Methods, information technology products, and manufacturing systems, performed by at least one computer, for optimizing the operation of at least one machine for the manufacture of liquid, semi-liquid, or semi-solid food products.

Abstract

The present invention provides a computer-operated method for optimizing the operation of machinery for the production of liquid or semi-liquid foods, as well as a system for the production of liquid or semi-liquid foods. [Solution] The manufacturing machine comprises a control unit 5 which includes a processing container 2, a stirrer 3, a plurality of sensors S, a plurality of actuators, and an artificial intelligence control module 6 which defines a classifier CL connected to the sensors and actuators, and the method executed by the computer includes the steps of: assigning labels to information received by the control module from the control unit, the labels representing the balance of components of the mixture being processed; performing a training procedure for the artificial intelligence control module using a second dataset containing information having predetermined labels; and sending an instruction to the control unit to replace the control module with an additional alternative artificial intelligence control model.

Description

Technical Field

[0001] The present invention relates to a method, an information technology product, and a system for the production of liquid or semi-liquid foods, which are executed by at least one computer to optimize the operation of at least one machine for the production of liquid or semi-liquid foods.

Background Art

[0002] In the field of handmade ice cream production, there is a growing interest in advanced machines that can reduce waste during the production process and constantly improve the quality of the final product. The production of handmade ice cream is actually a complex process that requires simultaneous control of many functional parameters, such as temperature, time, stirring speed, and indirect values of the viscosity of the product during processing. These factors directly affect the texture, creaminess, and density of the ice cream, and thus its sensory properties.

[0003] One of the most common problems when using current machines for ice cream production is the difficulty in maintaining high quality and uniform standards, especially when the mixture is unbalanced or not optimally prepared. Traditional machines do not actually have an advanced system that can immediately detect errors or abnormalities during the stirring process. This lack of monitoring and control can lead to the production of ice cream that does not meet the expectations of manufacturers and end consumers, which in turn can damage the quality of the final product and potentially cause an increase in waste, having an adverse economic impact on the artisan laboratory.

[0004] Therefore, it is particularly necessary to implement technical solutions that enable better control of operating parameters and thus ensure the optimization of the production process and the reduction of food losses.

[0005] Furthermore, the ongoing innovative research of ice cream makers, involving the development of increasingly sophisticated recipes and novel mixtures, necessitates ice cream production machinery that can adapt to new formulations without compromising the quality of the final ice cream. However, currently available machines lack the intelligence and flexibility to optimally handle new mixtures or non-standardized ingredients, such as those used in vegan ice cream that are lactose-free or low in sugar.

[0006] As a result, there is a need for a technical solution that offers greater adaptability and automated control, and ultimately guarantees the same results as traditional handcrafted production, regardless of the complexity of the recipe. [Overview of the project]

[0007] The object of the present invention is to satisfy the above needs and overcome the limitations of the prior art by providing a computer-operated method for optimizing the operation of at least one machine for manufacturing a product, an information technology product for performing such a method, and a manufacturing system.

[0008] Furthermore, an object of the present invention is to provide a computer-driven method, an information technology product for carrying out such a method, and a manufacturing system for optimizing the operation of at least one machine for manufacturing a product, which enables the processing of new mixtures in particular, thereby reducing waste and obtaining a particularly high-quality product. [Brief explanation of the drawing]

[0009] The technical features of the present invention can be clearly inferred from the following claims in accordance with the above objectives, and its advantages will become more apparent from the following detailed description with reference to the accompanying drawings illustrating exemplary and non-limiting embodiments. [Figure 1] A schematic diagram of a system for producing liquid, semi-liquid, or semi-solid products, which is the subject of this invention, is shown. [Figure 2]Figure 1 shows a schematic diagram of a specific part of the system for producing liquid and semi-liquid products, which is the subject of the present invention. [Figure 3] A schematic diagram of the information labeling step based on the method that is the subject of this invention is shown. [Modes for carrying out the invention]

[0010] According to the present invention, a method is provided, performed by at least one computer 7, for optimizing the operation of at least one machine 1A, 1B for the production of liquid or semi-liquid food products.

[0011] More generally, machines 1A and 1B are machines for the production of liquid or semi-liquid food products.

[0012] For the sake of brevity, the attached drawing shows machine 1 for making ice cream, but the drawing and this specification are not to be considered limiting.

[0013] It should be noted that Machine 1 is primarily designed to enable the production of products in the ice cream sector or cold desserts.

[0014] More preferably, machine 1 enables the production of homemade ice cream and / or soft ice cream, sorbet, cream, and granita.

[0015] The machines (1A, 1B) are preferably machines for processing products in the ice cream field (homemade ice cream, sorbet, soft serve ice cream, granita) or cold desserts.

[0016] The machines (1A, 1B) are - Processing container 2, -A stirrer 3 is placed inside the container 2, - Multiple sensors S, - Multiple actuators (11, 12, 13), - A control unit 5 comprising an artificial intelligence control module 6 configured to be connected to the sensor S and the actuators (11, 12, 13) and to define a pre-trained classifier CL.

[0017] According to another aspect, the machine 1 comprises an interface U connected to the control unit 5 for transmitting or receiving information.

[0018] The interface U preferably comprises control means (or any kind of, for example, tactile control means, voice control means or physical control means) and / or a display device.

[0019] The control module 6 is trained using a first data set.

[0020] The control module 6 is preferably pre-trained by collecting data from sensors (such as pressure, temperature, density, etc.) regarding the following mixtures. The mixtures are - A balanced mixture, - An unbalanced mixture with a high fat content, - An unbalanced mixture with a high sugar (sucrose) content, - An unbalanced mixture with a low sugar (sucrose) content, - An unbalanced mixture with a high water content.

[0021] More precisely, the mixtures used for training are as follows. That is, - A balanced mixture, - An unbalanced mixture with a high fat content (preferably cream) having a first predetermined ratio (preferably corresponding to three times the optimal balance), - An unbalanced mixture with a high fat content (preferably cream) having a second predetermined ratio (preferably corresponding to 4.4 times the optimal balance), -(Preferably corresponding to +1 / 3 with respect to the optimal balance) A mixture with an unbalanced high sugar content (sucrose), -(Preferably corresponding to +2 / 3 with respect to the optimal balance) A mixture with an unbalanced high sugar content (sucrose), -(Preferably corresponding to 100% with respect to the optimal balance) A mixture with an unbalanced high sugar content (sucrose), -(Preferably corresponding to -1 / 3 with respect to the optimal balance) A mixture with an unbalanced low sugar content (sucrose), -(Preferably corresponding to -2 / 3 with respect to the optimal balance) A mixture with an unbalanced low sugar content (sucrose), -(Preferably corresponding to +10% with respect to the optimal balance) A mixture with an unbalanced high moisture content.

[0022] According to one aspect, the first data collected in relation to the mixture is - The density or viscosity of the product being processed in container 2, - The operating temperature of container 2, - The operating temperature or pressure of the heat exchange fluid at a predetermined point in the thermodynamic system 4, - The operating temperature of the heat exchange fluid flowing into the evaporator 10 associated with container 2, - The operating temperature of the heat exchange fluid flowing out of the evaporator 10 associated with container 2, - The operating pressure of the heat exchange fluid flowing out of the compressor 11, - The rotational speed of the stirrer 3, - The operating pressure of the heat exchange fluid flowing out of the evaporator 10, - Including the opening degree of the throttle element 12.

[0023] The method which is the subject of the present invention includes the following steps. That is, - A step of labeling information received from the control unit 5 by assigning a predetermined label to the information, wherein the label represents the balance of components of the mixture being processed in the container 2. - A step of performing a training procedure for an artificial intelligence control module 6 outside the control unit 5 using a second dataset containing the information having predetermined labels, in order to derive an additional artificial intelligence control module that defines a classifier to replace the pre-trained artificial intelligence control module 6, - The step of sending a command to replace control module 6 with an additional artificial intelligence control model.

[0024] The term "information" refers to information transmitted by the control unit 5, for example, one or more of the following types: -Data from sensors connected to control unit 5, -Data from actuators connected / operated / controlled by control unit 5, - Processing by control unit 5, It should be noted that this means one or more types of information transmitted by the user interface U (which is part of machines 1A and 1B).

[0025] It should be noted that the definition of a “label indicating the balance of components” means a label that indicates whether a mixture is balanced with respect to one or more components (where balanced means having the optimal proportion of one or more components).

[0026] Such labels may, for example, be associated with or include multiple balance / unbalance classes.

[0027] Preferably, the method is carried out by a processing unit 7 which may be centralized and / or distributed.

[0028] It should be noted that some modules of such processing unit 7 may also be contained within machine 1 or control unit 5.

[0029] According to the present invention, the artificial intelligence control module 6 is essentially retrained using only the portion of the received information that has a predetermined label.

[0030] It should be noted that, according to the method described above, the control module 6 can be advantageously retrained using data (i.e., second data) that is considered to represent the optimal operation of the machine by the previous labeling step.

[0031] In this way, the control module 6 can be retrained very quickly. Retraining is actually performed using data with predetermined labels, starting from the current configuration of the control module 6 (for example, from the current weights if the control module 6 is defined by a neural network).

[0032] Preferably, it should be noted that the training of the control module 6 is not performed inside the control unit 5. The training of the control module 6 is performed by a processing unit 7 outside the control unit 5. In this way, advantageously, while the processing unit 7 is training a clone of the control module 6, the control module 6 located in the control unit 5 can continue to control the actuators of the machine 1.

[0033] It should be noted that, advantageously, the additional artificial intelligence control module is identical in its information technology structure to the pre-trained artificial intelligence control module 6.

[0034] In this way, training using the second set of data becomes significantly faster.

[0035] Model updates are basically performed through minor adjustments.

[0036] It should be noted that training the model starting from the original weights is a fairly advantageous approach, as it does not involve significant computational sacrifices.

[0037] Additional models trained on data with predetermined labels (for example, data with balanced mixtures, which is advantageous) can certainly improve the control accuracy of machines 1A and 1B.

[0038] Advantageously, this makes it possible to optimally handle changes in the operating conditions of machines 1A and 1B, thereby reducing processing waste.

[0039] for example, - The method allows for rapid retraining of the classification module 6 as a result of changes in the type of mixture being processed, and thus the machine can quickly adapt to new types of liquid or semi-liquid foods. - The method allows for rapid retraining of the classification module 6 as a result of changes that may occur, such as altering the environmental conditions in which the machine operates (outside temperature, humidity, etc.), in order to optimize the machine's operation. - The method allows for rapid retraining of classification module 6 in order to optimize the operation of the machine, as a result of a decrease in the performance of some elements of the machine that may affect the quality of the final product, or more generally, a change in performance (such as a change in one or more elements of the thermodynamic system).

[0040] In another embodiment, the step of sending an instruction to replace the control module 6 with an additional artificial intelligence control model includes the step of sending parameters representing the additional control module.

[0041] In this embodiment, only parameters representing the artificial intelligence mathematical model of the control module are transmitted, and the model architecture is not transmitted.

[0042] For example, in the case of a control module with a neural network, only the weights assigned to a single neuron are transmitted, and the network architecture (which remains identical between control module 6 and the alternative control module) is not transmitted.

[0043] According to one embodiment, the additional pre-trained artificial intelligence control module comprises a neural network, and the step of transmitting parameters representing the additional control module includes the step of transmitting a plurality of training weights of the neural network.

[0044] More generally, it should be noted that control module 6 could be any type of artificial intelligence classifier, for example, namely, -Neural networks, - Additional artificial intelligence submodules and neural networks configured to perform so-called “attention” according to the transformer architecture, - One or more decision trees, e.g., a random forest model, - Classification modules configured to define distance methods and kernel methods for time series, such as k-nearest neighbors (k-NN) or support vector machines (SVM), respectively. - A "support vector machine" type model, - It should be noted that this could be one or more kernel-based SVM-type binary classifiers.

[0045] In another embodiment, the step of labeling information received from the control unit 5 by assigning a predetermined label to the information includes the step of associating the information with a label that represents at least the following: -A balanced class corresponding to a mixture being processed, which is balanced with respect to one or more components. - The process includes the step of associating a label that represents at least one unbalanced class, corresponding to a mixture being processed that is unbalanced with respect to one or more components.

[0046] It should be noted that the definition of an unbalanced mixture is a mixture in which the proportions of its components deviate from the (theoretical) optimal proportions. On the other hand, the definition of a balanced mixture is a mixture in which the proportions of its components are (theoretical) optimal.

[0047] In another embodiment, the second dataset includes labels representing at least a balanced class corresponding to a mixture under processing that is balanced with respect to one or more components.

[0048] It should be noted that, advantageously, by training Model 6 with balanced class labels, the control module 6 is trained with data related to the optimal mixture of inputs.

[0049] In another embodiment, the machine 1 comprises a thermodynamic system 4 having a throttle element 12, a compressor 11, and an evaporator 10. Furthermore, the thermodynamic system 4 comprises a condenser 13.

[0050] In another embodiment, machine 1 includes a motor 20 connected to a stirrer 3.

[0051] Preferably, the labeling step, which involves assigning a predetermined label to the information received from the control unit 5, includes the step of assigning a label to the information based on one or more values ​​of the following parameters received from the control unit 5. The parameters are: - The density or viscosity of the product being processed in container 2, -Operating temperature of container 2, - The operating temperature or pressure of the heat exchange fluid at a predetermined point in the thermodynamic system 4. - The operating temperature of the heat exchange fluid flowing into the evaporator 10 associated with the container 2, - The operating temperature of the heat exchange fluid flowing out of the evaporator 10 associated with container 2, -Operating pressure of the heat exchange fluid flowing out of the compressor 11, -Operating pressure of the heat exchange fluid flowing out of the evaporator 10, - Rotation speed of agitator 3, -This is the opening degree of the throttle element 12.

[0052] In another embodiment, the labeling step includes the step of using an unsupervised type of machine learning algorithm.

[0053] In another embodiment, the labeling step includes using a clustering algorithm (e.g., K-means, DBSCAN) to identify patterns and groups in the information, which is advantageous as it does not require predetermined labels.

[0054] The advantage of this approach is that it eliminates the need for predetermined labels, such as labels previously defined by the user. Therefore, such a method is particularly useful for exploring large datasets and finding hidden patterns.

[0055] In this case, data labeling is performed by automatically grouping data into clusters based on similar features or patterns. Specifically, labeling by a clustering algorithm is carried out in the following steps: - A step to retrieve information about unlabeled test data and some metadata (provided by oversight module 9, which will be explained in more detail below). - A step of performing data preprocessing steps, preferably including data selection and normalization steps. This step may include steps such as: handling missing values, scaling variables, coding categories, and embedding transformations using a pre-trained controlled learning model. - A step of applying a clustering algorithm to the preprocessed data (such a step includes selecting an appropriate algorithm, including the selection of K-means as an illustrative and non-exclusive example). - A step to analyze data features and cluster groupings based on the similarity of those features (and also based on metadata provided by the oversight model, such as a predetermined number of clusters based on various unbalanced classes identified).

[0056] This method yields one or more of the following: - Assigned clusters (Each data is assigned to a specific cluster. For example, each test mixture is labeled as balanced, or belongs to unbalanced class 1, unbalanced class 2, unbalanced class 3, etc.) - One or more centroids or cluster representatives are obtained (information about average features or representing each cluster is provided to help interpret what distinguishes one group from others, and is used as metadata in the subsequent labeling process).

[0057] Furthermore, according to another embodiment, the algorithm is a clustering algorithm.

[0058] Furthermore, according to another embodiment, the labeling step includes using a clustering algorithm to identify patterns and / or groups of information.

[0059] Therefore, Figure 3 shows the steps of information clustering according to two dimensions (dimension 1 and dimension 2). As can be observed, the information is grouped according to various groups (G1, G2, G3), and each group is associated with a corresponding label.

[0060] In another embodiment, the labeling step of the received information includes a step of using a semi-supervised type machine learning algorithm.

[0061] According to such semi-supervised machine learning algorithms, the labeling step involves combining a small amount of labeled data with a large amount of unlabeled information.

[0062] The labeling step also includes identifying clusters, groups, or patterns on the labeled data and assigning labels to the information based on the identified clusters, groups, or patterns.

[0063] According to such semi-supervised machine learning algorithms, the labeling step includes a step of information preprocessing.

[0064] According to such a semi-supervised machine learning algorithm, the labeling step includes a step of generating or building autoencoders. In such a step, multiple autoencoders are created, each configured to recognize an imbalance of a particular type of mixture.

[0065] According to this embodiment, the autoencoder can detect anomalies or deviations that are not included in the recognized unbalance types.

[0066] According to one embodiment, the labeling step includes an autoencoder training step with labeled information relating to a specific type of imbalance that each autoencoder should recognize.

[0067] In another embodiment, the method includes an inference step in which unlabeled data is applied to each autoencoder.

[0068] In this step, each autoencoder reconstructs the unlabeled data. If the autoencoder can reconstruct the data correctly, it is labeled as belonging to an unbalanced type that the autoencoder has learned to recognize.

[0069] If an autoencoder shows a large reconstruction error, it means the data does not belong to that type of imbalance, and analysis continues using other autoencoders until the correct type is identified.

[0070] The advantage of this method is the expanded, labeled dataset; that is, it has the potential to significantly increase the available dataset, which in turn increases the amount of information available in the next training step.

[0071] The model can reliably recognize various types of imbalances and gradually improves as the amount of available labeled data increases.

[0072] From another perspective, the method is - A step of receiving information from the respective control units 5 of multiple machines 1A, 1B having the same artificial intelligence control module 6, wherein a second dataset includes the information having predetermined labels obtained from the multiple machines 1A, 1B.

[0073] A single machine 1A, 1B control model 6 can also be trained based on information received from multiple machines 1A, 1B (preferably of the same type, or having control models 6 of the same type).

[0074] In another embodiment, the second dataset includes at least a portion of the data from the first dataset.

[0075] Conveniently, in this way, the training step can also be performed by using a portion of the first data (or pre-training data).

[0076] Furthermore, according to another embodiment, the step of labeling the information received from the control unit 5 is performed on a centralized remote processor.

[0077] Furthermore, according to another embodiment, the step of labeling the information received from the control unit 5 is performed on a processor uniquely associated with the control unit 5. It should be noted that such a processor can also be located remotely from the control unit 5. In any case, such a processor is uniquely associated with machines 1A and 1B.

[0078] In another aspect, the present invention provides an information technology product that performs the method when loaded into a processing unit 7.

[0079] According to another aspect of the present invention, a system for producing liquid or semi-liquid food is provided, the system is - At least one machine 1A, 1B for the manufacture of liquid or semi-liquid food products, - Processing container 2, -A stirrer 3 is placed inside the container 2, - Multiple sensors S, - Multiple actuators 11, 12, 13, - A control unit 5 comprising an artificial intelligence control module 6 connected to the sensor S and actuator and configured to define a pre-trained classifier CL, wherein the control module 6 is trained using data from a first dataset, - The machine comprises machines 1A and 1B, each comprising at least one processing unit 7 configured to perform the steps of the method described above.

[0080] According to another embodiment of System 1, the processing unit 7 is connected to the control unit 5 and transmits commands to the control unit 5 of at least one machine 1A, 1B to replace the control module 6 with an additional artificial intelligence control model.

[0081] Furthermore, according to a particular embodiment, the processing unit 7 is defined by a plurality of processors that are distributed and connected to each other for data exchange.

[0082] In another embodiment, the processing unit 7 is centralized and remote from the control unit 5.

[0083] In another embodiment, the processing unit 7 and the control unit 5 can be at least partially identical, and some of the functions of the processing unit 7 can be basically performed by the control unit 5.

[0084] In another embodiment, it should be noted that the processing unit 7 includes a monitoring module 20 configured to receive the information from the control unit 5.

[0085] It should be noted that the monitoring module 20 is preferably placed on the board of the machine (1A, 1B).

[0086] According to one embodiment, the monitoring module 20 is preferably incorporated inside the control unit 5.

[0087] In another embodiment, it should be noted that the monitoring module 20 is located outside the machines (1A, 1B) and is operationally linked to more machines (1A, 1B).

[0088] The monitoring module 20 is configured to receive data from sensors and / or machine operating parameters.

[0089] Furthermore, according to another embodiment, the processing unit 7 includes a memory module 21 configured to perform the step of receiving a plurality of pieces of information transmitted by sensors and / or actuators and / or by the control units 5 of the machines 1A, 1B, and to store the information.

[0090] It should be noted that the memory module 21 comprises a first memory 21A and a second memory 21B. The first memory 21A is configured to store information received from the control unit 5. The second memory 21B is configured to store labels and their associated relationships with the information received from the control unit 5.

[0091] In another embodiment, the processing unit 7 includes a labeling module 22 configured to perform the step of labeling the received information.

[0092] According to one embodiment, the labeling module 22 is preferably configured to perform a labeling step of the information received based on the balance / unbalance class with respect to one or more components of the mixture being processed in the container 2.

[0093] In another embodiment, the processing unit 7 includes a training module 24 configured to perform the training procedure.

[0094] According to this embodiment, the processing unit 7 includes a supervisor module 23 configured to process the information and transmit a training procedure execution signal to the training module 24 based on the information.

[0095] Furthermore, the training module 24 is connected to the supervision module 23 and is configured to execute the training procedure based on the receipt of the training procedure execution signal.

[0096] It should be noted that the training module 24 is also connected to the memory module 21 and receives and stores a plurality of pieces of information transmitted by sensors and / or actuators and / or the control units of machines 1A and 1B.

[0097] The training module 24 is preferably located on a remote server.

[0098] Alternatively, the training module 24 can also be incorporated into the control unit 5 of specific machines (1A, 1B).

Claims

1. A method, performed by at least one computer (7), for optimizing the operation of at least one machine (1A, 1B) for the manufacture of liquid or semi-liquid food products, wherein the machine (1A, 1B) is - Processing container (2), - A stirrer (3) is placed inside the container (2), - Multiple sensors (S), - Multiple actuators (11, 12, 13), - A control unit (5) comprising a pre-trained artificial intelligence control module (6) configured to define a pre-trained classifier (CL) connected to the sensor (S) and actuator, wherein the control module (6) is trained using a first dataset, This method is - A step of labeling information received from the control unit (5) by assigning a predetermined label to the information, wherein the label represents the balance of components of the mixture being processed in the container (2), - A step of performing a training procedure for the artificial intelligence control module (6) using a second dataset containing the information having predetermined labels in order to derive an additional artificial intelligence control module that defines a classifier to replace the pre-trained artificial intelligence control module (6), A method comprising the step of sending an instruction to the control unit (5) to replace the control module (6) with the additional alternative artificial intelligence control model.

2. The method according to claim 1, wherein the additional artificial intelligence control module is identical in its information technology structure to the pre-trained artificial intelligence control module (6).

3. The method according to claim 1 or 2, wherein the step of sending an instruction to the control unit (5) to replace the control module (6) with the additional artificial intelligence control model includes the step of sending parameters representing the additional control module.

4. The method according to claim 3, wherein the additional pre-trained artificial intelligence control module comprises a neural network, and the step of transmitting parameters representing the additional control module includes the step of transmitting a plurality of training weights of the neural network.

5. The step of labeling the information received from the control unit (5) by assigning a predetermined label to the information is as follows: - A balance class corresponding to a mixture being processed, in which one or more components are balanced. The method according to any one of claims 1 to 4, comprising the step of associating the information with a label representing at least an unbalance class, which corresponds to a mixture being processed that is unbalanced with respect to one or more components.

6. The method according to any one of claims 1 to 5 and claim 5, wherein the second dataset includes labels representing at least the balance class corresponding to the mixture being processed, which is balanced with respect to one or more components.

7. The machine (1) comprises a thermodynamic system (4) having a throttle element (12), a compressor (11), a condenser (13), and an evaporator (10), and the labeling step, by assigning a predetermined label to the information received from the control unit (5), the label representing the balance of components of the mixture being processed in the container (2), includes the step of assigning the label to the information based on one or more values ​​of the following parameters received from the control unit (5), the parameters being: - The density or viscosity of the product being processed in the container (2), - Operating temperature of the container (2), - The operating temperature of the heat exchange fluid flowing into the evaporator (10) associated with the container (2), - The operating temperature of the heat exchange fluid flowing out of the evaporator (10) associated with the container (2), - The operating pressure of the heat exchange fluid flowing out of the compressor (11), - The operating pressure of the heat exchange fluid flowing out of the evaporator (10), - The rotational speed of the agitator (3), - The method according to any one of claims 1 to 6, wherein the opening degree of the throttle element (12).

8. The method according to any one of claims 1 to 7, wherein the labeling step includes using an unsupervised type machine learning algorithm.

9. The method according to claim 8, wherein the algorithm is a clustering algorithm.

10. The method according to claim 8 or 9, wherein the labeling step includes identifying patterns and / or groups in the information using the clustering algorithm.

11. The method according to any one of claims 1 to 7, wherein the step of labeling the received information includes the step of using a semi-supervised machine learning algorithm.

12. - The method according to any one of claims 1 to 11, comprising the step of receiving information from each control unit (5) of a plurality of machines (1a, 1b) having the same artificial intelligence control module (6), wherein the second dataset includes the information having predetermined labels obtained from the plurality of machines (1a, 1b).

13. The method according to any one of claims 1 to 12, wherein the second dataset includes at least a portion of the data of the first dataset.

14. The method according to any one of claims 1 to 13, wherein the step of labeling the information received from the control unit (5) is performed on a remote centralized processor.

15. The method according to any one of claims 1 to 13, wherein the step of labeling the information received from the control unit (5) is performed on a processor uniquely associated with the control unit (5).

16. A system for the production of liquid or semi-liquid food products, - comprising at least one machine (1A, 1B) for the manufacture of liquid or semi-liquid food products, wherein the machine (1A, 1B) - Processing container (2), - A stirrer (3) is placed inside the container (2), - Multiple sensors (S), - Multiple actuators (11, 12, 13), - A control unit (5) comprising an artificial intelligence control module (6) connected to the sensor (S) and actuator and configured to define a pre-trained classifier (CL), wherein the control module (6) is trained using a first dataset, A system comprising: at least one processing unit (7) configured to perform the steps of the method according to claims 1 to 15, wherein the processing unit (7) is connected to the control unit (5) to transmit instructions to the control unit (5) of the at least one machine (1A, 1B) to replace the control module (6) with the additional artificial intelligence control model.

17. The system according to claim 16, wherein the processing unit (7) is defined by a plurality of processors that are distributed and connected to each other for data exchange.

18. The system according to claim 16 or 17, wherein the processing unit (7) is centralized and located remotely from the control unit (5).

19. The system according to any one of claims 16 to 18, wherein the processing unit (7) comprises a monitoring module (20) configured to receive the information from the control unit (5).

20. The system according to any one of claims 16 to 19, wherein the processing unit (7) performs the step of receiving a plurality of pieces of information transmitted by a sensor (S) and / or an actuator and / or by the control unit of the machine (1A, 1B), and comprises a memory module (21) configured to store the information.

21. The system according to any one of claims 16 to 20, wherein the processing unit (7) comprises a labeling module (22) configured to perform the step of labeling the received information.

22. The system according to any one of claims 16 to 21, wherein the processing unit (7) comprises a training module (24) configured to perform the training procedure.

23. The system according to claim 22, wherein the processing unit (7) comprises a supervisor module (23) configured to process the information and transmit a training procedure execution signal to the training module (24) based on the information, and the training module (24) is connected to the supervisor module (23) and configured to execute the training procedure based on the receipt of the training procedure execution signal.

24. A computer program comprising instructions configured to perform the method described in any one of claims 1 to 15 when loaded into the processing unit (7) of the system according to any one of claims 16 to 23.

Images