Motor control method for chef machine

Abstract

A motor control method for a chef machine relates to the technical field of motor control. The method comprises the following steps: acquiring motor load data and vibration data in a dough stirring process; based on the motor load data and the vibration data, the state of the ice block in the dough stirring process is recognized, and the state comprises an ice block bottom sinking stage, an ice block wall throwing stage and a crushed ice wrapping stage; correspondingly implementing different motor control strategies according to the identified ice block state: in an ice block bottom sinking stage, adopting a first motor control strategy to prevent the ice blocks from continuously sinking to promote uniform hydration of flour at the bottom; in the ice block wall throwing stage, a second motor control strategy is adopted, so that the ice block crushing process is controlled, and motor load fluctuation is relieved; and in the crushed ice coercing stage, a third motor control strategy is adopted, so that crushed ice is controlled to be uniformly distributed, and excessive stirring is avoided. According to the motor control method for the chef machine provided by the invention, the dough can be better stirred.

Description

Technical Field

[0001] This application relates to the technical field of motor control, and particularly to a motor control method for a chef machine. Background Art

[0002] An important function of a household chef machine is to stir dough, and the quality of this function directly affects the product quality. The motors used in existing chef machines generally have multiple stirring speed gears, and each gear can drive the dough to rotate and stir at a fixed speed.

[0003] To control the dough temperature and prevent the dough from fermenting prematurely during stirring, users generally add ice water containing ice cubes when mixing water and flour.

[0004] However, due to the changes in ice cubes during this stirring process, there are still some problems in the existing technology, resulting in uneven dough stirring. Summary of the Invention

[0005] To solve the above technical problems or at least partially solve the above technical problems, this application provides a motor control method for a chef machine, which can make the effect of stirring dough by a household chef machine better.

[0006] This application provides a motor control method for a chef machine. The motor control method for a chef machine includes the following steps: Obtain the motor load data and vibration data during the dough stirring process; based on the motor load data and vibration data, identify the state of the ice cubes during the dough stirring process, including the ice cube sinking stage, the ice cube hitting the wall stage, and the crushed ice entrainment stage; According to the identified ice cube state, implement different motor control strategies correspondingly: In the ice cube sinking stage, adopt the first motor control strategy to prevent the ice cubes from continuously sinking to promote uniform hydration of the bottom flour; In the ice cube hitting the wall stage, adopt the second motor control strategy to control the ice cube crushing process and reduce the motor load fluctuation; In the crushed ice entrainment stage, adopt the third motor control strategy to control the uniform distribution of the crushed ice and avoid over-stirring.

[0007] Optionally, the motor control method for a chef machine further includes: Predict the ice cube crushing data in the ice cube hitting the wall stage; According to the predicted ice cube crushing data, adjust the signal recognition threshold in the crushed ice entrainment stage; Based on the adjusted signal recognition threshold, detect the current crushed ice entrainment state to adjust the third motor control strategy, so as to avoid over-stirring while ensuring the uniform distribution of the crushed ice.

[0008] Optionally, the first motor control strategy, the second motor control strategy, and the third motor control strategy are output through a preset motor control model of a cooking machine; The motor control model of the cooking machine includes: An input layer for receiving motor load data and vibration data; A feature extraction layer for receiving the data of the input layer and processing the data; A stage discrimination layer for receiving the output of the feature extraction layer to identify the stage where the ice cubes are in the dough mixing process; A pre-stage information memory layer for predicting and recording ice cube breakage prediction data according to the data of the feature extraction layer during the ice cube wall-slapping stage; A threshold adjustment layer for dynamically adjusting the signal recognition threshold in the ice cube entrainment stage according to the ice cube breakage prediction data stored in the pre-stage information memory layer; A control strategy layer for obtaining corresponding motor control parameters according to the identified ice cube stage using a corresponding expert neural network based on the output of the feature extraction layer; Wherein, in the ice cube entrainment stage, the ice cube entrainment intensity is also identified based on the adjusted signal recognition threshold, and the obtained motor control parameters are adjusted accordingly; An output layer for converting the motor control parameters in different stages into motor drive commands, thereby obtaining the first motor control strategy, the second motor control strategy, and the third motor control strategy respectively.

[0009] Optionally, the feature extraction layer includes: A time-frequency preprocessing unit for performing short-time Fourier transform on the motor load data and vibration data to generate a time-frequency spectrogram; A shared feature extraction unit, adopting a one-dimensional residual convolutional network structure, extracts cross-channel temporal features from the time-frequency spectrogram to form a shared feature vector.

[0010] Optionally, the stage discrimination layer adopts a Transformer structure for receiving the shared feature vector and outputting a stage probability vector characterizing the ice cube bottoming stage, the ice cube wall-slapping stage, and the ice cube entrainment stage.

[0011] Optionally, the pre-stage information memory layer adopts a fully connected neural network + LSTM structure, which is activated when it is determined to be the ice cube wall-slapping stage, and predicts the following indicators according to the shared feature vector in the ice cube wall-slapping stage: The size of the ice cube when it breaks; The degree of breakage of the ice cube.

[0012] Optionally, the threshold adjustment layer adopts a fully connected neural network structure. By receiving the size and degree of ice cube breakage predicted by the previous information memory layer, it outputs the motor load identification threshold and vibration identification threshold required for the crushed ice entrainment stage.

[0013] Optionally, the control strategy layer includes: A stage selection unit, adopting a gating logic structure, based on the stage probability vector output by the stage discrimination layer, activates the expert neural network corresponding to the ice cube stage with the highest current stage probability from the expert network unit in a hard gating manner; The expert network unit includes three groups of expert neural networks independently corresponding to the ice cube bottoming stage, the ice cube wall-slapping stage, and the crushed ice entrainment stage respectively, where: Both the expert neural network for the ice cube bottoming stage and the expert neural network for the ice cube wall-slapping stage adopt a fully connected neural network structure, take the shared feature vector output by the feature extraction layer as the input, and output the motor control parameters for the corresponding stage to the output layer; The expert neural network for the crushed ice entrainment stage includes a shallow convolutional autoencoder and a fully connected neural network structure. The shallow convolutional autoencoder is used to compress and reconstruct the input shared feature vector to obtain the abnormal residual features representing the crushed ice signal; Compare the abnormal residual features with the motor load identification threshold and vibration identification threshold provided by the threshold adjustment layer in real time, so as to quantify the current crushed ice entrainment intensity, and use the quantified crushed ice entrainment intensity and the shared feature vector together as the input of the expert neural network for the crushed ice entrainment stage, and output the motor control parameters for the crushed ice entrainment stage to the output layer.

[0014] Optionally, the output layer includes: A parameter safety domain trimming unit, which is used to receive the control parameters output by the control strategy layer and trim the control parameters within the safety domain range to ensure that the output control instruction is within the safe operation range of the chef machine motor; A control instruction generation unit, which is used to convert the control parameters trimmed by the safety domain into a chef machine motor drive instruction, and the chef machine motor drive instruction at least includes a target motor rotation time, a target motor speed instruction, and a target motor torque instruction.

[0015] The technical solution provided by this application has the following advantages compared with the prior art: One of its working principles and beneficial effects is that during the process of the chef machine kneading dough, users usually add water containing ice cubes to avoid the dough from fermenting prematurely during the kneading process by reducing the temperature of the dough. Due to the addition of ice cubes, during the dough kneading process, the state of the ice cubes in the kneading process will go through three stages: bottoming, wall-slapping, and crushed ice entrainment. Different stages have different effects on dough kneading. For example: When mixing flour and water at a low speed, due to the self-weight of the ice cubes, the ice cubes will gather at the bottom under the action of agitation to form a local low-temperature area, entering the bottom-settling stage.

[0016] The low temperature will inhibit the hydration of the flour. At this time, the ice cubes are large and smooth, and it is difficult to turn them over at a slow speed, thus preventing the flour under the ice cubes from turning over, resulting in insufficient hydration of the bottom flour and forming hard lumps at the bottom. These lumps will also cause uneven dough after subsequent mixing; After the initial mixing of flour and water, the speed of the stirring paddle will increase, and the ice cubes will be thrown towards the inner wall of the container due to centrifugal force, and the outer edge of the stirring paddle will randomly hit the ice cubes. If the motor adopts dynamic torque output, during the process of the stirring paddle hitting and knocking away the ice cubes, the motor torque will frequently jump between "high torque - low torque", resulting in a large fluctuation in the motor load, bringing fatigue to mechanical components and affecting the stability of the equipment; The ice cubes will be broken by the paddle blades during the wall-throwing stage, and the formed broken ice will be wrapped by the formed dough, and at this time, it enters the broken-ice entrapment stage.

[0017] If sufficient stirring is not carried out, local water spots will appear after the wrapped broken ice melts, which will affect the uniformity of the dough. Therefore, generally, the stirring time needs to be extended, but this will also cause the dough to be over-stirred, resulting in the gluten structure not conforming to the expectation or consuming more energy.

[0018] This application proposes a motor control method for a chef machine, which can identify different stages of ice cubes during the dough mixing process and adopt different motor control strategies for each stage. During the ice cube bottom-settling stage, this method adopts the first motor control strategy, and by reasonably adjusting the motor parameters, it prevents the ice cubes from settling at the bottom for a long time, promotes the full contact between the bottom flour and water, and realizes uniform hydration; when the ice cubes enter the wall-throwing stage, the system switches to the second motor control strategy to relieve the severe fluctuation of the motor torque output caused by hitting the ice cubes in this stage; finally, in the broken-ice entrapment stage, the third motor control strategy is started, and through an intelligent algorithm, it controls the uniform distribution of the broken ice in the dough and avoids over-stirring resulting in the destruction of the gluten structure.

[0019] Therefore, the motor control method for a chef machine provided by this application can better stir the dough.

[0020] The second working principle and its beneficial effect are that during the broken-ice entrapment stage, since the broken ice is wrapped by the dough and highly overlaps with the stirring resistance signal of the dough, it is difficult for traditional motor control methods to accurately identify the current degree of broken-ice entrapment at this stage, resulting in the inability to adjust the operation of the motor according to the current degree of ice cube breakage and the current state of the dough, and thus control the stirring time of the dough well.

[0021] In this application, by introducing a pre - information memory layer and a dynamic threshold adjustment mechanism, when the ice cubes enter the wall - throwing stage, the motor control system will predict and record the data of ice cube breakage through the pre - information memory layer. When the ice cube characteristics signals can be captured well in the early stage, the data of ice cube breakage predicted and recorded are used as a reference for determining the intensity of broken - ice entrainment in the subsequent broken - ice entrainment stage. By making corresponding judgments at the initial stage of ice cube breakage, a more accurate basis for signal recognition in the later broken - ice entrainment stage is provided, thus avoiding the problem in traditional methods that it is difficult to determine the intensity of broken - ice entrainment due to inaccurate separation threshold settings, resulting in insufficient or excessive dough mixing.

[0022] Therefore, the motor control method for a cook machine provided by this application can better mix the dough. Description of the Drawings

[0023] Figure 1 It is a schematic flowchart of the motor control method for a cook machine provided by an embodiment of this application. Detailed Embodiments

[0024] Next, the technical solutions in this application will be described in conjunction with the drawings.

[0025] Many specific details are set forth in the following description to facilitate a thorough understanding of this application, but this application may be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only a part of the embodiments of this application, rather than all of the embodiments. It should be noted that, without conflict, the embodiments of this application and the features in the embodiments may be combined with each other.

[0026] An embodiment of this application provides a motor control method for a cook machine, and the motor control method for a cook machine includes the following steps: S101: Obtain the motor load data and vibration data during the dough mixing process.

[0027] Specifically, in the embodiment of this application, the motor load data can be used to monitor the load condition of the motor in real - time through a current sensor installed in the motor control system.

[0028] The vibration data can be obtained by acquiring the vibration acceleration in real - time through a three - axis acceleration sensor coaxially installed with the motor.

[0029] S102: Based on the motor load data and vibration data, identify the state of the ice cubes during the dough mixing process, including the ice - cube sinking stage, the ice - cube wall - throwing stage, and the broken - ice entrainment stage.

[0030] S103: According to the identified ice - cube state, implement different motor control strategies correspondingly: During the ice sinking stage, a first motor control strategy is adopted to prevent the ice from continuously sinking to promote uniform hydration of the bottom flour; During the ice wall-slapping stage, a second motor control strategy is adopted to control the ice crushing process and reduce the motor load fluctuation; During the crushed ice entrainment stage, a third motor control strategy is adopted to control the uniform distribution of the crushed ice and avoid overmixing.

[0031] Specifically, the motor control method for the kitchen machine further includes: During the ice wall-slapping stage, predict the ice crushing data; According to the predicted ice crushing data, adjust the signal recognition threshold in the crushed ice entrainment stage; Based on the adjusted signal recognition threshold, detect the current crushed ice entrainment state to adjust the third motor control strategy, so as to avoid overmixing while ensuring the uniform distribution of the crushed ice.

[0032] In the embodiment of the present application, the steps of S102 - S103 are specifically implemented by a kitchen machine motor control model constructed as follows: The kitchen machine motor control model includes: An input layer for receiving motor load data and vibration data; A feature extraction layer for receiving the data from the input layer and processing the data; Specifically, the feature extraction layer includes: A time-frequency preprocessing unit for performing short-time Fourier transform on the motor load data and vibration data to generate a time-frequency spectrogram.

[0033] Specifically, the processing flow of the time-frequency preprocessing unit includes: Data segmentation and preprocessing, segmenting the continuously collected motor load data and vibration data according to a fixed time window (window size: 256 ms, step size: 64 ms); Multi-channel synchronous processing, simultaneously performing short-time Fourier transform on the motor load data (single channel) and vibration data (three-axis acceleration, a total of three channels) to form a four-channel time-frequency representation; Noise suppression, applying adaptive threshold filtering to the obtained time-frequency spectrogram to remove environmental interference and the inherent noise components generated by the motor operation itself; Band-selective enhancement, selectively enhancing the characteristic frequency bands generated by the ice in different states (ice sinking: 5 - 30 Hz, ice wall-slapping: 50 - 200 Hz, crushed ice entrainment: 20 - 80 Hz) to improve the saliency of the features; Time-frequency feature normalization, respectively performing amplitude normalization processing on the time-frequency spectrograms of different channels to make the amplitude ranges of the data in each channel consistent.

[0034] The shared feature extraction unit adopts a one-dimensional residual convolutional network structure to extract cross-channel temporal features from the time-frequency spectrogram and form a shared feature vector.

[0035] Specifically, the one-dimensional residual convolutional network structure includes: An input layer for receiving the normalized multi-channel time-frequency spectrogram; A channel fusion layer that receives the output of the input layer, uses 1×1 convolutional kernels to preliminarily fuse the time-frequency information of four channels, and outputs 32 feature channels; A residual block stacking layer that receives the output of the channel fusion layer, contains 5 consecutive residual blocks, each residual block consists of two one-dimensional convolutional layers (with convolutional kernel sizes of 3 and 5 respectively), a batch normalization layer, and a ReLU activation function, and has a skip connection structure to alleviate the problem of gradient disappearance; A global pooling layer that receives the output of the residual stacking layer and extracts global temporal information by means of adaptive average pooling; A feature dimension unification layer that receives the output of the global pooling layer and maps the extracted features to a 128-dimensional feature vector space through a fully connected layer to form the final shared feature vector.

[0036] A stage discrimination layer for receiving the output of the feature extraction layer to identify the stage of the ice cube during the dough mixing process; Specifically, the stage discrimination layer adopts a Transformer structure for receiving the shared feature vector and outputting a stage probability vector representing the ice cube sinking to the bottom stage, the ice cube hitting the wall stage, and the crushed ice entrainment stage.

[0037] In a specific implementation, the Transformer structure includes: A multi-head self-attention mechanism unit for capturing long-term dependencies in the shared feature vector, where 4 attention heads are set, and the dimension of each attention head is 64. By calculating the correlations between different time steps in the feature sequence, the perception ability of the continuous change process of the ice cube state is enhanced; A feed-forward neural network unit that contains two fully connected layers. The first layer uses 256 neurons and is equipped with a ReLU activation function, and the second layer uses 128 neurons for non-linearly transforming the features extracted by the multi-head self-attention mechanism; A layer normalization unit applied to the outputs of the multi-head self-attention mechanism unit and the feed-forward neural network unit; A residual connection unit for retaining the original feature information and preventing the problem of gradient disappearance in the deep network; The temporal encoding unit adds positional encoding to enable the model to distinguish features at different time steps, enhancing the ability to recognize changes in the ice cube state during the dough mixing process; The classification head unit consists of a fully connected layer with 3 neurons corresponding to three ice cube states, and outputs a vector representing the probabilities of the three states through the softmax function.

[0038] The training process of the stage discriminant layer includes: Collect an annotated dataset containing motor load data and vibration data during the dough mixing process of different users under various conditions, as well as the corresponding ice cube state annotations (sinking stage, wall-slapping stage, or ice cube entrainment stage); Use the cross-entropy loss function as the optimization objective, train with the Adam optimizer, set the initial learning rate to 0.001, and adopt a learning rate decay strategy to train the Transformer structure of the stage discriminant layer.

[0039] The pre-information memory layer is used to predict and record ice cube breakage prediction data according to the data of the feature extraction layer during the ice cube wall-slapping stage; Specifically, the pre-information memory layer adopts a fully connected neural network + LSTM structure, is activated when it is determined to be the ice cube wall-slapping stage, and predicts the following indicators according to the shared feature vector in the ice cube wall-slapping stage: The size of the ice cube when it breaks; The degree of ice cube breakage.

[0040] Specifically, the specific structure of the pre-information memory layer is as follows: The selective activation layer is activated when the probability of the ice cube wall-slapping stage in the output of the stage discriminant layer exceeds the preset activation threshold. The pre-information memory layer starts to receive the shared feature vector and perform the prediction task; The multi-layer perceptron structure is composed of multiple fully connected layers with a hierarchical structure of 128-256-128-64 units. The LeakyReLU activation function is used between each layer to receive the shared feature vector, process the data and then output; Temporal feature aggregation uses long short-term memory network (LSTM) units with a capacity of 64 to aggregate the feature information of multiple consecutive time windows within the wall-slapping stage, capturing the dynamic characteristic changes when the ice cube is impacted; The dual-task output head is divided into two parallel branches: The ice cube size prediction branch: used to output the estimated average size of the ice cube when it breaks; The breakage degree prediction branch: used to divide the ice cube breakage degree into 3 levels (sufficient, moderate, insufficient).

[0041] A threshold adjustment layer for dynamically adjusting the signal recognition threshold in the ice crushing entrainment stage according to the ice crushing prediction data stored in the previous information memory layer; Specifically, the threshold adjustment layer adopts a fully connected neural network structure. By receiving the size of the ice block when it breaks and the degree of ice block breakage predicted by the previous information memory layer, it outputs the motor load recognition threshold and vibration recognition threshold required in the ice crushing entrainment stage.

[0042] In the embodiment of the present application, the threshold adjustment layer adopts a dual-channel fully connected neural network structure, including: An input normalization unit that performs Min-Max normalization processing on the ice block size and breakage degree data respectively, and maps them to the 0-1 interval; A feature extraction dual-channel structure, where the size feature channel includes two layers of fully connected networks with 32 nodes, and the breakage degree feature channel includes two layers of fully connected networks from 16 to 32 nodes. The two channels extract relevant feature information respectively; A threshold generation unit that fuses the two-way features through a layer of fully connected network and generates the motor load recognition threshold and vibration recognition threshold.

[0043] A control strategy layer for obtaining corresponding motor control parameters according to the output of the feature extraction layer using the corresponding expert neural network according to the recognized ice block stage; Wherein in the ice crushing entrainment stage, based on the adjusted signal recognition threshold, the ice crushing entrainment intensity is recognized, and the obtained motor control parameters are adjusted accordingly; Specifically, the control strategy layer includes: A stage selection unit that adopts a gated logic structure. Based on the stage probability vector output by the stage discrimination layer, it activates the expert neural network corresponding to the ice block stage with the highest current stage probability in a hard gating manner from the expert network unit; The expert network unit includes three groups of expert neural networks corresponding independently to the ice block bottoming stage, the ice block wall-slapping stage, and the ice crushing entrainment stage respectively, where: Both the ice block bottoming stage expert neural network and the ice block wall-slapping stage expert neural network adopt a fully connected neural network structure. Taking the shared feature vector output by the feature extraction layer as the input, they output the motor control parameters of the corresponding stage to the output layer; Among them, the ice block bottoming stage expert neural network is specially trained to generate the first motor control instruction. The first motor control strategy is to adopt a variable-speed stirring mode in the ice block bottoming stage and periodically adjust the motor speed and steering switch.

[0044] The ice block bottoming stage expert neural network determines the specific values within the following motor parameter limit range, that is, the bracket range, according to the actually detected motor load data and vibration data: Base speed (50 - 80) RPM; The speed is increased for (0.5 - 1) second every (5 - 8) seconds (the increase amplitude is 30% - 50% of the base speed); After rotating in one direction for (3 - 5) seconds, it reverses; The value of torque output (70% - 85% of the rated torque).

[0045] Among them, the expert neural network in the ice throwing stage is specially trained to generate the second motor control instruction, and the second motor control strategy is to adopt an adaptive torque control mode with smooth torque change in the ice throwing stage, that is, when encountering a situation where the motor load increases and torque needs to be increased, the torque is increased slowly.

[0046] The expert neural network in the ice throwing stage determines the specific values within the following motor parameter limits according to the actually detected motor load data and vibration data: Base speed (100 - 150) RPM; Torque change limit (20% - 40% of the rated torque); Torque smoothing coefficient, that is, the change rate of torque (0.6 - 0.8).

[0047] The expert neural network in the ice crushing stage includes a shallow convolutional autoencoder and a fully connected neural network structure. The shallow convolutional autoencoder is used to compress and reconstruct the input shared feature vector to obtain the abnormal residual features representing the ice crushing signal; The abnormal residual features are respectively compared with the motor load recognition threshold and the vibration recognition threshold provided by the threshold adjustment layer in real time, so as to quantify the current ice crushing intensity.

[0048] Specifically, the shallow convolutional autoencoder includes an encoder part and a decoder part: The encoder part consists of two one-dimensional convolutional layers. The first layer uses 16 convolutional kernels with a kernel size of 3 and a stride of 1. The second layer uses 8 convolutional kernels with a kernel size of 3 and a stride of 2. Each layer is followed by a batch normalization layer and a ReLU activation function to compress the input 128-dimensional shared feature vector into a 32-dimensional latent representation; The decoder part consists of two one-dimensional transposed convolutional layers. The first layer uses 16 convolutional kernels with a kernel size of 3 and a stride of 2. The second layer uses convolutional kernels with the same number of channels as the input features, with a kernel size of 3 and a stride of 1. Each layer is followed by a batch normalization layer and a ReLU activation function to reconstruct the 32-dimensional latent representation into a feature vector with the same dimension as the original input; The autoencoder uses normal dough mixing data (without ice chips entrainment) for unsupervised learning during the training phase, and uses the mean squared error loss function to optimize the reconstruction error, enabling the network to learn the feature distribution of normal dough signals; When the input contains the feature vector of the ice chips entrainment signal, since the network only learns the feature distribution of normal dough, an error will occur during the reconstruction process, and this error is the abnormal residual feature RE(t).

[0049] Then, based on prior knowledge, the subspace division of the shared feature vector is obtained. For example, in the embodiment of the present application, the first 32 dimensions of the shared feature vector correspond to the motor load feature, and the last 96 dimensions are the vibration features.

[0050] The subspace division needs to be based on the characteristics of the one-dimensional residual convolution network structure in the shared feature extraction unit, because this network retains the basic correspondence of the original signal channels when extracting features.

[0051] Process the abnormal residual feature RE(t) according to the subspace division ratio of the reconstruction error: Motor load abnormal residual: RL(t) = ||RE(t)[0:32]||², that is, the sum of the squares of the reconstruction errors of the first 32 dimensions; Vibration abnormal residual: RV(t) = ||RE(t)[32:128]||², that is, the sum of the squares of the reconstruction errors of the last 64 dimensions.

[0052] After obtaining the motor load abnormal residual and the vibration abnormal residual, the ice chips entrainment intensity can be calculated through the following formula. The quantization calculation method of the ice chips entrainment intensity S(t) is: S(t) = RL(t) / TL + RV(t) / TV Where TL and TV are the motor load recognition threshold and the vibration recognition threshold provided by the threshold adjustment layer respectively; Take the quantized ice chips entrainment intensity and the shared feature vector as the input of the expert neural network in the ice chips entrainment stage, and output the motor control parameters in the ice chips entrainment stage to the output layer.

[0053] Among them, the expert neural network in the ice chips entrainment stage is specially trained to generate the third motor control instruction, and the third motor control strategy is to dynamically adjust the mixing duration in the ice chips entrainment stage.

[0054] The expert neural network in the ice chips entrainment stage determines the specific values within the following motor parameter limits according to the actually detected motor load data, vibration data and ice chips entrainment intensity: Base speed (120 - 180) RPM; Mixing duration (60 - 360) seconds.

[0055] An output layer, configured to convert motor control parameters at different stages into motor drive instructions, thereby obtaining a first motor control strategy, a second motor control strategy, and a third motor control strategy respectively.

[0056] Specifically, the output layer includes: A parameter safety domain clipping unit, configured to receive the control parameters output by the control strategy layer and clip the control parameters within the safety domain range to ensure that the output control instructions are within the safe operating range of the chef machine motor; A control instruction generation unit, configured to convert the control parameters after being clipped by the safety domain into a chef machine motor drive instruction, where the chef machine motor drive instruction at least includes a target motor rotation time, a target motor speed instruction, and a target motor torque instruction.

[0057] The technical solution provided by the embodiments of the present application has the following advantages compared with the prior art: One of its working principles and beneficial effects is that during the process of the chef machine mixing dough, users usually add water containing ice cubes to avoid premature fermentation of the dough during the mixing process by reducing the temperature of the dough. Due to the addition of ice cubes, during the dough mixing process, the state of the ice cubes during the mixing process will go through three stages: sinking to the bottom, being thrown against the wall, and being wrapped by broken ice. Different stages have different effects on dough mixing. For example: When mixing flour and water at a low speed, due to the self-weight of the ice cubes, the ice cubes will gather at the bottom under the action of stirring to form a local low-temperature area, entering the stage of sinking to the bottom.

[0058] Low temperature will inhibit the hydration of flour, and at this time, the ice cubes are large and smooth and difficult to turn over at a slow speed, preventing the flour under the ice cubes from turning over, resulting in insufficient hydration of the bottom flour and forming hard lumps at the bottom. These lumps will also cause uneven dough after subsequent mixing; After the flour and water are initially mixed, the speed of the stirring paddle will increase, and the ice cubes will be thrown against the inner wall of the container due to centrifugal force, and the outer edge of the stirring paddle will hit the ice cubes irregularly. If the motor uses dynamic torque output, the motor torque will frequently jump between "large torque - small torque" during the process of the stirring paddle hitting and knocking away the ice cubes, resulting in a large fluctuation in the motor load, bringing fatigue to mechanical components and affecting the stability of the equipment; The ice cubes will be broken by the paddle blades during the stage of being thrown against the wall, and the formed broken ice will be wrapped by the formed dough, and at this time, it enters the stage of being wrapped by broken ice.

[0059] If sufficient stirring is not carried out, local water spots will appear after the wrapped broken ice melts, which will affect the uniformity of the dough. Therefore, generally, the stirring time needs to be lengthened, but this will also lead to overmixing of the dough, resulting in a gluten structure that does not match the expectation or consuming more energy.

[0060] The embodiment of the present application proposes a motor control method for a chef machine, which can identify different stages of ice cubes during the dough mixing process and adopt different motor control strategies for each stage. In the stage where the ice cubes sink to the bottom, the method adopts the first motor control strategy, and by reasonably adjusting the motor parameters, it prevents the ice cubes from sinking to the bottom for a long time, promotes the full contact between the bottom flour and moisture, and realizes uniform hydration; when the ice cubes enter the stage of hitting the wall, the system switches to the second motor control strategy, and in this stage, it alleviates the severe fluctuation of the motor torque output caused by hitting the ice cubes; finally, in the stage of ice cubes being wrapped in the dough, the third motor control strategy is started, and through an intelligent algorithm, it controls the uniform distribution of the crushed ice in the dough and at the same time avoids overmixing resulting in the destruction of the gluten structure.

[0061] Therefore, the motor control method for a chef machine provided by the embodiment of the present application can better mix the dough.

[0062] The second working principle and its beneficial effect are that in the stage of ice cubes being wrapped in the dough, since the crushed ice is wrapped by the dough and highly overlaps with the mixing resistance signal of the dough, it is difficult for the traditional motor control method to accurately identify the current degree of ice cube wrapping in this stage, resulting in the inability to adjust the operation of the motor according to the current degree of ice cube crushing and the current state of the dough, thus controlling the mixing time of the dough well.

[0063] The embodiment of the present application introduces a pre-information memory layer and a dynamic threshold adjustment mechanism. When the ice cubes enter the stage of hitting the wall, the motor control system will predict and record the data of ice cube crushing through the pre-information memory layer. When the ice cube characteristic signals can be captured well in the early stage, these predicted and recorded ice cube crushing data are used as a reference for determining the intensity of ice cube wrapping separation in the subsequent stage of ice cubes being wrapped in the dough. By making corresponding judgments in the initial stage of ice cube crushing in this way, a more accurate basis for signal recognition in the later stage of ice cubes being wrapped in the dough is provided, thus avoiding the problem that it is difficult to determine the intensity of ice cube wrapping due to inaccurate setting of the separation threshold in the traditional method, resulting in insufficient or excessive mixing of the dough.

[0064] Therefore, the motor control method for a chef machine provided by the embodiment of the present application can better mix the dough.

[0065] It should be noted that in this text, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Additionally, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements not only includes those elements but also includes other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not preclude the existence of additional identical elements in the process, method, article or device comprising the element. Moreover, in the description of the embodiments of the present application, unless otherwise specified, " / " means "or", for example, A / B can mean A or B; "and / or" in this text is merely a description of the relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. And, in the description of the embodiments of the present application, "a plurality of" means two or more than two.

[0066] The above are only specific embodiments of the present application, enabling those skilled in the art to understand or implement the present application. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments described herein, but rather to the broadest scope consistent with the principles and novel features disclosed herein.

Claims

1. A motor control method for a chef machine, characterized in that, The motor control method for a cooking machine includes the following steps: Obtain the motor load data and vibration data during dough mixing; based on the motor load data and vibration data, identify the state of ice cubes during dough mixing, including the ice cube sinking stage, the ice cube wall-slapping stage, and the crushed ice entrainment stage; According to the identified ice cube state, implement different motor control strategies correspondingly: In the ice cube sinking stage, adopt the first motor control strategy to prevent the ice cubes from continuously sinking to promote uniform hydration of the bottom flour; In the ice cube wall-slapping stage, adopt the second motor control strategy to control the ice cube crushing process and reduce the motor load fluctuation; In the crushed ice entrainment stage, adopt the third motor control strategy to control the uniform distribution of crushed ice and avoid overmixing.

2. The motor control method for a chef machine according to claim 1, wherein The motor control method for a cooking machine further includes: In the ice cube wall-slapping stage, predict the ice cube crushing data; According to the predicted ice cube crushing data, adjust the signal recognition threshold in the crushed ice entrainment stage; Based on the adjusted signal recognition threshold, detect the current crushed ice entrainment state to adjust the third motor control strategy, so as to avoid overmixing while ensuring the uniform distribution of crushed ice.

3. The motor control method for a cooking machine according to claim 2, characterized in that, The first motor control strategy, the second motor control strategy, and the third motor control strategy are output through a preset cooking machine motor control model; The cooking machine motor control model includes: An input layer for receiving the motor load data and vibration data; A feature extraction layer for receiving the data of the input layer and processing the data; A stage discrimination layer for receiving the output of the feature extraction layer to identify the stage of ice cubes during dough mixing; A pre-stage information memory layer for predicting and recording the ice cube crushing prediction data according to the data of the feature extraction layer in the ice cube wall-slapping stage; A threshold adjustment layer for dynamically adjusting the signal recognition threshold in the crushed ice entrainment stage according to the ice cube crushing prediction data stored in the pre-stage information memory layer; A control strategy layer for obtaining corresponding motor control parameters according to the identified ice cube stage using a corresponding expert neural network based on the output of the feature extraction layer; Wherein in the crushed ice entrainment stage, the crushed ice entrainment intensity is also identified based on the adjusted signal recognition threshold, and the obtained motor control parameters are adjusted accordingly; An output layer for converting the motor control parameters in different stages into motor drive commands, thereby obtaining the first motor control strategy, the second motor control strategy, and the third motor control strategy respectively.

4. The motor control method for a cooking machine according to claim 3, characterized in that, The feature extraction layer includes: A time-frequency preprocessing unit for performing short-time Fourier transform on the motor load data and vibration data to generate a time-frequency spectrogram; A shared feature extraction unit adopting a one-dimensional residual convolution network structure to extract cross-channel time-series features from the time-frequency spectrogram to form a shared feature vector.

5. The motor control method for a cooking machine according to claim 4, wherein The stage discrimination layer adopts a Transformer structure for receiving the shared feature vector and outputting a stage probability vector representing the ice cube sinking stage, the ice cube wall-slapping stage, and the crushed ice entrainment stage.

6. The motor control method for a cooking machine according to claim 5, wherein The pre-stage information memory layer adopts a fully connected neural network + LSTM structure, which is activated when it is determined to be the ice cube wall-slapping stage, and predicts the following indicators according to the shared feature vector in the ice cube wall-slapping stage: Size when the ice cube breaks; Degree of ice cube breakage.

7. The motor control method for a cooking machine according to claim 6, characterized in that, The threshold adjustment layer adopts a fully connected neural network structure. By receiving the size when the ice cube breaks and the degree of ice cube breakage predicted by the previous information memory layer, it outputs the motor load identification threshold and vibration identification threshold required for the ice crushing entrainment stage.

8. The motor control method for a cooking machine according to claim 7, wherein The control strategy layer includes: A stage selection unit, adopting a gated logic structure. Based on the stage probability vector output by the stage discrimination layer, it activates the expert neural network corresponding to the ice cube stage with the highest current stage probability from the expert network units in a hard-gated manner. The expert network units include three groups of expert neural networks independently corresponding to the ice cube bottoming stage, the ice cube wall-slapping stage, and the ice crushing entrainment stage respectively, where: The expert neural network for the ice cube bottoming stage and the expert neural network for the ice cube wall-slapping stage both adopt a fully connected neural network structure. Taking the shared feature vector output by the feature extraction layer as the input, they output the motor control parameters for the corresponding stage to the output layer. The expert neural network for the ice crushing entrainment stage includes a shallow convolutional autoencoder and a fully connected neural network structure. The shallow convolutional autoencoder is used to compress and reconstruct the input shared feature vector to obtain the abnormal residual features representing the ice crushing signal. The abnormal residual features are respectively compared with the motor load identification threshold and vibration identification threshold provided in real time by the threshold adjustment layer, so as to quantify the current ice crushing entrainment intensity. The quantified ice crushing entrainment intensity and the shared feature vector are used together as the input of the expert neural network for the ice crushing entrainment stage, and the motor control parameters for the ice crushing entrainment stage are output to the output layer.

9. The motor control method for a cooking machine according to claim 8, characterized in that, The output layer includes: A parameter safety domain clipping unit, which is used to receive the control parameters output by the control strategy layer and clip the control parameters within the safety domain range to ensure that the output control instruction is within the safe operation range of the chef machine motor. A control instruction generation unit, which is used to convert the control parameters after safety domain clipping into a chef machine motor drive instruction. The chef machine motor drive instruction at least includes the target motor rotation time, the target motor speed instruction, and the target motor torque instruction.

Images