottom temperature. Then, the LSTM layer models the continuous temporal data, constructing a temporal linkage relationship between temperature, soy milk concentration, and foam state, autonomously learning and precipitating latent precursor feature patterns before overflow and burnt-bottom abnormalities occur.

[0031] The model ultimately outputs two core predictive indicators: the foam rise rate and the probability of bottom scorching. The probability of bottom scorching is calculated by coupling the slurry concentration gradient and the heating rate at the bottom of the pot, providing a comprehensive quantification of abnormal risk conditions in the slurry-making process. This embodiment, through the dual fusion of spatial feature extraction and temporal trend modeling, can deeply explore the dynamic evolution patterns of multi-dimensional operating parameters, capturing early abnormal signs that traditional fixed threshold detection cannot identify. Compared to the passive judgment and delayed response control method of traditional soybean milk makers with a single threshold, this model has the ability to predict and forecast trends in advance, resulting in higher risk identification accuracy and stronger response time. It fundamentally solves the problems of control lag, misjudgment, and missed judgment in traditional passive detection mechanisms, effectively ensuring the safe and stable operation of the slurry-making process.

[0032] In one embodiment, such as Figure 4 As shown, step S103 includes steps S401 to S402.

[0033] Step S401: Compare the foam rising rate of the slurry with a preset dynamic threshold, and determine that the slurry has a risk of overflow when the foam rising rate of the slurry reaches the preset dynamic threshold. Step S402: Compare the probability value of the scorching risk with a preset scorching threshold, and determine that the soymilk maker has a risk of scorching when the probability value of the scorching risk reaches the preset scorching threshold.

[0034] This embodiment uses the foam rising rate as the core indicator for determining the risk of overflow, and configures an adaptively adjustable dynamic threshold. This dynamic threshold can be dynamically corrected by combining real-time collected slurry concentration gradient and initial water addition, and can adapt to the fluctuations in foam dynamic characteristics caused by different food properties, differences in water addition, and changes in altitude and air pressure. Compared with the traditional lagging judgment scheme that relies on anti-overflow electrode touch detection, this embodiment can complete the overflow risk identification 10-15 seconds in advance without waiting for the foam to accumulate to a fixed detection height, reserving sufficient response time for the subsequent anti-overflow control strategy, truly realizing the upgrade from traditional passive lagging protection to predictive active protection. For example, for low-viscosity pure soybean slurry, the dynamic threshold for the foam rising rate is set to 0.5 mm / s; for high-viscosity mixed grain slurry, where foam is prone to continuous accumulation, the corresponding dynamic threshold is adaptively adjusted to 0.7 mm / s, which can accurately adapt to the foaming patterns of different foods and effectively avoid the problems of misjudgment of overflow and control lag caused by differences in food and fluctuations in operating conditions.

[0035] Furthermore, this embodiment combines multi-dimensional operating parameters to achieve accurate quantitative judgment of the risk of bottom scorching. It calculates the probability of bottom scorching risk based on the coupled correlation between the heating rate of the bottom temperature and the slurry concentration gradient, and the judgment logic closely matches the actual bottom scorching formation mechanism. During the slurry preparation process, when the bottom temperature rise rate is ≥0.8℃ / s and the slurry concentration gradient is ≤0.3NTU, it indicates that the slurry medium at the bottom of the pot is thin, and solid particles are easily deposited and adhered, leading to a sharp increase in the risk of bottom scorching. At this time, the probability of bottom scorching risk output by the model will exceed 0.7, and the system immediately determines that there is a risk of bottom scorching and triggers corresponding protective control commands. The judgment method provided in this embodiment quantifies the risk level through multi-parameter coupling, effectively solving the technical shortcomings of traditional single temperature detection, which cannot identify local dry burning and particle deposition and pasting at the bottom of the pot in advance, and significantly improving the accuracy and timeliness of bottom scorching risk prediction.

[0036] In one embodiment, such as Figure 5 As shown, step S105 includes steps S501 to S502.

[0037] Step S501: Based on the predictive anti-overflow command, invoke the preset anti-overflow adjustment strategy; Step S502: Control the soymilk maker's soymilk making operation according to the anti-overflow adjustment strategy; wherein, the anti-overflow adjustment strategy includes reducing the heating power of the soymilk maker and increasing the stirring speed of the soymilk maker.

[0038] In this embodiment, when executing anti-overflow control through predictive anti-overflow commands, a two-dimensional collaborative control strategy is specifically adopted, combining fine-tuning of heating power with dynamic increase of agitation speed, to achieve stable suppression and rapid dissipation of slurry foam. During the specific control process, the IGBT stepless power adjustment heating control unit can reduce the real-time heating power to 70%~85% of the current operating power within 0.5 seconds, slightly reducing the power to mitigate the violent boiling state of the slurry and reduce the foam generation rate from the source. Simultaneously, the motor drive unit dynamically increases the agitation speed, relying on the shearing action of high-speed water flow and mechanical impact to break up the foam already generated on the liquid surface, quickly curbing the continuous rise of foam.

[0039] In practical applications, different speed adjustment parameters can be configured for different pulping ingredients. For example, the beating speed for soybean pulp can be increased to 1200~1500rpm, while for mixed grain pulp with higher viscosity and denser foam, the beating speed can be increased to 1600rpm to meet the defoaming needs of different ingredients.

[0040] This embodiment maintains the soy milk at a constant boil throughout the process, eliminating the need for the traditional passive anti-overflow mode of turning off the heat. This effectively avoids problems such as increased soy milk preparation time, insufficient nutrient extraction, and deterioration in flavor and taste caused by heating interruptions. Furthermore, through precise power adjustment and dynamic agitation and defoaming, compared to control methods relying solely on high-speed agitation and defoaming, it effectively reduces equipment energy consumption. While ensuring excellent anti-overflow performance and stable soy milk quality, it maximizes soy milk preparation efficiency, achieving a balance between safe anti-overflow, efficient soy milk preparation, and superior taste.

[0041] In one embodiment, such as Figure 6 As shown, step S105 further includes steps S601 to S602.

[0042] Step S601: Based on the predictive anti-blurring instruction, invoke the preset anti-blurring adjustment strategy; Step S602: Control the soymilk making operation of the soymilk maker according to the anti-smearing adjustment strategy; wherein, the anti-smearing adjustment strategy includes reducing the heating power of the outer perimeter of the bottom of the soymilk maker and enhancing the central turbulence intensity of the soymilk maker.

[0043] This embodiment employs a synergistic control strategy combining zoned temperature control and dynamic turbulence when controlling for anti-scorching via predictive anti-scorching commands. This optimizes the in-pot slurry preparation conditions from both temperature and flow field dimensions, fundamentally preventing scorching at the bottom of the pot. This embodiment implements differentiated power control for the concentric inner and outer ring heating elements configured at the bottom of the pot. When the anti-scorching strategy is triggered, the heating power of the outer ring at the bottom of the pot is significantly reduced, decreasing the outer ring heating output power by 60%~75%, preventing excessively high temperatures in localized areas at the bottom of the pot and causing the slurry to dry-burn and solidify. Simultaneously, the basic heating power of the central area is retained to maintain the overall cooking temperature environment within the pot, ensuring continuous slurry preparation without affecting the extraction of nutrients and the flavor of the soy milk due to significant temperature drops.

[0044] Meanwhile, the blade drive unit dynamically increases the stirring intensity, raising the central slurry turbulence intensity to 80%~90%, enhancing the convective disturbance effect of the slurry inside the pot. This continuously drives the solid particles in the slurry to be uniformly suspended and dynamically flowed, effectively preventing the deposition and accumulation of high-protein and high-starch particles at the bottom of the pot at high temperatures. This solves problems such as localized overheating and scorching, and scale and blackening of the pot bottom that easily occur under static cooking conditions. This embodiment forms a temperature-flow field coupled control mechanism of "low-power temperature control in the outer ring to prevent overheating, strong turbulence in the center to prevent deposition, and constant temperature throughout the entire range to ensure cooking." While efficiently avoiding the risk of scorching and improving the stability of equipment operation, it effectively ensures the consistency of the cooking quality and taste of the finished soy milk.

[0045] In one embodiment, the soy milk maker's pulping control method further includes: Obtain user feedback data; Based on the user feedback data, an optimizer is used to update the parameters of the cooking state recognition model.

[0046] In this embodiment, by collecting user feedback data on taste after making soy milk, and combining it with a cloud-based incremental learning mechanism, the cooking state recognition model is dynamically iterated and updated. This allows the model to continuously adapt to different users' taste preferences, common ingredient types, and regional environmental conditions, thereby continuously optimizing the model's generalization ability and working condition recognition accuracy. This effectively improves the problem of insufficient adaptability of fixed model parameters in complex and changing scenarios, further reduces the probability of misjudging the risk of overflow and burning under different working conditions, and comprehensively enhances the stability and intelligence level of the soy milk maker's adaptive control.

[0047] Specifically, user feedback and evaluations of taste during use can be uploaded to a cloud server in real time via the device's built-in Wi-Fi module for aggregation and analysis. A lightweight incremental learning strategy is deployed in the cloud, using the Adam optimizer to perform small-batch iterative updates of model parameters. The parameter update process employs a localized update mechanism, fine-tuning only a subset of weights directly related to taste quality, such as the correlation coefficient between concentration gradient and beating speed. This avoids core weights related to spill prevention and anti-burning safety controls, thus preventing the computational overhead and disruption of existing safety control performance caused by retraining the entire model.

[0048] For example, if a user reports that the finished product's smoothness is too low after making mixed grain soy milk, the cloud platform, based on the data from this batch of soy milk making, identifies that the average particle size of the soy milk is too large. It can then automatically increase the concentration gradient-beating speed correlation coefficient corresponding to the mixed grain processing condition by 0.2. This allows the equipment to adaptively increase the beating speed or extend the beating time in the next mixed grain soy milk making scenario, thereby reducing the particle size of the soy milk and improving the smoothness of the finished product. This embodiment uses a local incremental update method, which significantly reduces the computational cost of model iteration while achieving personalized taste optimization for the user. It also effectively retains the model's original risk prediction and safety control capabilities, balancing the equipment's personalized adaptability with the safety of the soy milk making operation.

[0049] In practical applications, taking the multigrain soy milk making scenario as an example, the soy milk maker's pulping control method provided in this embodiment will be described in detail. After the user adds multigrain ingredients and completes the water addition, the pulping program is started, and the equipment immediately enters the adaptive sensing and intelligent control mode. Specifically, firstly, the initial turbidity parameters of the pulp are collected by an optical turbidity sensor to identify the material characteristics of multigrain ingredients with an average particle size greater than 30μm. Based on the preset material property mapping relationship, the dynamic threshold for preventing foam rise rate overflow corresponding to the multigrain working condition is automatically and adaptively adjusted to 0.7mm / s, and the basic stirring speed of the equipment is pre-adjusted to 1600rpm to adapt to the pulping characteristics of multigrain's high viscosity, easy foaming, and difficulty in fine pulping.

[0050] Throughout the subsequent heating and stirring process, the top image acquisition module captures real-time images of the foam on the liquid surface, continuously monitors the dynamic changes in foam height, and calculates the foam rising rate. When the foam rising rate reaches 0.6 mm / s, approaching the preset dynamic threshold, the cooking status recognition model predicts in advance that the slurry is about to overflow. It then outputs a predictive anti-overflow control command and simultaneously executes a combined control strategy, reducing the real-time heating power to 65% of the rated power while maintaining a high stirring speed of 1600 rpm. Through the synergistic effect of cooling and foam suppression and high-speed foam breaking, the foam growth trend is suppressed in advance before it touches the lid and overflows, achieving lag-free, preventative anti-overflow control.

[0051] After a single batch of mixed grain soy milk is made, users can quantitatively evaluate the taste of the finished product via a mobile app. When a user reports that the texture is "too coarse," the cloud server automatically collects this feedback and initiates an incremental learning mechanism. By retrospectively analyzing the data from the entire soy milk making process, it was confirmed that the particle size distribution of the mixed grain soy milk was between 20 and 38 μm, indicating insufficient particle refinement. Based on long-term user taste preference data, the correlation parameters of the cooking state recognition model were fine-tuned, increasing the weight of the concentration gradient-beating speed correlation in the mixed grain soy milk making scenario by 0.15, providing a basis for parameter iteration for subsequent soy milk making process optimization.

[0052] When the user made the same mixed grain soy milk for the tenth time, the iteratively optimized model parameters were called to automatically increase the mixing intensity of the mixed grains to 1650rpm, which enhanced the particle refinement effect of the soy milk. Ultimately, the particle size of the finished soy milk was optimized to 15~35μm, and the fineness of the finished product was significantly improved. The user's taste score increased from 3.8 points before optimization to 4.2 points.

[0053] After long-term practical testing and verification by 100 groups of users, the adaptive predictive control scheme of this embodiment has been successfully implemented, resulting in a stable overflow rate of less than 2% and a burnt-on-bottom rate of less than 3%. This completely solves the problems of poor adaptability and frequent safety malfunctions in traditional soy milk makers. Simultaneously, overall user satisfaction with the taste has increased by 40%, and after more than 10 iterations of user data learning, the accuracy of predicting and adapting to personalized taste needs has increased from 65% to over 80%. This fully demonstrates that the technical solution proposed in this embodiment possesses excellent adaptability to operating conditions, stable safety control, and personalized iterative optimization capabilities.

[0054] Combination Figure 9 and Figure 10As shown in the architecture, the soybean milk maker's pulping control system in this embodiment is composed of a pulp sensing module, a control decision module, an execution module, and a user feedback module. These functional units can achieve bidirectional interaction of data and control commands, jointly constructing a fully closed-loop intelligent pulping system of "sensing-recognition-regulation-learning." Logically, the entire system can be divided into an equipment-side sensing layer, a time-series fusion decision layer, and a cloud-based incremental learning layer. Relying on a complete technical chain of multimodal operating condition acquisition, intelligent risk prediction, dynamic collaborative regulation, and iterative optimization based on user preferences, it achieves full-condition adaptive and precise control of the soybean milk maker's pulping process. The equipment-side perception layer relies on various high-precision sensing hardware to complete comprehensive real-time acquisition of pulping conditions. Among them, the TSL2561 optical turbidity sensor, with a working wavelength of 550nm, is vertically deployed on the side wall of the heating vessel. By collecting the transmitted light signal of the slurry, it accurately calculates the particle size distribution and concentration gradient parameters of the slurry particles. Combined with the physical characteristics of the ingredients, it completes the identification of the ingredient type. It can also link with the image detection unit to dynamically correct the foam rate judgment threshold. At the same time, the sensor can adaptively adjust the sampling frequency according to the instructions of the decision layer. The 1080P miniature image acquisition module mounted on the top of the vessel is equipped with the YOLOv5 lightweight semantic segmentation model, which can perform pixel-level accurate analysis of the foam area on the liquid surface and continuously acquire the temporal change data of foam height and foam rising rate. The high-precision temperature and pressure sensor integrated in the center of the bottom of the vessel can simultaneously collect the bottom temperature and the internal air pressure. The temperature detection accuracy can reach ±0.1℃ and the pressure detection accuracy can reach ±0.01kPa. All the multi-dimensional raw operating condition data collected by the sensing modules are uniformly uploaded to the control decision module, providing comprehensive and accurate data support for subsequent model inference and risk assessment.

[0055] As the core computational and decision-making hub of the system, the control decision module first uses a sliding window algorithm in the data preprocessing unit to smooth and filter time-series data, effectively eliminating acquisition noise and data fluctuations caused by instantaneous boiling of the slurry. The preprocessed three-dimensional feature vectors of concentration gradient, bottom temperature, and foam height are input into the CNN-LSTM fusion model inference engine to simultaneously mine the local spatial features of the pulping process and the temporal linkage change patterns in the long-term dimension, accurately outputting two core prediction results: the probability of bottom sludge accumulating and the early warning of excessive foam rise. After the risk feature parameters output by the model are sent to the dynamic decision logic unit, predictive anti-overflow control commands and zoned anti-sludge control commands can be generated and sent to the execution module. At the same time, real-time temperature and pressure data at the bottom of the slurry are continuously transmitted back to the decision logic unit, realizing dynamic correction and accurate judgment of risk thresholds. This system innovatively adopts a non-stop predictive control logic. The anti-overflow control process suppresses foam growth by slightly reducing the heating power and simultaneously increasing the stirring speed, maintaining the slurry in a boiling state throughout the process. The anti-scorching control adopts a temperature field-flow field coupling control strategy that combines bottom zone temperature control with high-intensity turbulence in the center of the slurry, avoiding the risk of scorching at the source. At the same time, the decision layer can send the ingredient adaptation parameters and sampling frequency adjustment instructions back to the sensing layer, and upload the basic parameters of equipment safety control to the cloud, realizing data linkage and interaction between the equipment and the cloud.

[0056] The execution module mainly includes a drive execution unit and a human-machine interaction unit, responsible for implementing various control commands issued by the decision-making level and completing user interactions. The drive execution unit is equipped with high-precision stepless control hardware. The IGBT stepless heating unit can respond to control commands quickly within 0.5 seconds, achieving fine-grained continuous power adjustment with a precision of ±5W, ensuring a stable and controllable pulping temperature field. The 0~1500rpm full-range stepless speed-regulating stirring motor, paired with a Hall sensor, can collect and feedback the actual motor speed in real time, achieving closed-loop precise control of stirring intensity, adapting to the refined pulping needs of different ingredients. The human-machine interaction unit can build a visual interactive interface based on mobile APP and other terminal devices, allowing users to pre-customize three parameters for soy milk smoothness (high, medium, and low) and their preferred concentration and taste. After the pulping process is completed, users can quantitatively rate the smoothness and concentration of the finished product from 1 to 5 points, forming standardized, traceable user feedback data, which is simultaneously uploaded to the cloud-based user feedback module.

[0057] The cloud-based user feedback module is equipped with an incremental learning algorithm based on the Adam optimizer, specifically designed to enable user-driven adaptive iterative optimization of the model. After receiving quantified user ratings, the system only iteratively updates weight parameters highly correlated with taste, such as the concentration gradient-whipping speed correlation coefficient, without requiring a complete retraining of the CNN-LSTM model, significantly reducing computational overhead. The optimized taste-related weight parameters are fed back in real-time to the CNN-LSTM fusion inference engine in the control decision layer, continuously improving the model's operational condition prediction accuracy and personalized taste adaptation capabilities. Simultaneously, the cloud distributes long-term iteratively optimized ingredient-specific pulping strategies to the equipment, enabling continuous adaptive upgrades to the pulping process. The entire system features a unique safety and personalized weight isolation mechanism. Model iterative optimization only fine-tunes taste parameters, while core safety control parameters such as anti-overflow and anti-burning remain stable. While achieving personalized taste adaptation and iterative optimization, it fully ensures the safety and stability of the equipment's pulping operation, ultimately forming an integrated intelligent pulping control system with multimodal accurate perception, intelligent risk prediction, dynamic closed-loop control, and autonomous iterative learning.

[0058] In summary, the key feature of this embodiment is the establishment of an integrated, fully closed-loop intelligent pulping control system encompassing "sensing-recognition-control-learning." This embodiment incorporates a dedicated multimodal sensing module adapted to the soybean milk pulping process. An optical turbidity sensor with a center wavelength of 550nm±10nm, vertically positioned on the side wall of the heating vessel, is combined with a miniature image acquisition module mounted downwards from the top of the vessel. A lightweight semantic segmentation model enables pixel-level precise segmentation of the foam. By detecting the intensity of transmitted light in the pulp and combining this with the turbidity-particle size mapping established through extensive prior calibration experiments, the particle size distribution and concentration gradient of the pulp within the 2-50μm range can be accurately inferred in real time. Simultaneously, a sliding window algorithm is used to extract the temporal variation characteristics of foam height and accurately calculate the foam rise rate, achieving comprehensive and high-precision quantitative sensing of the multidimensional state of the pulp.

[0059] Building upon this foundation, this embodiment introduces a pre-trained CNN-LSTM fusion model. Using a three-dimensional feature vector composed of concentration gradient, bottom temperature, and foam height as input, it simultaneously mines the spatial characteristics and temporal evolution patterns of the working conditions, accurately outputting predictions of overflow and scorching risks, and generating predictive anti-overflow and zoned anti-scorching coordinated control commands. The anti-overflow strategy dynamically adjusts the heating power and stirring speed based on the foam rise rate, while the anti-scorching strategy relies on the probability of bottom scorching risk to achieve zoned temperature control of the bottom heating element and coordinated control of the central slurry turbulence, thus constructing a real-time closed-loop control system for the pulping process.

[0060] Meanwhile, this embodiment also adds a user-evaluation-driven cloud-based incremental learning mechanism, abandoning the traditional fixed-parameter control mode. This embodiment can perform local iterative updates only on weighted parameters that are strongly related to taste, such as the concentration gradient-whipping speed correlation coefficient, without retraining the entire model. Under the premise of ensuring the stability and reliability of the equipment's safety control logic, it continuously iterates and optimizes the pulping process parameters, enabling the equipment to dynamically adapt to different user taste preferences. This truly achieves adaptive optimization and personalized customization of the pulping process, significantly improving the intelligent adaptive level of the soy milk maker.

[0061] Based on the above complete control scheme, this embodiment can specifically solve the relevant problems of existing household soy milk makers: First, it breaks free from the constraints of factory-fixed timing programs, can match the different physical properties of various ingredients, solves the problem of homogenized taste of finished products, and fully releases the internal nutrients of ingredients; Second, it abandons the passive detection and protection mode relying on electrodes and temperature sensors, and achieves adaptive control by relying on the dynamic threshold of foam rising rate, adapting to the fluctuation of water volume and foam fluctuations caused by different altitude air pressures, reducing overflow failures from the source; Third, it builds a closed-loop control architecture for real-time perception and timing prediction of multi-dimensional soy milk state, making up for the shortcomings of traditional open-loop control without real-time feedback, and intervening in advance for the problem of scorching caused by overheating of the bottom of the pot; Fourth, it establishes a user taste feedback iterative learning link, giving the equipment the ability to autonomously optimize process parameters, making up for the shortcomings of traditional equipment that cannot adapt to personalized drinking needs.

[0062] This embodiment relies on an integrated technical solution of multimodal fusion perception, temporal intelligent prediction, closed-loop dynamic control, and incremental learning iteration, achieving significant technological progress and application results. First, through precise monitoring of the all-dimensional soy milk state and proactive predictive anti-overflow and anti-scorching control, the overflow rate of the soy milk maker can be stably controlled below 2%, and the scorching rate can be reduced to below 3%. This completely solves the industry pain points of traditional soy milk makers, such as easy overflow, easy scorching, and uneven heating, and greatly improves the stability of equipment operation and the safety of use.

[0063] Secondly, this embodiment can dynamically adapt core process parameters such as heating power and stirring speed according to the differences in physical properties of different ingredients such as soybeans and grains, achieving adaptive pulping control for all types of ingredients. Precise dynamic stirring and emulsification control keep the particle size of soy milk stable within a uniform range of 2~50μm. Compared with traditional pulping methods, the finished product has a 40% smoother texture and a 25% higher release efficiency of nutrients such as protein and trace elements, significantly optimizing the quality of the finished soy milk.

[0064] Finally, the cloud-based incremental learning mechanism based on user feedback can continuously improve the accuracy of the device's taste prediction by more than 15%, achieving an adaptive iteration effect where the soy milk making effect "becomes more and more in line with the user's personalized taste the more it is used." This effectively solves the problem of traditional soy milk makers having a single taste and being unable to adapt to personalized needs, greatly improving user satisfaction and the product's core competitiveness in the market.

[0065] Figure 11 This is a schematic block diagram of a soymilk maker's pulping control device 1100 provided in an embodiment of the present invention. The device 1100 includes: The feature acquisition unit 1101 is used to acquire multi-dimensional feature vectors of the soy milk in the soy milk maker through a multi-modal sensing module after the soy milk maker starts the soy milk making operation; wherein, the multi-dimensional feature vectors include the concentration gradient of the soy milk, the foam height and the bottom temperature of the soy milk maker. The trend prediction unit 1102 is used to predict the trend of the slurry based on the multidimensional feature vector and using a pre-built cooking state recognition model to obtain the predicted trend of the slurry. The risk assessment unit 1103 is used to determine whether there is a risk of overflow of the slurry based on the predicted trend of the slurry, and to determine whether there is a risk of the soy milk maker burning at the bottom. The instruction generation unit 1104 is used to generate a predictive anti-overflow instruction when it is determined that there is a risk of the slurry overflowing; and to generate a predictive anti-sticking instruction when it is determined that there is a risk of the soy milk maker sticking to the bottom. The operation control unit 1105 is used to control the soy milk making operation of the soy milk maker based on the predictive anti-overflow command and / or the predictive anti-smear command.

[0066] In one embodiment, the soymilk maker is equipped with a heating pot, and the multimodal sensing module includes an optical turbidity sensor, an image acquisition module, and a temperature and pressure sensor, such as... Figure 12 As shown, the feature acquisition unit 1101 includes: The concentration acquisition unit 1201 is used to acquire the transmitted light intensity of the slurry using an optical turbidity sensor, obtain the corresponding turbidity based on the transmitted light intensity, and obtain the concentration gradient of the slurry based on a pre-established average particle size-turbidity mapping database. The height acquisition unit 1202 is used to acquire the foam area image of the slurry through the image acquisition module, and to segment the foam area image using a lightweight semantic segmentation model in order to extract the foam height of the foam area image. Temperature acquisition unit 1203 is used to acquire the bottom temperature of the heating vessel through a temperature and pressure sensor.

[0067] In one embodiment, such as Figure 13 As shown, the trend prediction unit 1102 includes: The feature input unit 1301 is used to input the multidimensional feature vector into the cooking state recognition model; wherein, the cooking state recognition model is a CNN-LSTM fusion model; Pattern extraction unit 1302 is used to extract local spatial change patterns from the multidimensional feature vector using the CNN layer in the CNN-LSTM fusion model; The relationship establishment unit 1303 is used to establish a temporal linkage relationship between the multidimensional feature vectors using the LSTM layer in the CNN-LSTM fusion model, and to learn the early warning patterns of overflow and burnt bottom accidents. The prediction setting unit 1304 is used to combine the local spatial change mode and the precursor mode to output the foam rising rate and the probability value of the risk of sludge sticking to the bottom, and to use the foam rising rate and the probability value of the risk of sludge sticking to the bottom as the predicted trend of the slurry; wherein, the probability value of the risk of sludge sticking to the bottom is obtained based on the concentration gradient of the slurry and the rising rate of the bottom temperature of the vessel.

[0068] In one embodiment, such as Figure 14 As shown, the risk assessment unit 1103 includes: The overflow determination unit 1401 is used to compare the foam rising rate of the slurry with a preset dynamic threshold, and determine that the slurry has an overflow risk when the foam rising rate of the slurry reaches the preset dynamic threshold. The bottom-sticking determination unit 1402 is used to compare the bottom-sticking risk probability value with a preset bottom-sticking threshold, and determine that the soymilk maker has a bottom-sticking risk when the bottom-sticking risk probability value reaches the preset bottom-sticking threshold.

[0069] In one embodiment, such as Figure 15 As shown, the operation control unit 1105 includes: The first invocation unit 1501 is used to invoke a preset anti-overflow adjustment strategy based on the predictive anti-overflow instruction; The first control unit 1502 is used to control the soymilk maker's soymilk-making operation according to the anti-overflow adjustment strategy; wherein the anti-overflow adjustment strategy includes reducing the heating power of the soymilk maker and increasing the stirring speed of the soymilk maker.

[0070] In one embodiment, such as Figure 16 As shown, the operation control unit 1105 further includes: The second invocation unit 1601 is used to invoke a preset anti-blurring adjustment strategy based on the predictive anti-blurring instruction. The second control unit 1602 is used to control the soymilk making operation of the soymilk maker according to the anti-smearing adjustment strategy; wherein the anti-smearing adjustment strategy includes reducing the heating power of the outer perimeter of the bottom of the soymilk maker and enhancing the central turbulence intensity of the soymilk maker.

[0071] In one embodiment, the soy milk maker's pulping control device 1100 further includes: The feedback acquisition unit is used to acquire user feedback data; The parameter update unit is used to update the parameters of the cooking state recognition model using an optimizer based on the user feedback data.

[0072] Since the embodiments of the apparatus and the embodiments of the method correspond to each other, please refer to the description of the embodiments of the method for the embodiments of the apparatus, which will not be repeated here.

[0073] Please see Figure 17 The present invention also provides a soymilk maker 1700, which is a device with both wireless and wired communication capabilities.

[0074] The soymilk maker 1700 includes a processor 1702, a memory, and a network interface 1705 connected via a system bus 1701. The memory may include a non-volatile storage medium 1703 and internal memory 1704.

[0075] The non-volatile storage medium 1703 can store an operating system 17031 and a computer program 17032. When the computer program 17032 is executed, it causes the processor 1702 to execute a soy milk making control method.

[0076] The processor 1702 provides computing and control capabilities to support the operation of the entire soymilk maker 1700.

[0077] The internal memory 1704 provides an environment for the execution of the computer program 17032 in the non-volatile storage medium 1703. When the computer program 17032 is executed by the processor 1702, the processor 1702 can execute a soy milk making control method.

[0078] This network interface 1705 is used for network communication with other devices. Those skilled in the art will understand that... Figure 17 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the soymilk maker 1700 to which the present invention is applied. The specific soymilk maker 1700 may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0079] The processor 1702 is used to run a computer program 17032 stored in a memory to implement any embodiment of the soy milk maker's control method described above.

[0080] It should be understood that, in this embodiment of the invention, the processor 1702 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0081] It will be understood by those skilled in the art that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program may be stored in a storage medium, which is a computer-readable storage medium. The computer program is executed by at least one processor in the computer system to implement the process steps of the embodiments of the above methods.

[0082] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed, can perform the steps provided in the above embodiments. The storage medium may include various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0083] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to in the method section. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of the claims of this application.

[0084] It should also be noted that, in this specification, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A method for controlling the pulping process in a soy milk machine, characterized in that, include: After the soymilk maker starts the soymilk making operation, a multi-dimensional feature vector of the soymilk liquid in the soymilk maker is collected through a multi-modal sensing module; wherein, the multi-dimensional feature vector includes the concentration gradient of the soymilk liquid, the foam height and the bottom temperature of the soymilk maker. Based on the multidimensional feature vector, the trend of the slurry is predicted using a pre-built cooking state recognition model to obtain the predicted trend of the slurry. Based on the predicted trend of the slurry, determine whether there is a risk of the slurry overflowing, and determine whether there is a risk of the soy milk maker burning at the bottom; When it is determined that there is a risk of overflow of the slurry, a predictive anti-overflow command is generated; and when it is determined that there is a risk of scorching at the bottom of the soy milk maker, a predictive anti-scorching command is generated. The soymilk making operation of the soymilk maker is controlled based on the predictive anti-overflow command and / or the predictive anti-smear command.

2. The soybean milk making control method according to claim 1, characterized in that, The soymilk maker is equipped with a heating pot. The multimodal sensing module includes an optical turbidity sensor, an image acquisition module, and a temperature and pressure sensor. The acquisition of multidimensional feature vectors of the soymilk liquid in the soymilk maker via the multimodal sensing module includes: The intensity of transmitted light in the slurry is collected using an optical turbidity sensor. The corresponding turbidity is obtained based on the intensity of transmitted light, and the concentration gradient of the slurry is obtained based on a pre-established average particle size-turbidity mapping database. The foam area image of the slurry is acquired by the image acquisition module, and the foam area image is segmented using a lightweight semantic segmentation model to extract the foam height of the foam area image. The temperature at the bottom of the heating vessel is collected using a temperature and pressure sensor.

3. The soybean milk making control method according to claim 1, characterized in that, The step of predicting the trend of the slurry based on the multidimensional feature vector and using a pre-built cooking state recognition model to obtain the predicted trend of the slurry includes: The multidimensional feature vector is input into the cooking state recognition model; wherein, the cooking state recognition model is a CNN-LSTM fusion model; The CNN layer in the CNN-LSTM fusion model is used to extract local spatial variation patterns from the multidimensional feature vector; The LSTM layer in the CNN-LSTM fusion model is used to establish a temporal linkage relationship between the multidimensional feature vectors, and the early warning patterns of overflow and burnt bottom accidents are learned. The foam rising rate and the probability value of bottom sludge risk are output by combining the local spatial change mode and the precursor mode, and the foam rising rate and the probability value of bottom sludge risk are used as the predicted trend of the slurry; wherein, the probability value of bottom sludge risk is obtained based on the concentration gradient of the slurry and the rising rate of the bottom temperature.

4. The soybean milk making control method according to claim 3, characterized in that, The steps of determining whether there is a risk of overflow of the slurry based on the predicted trend of the slurry, and determining whether there is a risk of the soy milk maker burning at the bottom, include: The foam rising rate of the slurry is compared with a preset dynamic threshold, and when the foam rising rate of the slurry reaches the preset dynamic threshold, it is determined that the slurry has a risk of overflow. The probability value of the risk of sticking to the bottom is compared with a preset threshold for sticking to the bottom, and when the probability value of the risk of sticking to the bottom reaches the preset threshold for sticking to the bottom, it is determined that the soymilk maker has a risk of sticking to the bottom.

5. The soybean milk making and pulping control method according to claim 1, characterized in that, The control of the soymilk maker's soymilk-making operation based on the predictive anti-overflow command and / or the predictive anti-smear command includes: Based on the predictive anti-overflow command, a preset anti-overflow adjustment strategy is invoked; The soymilk maker's soymilk-making operation is controlled according to the anti-overflow adjustment strategy; wherein, the anti-overflow adjustment strategy includes reducing the heating power of the soymilk maker and increasing the blending speed of the soymilk maker.

6. The soybean milk making control method according to claim 1, characterized in that, The method of controlling the soymilk maker's soymilk-making operation based on the predictive anti-overflow command and / or the predictive anti-smear command further includes: Based on the predictive anti-blurring command, a preset anti-blurring adjustment strategy is invoked; The soymilk maker's soymilk-making operation is controlled according to the anti-smearing adjustment strategy; wherein, the anti-smearing adjustment strategy includes reducing the heating power of the outer perimeter of the soymilk maker's pot and enhancing the central turbulence intensity of the soymilk maker.

7. The soybean milk making control method according to claim 1, characterized in that, Also includes: Obtain user feedback data; Based on the user feedback data, an optimizer is used to update the parameters of the cooking state recognition model.

8. A soybean milk making machine control device, characterized in that, include: The feature acquisition unit is used to acquire multi-dimensional feature vectors of the soy milk maker through a multi-modal sensing module after the soy milk maker starts the soy milk making operation; wherein, the multi-dimensional feature vectors include the concentration gradient of the soy milk maker, the foam height and the bottom temperature of the soy milk maker. The trend prediction unit is used to predict the trend of the slurry based on the multidimensional feature vector and using a pre-built cooking state recognition model to obtain the predicted trend of the slurry. The risk assessment unit is used to determine whether there is a risk of overflow of the slurry based on the predicted trend of the slurry, and to determine whether there is a risk of the soy milk maker burning at the bottom. The instruction generation unit is used to generate a predictive anti-overflow instruction when it is determined that there is a risk of the slurry overflowing; and to generate a predictive anti-sticking instruction when it is determined that there is a risk of the soy milk maker sticking to the bottom. An operation control unit is used to control the soymilk making operation of the soymilk maker based on the predictive anti-overflow command and / or the predictive anti-smear command.

9. A soy milk maker, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the soy milk maker's soy milk control method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the soy milk making control method as described in any one of claims 1 to 7.