Machine for making liquid, semi-liquid or semi-solid food products and associated control method

Abstract

A machine (1) for making liquid or semi-liquid or semi-solid food products, comprising: a processing container (2) for processing a basic product; an agitator (3) adapted to conduct the product within the processing container (2) for processing the base product; a heat treatment system (4) configured to exchange heat with the processing vessel (2) processing the base product; a control unit (5) connected to the heat treatment system (4) and to the agitator (3) to drive them, at least one sensor (S) configured to detect an operating parameter of the machine; a first artificial intelligence control module (6), the blocks of which are configured to define a pre-trained classifier (CL) having at least one input (I1, I2, I3, I4, I5, I6, I7) and at least one output (U11), where the at least one input (I1) is connected to the sensor (S); a drive module (7) connected to the output (U11) of the first control module (6) and at least to the heat treatment system (4) and / or the stirrer (3).

Description

[0001] The present invention relates to a machine for making liquid or semi - liquid foods, in particular for products in the ice - cream, pastry or catering industries, and to a method for controlling such a machine.

[0002] In the prior art of the relevant industries, machines are known for the thermal processing and / or machining of base products, which preferably have at least one liquid or semi - liquid component for making finished products of the liquid or semi - liquid or semi - solid type.

[0003] In particular, machines known in the ice - cream industry include a processing container, an agitator installed inside the container, and a thermal system associated with the container to exchange heat with the walls of the container.

[0004] In this type of machine, the base product is thus batch - frozen inside the container, that is, cooled by the thermal system and simultaneously agitated by the agitator in order to transform the base product into the final ice - cream product.

[0005] Generally speaking, in machines for making traditional artisanal ice - cream, batch - freezing is achieved by the combined use of a refrigeration cycle and mechanical agitation. The batch - freezing of the base food at variable temperatures depends on the type of process and the composition of the mixture itself (usually in the range of - 11 < T(°C) < - 6). The characteristic of this stage of the process is that, through the rapid cooling of the mixture, it is possible to distribute the solid crystalline phase and the gas phase into the structure of the ice - cream without separating the solutes.

[0006] As is known, ice - cream is a fluffy emulsion, characterized by a highly specific colloidal structure, which gives it its unique texture and sensory properties.

[0007] More precisely, as a variant of ice - cream, Gelato, is a solution characterized by a liquid dispersing phase and simultaneously has the following characteristics:

[0008] - An emulsion (which contains fat globules);

[0009] - A colloid (which is a non - homogeneous phase dispersant)

[0010] - A foam (which contains air bubbles).

[0011] Furthermore, it should be noted that ice - cream requires the simultaneous presence of three states of matter (solid, liquid, gas).

[0012] In addition, it is noted that ice - cream is a non - Newtonian fluid.

[0013] As is clear from the above, ice cream is an extremely complex food product consisting of several ingredients bound by different interacting forces and exhibiting different physical states that affect its finished product quality, stability and properties.

[0014] The ingredients of the initial mixture are water, fat, lean milk solids, sugar, stabilizers and emulsifiers, and as they gradually cool and undergo the mixing action of a blender, they become "ready ice cream", a product resulting from the dispersion of each ingredient in a different phase.

[0015] The structure of ice cream is formed by a dispersion of air in a frozen liquid, which is about two-thirds water. The aqueous phase includes sucrose, lactose, other sugars and mineral salts, in addition to microscopic particles of air, fat and ice crystals trapped within it.

[0016] Thus, broadly speaking there are four main identifiable component phases: ice crystals, gas bubbles, fat globules and a non-frozen viscous liquid matrix (monosaccharides, polysaccharides, proteins and emulsifiers) commonly referred to as cheese.

[0017] Close and precise monitoring during the processing of the four phases has a significant impact on the physical, textural and sensory properties.

[0018] Furthermore, the cheese acts as a continuous phase which interacts with the other phases present in the ice cream, keeping the structure of the ice cream as a whole and its properties dependent on the composition of the mix and on the temperature.

[0019] The proper consistency and flow shape of ice cream are fundamental aspects of ice cream quality. In addition to being a non-Newtonian fluid and exhibiting thixotropic behavior (i.e., its apparent viscosity decreases with time of applied force), ice cream is actually considered a pseudoplastic fluid, whose behavior can only be described experimentally on a macroscopic scale by specific mathematical models.

[0020] On the other hand, the number and size of ice crystals affect many properties of the finished product, such as spoonability, hardness and melting rate, which in turn can be affected by clustered fat globule aggregates.

[0021] The purpose of the air bubbles is to provide a light, fluffy consistency.

[0022] From a practical standpoint, determining when the mixture has reached the exact consistency (ie, texture) for an optimal finished product is critical.

[0023] Finally, given the complexity of ice cream, based on the above, it is important to emphasize that even small changes in the mix recipe can lead to various problems in the finished product.

[0024] As for methods used to measure or quantify the quality of finished ice cream, please note the following.

[0025] Some conventional methods can be used to define the properties of the finished product (in terms of quality), such as determining the overrun, assessing the temperature, melting or melting rate, and panel testing (sensory analysis).

[0026] In detail, the overrun is determined by comparing the weight of the mix with the weight of ice cream in a suitable calibrated container. Overrun refers to the percentage of air incorporated into the ice cream and provides an indication of consistency, softness and appearance.

[0027] Melting or thawing rate is usually quantified by measuring the mass that drips from a product through a screen as a function of time at a given temperature, and is related to the properties of the ice crystals, and therefore the structure of the ice cream.

[0028] In panel testing, specially qualified panelists with specific experience assess the overall acceptability of a product, i.e. its quality, based on different aspects of sight, smell and taste.

[0029] The production of ice cream is obtained through a complex interaction between different process parameters and conditions: indeed, structural changes and physical variations affect the sensory and melting characteristics, which means that it is not easy to determine exactly when the product is "ready".

[0030] Prior art machines implement various strategies that have been developed for monitoring the production process to define when the desired consistency has been achieved and the ice cream is ready to be removed from the batch freezing cylinder.

[0031] In a first class of state-of-the-art machines, a timer is used to control the production time. This type of machine has the obvious disadvantage that not all ice cream mixes allow the same consistency and viscosity to be obtained in the same length of time, so optimizing the production of ice cream depends largely on the operator's experience in using the machine and managing the recipe.

[0032] In order to overcome this apparent drawback, a second type of machine for making ice cream comprises a product temperature sensor in order to be able to determine, based on this parameter, when the ice cream has reached its optimum consistency.

[0033] It should be noted that the temperature of the product is a quantity that is difficult to determine uniformly in a processing chamber and, moreover, it is not possible to identify an optimum value for all types of ice cream.

[0034] This means that optimizing ice cream production in this case also depends heavily on the operator’s experience in using the machine and managing the recipe.

[0035] A third type of machine includes a device configured to detect the resistance of the ice cream to the rotation of the churn (thereby indirectly measuring the hardness and viscosity of the ice cream).

[0036] The means may be mechanical or electronic, ie based on the power input of the stirrer motor.

[0037] In such electronic devices, the value of the current or voltage is processed to obtain an indirect measure of the torque delivered by the stirrer drive motor.

[0038] In another prior art solution, the ice cream comprises a level sensor capable of detecting the presence of the mix in the batch freezing chamber and of performing a conductivity measurement to see if the ice cream has reached the correct consistency.

[0039] The solution is based on the principle that due to the reduced temperature in the batch freezing chamber, the water in the mix becomes solid during processing, resulting in a lower concentration of solutes and thus a lower conductivity, which can be detected by conductivity measurement.

[0040] The Applicant has discovered that, whatever their kind, the machines for making liquid or semi-liquid food products that exist to date are not able to act substantially in real time on the ice cream production process currently applied, in order to overcome drawbacks such as:

[0041] - Lengthy processing time;

[0042] - It is not possible to batch freeze unbalanced mixes (that is, mixes in which the percentage of some ingredients, such as sugar, water or fat, differs from the optimal percentage for processing);

[0043] - Heat exchange problems with mixtures with high fat content;

[0044] - Ice cream lumps may form inside the mixer;

[0045] - Problems with repeatability in defining the correct consistency.

[0046] Furthermore, with the machines that currently exist for the manufacture of liquid or semi-liquid foods, it is not possible to manufacture under all conditions a finished product of optimal quality, as far as the balance of the ingredients is concerned.

[0047] Therefore, the need is particularly strongly felt in the relevant industry for a machine for processing liquid or semi-liquid products, in particular ice cream, which is able to carry out the batch freezing process in an optimal manner and at the same time produce a product of optimal quality.

[0048] The object of the present invention is to meet the above needs and to overcome the above mentioned drawbacks of the prior art by providing a machine for processing liquid and / or semi-liquid products, in particular ice cream, and by providing a method for controlling such a machine, which machine is able to optimize the batch freezing process and at the same time produce a product of the best quality.

[0049] Another object of the invention is to provide a machine for processing liquid and / or semi-liquid products which allows batches of ice cream to be frozen in the shortest possible time without in any way negatively affecting the quality of the finished product, and also a method for controlling the machine.

[0050] Another object of the present invention is to provide a machine for processing liquid, semi-liquid or semi-solid products which allows to prevent potential problems in the processing cycle and also to provide a method for controlling the machine.

[0051] Another object of the present invention is to provide a machine for processing liquid, semi-liquid or semi-solid products which allows to accurately determine the right time to extract the ice cream, that is to say to determine when the mixture has reached the optimum consistency.

[0052] In accordance with the foregoing objects, the technical features of the present invention are clearly described in the following claims, while their advantages become more apparent from the following detailed description, with reference to the accompanying drawings showing purely exemplary embodiments, in which:

[0053] - Figure 1 A machine for making a liquid, semi-liquid or semi-solid according to the invention is schematically shown;

[0054] - Figure 2 A machine for making a liquid, semi-liquid or solid product according to the invention is schematically shown;

[0055] - Figure 3 Schematically shows Figure 1 Details of the machine's control unit;

[0056] - Figure 4 Schematically represented by Figure 1 The control process implemented by the control unit of the machine;

[0057] - Figure 5 Schematically represents the architecture of the control module of the machine 1 of the invention in an embodiment comprising a neural network;

[0058] - Figure 6 The composition of some mixtures used in the training step is schematically shown;

[0059] - Figures 7 to 9Schematically represents the trends of some parameters in a machining cycle in the machine of the invention during a training step, that is, when the control module has not yet been trained;

[0060] - Fig.10 Schematically represented is a processing cycle in the machine of the invention during a training step, ie when the control module has not yet been trained. According to the invention is defined a machine 1 for preparing liquid or semi-liquid or semi-solid food products.

[0061] It should be noted that the machine 1 allows mainly ice cream and cold dessert products to be prepared.

[0062] More preferably, the machine 1 allows the production of ice cream, handmade gelato and / or soft serve ice cream, sorbet, cream and gelato.

[0063] The machine 1 comprises:

[0064] - Processing vessels for processing basic products2;

[0065] - a stirrer 3 suitable for mixing the product inside the base product processing container 2 and rotating inside the processing container 2;

[0066] a heat treatment system 4 configured to heat treat the product inside the container 2;

[0067] - a control unit 5 connected to the thermal treatment system 4 and to the stirrers 3 in order to drive them;

[0068] - At least one sensor S, which is configured to detect an operating parameter of the machine and / or a state of a part thereof.

[0069] According to the present invention, the control unit 5 comprises:

[0070] a first artificial intelligence control module 6 configured to define a pre-trained classifier CL having at least one input ( I1 , I2 , I3 , I4 , I5 , I6 , I7 ), wherein at least one input I1 is connected to a sensor S;

[0071] - A drive module 7, which is connected to the output U11 of the first control module 6 and is at least connected to the heat treatment system 4 and / or to the agitator 3, so as to adjust at least one operating parameter related to the heat treatment system 4 and / or the agitator 3 according to the value of at least one output U11 of the first control module 6.

[0072] According to one aspect, the first control module 6 includes a plurality of inputs ( I1 , I2 , I3 , I4 , I5 , I6 , I7 ).

[0073] According to another aspect, at least one output U11 of the first control module 6 represents a classification of the basic product mixture.

[0074] Basically, the first control module 6 is configured to classify the mixture.

[0075] It should be noted that, in one embodiment, the first control module 6 comprises a classifier CL based on a machine learning (artificial intelligence) solution.

[0076] It should be noted that in one embodiment, the first control module 6 comprises a neural network RN comprising a plurality of layers (S1, S2, S3, S4, S5, S6) and having a plurality of inputs (I1, I2, I3, I4, I5, I6, I7).

[0077] Described below in detail without loss of generality is an embodiment in which the first control module 6 comprises a neural network RN.

[0078] However, other exemplary embodiments are also described below which do not have a neural network RN as the first control module 6 .

[0079] Preferably, according to another aspect, the machine 1 comprises a dispenser 21 connected to the first container 2 to dispense the product.

[0080] According to another aspect, the machine 1 preferably comprises an interface U connected to the control unit 5 to allow a user to send instructions / settings to the control unit 5 .

[0081] Preferably, the machine 1 comprises an electric motor 20 connected to the stirrer 3 to drive the same in rotation.

[0082] Preferably, the neural network RN comprises at least two hidden layers, the number of neurons in the hidden layers is preferably greater than 10 (preferably activated by a rectified linear unit (ReLU) or a hyperbolic tangent function (tanh hyperbolic tangent) or a leaky rectified linear unit (LeakyReLU) activation function).

[0083] Preferably, the last layer S6 of neurons, ie the layer at the output of the neural network RN, has a sigmoid activation function (in the case of two classes) or a softmax activation function (in the case of three or more classes).

[0084] Referring to the activation functions of the neurons that make up the different layers (S1, S2, S3, S4, S5, S6), note that these activation functions can be of the following types:

[0085] - S-shaped (compresses input values ​​between 0 and 1);

[0086] - ReLU (rectified linear unit), which is a linear activation function defined as f(x) = max(0, x);

[0087] - Softmax (preferably used in the output state: the sum of the outputs is 1; advantageously, it allows to represent probability distributions);

[0088] - Tanh (double tangent function, which has the shape of an S curve, similar to the sigmoid function, but compresses input values ​​between -1 and 1; usually, it is often used in hidden layers).

[0089] It should be noted that the container 2 and the blender 3 together define a batch freezing unit.

[0090] As is known, a neural network RN is a computational model whose purpose is to determine output values ​​based on input values.

[0091] The neural network RN consists of several neuron layers (in Figure 5 In the example of , there are 6 layers: S1, S2, S3, S4, S5, S6; it should be understood that the number of layers can be more or less than the number of layers shown), where each layer performs a weighted sum operation on the input and passes the result to an activation function, which is used to keep it within a specific range and approximate it.

[0092] According to one aspect, neurons in a hidden layer pass their results to neurons in the next layer.

[0093] The result will be given by a linear combination of the outputs of the last hidden layer (through its activations).

[0094] According to one aspect, at least some of the neural network layers S2, S3, S4, S5 are hidden layers and the neural network is a deep neural network.

[0095] According to another aspect, the neural network RN is a recurrent neural network (RNN).

[0096] According to another aspect, the neural network RN is a convolutional neural network (CNN).

[0097] Alternatively, the neural network RN is any variant of the RNN and CNN types or any hybrid version given by a combination of RNN and CNN and their variants.

[0098] According to another aspect, the neural network RN is assisted by means implementing so-called "attention" according to known converter architectures.

[0099] It should be noted that in this case the first control module 6 comprises a neural network RN and other artificial intelligence submodules configured to implement the so-called “attention” according to a known converter architecture.

[0100] In experimental tests performed by the applicant using a neural network RN on machine 1, it was observed that a network of the CNN type achieved the best results in terms of performance.

[0101] The following will better describe the Figure 5 The structure of the neural network RN is schematically shown in FIG. , in particular with respect to its input and output.

[0102] It should be noted that the number of neurons per layer as well as the number of layers may vary without thereby departing from the scope of the inventive concept.

[0103] Preferably, the network comprises more than four layers.

[0104] Focusing our attention on the architecture of the control unit 5, the following should be noted.

[0105] According to one aspect, the machine 1 comprises an additional second control module 8 configured to carry out machining and / or detect anomalies based on data from at least one sensor S.

[0106] The second module 8 can also be used with a different type of first control module 6, that is, with a first artificial intelligence control module 6 that does not include only one neural network. Therefore, what is described with reference to the second module 8 also explicitly applies to a first artificial intelligence control module 6 that does not include only one neural network.

[0107] The second module 8 thus receives data from at least one sensor S.

[0108] The second control module 8 has at least one first output U21 connected to the drive module 7 so as to be able to transmit control information to the drive module 7 .

[0109] The control information is information derived by the second control module 8 based on data provided by at least one sensor S as input to the second control module 8 .

[0110] In addition, according to another aspect, the drive module 7 is configured to adjust at least one operating parameter related to the heat treatment system 4 and / or the agitator 3 according to the value of at least one output U11 of the first control module 6 and / or the value of the output U21 of the second control module 8.

[0111] This concept is explained in more detail below.

[0112] Basically, the drive module 7 can operate in two different modes, that is, it can adjust at least one operating parameter related to the thermal treatment system 4 and / or the stirrer 3 according to:

[0113] 1) The value of the output U11 of the first control module 6 only;

[0114] 2) The values ​​of the outputs ( U11 , U21 ) of both the first control module 6 and the second control module 8 .

[0115] Obviously, the invention involves the use of the first control module 6; the value of the output U21 of the second control module 8 used only for the drive module 7 can be used in certain special cases and only in the initial phase, for reasons better explained below.

[0116] Therefore, it should be noted that during normal machine operation there may be times when the drive module 7 drives the heat treatment system 4 and / or the stirrer 3 based on only one output (U11 or U21) or based on both outputs.

[0117] It should also be noted that the values ​​of the output U11 of the first control module 6 and the output U21 of the second control module 8 already take into account (ie are based on) the value of the signal of the at least one sensor S.

[0118] It should be noted that the term "output" is used in this disclosure to denote a software variable or parameter that provides a (numeric and / or alphanumeric) value.

[0119] In practice, the control unit 5 comprises hardware and software elements defining the above-mentioned modules (6, 7 and 8).

[0120] According to another aspect, the second control module 8 comprises a second output U22 connected to one I2 of the inputs of the first layer among the layers S1, S2, S3, S4, S5, S6 of the neural network RN to send control information to the first neural network control module 6.

[0121] Basically, according to this aspect, the second control module 8 is coupled to the first control module 6 to exchange information therewith.

[0122] In particular, the output U22 of the second control module 8 constitutes the input I2 of the first control module 6 .

[0123] With specific reference to the case where the first control module 6 is a neural network RN, it should be noted that the output U22 of the second control module 8 constitutes the input I2 of the first layer S1 among the layers S1, S2, S3, S4, S5, S6 of the neural network RN.

[0124] According to one aspect, at least one sensor S is configured to detect a value related to the power supply of a drive motor 20 connected to the stirrer 3 to drive the same in rotation.

[0125] In other words, the sensor S may detect the current and / or voltage and / or power value of the motor 20 driving the stirrer 3 .

[0126] The sensor S may be of any type as long as it is compatible and can detect one or more of the current and / or voltage and / or power of the motor 20 driving the stirrer 3 .

[0127] It can also be integrated in an additional control and drive unit (drive) of the drive motor of the stirrer 3 .

[0128] According to another aspect, the second control module 8 is configured to derive parameters related to the consistency and / or viscosity of the product being processed in the container 2 from values ​​related to the power supply of the motor 20 driving the stirrer 3 .

[0129] Preferably, the parameter related to the consistency and / or viscosity of the product being processed in the container 2 is comprised between a minimum value corresponding to the product to be processed and a maximum value corresponding to the finished product.

[0130] Preferably, the parameter related to the consistency and / or viscosity of the product being processed in the container 2 is a value between 0 and 100, wherein 0 corresponds to the product at the beginning of the cycle, i.e. when the base ingredients are not yet batch frozen, and 100 corresponds to the final ice cream product at the end of processing.

[0131] According to another aspect, the drive module 7 is configured to adjust at least one operating parameter related to the heat treatment system 4 and / or the stirrer 3 only according to the value of the output U21 of the second control module 8 within a preset time length at the beginning of the processing cycle.

[0132] Basically, according to this aspect, the drive module 7 is activated only on the basis of the value of the output U21 of the second control module 8 for a preset length of time (preferably less than 30 seconds) at the beginning of a machining cycle.

[0133] Advantageously, the fact that only the value of the output U21 of the second control module 8 is used as the basis for the drive module 7 allows capturing multiple samples of the signal of the sensor S which, together with other information, will constitute the value of the input for the neural network RN; only at this point, when multiple input values ​​have been sampled, will it be possible to use the first control module 6 (by itself or in combination with the second module 8) to send control information to the drive module 7.

[0134] In fact, it should be noted that the first control module 6 needs to have as input a certain number of values ​​captured at different times in order to be able to calculate a value for at least one output U11 .

[0135] Basically, when a sufficient, predetermined amount of information (i.e. the number of data samples) is statistically available, input data can be provided to the control module 6 so as to be able to calculate at least one output U11 (which can therefore be used by the drive module 7 to allow adjustment of the heat treatment system 4 and / or the agitator 3 when necessary).

[0136] With reference to this aspect, note the following aspects.

[0137] It should be noted that the first control module 6 preferably allows to perform fast classification by iterative refinement. In other words, early classification is performed based on different time intervals (unlike that performed by the classifiers of the prior art) without waiting for the end of the processing cycle, and the first control module 6 can perform classification at different time intervals, wherein the number of data samples captured by the sensor S gradually increases in number. This has the advantage of performing an initial classification at the beginning of a batch freezing process, giving the machine a lot of room to correct possible mixture imbalances, for example by implementing a slower but more "robust" production process.

[0138] Basically, the first control module 6 can increase its classification reliability as more and more time passes after the start of processing, and thus as the number of data samples captured by the sensor S and used as input increases; however, it is activated almost immediately after the start of the cycle and does not need to wait for the end of the cycle.

[0139] According to another aspect, the heat treatment system 4 is a thermodynamic system.

[0140] Preferably, the heat treatment system 4 is a system of the thermodynamic type.

[0141] In other words, the heat treatment system 4 comprises a circuit using a heat exchanger fluid and includes: an evaporator 10 , a compressor 11 , a throttling element 12 and a condenser 13 .

[0142] Preferably, the circuit is implemented as a thermodynamic vapour compression cycle.

[0143] The heat exchanger fluid is a refrigerant fluid.

[0144] According to another aspect, the drive module 7 is configured to adjust one or more of the following operating parameters:

[0145] - driving power of the compressor 11;

[0146] - the opening of the throttling element 12;

[0147] - rotation speed of the stirrer 3;

[0148] - Processing cycle time (i.e. the moment when the processing cycle stops).

[0149] It should be noted that, according to the above described architecture, the loop configurable to control the machine 1 allows the use of the output of the first neural network control module 6 as input to the drive module 7 .

[0150] As better explained below, when describing a specific mode of training a neural network, it is advantageous that the machine 1 can be trained in such a way that it effectively completes an ice cream production cycle even when the mixture is not perfectly balanced, i.e. the ratio of one or more ingredients deviates from the optimal (theoretical) ratio.

[0151] According to one aspect, the at least one sensor S is configured to detect at least one of the following parameters:

[0152] - the operating pressure of the heat exchanger fluid at a predetermined point in the circuit of the heat treatment system 4;

[0153] - the operating temperature of the heat exchanger fluid at a predetermined point in the circuit of the thermal treatment system 4;

[0154] - operating temperature of container 2;

[0155] - the operating temperature of the heat exchanger fluid exiting the evaporator 10 associated with the container 2;

[0156] - the operating temperature of the heat exchanger fluid flowing into the evaporator 10 associated with the container 2;

[0157] - operating pressure of the heat exchanger fluid leaving the compressor 11;

[0158] - the operating pressure of the heat exchanger fluid leaving the evaporator 10;

[0159] - Processing cycle time;

[0160] - The temperature of the ambient air near machine 1.

[0161] It should be noted that, more generally, the machine 1 may be equipped with a plurality of sensors S, each configured to detect a parameter from among those previously mentioned.

[0162] Thus, in particular, the machine 1 may comprise a pressure sensor S configured to detect the operating pressure of the heat exchanger fluid at a predetermined point in the circuit of the heat treatment system 4 .

[0163] According to another aspect, the machine 1 may comprise a temperature sensor S configured to detect the operating temperature of the heat exchanger fluid at a predetermined point in the circuit of the heat treatment system 4 .

[0164] According to another aspect, the machine 1 may include a sensor S configured to detect the operating temperature of the heat exchanger fluid flowing out of the evaporator 10 , wherein the evaporator 10 is associated with the container.

[0165] According to another aspect, the machine 1 may comprise a sensor S configured to detect the operating temperature of the container 2 .

[0166] According to another aspect, the machine 1 may comprise a sensor S configured to detect the operating temperature of the heat exchanger fluid exiting the evaporator 10 associated with the container 2 .

[0167] According to another aspect, the machine 1 may include a sensor S configured to detect the operating temperature of the heat exchanger fluid flowing into the evaporator 10 , wherein the evaporator 10 is associated with the container 2 .

[0168] According to another aspect, the machine 1 may include a sensor S configured to detect the operating pressure of the heat exchanger fluid exiting the compressor 11 .

[0169] According to another aspect, the machine 1 may include a sensor S configured to detect the operating pressure of the heat exchanger fluid exiting the evaporator 10 .

[0170] According to another aspect, the machine 1 may comprise a sensor S configured to detect the machining cycle time (a sensor of this type may also be a software variable of the control unit 5 ).

[0171] According to another aspect, at least one sensor S is configured to detect an acceleration value of at least a portion of the machine.

[0172] According to another aspect, the inputs I1, I2, I3, I4, I5, I6, I7 include one or more of the following inputs:

[0173] - a first input (I1) associated with (ie representing) a value representative of the consistency or viscosity of the product being processed in the container 2;

[0174] a second input I2 associated with (i.e. representing) a value representative of the operating temperature of the container 2;

[0175] a third input I3 associated with (ie representing) a value representative of the operating temperature of the heat exchanger fluid flowing into the evaporator 10 associated with the container 2 ;

[0176] a fourth input I4 associated with (ie representing) a value representative of the operating temperature of the heat exchanger fluid exiting the evaporator 10 associated with the container 2 ;

[0177] a fifth input I5 associated with (ie representing) a value representative of the operating pressure of the heat exchanger fluid exiting the compressor 11 ;

[0178] a sixth input I6 associated with (ie representing) a value representative of the operating pressure of the heat exchanger fluid exiting the evaporator 10 ;

[0179] A seventh input I7 associated with (ie representing the value of) the opening degree of the throttle element 12 .

[0180] In case the first module 6 comprises a neural network RN, these inputs are inputs to the first of the layers S1 , S2, S3, S4, S5, S6 of the neural network RN.

[0181] It should be noted that these inputs I1 , I2, I3, I4, I5, I6, I7 are chosen because for the ice cream industry they are considered to be the most important parameters for controlling and monitoring the process cycle, that is to say for the machine 1 to be suitable for making ice cream.

[0182] It should be understood that the present description covers embodiments of the machine 1 in which there is a neural network RN having only one input I1, or having n of two to seven of the above inputs selected in any manner.

[0183] It is particularly important to emphasize that among the above-mentioned inputs I1, I2, I3, I4, I5, I6, and I7, the first input I1 and the sixth input I6 appear to be the most important or have the greatest weight (i.e., influence) in determining at least one output U11 of the first control module 6 (determined based on the training tests conducted).

[0184] According to another aspect, the first control module 6 is configured to classify the mixture being processed in the container 2 based on a balance of its components, at least one output U11 of the first control module 6 representing the classification of the mixture being processed in the container 2 .

[0185] In other words, according to this aspect, at least one output U11 represents the probability that the mixture being processed is balanced with respect to one or more of the base components or the totality of the base components.

[0186] As is known, in order to obtain an ice cream of optimal quality, balanced ingredients are essential. Therefore, information on whether the ingredients are balanced (or unbalanced) is essential to be able to take corrective actions if necessary (i.e. control the thermal treatment system 4 or the stirrer 3 accordingly) so that the base mix can be processed in an optimal manner.

[0187] Advantageously, according to another aspect, classifying the type of unbalanced mixture becomes the basis for being able to identify the optimal freezing point, thereby increasing the probability that the batch freezing process can be completed.

[0188] Firstly, for the operation of the control unit it is sufficient to observe whether the mixture being processed is balanced or unbalanced with respect to one or more base components or the totality of the base components.

[0189] According to another aspect, the first control module 6 is configured to perform a multi-class classification of the mixture based on a balance of components in the mixture being processed in the container 2 .

[0190] According to this aspect, the first control module 6 comprises a plurality of outputs U11 , U12 , U13 , U14 , U15 , U16 , U17 , U18 , U19 .

[0191] According to this aspect, in the case where the first control module 6 comprises a neural network RN, the last of the layers S1, S2, S3, S4, S5, S6 of the neural network RN comprises a plurality of outputs U11, U12, U13, U14, U15, U16, U17, U18, U19.

[0192] Each output U11 , U12 , U13 , U14 , U15 , U16 , U17 , U18 , U19 is associated with (ie represents) a class of mixtures in which the base components are balanced or unbalanced in a predetermined manner.

[0193] In a basic embodiment, the first control module 6 has only one output U11 which indicates the degree to which one or all of the basic components are balanced or unbalanced.

[0194] More precisely, each output U11, U12, U13, U14, U15, U16, U17, U18, U19 provides a value between a minimum and a maximum value, preferably between 0 and 1, representing the probability that the product processed in the container 2 corresponds to a preset category.

[0195] According to one aspect, the aforementioned outputs U11, U12, U13, U14, U15, U16, U17, U18, U19 of the first control module 6 are collectively associated with (i.e., they represent) two or more of the following categories of mixtures being processed:

[0196] - the product being processed in container 2 is balanced with respect to at least one basic ingredient;

[0197] - the product being processed in vessel 2 is unbalanced with respect to water content;

[0198] - the product being processed in vessel 2 is unbalanced with respect to fat;

[0199] - The product being processed in vessel 2 is unbalanced with respect to sugars.

[0200] According to a particular embodiment, the aforementioned outputs U11, U12, U13, U14, U15, U16, U17, U18, U19 of the first control module 6 are collectively associated with (i.e., they represent) two or more of the following categories of mixtures being processed:

[0201] - the product being processed in container 2 is balanced with respect to at least one basic ingredient;

[0202] - the product being processed in vessel 2 is unbalanced with respect to water content;

[0203] - the product being processed in container 2 is unbalanced due to excess fat;

[0204] - the product being processed in container 2 is unbalanced due to lack of fat;

[0205] - the product being processed in vessel 2 is unbalanced due to lack of sugar;

[0206] - The product being processed in vessel 2 is unbalanced due to excess sugar.

[0207] According to one aspect, the control unit 5 is configured to transmit data to at least one I1 of the inputs of the first control module 6 at preset time intervals preferably less than 30 seconds, more preferably still less than 15 seconds, or advantageously less than 5 seconds.

[0208] The above is a description of the structure of the control unit 5 with particular reference to the first control module 6 .

[0209] The process for training the neural network RN is now described below, that is to say how the above-mentioned neural network is trained.

[0210] It should be understood that what is described below applies to the training of a first control module 6 comprising a neural network RN, although some aspects of the described process, particularly with regard to the type of mixture used, also apply to a control module 6 comprising another type of artificial intelligence classifier CL.

[0211] Training a neural network RN means finding the weights of the activation functions to give the best possible performance.

[0212] In general, training a neural network involves adjusting the weights of the neuron-to-neuron connections to minimize the prediction error of the neural network model.

[0213] As is known, the process for training a neural network occurs in several steps:

[0214] -1) Weight initialization: Initialize the weights of neuron-to-neuron connections with random values;

[0215] -2) Feedforward: The input is presented to the neural network, which processes the input by passing it through different layers of neurons, each of which has an activation function on its weighted input.

[0216] -3) Error calculation: The error is calculated by comparing the output produced by the neural network with the expected value.

[0217] -4) Back Propagation: The error is propagated backwards through the neural network, modifying the weights of the neuron-to-neuron connections to reduce the error.

[0218] Repeat steps 2 to 4 a fixed number of times or until the error reaches a predetermined acceptable value.

[0219] Preferably, after being trained, the neural network RN is validated on an additional data set (test data) to evaluate its performance on new data.

[0220] To train the neural network RN, the machine 1 is put into operation, ie processing cycles are performed with different kinds of mixtures loaded in the container 2, balanced to different degrees, that is to say with different classifications as to the balance of the components, as described in more detail below.

[0221] For example, with reference to one type of mixture, whether balanced or unbalanced with respect to its base components, multiple tests are performed (preferably more than 5 times for each type of mixture), and during training, a probability of 1 is assigned to the output associated with the desired class of the mixture, and a probability of 0 is assigned to the output associated with the class of the other mixture.

[0222] The mixtures used for training are as follows (taking into account that for each type of mixture, several batches are used in order to train Machine 1 with respect to small changes in the mixture composition):

[0223] - Balanced mixtures;

[0224] - Unbalanced mixture with excess fat;

[0225] - Unbalanced mixtures with excess sugar (sucrose);

[0226] - An unbalanced mixture lacking sugar (sucrose);

[0227] - Unbalanced mixture with excess water.

[0228] More precisely, the following 9 mixtures were used for training:

[0229] - Balanced mixtures;

[0230] - an unbalanced mixture with a first predetermined percentage excess of fat (preferably cream) (preferably corresponding to 3 times the optimal balance);

[0231] - an unbalanced mixture having an excess of fat (preferably cream) by a second predetermined percentage (preferably corresponding to 4.4 times the optimal balance);

[0232] - an unbalanced mixture with a first predetermined percentage excess of sugar (sucrose) (preferably corresponding to +1 / 3 of the optimal balance);

[0233] - an unbalanced mixture with a second predetermined percentage excess of sugar (sucrose) (preferably corresponding to +2 / 3 of the optimal balance);

[0234] - an excess of sugar (sucrose) in excess of a third predetermined percentage of the unbalanced mixture (preferably corresponding to 100% of the optimal balance);

[0235] - an unbalanced mixture lacking a first predetermined percentage of sugar (sucrose) (preferably corresponding to -1 / 3 of the optimal balance);

[0236] - an unbalanced mixture lacking sugar (sucrose) having a second predetermined percentage (preferably corresponding to -2 / 3 of the optimal balance);

[0237] - an unbalanced mixture lacking a second predetermined percentage of sugar (sucrose) (preferably corresponding to +10% compared to the optimal balance);

[0238] For example, Figure 6 The composition of the test mixtures (in percent) is shown for the following mixtures respectively:

[0239] - Balance (first row),

[0240] - Imbalance, excess water (second row),

[0241] - Unbalanced, excess fat (cream), more than three times the optimal balance (third row),

[0242] - Unbalanced, excess fat (cream), more than 4.4 times the optimal balance (fourth row),

[0243] - Imbalance, lack of sugar (sucrose), less than 2 / 3 of the optimal balance (fifth row),

[0244] - Imbalance, lack of sugar (sucrose), less than 1 / 3 of the optimal balance (sixth row),

[0245] - Imbalance, excess sugar (sucrose), more than 1 / 3 of the optimal balance (the seventh row),

[0246] - Imbalance, excess sugar (sucrose), more than 2 / 3 of the optimal balance (eighth row),

[0247] - Imbalance, excess sugar (sucrose), greater than 100% of the optimal balance (ninth row).

[0248] The abbreviation "LMS" means lean milk solids.

[0249] It should be noted that greater accuracy may be achieved by using more mixtures, such as more than one unbalanced mixture with excess water, and / or an unbalanced mixture with insufficient fat.

[0250] As can be easily understood, a test mix is ​​loaded into the container 2 and the machine 1 is started so as to carry out an ice cream production cycle on the test mix.

[0251] Thus, during this cycle, the neural network RN receives as input the values ​​coming from the sensor S, ie at its input as described above, and sets at the output of the neural network RN a label corresponding to the correct classification of the mixture being processed in the container 2 .

[0252] It should be noted that, in practice, the neural network RN is constructed to act as a classifier.

[0253] Preferably, training is performed for between 100 and 600 epochs, more preferably, training is performed for less than 250 epochs.

[0254] Preferably, training is performed by an Adam optimizer.

[0255] More preferably, 12 samples are used in each batch for training.

[0256] In the tests conducted, it was observed that sugar had the effect of lowering the freezing point of the processed mixture.

[0257] Thus, a mix that is only lightly sweetened (low in sugar) will tend to freeze faster, risking ice cubes forming before the desired consistency is reached, whereas a mix that is high in sugar will tend to retain more liquid because it is frozen at a lower temperature.

[0258] As described in more detail below, it should be noted that knowledge of the type of mix, and in particular the sugar balance, is obtained by classifying the mix via the first control module 6 , thus allowing the actions aimed at completing the batch freezing cycle to be implemented in an optimal way according to the type of product processed.

[0259] Indeed, as mentioned above, for example in the case of sugar, knowing by means of the first module 6 whether the mixture is unbalanced (preferably but not necessarily in excess or in deficiency) allows appropriate corrective measures to be taken.

[0260] Figure 7 Shown are the trends of some parameters captured by the sensor S during a production process with a balanced mixture.

[0261] More precisely, in Figure 7 In the graph shown, R1 is the trend over time (x-axis) of a parameter associated with the consistency and / or viscosity of the product being processed in the container 2, which parameter is derived from the value related to the power supply of the motor of the agitator 3; this parameter is represented by the value on the y-axis and varies from 0 to 100.

[0262] exist Figure 7 , R2 refers to the temperature on the inlet side of the evaporator 10, R3 refers to the temperature on the outlet side of the evaporator 10, and R4 refers to the difference between the two temperatures.

[0263] It can be inferred from the graph that the parameter associated with the consistency and / or viscosity of the product being processed in the container 2 tends to increase during processing until reaching a maximum value (corresponding to 100) when the product is ready.

[0264] Experiments have shown that parameters related to the consistency and / or viscosity of the product processed in container 2 have different trends, which are a function of the balance of the mixture type relative to one or more ingredients, or of anomalies that may occur during processing (e.g. frozen blocks).

[0265] refer to Figure 8 , derived during supervised training by sampling trends in parameters associated with the consistency and / or viscosity of the products of different mixtures processed in the vessel 2, noting that:

[0266] -Mixture M1 is unbalanced, with 3 times the amount of cream;

[0267] -Mixture M2 is unbalanced, with 4.4 times the amount of cream;

[0268] -Mixture M3 is unbalanced and has 10% water;

[0269] -Mixture M4 is balanced.

[0270] As can be observed, in terms of equilibrium, the parameters related to the consistency and / or viscosity of the product processed in the container 2 reach a consistency corresponding to the finished product (ie substantially equal to 100) at different times for different types of mixtures.

[0271] It can be noted that the unbalanced mixtures (M1, M2, M3) reach the maximum consistency in less time than the balanced mixture (M4).

[0272] It can therefore be easily understood from this example that the knowledge of the classification of the mixture, through the control module 6, allows the drive module 7 to control the machine in an optimal way so as to allow ending a batch freezing cycle with a product of the highest quality, based on the combination of the output of the first control module 6 (the classification of the mixture) and the output of the second control module 8 (the parameter value related to the consistency and / or viscosity of the product), regardless of the balance of the ingredients.

[0273] refer to Fig. 9 , derived during the supervised training by sampling the trend of the pressure values ​​of the heat exchanger fluid of the thermodynamic system 4 downstream of the compressor 11, it can be observed that:

[0274] - Mixture M1 is unbalanced, with 3 times the amount of cream;

[0275] -Mixture M2 is unbalanced, with 4.4 times the amount of cream;

[0276] -Mixture M3 is unbalanced and has 10% water;

[0277] -Mixture M4 is balanced.

[0278] from Fig. 9 It is evident that, for different mixtures, the pressure of the heat exchanger fluid of the thermodynamic system 4 downstream of the compressor is different along the cycle.

[0279] For the first control module 6 , this input therefore constitutes a value that can be reliably used for classification purposes.

[0280] The graph also shows that for the unbalanced mixture M3 with respect to water, the pressure during the processing cycle is lower than that of the balanced mixture M4.

[0281] Fig.10 An unbalanced mixture ("B") is shown compared to an equilibrium mixture ("A"):

[0282] - Curve X1 represents the consistency or viscosity of the mixture measured over time;

[0283] - Curve X2 represents the opening of the throttling element 12;

[0284] - curve X3 represents the temperature of the heat exchanger fluid flowing out of the evaporator 10 of the thermodynamic system 4;

[0285] The curve X4 represents the temperature of the heat exchanger fluid flowing into the evaporator 10 of the thermodynamic system 4 .

[0286] exist Fig.10 , the areas marked P1 and P2 correspond to the shut-down of the compressor 11, which corresponds to the end of the batch freezing cycle.

[0287] It should be noted that if the ice cream is not removed from the container 2 when the compressor 11 is switched off, the machine 1 restarts the compressor 11 and the stirrer 3 (areas P3 and P4, not relevant here) in order to keep the ice cream at a predetermined temperature.

[0288] It should be noted that Fig.10 Corresponds to the tests performed on the machine 1 during the training of the first neural network control module 6 .

[0289] Referring to the unbalanced mix "B" and to line LO, which represents the optimum consistency, it is noted that the consistency curve X1, ie the viscosity of the mix, does not reach the optimum consistency LO and therefore the ice cream produced is of sub-optimal quality.

[0290] In this case, in the case of an unbalanced mixture “B”, the presence of a special control cycle with the first control module 6 will allow the drive module 7 to stop the processing cycle at point P5 , which definitely corresponds to a more stable consistency than at point P2 .

[0291] The above is clear evidence of the technical performance enhancements that can be achieved with the machine 1 of the present invention.

[0292] The control actions to be implemented by the drive module 7 will be described below by way of non-limiting examples.

[0293] Firstly, the expression "adjusting at least one operating parameter related to the heat treatment system (4) and / or the agitator (3)" is used to indicate that the drive module 7 operates on the heat treatment system 4 and / or the agitator 3 (specifically, on the motor connected to the agitator 3) to control their functions.

[0294] It should be noted that the drive module 7 may also operate on other components of the machine 1 .

[0295] It should also be noted that the drive module 7 can advantageously operate on one or more of the following components of the heat treatment system 4 : the compressor 11 and / or the throttling element 12 and / or components acting on the condenser 13 (cooling fan and / or cooling water pump).

[0296] Described below are some situations that may occur during operation of the machine 1 due to the composition of the mixture and / or specific process conditions.

[0297] In these cases, the manner in which the control unit 5 regulates operating parameters associated with the heat treatment system 4 and / or the stirrer 3 will be described.

[0298] Under certain conditions during processing, a frozen mass of ice cream may be produced in the middle of the mixer 3 .

[0299] The control module 8 may be configured to detect such potential anomalies, that is, anomalies involving the frozen ice cream mass in the middle of the blender 3, by one or more of the following sensor signals:

[0300] - the value associated with the consistency of the product decreases;

[0301] - and / or increased superheat at the suction end of the refrigeration compressor.

[0302] In these cases, if the first control module 6 classifies the mixture as an unbalanced mixture (if the classification is multi-class, preferably an imbalance with excess fat), the drive module 7 can be configured to command one or more of the following actions (so that the batch freezing cycle can be completed in an optimal manner):

[0303] - Increase the opening of the pressure reducing element (throttle valve) to reduce overheating at the inlet of the compressor 11;

[0304] - intermittently, preferably in pulse mode (preferably at high speed) driving the motor of the stirrer 3;.

[0305] - Reverse the direction of rotation of the motor of the stirrer 3

[0306] Advantageously, thanks to these actions, the processing cycle can be completed. It should be noted that the fact that the first control module 6 has classified the mixture as an unbalanced mixture advantageously confirms the anomaly detected by the drive module 7: in fact, experiments have shown that when the mixture is unbalanced, it generally tends to form a frozen mass of ice cream in the middle.

[0307] In this way, the corrective actions taken allow the processing cycle to be completed efficiently, resulting in the best quality product regardless of whether the base ingredients are balanced.

[0308] In another example, under certain conditions, a layer of ice may form on the inner wall of the container 2 .

[0309] The control module 8 may be configured to detect such potential anomalies, that is, anomalies involving the ice layer on the inner wall of the container 2, by one or more of the following sensor signals:

[0310] - a change in the temperature at the inlet and outlet of the evaporator 10 (ie these values ​​differ from the expected or desired or previous values).

[0311] In these cases, if the first control module 6 classifies the mixture as an unbalanced mixture (if the classification is multi-class, preferably, an imbalance with a lack of sugar and / or fat), the drive module 7 can be configured to command one or more of the following actions (so that the batch freezing cycle can be completed in an optimal way):

[0312] - regulating (reducing) the driving power of the compressor 11;

[0313] - Activating a mode for injecting hot gas into the evaporator 10 , that is, a mode for heating the evaporator 10 .

[0314] It should be noted that in order to inject hot gas into the evaporator 10 to heat the evaporator 10, that is, in order to operate the evaporator 10 for a certain length of time as a heat exchanger that delivers heat (heat) rather than absorbs heat (cooling), the thermal treatment system 4 must be configured so that the evaporator 10 can operate in two alternative modes:

[0315] - (1) A configuration known as a hot gas cycle (not shown because it is known to anyone familiar with the industry)

[0316] - or (2) operate like a heat pump (not shown as it is known to anyone familiar with the industry).

[0317] In these configurations, the heat treatment system 4 is provided with one or more valves operable by the control unit 5 to enable switching between a cooling mode of the evaporator 10 and an operating mode in which the evaporator 10 is heated by injecting hot gas. In another example, during processing, the product may reach an optimum consistency in a time that is too short compared to the expected solidification time.

[0318] This usually means that the final temperature of the ice cream is incorrect, that is, it is not as cold as desired for high-quality ice cream.

[0319] The control module 8 may be configured to detect such potential anomalies, that is, anomalies involving a value of consistency being reached in a time shorter than a predetermined set time, by one or more of the following sensor signals:

[0320] - a signal of the consistency of the product compared to the start time of the processing cycle (for example, a signal detected indirectly by the power input value of the motor 20 of the stirrer 3);

[0321] - The temperature signal of process vessel 2 deviates in time from the expected time value or curve.

[0322] In these cases, if the first control module 6 classifies the mixture as an unbalanced mixture (if the classification is multi-class, preferably, an imbalance with a sugar deficiency and / or fat excess), the drive module 7 can be configured to command one or more of the following actions (so that the batch freezing cycle can be completed in an optimal manner):

[0323] - increase the value of the consistency associated with the product, which corresponds to the end of the processing cycle (e.g. allowing stopping at a standard level above the value of the consistency);

[0324] - increasing the opening of the throttle element 12 (this advantageously allows making the inner wall of the container 2 less cold and / or reducing the consumption of the stirrer 3 required to scrape the ice from the surface of the inner wall of the container 2, which may change the consistency value);

[0325] - Reducing the cooling capacity of the thermal system 4 , in particular reducing the capacity of the compressor 11 .

[0326] In another example, as the product is processed, the detected value related to the consistency of the product, which is known to be one of the main parameters affecting the quality of ice cream products, may become lower.

[0327] The control module 8 may be configured to detect such potential anomalies, that is, anomalies involving a decrease in the consistency of the product compared to an expected value or an expected trend over an expected (set) time, by one or more of the following sensor signals:

[0328] - A signal related to the power input (voltage and / or current) of the motor 20 of the stirrer 3 .

[0329] In these cases, if the first control module 6 classifies the mixture as an unbalanced mixture, the drive module 7 may be configured to command one or more of the following actions (so that the batch freezing cycle can be completed in an optimal manner):

[0330] - reducing the cooling power of the thermal system 4, in particular reducing the power of the compressor 11;

[0331] - Injecting hot gas into the evaporator 10 , that is to say heating the evaporator 10 .

[0332] As regards the configuration of the thermal system 4 which enables the injection of hot gas into the evaporator 10 , this also applies as described above with reference to the frozen ice cream mass in the middle of the mixer 3 .

[0333] In another example, the detected value related to product consistency, which is one of the main parameters affecting the quality of the ice cream product, is known and may not vary over time while the product is processed.

[0334] The control module 8 may be configured to detect such potential anomalies, that is, anomalies involving consistency that is constant over time compared to an expected value or expected trend at an expected (set) time, by one or more of the following sensor signals:

[0335] - a signal related to the power input (voltage and / or current) of the motor of the stirrer 3;

[0336] a temperature signal representative of the temperature of the refrigerant fluid flowing into and / or out of the evaporator 10 .

[0337] In these cases, if the first control module 6 classifies the mixture as an unbalanced mixture, the drive module 7 can be configured to command one or more of the following actions (so that the batch freezing cycle can be completed in an optimal manner):

[0338] - regulating (reducing) the cooling power of the thermal system 4, in particular reducing the power of the compressor 11;

[0339] - Injection of hot gas into the evaporator 10.

[0340] As regards the configuration of the thermal system 4 which enables the injection of hot gas into the evaporator 10 , the configuration described above with reference to the frozen ice cream mass in the middle of the mixer 3 also applies in this case.

[0341] In another example, the temperature of the heat exchanger fluid flowing out of the evaporator 10 may drop suddenly, that is, the temperature value of the heat exchanger fluid flowing out of the evaporator 10 may drop below a certain threshold value in less than a preset time.

[0342] The control module 8 may be configured to detect such potential anomalies through one or more of the following sensor signals, that is,

[0343] Abnormality that the temperature of the heat exchanger fluid flowing out of the evaporator 10 drops below a preset value within a preset time:

[0344] - A signal related to the power input (voltage and / or current) of the motor of the stirrer 3 .

[0345] a temperature signal representative of the temperature of the refrigerant fluid flowing into and / or out of the evaporator 10 .

[0346] In these cases, if the first control module 6 classifies the mixture as an unbalanced mixture, the drive module 7 may be configured to command one or more of the following actions (so that the batch freezing cycle can be completed in an optimal manner):

[0347] - reducing the cooling power of the thermal system 4, in particular reducing the power of the compressor 11;

[0348] - Injecting hot gas into the evaporator 10, that is to say operating the evaporator 10 in heating mode.

[0349] As for the structure of the thermal system 4 which enables the injection of hot gas into the evaporator 10, the structure described above with reference to the frozen ice cream mass in the middle of the mixer 3 also applies in this case.

[0350] In another example, processing may continue too long.

[0351] The control module 8 may be configured to detect such potential anomalies, that is, anomalies involving a processing time length greater than a set or expected time length, through one or more of the following sensor signals:

[0352] - a processing time signal (e.g. obtainable via a timer of the machine 1);

[0353] - a signal representing the trend over time of the consistency of the product being processed in the container 2;

[0354] - the pressure value of the refrigerant fluid downstream of the compressor 11;

[0355] - a temperature signal representative of the temperature of the air of the external environment in the vicinity of the machine 1;

[0356] - Temperature signal at the outlet of compressor 11.

[0357] In these cases, if the first control module 6 classifies the mixture as an unbalanced mixture (if the classification is multi-class, in particular, an imbalance with an excess of fat or sugar), the drive module 7 can be configured to command one or more of the following actions (so that the batch freezing cycle can be completed in an optimal way):

[0358] - regulating (increasing) the cooling power of the thermal system 4, in particular increasing the power of the compressor 11;

[0359] - For an air condenser, adjust (increase) the flow rate of cooling air towards the condenser 13, or for a water condenser, adjust (increase) the flow rate of water towards the condenser 13.

[0360] Advantageously, as can be inferred from the numerous examples listed above, the machine 1 of the invention allows to reliably process ice cream made from any kind of mix, whether balanced or not with respect to one or more of the ingredients, and to obtain optimal results at any time and under any conditions, since the machine 1 of the invention is able to quickly identify and correct the main anomalies that could adversely affect the quality of the ice cream.

[0361] It is also advantageous that the machine 1 of the invention makes it possible to assess when it is indeed the right time to take out the ice cream and / or to signal “ice cream is ready”, ie when the texture of the mix is ​​at its optimum.

[0362] It should also be considered that, under certain conditions, the ice cream machine intentionally uses an unbalanced mixture (because certain recipes require that certain types of ingredients be loaded into the container 2 in less than optimal proportions, these ingredients being sugar, water or fat); therefore, under such conditions, the machine 1 can produce ice cream with optimal characteristics even if the mixture that must be used is not perfectly balanced.

[0363] The machine of the invention thus makes it possible to manufacture products of optimal quality while offering the possibility of creating new products in order to expand the product range of the ice cream industry.

[0364] It should also be considered that the machine 1 of the invention advantageously allows to process ice cream in the shortest possible time and to stop the batch freezing cycle when the ice cream is really ready, that is to say when its consistency and texture are just right to optimize its quality.

[0365] Furthermore, the machine 1 of the present invention advantageously allows for prevention of a number of problems that may arise during processing and for prompt action to be taken to resolve these problems.

[0366] Furthermore, the machine 1 of the invention is very easy to use, since it requires no special settings, and also compensates for any errors that the operator may make in balancing the ingredients, always producing ice cream of optimal quality.

[0367] According to another aspect, a method for controlling a machine 1 for preparing liquid or semi-liquid food products is also defined, the method comprising the following steps:

[0368] - A machine 1 for preparing a liquid or semi-liquid food product is provided, the machine comprising:

[0369] - Processing vessels for processing basic products2;

[0370] - a stirrer 3 suitable for mixing the product in the processing vessel 2 in which the base product is processed;

[0371] a thermal treatment system 4 configured to exchange heat with the process vessel 2 in which the base product is processed;

[0372] - at least one sensor S, which is configured to detect an operating parameter of the machine;

[0373] - providing a first artificial intelligence control module 6 configured to define a classifier having at least one input I1, I2, I3, I4, I5, I6, I7 and at least one output U11, wherein at least one input I1 is connected to at least one sensor S;

[0374] - training the first artificial intelligence control module 6 based on the processing of different mixtures in the container 2, thereby obtaining a first trained control module 6;

[0375] - feeding the mixture into container 2 and starting the processing cycle;

[0376] - driving at least the heat treatment system 4 and / or the stirrer 3 according to the value of at least one output U11 of the first control module 6 to adjust at least one operating parameter related to the heat treatment system 4 and / or the stirrer 3.

[0377] According to one aspect, the first control module 6 comprises a neural network RN.

[0378] According to another aspect, at least one output U11 represents a classification of the mixture of base products and the method comprises a step of classifying, via the first control module 6 , the mixture processed in the container 2 into one or more preset categories with respect to the balance of the base ingredients.

[0379] According to another aspect, the inputs I1, I2, I3, I4, I5, I6, I7 of the first control module (6) include one or more of the following inputs:

[0380] - a first input I1 associated with a value representative of the consistency and / or viscosity of the product being processed in the container 2;

[0381] - a second input I2 associated with a value representative of the operating temperature of the container 2;

[0382] a third input I3 associated with a value representative of the operating temperature of the heat exchanger fluid flowing into the evaporator 10 associated with the container 2 ;

[0383] a fourth input I4 associated with a value representative of the operating temperature of the heat exchanger fluid exiting the evaporator 10 associated with the container 2 ;

[0384] a fifth input I5 associated with a value representative of the operating pressure of the heat exchanger fluid leaving the compressor 11 ;

[0385] a sixth input I6 associated with a value representative of the operating pressure of the heat exchanger fluid leaving the evaporator 10 ;

[0386] A seventh input I7 associated with a value representative of the degree of opening of the throttle element 12 .

[0387] According to another aspect, the method comprises a step of monitoring the signal of at least one sensor S to determine events corresponding to anomalies in the production process.

[0388] According to this aspect, the step of driving at least the heat treatment system 4 and / or the agitator 3 includes the step of driving at least the heat treatment system 4 and / or the agitator 3 to adjust at least one operating parameter related to the heat treatment system 4 and / or the agitator 3 according to a mixture classification value corresponding to at least one output U11 of the first control module (6), and a machine process fault previously determined during the monitoring step.

[0389] That is, according to this aspect, a combination of the determined process anomaly and the classification performed by the first control module 6 adjusts one or more of the control parameters.

[0390] According to another aspect, the step of training the first control module 6 in supervised mode based on the processing of mixtures with differently balanced base ingredients inside the container 2 in order to obtain a first control module 6 trained for classifying said mixture of base products comprises the following steps:

[0391] - Processing of the following types of mixtures:

[0392] - mixtures with regard to the balance of the basic ingredients;

[0393] - Mixtures with unbalanced water content;

[0394] -About the unbalanced mixture of fats;

[0395] -About unbalanced mixtures of sugars;

[0396] - During training, the first control module 6 is provided with labels associated with the desired category of each mixture processed.

[0397] It should be noted that for all intents and purposes, the labels correspond to the type of mixture being processed, that is, for a balanced mixture, the assigned labels are used to set the output (e.g., U11) to a specific value that corresponds to the desired classification of the type of mixture being processed.

[0398] According to another aspect, the step of driving at least the heat treatment system 4 and / or the stirrer 3 to adjust at least one operating parameter associated with the heat treatment system 4 and / or the stirrer 3 according to the mixture classification value comprises the following steps:

[0399] - Check whether the determined mixture classification value corresponds to an abnormal state of the mixture, i.e. an unbalanced mixture;

[0400] - Commands are also issued to the heat treatment system 4 and / or the stirrer 3 based on the results of the check.

[0401] According to another aspect, the step of driving at least the heat treatment system 4 and / or the stirrer 3 to adjust at least one operating parameter associated with the heat treatment system 4 and / or the stirrer 3 comprises the following steps:

[0402] - comparing the detected fault with the determined classification value of said mixture;

[0403] - Based on the result of the comparison, commands are also issued to the heat treatment system 4 and / or the stirrer 3 .

[0404] According to another aspect, the step of driving at least the heat treatment system 4 and / or the stirrer 3 according to the mixture classification value to adjust at least one operating parameter associated with the heat treatment system 4 and / or the stirrer 3 comprises one or more commands for:

[0405] - varying the opening of a throttling element 12 forming part of the heat treatment system 4 (thus adjusting the throttling pressure);

[0406] - regulating the power of the compressor 11 forming part of the thermal treatment system 4;

[0407] - regulating the flow rate of the cooling fluid, air or water +, at the condenser 13 forming part of the thermal treatment system 4;

[0408] - switching the heat treatment system 4 between a cooling mode for cooling the container 2 and a heating mode for heating the container 2, the heating mode comprising an operating mode in which the heat treatment system 4 is operated by injecting hot gas, known as a hot gas cycle or a heat pump cycle;

[0409] - Adjust the speed and direction of rotation of the stirrer 3.

[0410] The control method of the present invention can prevent a number of problems that may occur during the processing of a product, in particular during the processing of ice cream, and may include, for example, the following problems:

[0411] - ice cubes in the middle of the blender 3;

[0412] - The ingredients are too cold;

[0413] - The final product has the wrong consistency and texture.

[0414] Basically, the proposed control method advantageously allows implementing the most effective corrective action depending on the characteristics of the product being processed, thus not only allowing any potential problems to be quickly resolved, but also ensuring that the manufactured product is of optimal quality, regardless of whether the base mix is ​​balanced or not.

[0415] As pointed out in the introduction, ice cream is an extremely complex product and its production therefore requires an extremely complex control system also based on the composition of the mix itself; advantageously, the proposed method allows controlling the ice cream production process in an extremely efficient and reliable manner.

[0416] According to another aspect, the machine 1 comprises a sensor S configured to detect the amount of base product placed in the process container 2 .

[0417] Alternatively, the machine 1 may comprise a user interface configured to detect the amount of base product placed in the process container 2 .

[0418] It should therefore be noted that one of the inputs ( I1 , I2 , I3 , I4 , I5 , I6 , I7 ) of the neural network RN, for example an additional input (not shown in the figures), can be associated with the value of the basic product quantity detected by the sensor S.

[0419] It should also be noted that, according to another aspect, the training process can include training the machine to recognize different levels of basic product amounts within the container 2; indeed, in this way, the first control module 6 can classify the mixture more accurately even for different amounts loaded in the container 2.

[0420] Other embodiments of the first control module 6 described below are also based on machine learning, but are a replacement for the first control module 6 being a neural network.

[0421] Where possible, all that is described above in connection with the first control module 6 (in general or with specific reference to the neural network RN) also applies to these other embodiments.

[0422] In one embodiment, the first control module 6 includes one or more decision trees.

[0423] According to this aspect, for example, the first control module 6 includes a random forest (RF) model.

[0424] A random forest model is a model composed of any number of decision trees; it should be noted that in this case the assignment to a class, i.e., the classification, as output is done by combining all the outputs (i.e., the decisions or "votes") of all the decision trees to obtain a single result.

[0425] Basically, the class provided as output by the first control module 6 is the class that obtains the majority vote.

[0426] When training the first control module 6, preferably a so-called "time series forest" is used.

[0427] It should be noted that such a time series is actually defined by a sequence of data values ​​derived (sampled) from at least one sensor S (or multiple sensors S).

[0428] Time Series Forest allows training a random forest model not with the original data but with features extracted from the time series: for example, mean, standard deviation, and slope.

[0429] Alternatively, other algorithms based on tree models can be used instead of time series forests: for example, extreme gradient boosting (XGB) or time series feature bag.

[0430] Alternatively, the first control module 6 includes a classifier configured to define a time series distance measure and a kernel, respectively, such as k-nearest neighbor (k-NN) or a support vector machine (SVM).

[0431] In another embodiment, the first control module 6 comprises an algorithm for learning a model of the “support vector machine” type.

[0432] Advantageously, a support vector machine is a model for performing binary classification, but can also be used for multi-class classification (by subdividing the first module 6 into interconnected submodules, each submodule being configured to perform binary classification).

[0433] In the case of binary classification, the first control module 6 of the support vector machine type preferably receives as input a training data set of a plurality of input time series belonging to two classes and the associated labels.

[0434] Basically, during the training process, several time series may be fed as input, each of which is associated with a specific class; as described above, these classes may include mixtures in which one or more basis components are balanced to varying degrees.

[0435] Considering a training input dataset {(Xt, yt)}, t=1, 2, ..., T, where y belongs to {0, 1}, the training process allows the construction of a decision function f(X) that correctly classifies time series that do not form part of the training time series.

[0436] During the training process, the support vector machine determines the hyperplane that separates the input data belonging to the two classes.

[0437] If the training sequences are linearly separable, then the decision function f(X) will be a linear function called the “decision boundary” (linear SVM classifier).

[0438] The decision boundary is found in such a way that the distance between the decision boundary and the nearest data item is as large as possible.

[0439] For many datasets, a hyperplane will not be a good classifier, so more geometric shapes are often employed, mapping the training data in a higher dimensional space and finding the hyperplane in the new space.

[0440] In this case, the first control module 6 comprises a kernel-based SVM: the classification of points in a multidimensional space is performed using a “kernel function” (typical choices are the polynomial kernel or RBF, ie radial basis function).

[0441] In case of multi-class classification, the first control module 6 comprises a plurality of binary SVM classifiers working in coordination; in particular, for n classes, (n(n-1)) / 2 binary classifiers must be used.

[0442] Therefore, in this case, the first classification module 6 will include a plurality of binary SVM classifiers.

[0443] This approach is called "one-vs-one multi-class classification"; according to this strategy, each trained SVM classifier is able to classify a point into two different classes.

[0444] In another embodiment, the first control module 6 comprises a k-NN (k-nearest neighbor) classifier configured to “include” new samples in the training space before classifying the new samples.

[0445] According to this aspect, the input of the first control module 6 consists of k training examples and the output is the class to which they belong.

[0446] It should therefore be noted that, according to this aspect, the first control module 6 comprising a k-NN classifier does not comprise any step of training the model.

[0447] A point is classified as the majority vote of its neighbors, that is, the class assigned to it is the most common among its “k-nearest” neighbors. For time series classification, it is necessary to verify the similarity between time series.

[0448] A "majority vote" technically requires a majority greater than 50%, which works mostly when there are only two categories. If there are more than two categories, such as four categories, then the class label with a vote greater than 25% can be assigned.

[0449] According to a further aspect, the first classification module 6 is designed to implement a so-called “bag-of-words” (dictionary-based) method.

[0450] According to this method, the time series is discretized into symbol sequences, then words are extracted from these sequences through a sliding window, and finally the number of words is counted for all words in the dictionary.

[0451] The first classification module 6 is configured to implement the above steps.

[0452] In one embodiment, the first classification module 6 is configured to implement a so-called Symbolic Aggregate Approximation (SAX): According to this aspect, the time series is discretized into symbolic sequences in order to subsequently apply sequence data mining methods.

[0453] In another embodiment, the first classification module 6 is configured to implement a symbolic Fourier approximation: the representation of the sequence is produced by representing the signal as a discrete Fourier transform, thus in the frequency domain instead of in the time domain.

[0454] It has been emphasized above that, according to the present invention, the first control module 6 can implement different classification strategies (for classifying the mixture) without departing from the specific principles of the present invention itself.

Claims

1. A machine (1) for preparing liquid or semi-liquid or semi-solid food, comprising: - a processing vessel (2) for processing the base product; - a stirrer (3) suitable for mixing the product in the processing vessel (2) in which the base product is processed; - a thermal treatment system (4) configured to exchange heat with the processing vessel (2) in which the base product is processed; - a control unit (5) connected to the thermal treatment system (4) and the stirrer (3) in order to drive them; - at least one sensor (S) configured to detect an operating parameter of the machine; The machine (1) is characterized in that the control unit (5) comprises: a first artificial intelligence control module (6) configured to define a pre-trained classifier (CL) having at least one input (I1, I2, I3, I4, I5, I6, I7) and at least one output (U11), wherein the at least one input (I1) is connected to the sensor (S); - A drive module (7), which is connected to the output (U11) of the first control module (6) and is at least connected to the heat treatment system (4) and / or the agitator (3) to adjust at least one operating parameter related to the heat treatment system (4) and / or the agitator (3) according to the value of the at least one output (U11) of the first control module (6).

2. The machine (1) according to claim 1, characterized in that The first control module (6) comprises a plurality of inputs (I1, I2, I3, I4, I5, I6, I7).

3. The machine (1) according to claim 1 or 2, characterized in that Said at least one output (U11) of said first control module (6) represents the classification of the mixture of said basic products.

4. The machine (1) according to any one of claims 1 to 3, characterized in that The first control module (6) is configured to perform classification at multiple time intervals including between the start and end of processing, thereby performing a so-called "early classification", and gradually improve the accuracy of the classification by iterative refinement at different time intervals.

5. The machine (1) according to any one of claims 1 to 4, characterized in that The first control module (6) comprises a neural network (RN), which comprises a plurality of layers (S1, S2, S3, S4, S5, S6) and has a plurality of inputs (I1, I2, I3, I4, I5, I6, I7).

6. Machine (1) according to any one of the preceding claims, characterised in that It also includes a second control module (8) connected to the at least one sensor (S), the second control module (8) being configured to process and / or detect faults based on data from the at least one sensor (S), the second control module (8) having at least a first control output (U21) connected to the drive module (7).

7. Machine (1) according to the preceding claim, characterised in that The drive module (7) is configured to adjust at least one operating parameter related to the heat treatment system (4) and / or the agitator (3) according to the value of the output (U11) of the first control module (6) and the value of the output (U21) of the second control module (8).

8. The machine (1) according to any one of claims 1 to 7 and according to claim 2, characterized in that The second control module (8) also includes a second control output (U22), which is connected to one (I2) of the inputs (I1, I2, I3, I4, I5, I6, I7) of the first control module (6) to transmit control information to the first control module (6).

9. Machine (1) for preparing liquid or semi-liquid or semi-solid food products according to any of the preceding claims and claim 6, characterized in that The at least one sensor (S) is configured to detect values ​​related to the power supply of a motor connected to the agitator (3) to drive it, and the second control module (8) is configured to derive parameters related to the consistency and / or viscosity of the product processed in the container (2) from the values ​​related to the power supply of the motor.

10. Machine (1) for preparing liquid or semi-liquid or semi-solid food products according to the preceding claim, characterised in that The parameters associated with the consistency and / or viscosity of the product processed in said container (2) are comprised between a minimum value corresponding to the product to be processed and a maximum value corresponding to the finished product.

11. The machine (1) according to any one of the preceding claims and according to claim 2, characterized in that The inputs (I1, I2, I3, I4, I5, I6, I7) of the first control module (6) include one or more of the following inputs: - a first input (I1) associated with a value representative of the consistency and / or viscosity of the product being processed in said container (2); - a second input (I2) associated with the value of the operating temperature of the container (2); - a third input (I3) associated with a value representative of the operating temperature of the heat exchanger fluid flowing into the evaporator (10) associated with the container (2); - a fourth input (I4) associated with a value representative of the operating temperature of the heat exchanger fluid exiting the evaporator (10) associated with the container (2); a fifth input (I5) associated with a value representative of the operating pressure of the heat exchanger fluid exiting the compressor (11); a sixth input (I6) associated with a value representative of the operating pressure of the heat exchanger fluid exiting the evaporator (10); - a seventh input (I7) associated with a value representative of the degree of opening of the throttling element (12).

12. Machine (1) for preparing liquid or semi-liquid or semi-solid food products according to any one of the preceding claims, characterised in that The first control module (6) is configured to classify the mixture being processed in the container (2) based on the balance of components in the mixture, at least between a balanced category and an unbalanced category, the balanced category corresponding to the processed mixture being balanced with respect to one or more components, and the unbalanced category corresponding to the processed mixture being unbalanced with respect to one or more components; the at least one output (U11) of the first control module (6) represents the classification of the mixture being processed in the container (2).

13. Machine (1) for preparing liquid or semi-liquid or semi-solid food products according to the preceding claim, characterised in that The first control module (6) is configured to perform multi-type classification of the mixture based on the balance of components in the mixture being processed in the container (2), and the first control module (6) includes multiple outputs (U11, U12, U13, U14, U15, U16, U17, U18, U19).

14. Machine (1) for preparing liquid or semi-liquid or semi-solid food according to claim 13, characterized in that The outputs (U11, U12, U13, U14, U15, U16, U17, U18, U19) of the first control module (6) are generally associated with two or more of the following types of processed mixtures: - the product processed in said container (2) is balanced with respect to at least one basic ingredient; - the product processed in said container (2) is unbalanced with respect to the water content; - the product processed in said container (2) is unbalanced with respect to fats; - The product processed in said container (2) is unbalanced with respect to sugars.

15. A method for controlling a machine (1) for preparing a liquid or semi-liquid food product, the method comprising the following steps: - A machine (1) for preparing a liquid or semi-liquid food product is provided, the machine comprising: - a processing vessel (2), which is used to process the base product; - a stirrer (3) suitable for mixing the product in the processing vessel (2) in which the base product is processed; a thermal treatment system (4) configured to exchange heat with a processing vessel (2) in which the base product is processed; - at least one sensor (S) configured to detect an operating parameter of the machine; - providing a first artificial intelligence control module (6) configured to define a classifier (CL) having at least one input (I1, I2, I3, I4, I5, I6, I7) and at least one output (U11), wherein the at least one input (I1) is connected to the at least one sensor (S); - training a first artificial intelligence control module (6) based on the processing of different mixtures inside the container (2) in order to obtain a first trained control module (6); - feeding the mixture into the container (2) and starting the processing cycle; - at least driving the heat treatment system (4) and / or the stirrer (3) to adjust at least one operating parameter related to the heat treatment system (4) and / or the stirrer (3) according to the value of the at least one output (U11) of the first trained control module (6).

Images