Smart freezer

Abstract

A method and a freezer for adjusting an ice cream composition processed by a smart freezer comprising: providing ice cream ingredients to said smart freezer, freezing and processing said ice cream ingredients into an ice cream composition, observing one or more ice cream characteristics of said ice cream composition, automatically analyzing said one or more ice cream characteristics, automatically providing a smart freezing setting comprising a plurality of adjustable smart freezer parameters based on said analyzed observed ice cream characteristics and a desired ice cream quality of said ice cream composition.

Description

SMART FREEZERField of the invention

[0001] The present invention relates to smart freezer and a method for adjusting a smart freezer.Background of the invention

[0002] In the production industry of ice cream products, it is critical to optimize the quality of the ice cream products to be able to deliver ice cream products with the correct amount of coating, the right amount of volume or the right creaminess, etc. It is also critically to optimize the yield of the ice cream products produced in a production line to avoid too much waste of ice cream during the production and prevent breakdowns in the production line. Traditionally a lot of ice cream products are being discarded because of not being shaped correctly at a freezer along the production line. Another problem would be that the ice cream products would also not be suitable for sale due to a low quality, or an incorrect volume compared to the saying on the packaging.

[0003] An example of such a freezer / regulation is disclosed in US 4,316,490 where an amount of air in ice cream products is changed based on weight measurements.

[0004] A challenge of the disclosed freezer is that it is difficult to control and adjust the freezer for obtaining e.g. a desired ice cream quality.Summary of the invention

[0005] The inventors have identified challenges and problems related to ice cream production and subsequently made the following advantageous invention, which in one aspect relates to a method for adjusting an ice cream composition processed by a smart freezer comprising: providing ice cream ingredients to said smart freezer, freezing and processing said ice cream ingredients into an ice cream composition, observing one or more ice cream characteristics of said ice cream composition, automatically analyzing said one or more ice cream characteristics,automatically providing a smart freezing setting comprising a plurality of adjustable smart freezer parameters based on said analysed observed ice cream characteristics and a desired ice cream quality of said ice cream composition.

[0006] Ice cream composition may to be understood as the part where ice cream ingredients have entered a smart freezer. The ice cream composition is further to be understood as a composition (ice cream ingredients including air) in the smart freezer which continuously is being processed and changing its composition through the smart freezer. The ice cream composition is being cooled through the smart freezer and partly frozen while air optionally is being distributed in the ice cream composition. The ice cream composition is being processed through the smart freezer until an ice cream former is reached, where the ice cream composition is being shaped and divided into ice cream items. The ice cream composition may vary according to any of the ice cream characteristics. The ice cream composition may change according to what type of ice cream item type is being produced at the specific production line or at the specific day. The ice cream composition may be a mass of ice cream manufactured prior to the ice cream former. The ice cream item type could be any of the following: ice cream, lemonade ice, water ice, popsicle, ice cream-sandwich, ice cream-cone, a lolly, gelato, frozen dessert, frozen yogurt, granita, sorbet, kulfi, dondurma or any combination thereof. The ice cream item type could be in the size of a popsicle stick, a sandwich, a cone, an ice cream boat or an ice cream cake. The ice cream composition could further comprise other edible parts like caramel, chocolate, fruit juice, edible decoration, jam or any combination thereof. The ice cream composition could further be a vegan produced ice cream composition. The ice cream composition could also be several different flavours of ice cream like chocolate, strawberry, vanilla, raspberry, stracciatella, coffee, tutti frutti, or any other ice cream flavour. The ice cream composition may be any kind of colour depending on what type of ice cream item type is being made.

[0007] Furthermore, the ice cream composition may be used for an ice cream item which could comprise a stick, a container, a cone, or any other relevant part for holding the ice cream item when eaten. The parts related for holding the ice cream item wheneaten could be of an edible material like e.g., a waffle or cookie. The part related for holding the ice cream items could also be a non-edible like a stick or a cup of plastic or wood. The ice cream item type could be an ice cream item type without any holding means which is only wrapped in paper, foil, plastic bag, recycled material, cardboard, carton, bamboo or directly in a box. The ice cream items may also be double wrapped in both a foil, paper or bag and after that being placed in a box, where the box may comprise multiple ice cream items. The ice cream item types without any means for holding the ice cream item could be like a sandwich, a boat, a bar, or bites which would typically be eating by hand. The ice cream item type without any means for holding the ice cream item could also be in the size of a cake which would typically be eaten by flatware.

[0008] The ice cream composition may vary both in a macroscopic and a microscopic size. The macroscopic characteristics of the ice cream composition may be weight, density, amount of air, viscosity, stickiness, rigidness, consistency (not full of larger air bubbles), breaking up, or any other parameter of the ice cream composition which may be measured in a macroscopic size.

[0009] Ice cream characteristics may be understood as the characteristics of the ice cream composition. The ice cream characteristics of the ice cream composition may be how airy, creamy or sticky the ice cream composition is. The ice cream characteristics may also be the dimensions, temperature, amount of air or weight of the ice cream composition. The ice cream characteristics may be observed with different sensors and / or an operator according to what type of ice cream characteristics is to be observed. The ice cream characteristics of the ice cream composition may be observed by a worker or operator who can observe e.g., the mouthfeel of the ice cream composition or the stickiness of the ice cream composition to any parts of the machinery along the production line.

[0010] Microscopic ice cream characteristics may be understood as how the foamstructure of the ice cream composition is made. The crystal-structure may vary in size of each crystal or the distance between each crystal. The characteristics of the ice cream composition in a microscopic structure may also be referred to how many fatparticles is with a certain distance of each ice crystal. Fat globules may be in the size of Ip and the fat globules agglomerate to form a fat structure (l-100p) which is the foundation of the foam structure holding the ice crystals and the air cells in place. The foam structure is embedded in the liquid phase called serum. The different types of cells or particles in the microscopic size of the ice cream composition may vary or comprise other cells which may affect the quality of the ice cream composition e.g., the mouthfeel when eating the ice cream composition. The different structures in a microscopic order of size may affect the different macroscopic ice cream characteristics.

[0011] Smart Freezer is to be understood as a freezer along a production line for producing ice cream products. The smart freezer may typically be placed in the beginning of the production line before hardening tunnel, coating station and an ice cream former. Prior to the smart freezer in the production line a mixer may typically be placed. Further a feeder may also be connected to the smart freezer. The smart freezer is to be understood as the part of the production line where ice cream typically is made. The smart freezer may freeze and cool the different ingredients being feed to the smart freezer and make an ice cream composition.

[0012] The smart freezer comprises a freezing cylinder, one or more scraper blades and a dasher for stirring the ice cream composition while freezing the ice cream composition. The scraper blades of the dasher in the smart freezer scratches the inside of the surface of the freezing cylinder to mix the cooled and crystalized ice cream into the ice cream composition. The process keeps going while the ice cream composition is located within the freezer and the ice cream composition is getting a finer crystal structure. Different parts of the ice cream composition are freezing at the edge of the freezer and a scraper blade keeps scraping the inside of the freezer cylinder wall. The ice cream composition which has been scraped off is being transferred to the centre of the freezer in a longitudinal direction while new parts of the ice cream composition is being cooled down along the edges of the freezer. A compressor may also fill air into the ice cream freezer for generating a higher quality of the ice cream composition when air is mixed with the ingredients of the ice cream composition in the freezer. Theprocess of adding air to the ice cream composition is measured as overrun defined as the amount of air in relation to the amount of ice cream composition. The smart freezer crystallizes the ice cream composition and distribute the air to get the desired ice cream quality.

[0013] Ice cream quality (DICQ) of an ice cream composition may to be understood as a parameter to describe how good or bad the ice cream composition is. The ice cream quality may be depending on both microscopic and / or macroscopic ice cream characteristics. The macroscopic characteristics may be weight, air, volume, amount of air, viscosity, flow, visual-characteristics, process characteristics (like stickiness, consistency, rigidness), or any other relatable parameter to the ice cream composition. Microscopic characteristics is the structure, size and construction of the ice cream crystals in the ice cream compositions e.g., how many fat cells, how many air cells, distance of the crystal-structure and so on. The different ice cream characteristic of the ice cream composition may affect each other. When adjusting one ice cream characteristics of the ice cream composition it may adjust the ice cream quality of the ice cream composition, but typically other ice cream characteristics may also be affected in the process and thereby changing the ice cream quality again.

[0014] The ice cream quality may be changed from production to production depending on what type of ice creams are being made and / or based on a costumer demand. This may be advantageous when making either discount or premium ice cream products that the ice cream quality may be changed.

[0015] Adjustable smart freezer parameter(s) (SFP) may be understood as the different parameters adjustable for changing the process for the smart freezer. The smart freezer parameters are to be set for a predetermined value which may define a smart freezer setting. Changing the different smart freezer parameters is a way to adjust to the ice cream quality of the ice cream composition. Changing one smart freezer parameter may be a temperature of the smart freezer and keeping all other parameters at the same level. Changing the temperature may help freeze the ice cream composition. The smart freezer parameters may typically affect each other when one smart freezer parameter is adjusted it may affect another smart freezer parameter thatneeds to be adjusted in order to obtain the desired ice cream quality or minimize the deviation from the desired ice cream quality.

[0016] The adjustable smart freezer parameters may be any of the following: temperature of freezer, humidity of the freezer, pressure in the freezer, flow of ice cream composition, flow of ice cream ingredients, rotation speed of the dasher, flow of refrigerant, temperature of refrigerant and operation of compressor. Other smart freezer parameters may be applied. It should be noted that the provided smart freezing setting comprising a plurality of adjustable smart freezer parameters based on said analysed observed ice cream characteristics and a desired ice cream quality of said ice cream composition may typically refer to smart freezer parameters SFP which have been set to a value. The setting, these values, may then be set manually by an operator of the smart freezer or the setting may be automatically be applied in a control loop, thereby updating the smart freezer adjustable parameters.

[0017] External “operating” parameters may be any of ambient humidity, ambient temperature and ambient pressure which all are operating conditions for the smart freezer rather than operating parameters. The external “operating” parameters may still affect the operation of the smart freezer. If atmospheric pressure is varying due to weather or heights it will affect the amount of air injected into the ice cream composition.

[0018] Smart freezer setting may be understood as a configuration for a smart freezer, where a plurality of different smart freezer parameters has been set to provide a specific output from the smart freezer. A change in one or more of the adjustable smart freezer parameters may change the complete smart freezer setting. The smart freezer may be suitable for being set in multiple different smart freezer settings. A smart freezer may comprise e.g., 17 different adjustable smart freezer parameters where a smart freezer setting is defined as each of the 17 adjustable smart freezer parameters is set. The smart freezer setting may be changed by changing only one of the 17 adjustable smart freezer parameters from a previous smart freezer setting. The other 16 adjustable smart freezer parameters have to be taken into account for the smart freezer setting even though only one adjustable smart freezer parameters is changed.

[0019] Ice cream ingredients are to be understood as different types of mix of ice cream types. The ice cream composition may comprise one or more ice cream ingredients with types of the same ice cream and / or different types of ice cream types. Ice cream ingredients may also be air which is part of the ice cream composition and added as an ice cream ingredient. Ice cream ingredients may also be referred to as different types of solid pieces e.g., chocolate or berries being mixed into the ice cream composition to make a specific ice cream product. The ice cream ingredients may be added by the mixer or may be added after the smart freezer by an ingredients feeder.

[0020] It may be advantageous to have a smart freezer where a certain and predefined quality of ice cream is provided. The ice cream quality may be set up by a person not skilled in the art of making ice cream but can be guided by the smart freezer to set the adjustable smart freezer parameters. It is further advantageous to have a smart freezer for making a more uniformly ice cream from production to production and day to day. The smart freezer may adapt to changes in the weather regarding e.g., humidity, temperature and / or pressure and still make a uniformly ice cream quality. The smart freezer may also maintain a uniformly ice cream quality whether the ice cream is produced in e.g., Denmark, India or USA.

[0021] It may be advantageous to adjust parameters without a skilled person when the smart freezer guides the non-skilled person. The smart freezer may help guiding with the correct adjustable parameters to have the correct setting for a specific ice cream production by displaying a setting comprising a combination of smart freezer parameters an operator may use for a manual setting of the freezer and optionally also other settings related to equipment of the production line upstream or downstream the smart freezer.

[0022] It may be advantageous to have a smart freezer to produce a higher yield of ice cream items in the production. The ice cream composition leaving the freezer to the ice cream former may leave the freezer at a more optimal quality. The quality of the ice cream composition may be the best ice cream composition for the rest of the production line to generate the highest yield of ice cream products. The smart freezermay also deliver a greater quality of ice cream composition related to the mouthfeel when eating the ice cream product.

[0023] It may be advantageous to have a smart freezer which may adapt to local geographical conditions where the temperature or humidity may affect the process of producing ice cream products. The conditions for a smart freezer may be different for a smart freezer in a tropical country compared to a tempered country. The freezer may further adapt to weather or indoor climate changes. The smart freezer may adapt to different air-types inside a production hall where the temperature, humidity and / or pressure may change from day to day.

[0024] It may be advantageous to adjust ice cream composition before problems is visible to the human eye. The drifting of an ice cream production may be adjusted before being visible to the operator at the freezer. The smart freezer may provide one or more adjustable parameters to set the right setting for the smart freezer to avoid the drifting of the quality of ice cream composition. Quality can e.g. drift if there is variation in the ingredient properties. For instance, vegan ingredients can have changing properties, depending on where they are grown and who deliver them.

[0025] The maturing of the mix will can also causing different properties of the ingredients, for instance if the maturing time or the process temperatures changes

[0026] It may be advantageous to have a more optimal number of turns for the dasher in the smart freezer. The number of turns may be limited to save energy without turning to many times and still keep the desired ice cream quality. The optimal number of turns to save energy and still achieve the desired ice cream quality may be easier to adapt to different types of ice cream compositions.

[0027] It may be advantageous to have a smart freezer where the smart freezer setting is analyzed based on the type of ice cream composition. The different smart freezer parameters according to if the ice cream composition is a creamy ice cream or maybe a vegan ice cream.

[0028] According to an embodiment of the invention, adjusting of said ice cream composition is performed by using the provided smart freezer setting as a basis for adjusting the smart freezer parameters defined by said provided smart freezer setting.

[0029] It may be advantageous to adjust the ice cream composition to achieve the desired ice cream quality by providing said smart freezer setting. The smart freezer setting may be provided either as a guide for an operator or may be set automatically in the smart freezer. The manually guide is advantageous when a non-experienced operator is working e.g., during holidays or weekends where the skilled operator is having a day off.

[0030] According to an embodiment of the invention, said desired ice cream quality of said ice cream composition is defined prior to said step of providing said ice cream ingredients to said smart freezer.

[0031] According to an embodiment of the invention, said desired ice cream quality is based on an ice cream composition type.

[0032] According to an embodiment of the invention, said observing is a measuring performed by one or more sensors.

[0033] It may be advantageous to measure different ice cream characteristics to adjust the smart freezer to optimize the desired quality of the ice cream products for sale or further processing.

[0034] According to an embodiment of the invention, said method comprises an additional step of measuring the smart freezer parameters as they are set prior to automatically analysing said one or more ice cream characteristics.

[0035] It may be advantageous to measures / monitor the smart freezer parameters and use them as input when automatically analysing the ice cream characteristics to adjust the smart freezer with new smart freezer parameters.

[0036] According to an embodiment of the invention, said observing is done by one or more sensor(s).

[0037] According to an embodiment of the invention, said ice cream characteristics are macroscopic characteristics.

[0038] According to an embodiment of the invention, said ice cream characteristics are microscopic characteristics.

[0039] It may be advantageous to observe different types of ice cream characteristics when the desired ice cream quality is to be achieved.

[0040] According to an embodiment of the invention, said automatically analysing said one or more ice cream characteristics is based on said ice cream type being made.

[0041] According to an embodiment of the invention, said smart freezer setting relates to a desired ice cream composition type.

[0042] It may be advantageous to use the different type of ice cream as a parameter for the smart freezer setting to optimize the quality of the specific type of ice cream product. It may also be advantageous to change and adjust the smart freezer according to what type of ice cream product being made which may be affected differently along the production line due to different compositions of ice cream.

[0043] According to an embodiment of the invention, said ice cream ingredients are continuously added to the smart freezer.

[0044] According to an embodiment of the invention, said ice cream composition is based on said ice cream ingredients being added to said smart freezer.

[0045] According to an embodiment of the invention, multiple types of ice cream ingredients are added to said smart freezer.

[0046] According to an embodiment of the invention, said ice cream composition is adapted to a ice cream type.

[0047] According to an embodiment of the invention, said ice cream composition is made from one or more types of ice cream types with different flavours.

[0048] According to an embodiment of the invention, said provided smart freezing setting comprising a plurality of set adjustable smart freezer parameters based on said analysed observed ice cream characteristics and a desired ice cream quality of said ice cream composition is correlated to power consumption of the smart freezer.

[0049] According to an embodiment of the invention, said provided smart freezing setting comprising a plurality of set adjustable smart freezer parameters based on said analysed observed ice cream characteristics and a desired ice cream quality of said ice cream composition is correlated to power consumption of a dasher of the smart freezer.

[0050] When correlating the power consumption of the smart freezer as total power consumption or e.g. just as the power consumption of the dasher of the smart freezer, it will be possible to minimize energy consumption e.g. by rating energy consumption above a desired ice cream quality. In other words, the operator may define a desired ice cream quality as best quality irrespective of power consumption. It may also be possible to us a setting defining a lower quality of the ice cream quality but now with a minimized power consumption.

[0051] According to an embodiment of the invention, said automatically providing a smart freezing setting is performed by a controller configured to implement a closed- loop feedback control comprising the steps of: establishing an error as the difference between said desired ice cream quality and said observed ice cream characteristics, processing said error to obtain said smart freezing setting, communicating said smart freezing setting to a user interface.

[0052] By communicating said smart freezing setting to a user interface it may thereafter be possible for an operator to adjust the smart freezer without having complex knowledge about how the complex interactions are in the smart freezer, when the smart freezer is e.g. feeding an ice cream former of an ice cream production line.

[0053] According to an embodiment of the invention, said automatically providing a smart freezing setting is performed by a controller configured to implement a closed- loop feedback control comprising the steps of: establishing an error as the difference between said desired ice cream quality and said observed ice cream characteristics,processing said error to obtain said smart freezing setting, providing said smart freezing setting as an input to said smart freezer.

[0054] According to an embodiment of the invention, said automatically providing a smart freezing setting is performed by a controller configured to implement a closed- loop feedback control comprising the steps of: establishing an error as the difference between said desired ice cream quality and said observed ice cream characteristics, processing said error to obtain said smart freezing setting, providing said smart freezing setting as an input to said smart freezer for automatic adjustment of a present setting of the smart freezer.

[0055] It may be advantageous to have a controller implementing a closed-loop negative feedback control system because the dynamics of the smart freezer can be adapted to obtain the desired ice cream quality by measuring one or more ice cream characteristics and using as an input to the controller the difference between the desired ice cream quality and the observed ice cream characteristics.

[0056] It may be advantageous to have a controller implementing a closed-loop negative feedback control system because it can dynamically respond to observed ice cream characteristics and respond appropriately by adapting the smart freezer parameters such that the desired ice cream quality is obtained based on difference between the desired ice cream quality and the observed ice cream characteristics.

[0057] It may be advantageous to have a controller implementing a closed-loop negative feedback control system because it can be adapted according to a plurality of parameterizations of desired ice cream quality, a plurality of smart freezer settings, and a plurality of observed ice cream characteristics.

[0058] According to an embodiment of the invention, said error further comprises a difference between said desired ice cream quality and measured smart freezer parameters.

[0059] It may be advantageous to include measured smart freezer parameters in the error signal that is provided to the controller because additional measurements canprovide for improved performance of the closed-loop negative feedback control system. The measured smart freezer parameters can improve the observability of the internal states of the smart freezer and thereby provide for design of a controller that more effectively controls the process according to the goal of obtaining a desired ice cream quality.

[0060] According to an embodiment of the invention, said controller is a proportional-integral-derivative PID controller.

[0061] It may be advantageous to have a controller implementing a PID controller because the PID controller can be implemented digitally on a data processor. The PID controller is therefore able to provide desired ice cream characteristics based on observed ice cream characteristics without introducing undue cost other than the processing means by which the PID controller is implemented.

[0062] It may be advantageous to have a controller implementing a PID controller because the PID controller can be designed by experimental means i.e. without deriving an analytical model of the smart freezer. The PID controller is therefore simple and fast to implement when data comprising at least observed ice cream characteristics and adjustable smart freezer parameters are available.

[0063] According to an embodiment of the invention, said controller is a lead-lag compensator.

[0064] It may be advantageous to have a controller implementing a lead-lag compensator because the lead-lag compensator is designed such that the smart freezer obtains dynamic and static parameters that are required to obtain a desired ice cream quality based on observed ice cream characteristics. The lead-lag compensator is typically designed based on an analytical or derived model of the smart freezer, and therefore gives assurance as to the stability and performance of the closed-loop feedback system such that the smart freezer provides a consistent desired ice cream quality.

[0065] It may be advantageous to have a controller implementing a lead-lag compensator because the lead-lag compensator can be implemented in continuous time and thus ensure a minimum of time delay, high accuracy, and precision. The lead-lag compensator can easily be adapted to account for e.g. disturbances or sensor output models.

[0066] According to an embodiment of the invention, wherein said controller is configured for training based on said observed ice cream characteristics.

[0067] According to an embodiment of the invention, wherein said controller comprises adaptive controller parameters.

[0068] Adaptive controller parameters provide the advantage that the controller can be adapted to account for time-varying phenomena such as ambient conditions, changes in the dynamics of the smart freezer, or changes in the dynamics of the sensors providing the observed ice cream characteristics. Adaptive controller parameters are especially advantageous if the smart freezer cannot be modelled as a time-invariant system.

[0069] According to an embodiment of the invention, said adaptive controller parameters are configured for manual adjustments.

[0070] It may be advantageous that the adaptive controller parameters are configured for manual control because it allows the controller to be tuned by a worker or operator during e.g. a running-in period such as in a period of time following installation of the smart freezer.

[0071] It may be advantageous that the adaptive controller parameters are configured for manual control because it allows the controller to be tuned by a worker or operator in response to observable phenomena to which the controller itself may not be able to respond. This situation may arise if the dynamics of the system change in unexpected ways such that the controller can no longer maintain the desired ice cream quality based on observed ice cream characteristics.

[0072] According to an embodiment of the invention, said automatically analysing is analysed based on machine learning.

[0073] According to an embodiment of the invention, said automatically analysing is analysed based on a machine learning control model.

[0074] According to an embodiment of the invention, said machine learning control model is a reinforcement learning model.

[0075] Sometimes reinforcement learning model is referred to as a reinforcement learning control model.

[0076] According to an embodiment of the invention, said machine learning control model is trained at least based on ice cream characteristics.

[0077] According to an embodiment of the invention, said machine learning control model is trained and wherein said training includes reward shaping.

[0078] Advantageously, this may have the effect of minimizing the time and / or iterations required to train the machine learning control model to perform well enough to provide automated control of the smart freezer.

[0079] According to an embodiment of the invention, said machine learning control model is trained and wherein said training of said machine learning control model includes imitation learning.

[0080] Advantageously, this may have the effect of minimizing the time and / or iterations required to train the machine learning control model to perform well enough to provide automated control of the smart freezer. Imitation learning includes learning from feedback from one or more human expert in determining smart freezing settings to achieve a desired ice cream quality.

[0081] According to an embodiment of the invention, said smart freezer is controlled by a controller comprising a machine learning control model.

[0082] According to an embodiment of the invention, said an additional step of measuring is done prior to defining said desired ice cream quality.

[0083] According to an embodiment of the invention, the controller is configured for running the smart freezer in an assistance mode via the user interface, using input from the user interface in the form of observed ice cream characteristics and / or a desired ice cream quality and outputting at least one smart freezer setting which the user may set by manually setting the adjustable smart freezer parameters on the smart freezer.

[0084] According to an embodiment of the invention, the controller is configured for running the smart freezer in an automatic mode via the user interface, using input from the user interface in the form observed ice cream characteristics and / or a desired ice cream quality and automatically applying a smart freezer setting by automatically setting the adjustable smart freezer parameters on the smart freezer.

[0085] According to an embodiment of the invention, said smart freezer setting is also depending on the power consumption of the smart freezer.

[0086] The invention further relates to a smart freezer comprising: one or more sensor(s) for observing ice cream characteristics, a controller configured for receiving ice cream characteristics, a freezer and a mixer for freezing and processing ice cream ingredients, a user interface for setting a plurality of adjustable smart freezer parameters, wherein said plurality of adjustable smart freezer parameters defines a smart freezer setting for said smart freezer.

[0087] It is to be understood that the smart freezer may comprise a controller as a part of the smart freezer or may be connected to a smart freezer.

[0088] It may be advantageous to have a smart freezer for making an ice cream composition with a desired ice cream quality and minimize the deviation from the desired ice cream quality. It is further advantageous to be able to reach the desired ice cream quality by guiding a non-skilled person.

[0089] According to an embodiment of the invention, said smart freezer is upstream mechanically connected to a mixer.

[0090] According to an embodiment of the invention, said smart freezer is downstream mechanically connected to one or more an ice cream formers.

[0091] According to an embodiment of the invention, said smart freezer comprises a communication module.

[0092] According to an embodiment of the invention, the smart freezer comprises a user interface configured for visualizing e.g. by means of a display, the current smart freezer setting and / or an updated smart freezer setting in response to a received observed ice cream characteristics and / or a desired ice cream quality .

[0093] According to an embodiment of the invention, said smart freezer is connected to a server / cloud.

[0094] According to an embodiment of the invention, the controller is configured for running the smart freezer in an assistance mode via the user interface, using input from the user interface in the form of observed ice cream characteristics and / or a desired ice cream quality and outputting at least one smart freezer setting which the user may set by manually setting the adjustable smart freezer parameters on the smart freezer.

[0095] According to an embodiment of the invention, the controller is configured for running the smart freezer in an automatic mode via the user interface, using input from the user interface in the form observed ice cream characteristics and / or a desired ice cream quality and automatically applying a smart freezer setting by automatically setting the adjustable smart freezer parameters on the smart freezer.

[0096] According to an embodiment of the invention, said smart freezer is operated according to the method for adjusting an ice cream composition according to embodiments of the invention.The drawings

[0097] Various embodiments of the invention will in the following be described with reference to the drawings where: figs, la-c illustrate a smart freezer and ice cream former, figs. 2a and 2b illustrate a smart freezer principle, fig. 3 illustrates a production line for ice cream products, figs. 4a-b illustrate a block diagram of the process of making ice cream products, figs. 5a-c illustrate closed-loop feedback control of a smart freezer, fig. 6 illustrates a training of a machine learning model, fig. 7 illustrates the principles of smart freezer settings, and figs. 8a-c illustrate closed-loop feedback control of a smart freezer.Detailed description

[0098] The following description comprises nonlimiting examples of embodiments of the invention. Details such as specific structures, arrangements and methods are provided to give an understanding of embodiments of the invention. Note that detailed descriptions of well-known methods, systems, apparatuses, circuits, parameters, known sensors, actuators, signal paths, components, control leads, ingredients, algorithm training methods and architectures, etc. have been omitted to not obscure the description of the invention with unnecessary details. Non-limiting examples of such components that will not be described in detail, but which is typically included in ice cream production systems and / or freezing system designs include, e.g., means for freezing, including, cooling systems, pumps, and means for driving a conveyer belt etc. Further notice that the invention is not limited to the specific examples described below, and a person skilled in the art may choose to implement the invention in other embodiments without these specific details. E.g., the invention may be utilized in various types of ice cream productions lines which are not necessarily detailed in this disclosure. Furthermore, a skilled person in the field of the invention may choose to combine features of the described embodiments and of the illustrated embodiments of the invention. As such, the invention may be designed and altered into a multitude of varieties within the scope of the invention, as specified in the claims.

[0099] The following section comprises a description of various embodiments of the invention with references to the figures.

[0100] Fig. la illustrates a smart freezer SF which is connected to a mixer MIX and an ice cream former ICF. Between the smart freezer SF and the ice cream former ICF there may optionally be an ingredients feeder INF where caramel, chocolate or minor pieces of berries or other edible pieces may be added to the ice cream composition. The smart freezer SF is also connected to a compressor COP, supplying air into an ice cream composition ICO inside the smart freezer SF. The amount of air input to be a part of the ice cream composition may be described as an amount of overrun (amount of air compared to amount of ice cream composition). The smart freezer SF is connected to one mixer MIX but may in other variations of the invention be connectedto multiple mixers. The multiple mixers may be used with a valve in order to level out two similar types of ice cream mix in a smart freezer SF, but it is not preferable. The smart freezer SF further comprises multiple sensor SENS where the one sensor SENS is located within the smart freezer SF. The sensor located inside may observe temperature of ice cream composition ICO, mix flow, air flow, power of dasher or any other related characteristics or parameters for observing the smart freezer SF and the ice cream composition ICO. Another sensor SENS between the smart freezer SF and the ice cream former ICF to measure the flow of the ice cream composition ICO. The smart freezer SF and the ambient surroundings may comprise more sensor in order to measure the desired ice cream characteristics, external operating parameters and / or smart freezer parameters.

[0101] The smart freezer SF illustrated in fig. la is connected to a mixer MIX and a compressor COM. The mixer MIX comprises a mix pump (not shown) for pushing the mix and / or ice cream ingredients into the smart freezer SF. In another variant of the embodiment, the mix pump may also be implemented as part of the freezer. The same applied for the product pump, pumping ice cream out of the freezer. This may also be included as a part of the freezer

[0102] Figure la further illustrates the ice cream composition ICO being lead from the smart freezer SF to the ice cream former ICF. An additional pipe is leading towards the ice cream former ICF where additional ingredients are being added from the ingredients feeder INF to the ice cream composition ICO. The ingredient feeder INF may add chunks like berry, cookie, chocolate to the flow of ice cream before the ice cream former ICF. The ingredient feeder INF may be regarded as a machine e.g. having a size of approximately 1x0.6x1.5 m. There will typical be some meters of piping between the smart freezer SF and the ingredient feeder INF. The same between ingredient feeder INF and the ice cream former ICF. Sauces like caramel, chocolate may be added the ingredient feeder INF directly, via piping from a tank or other types of process equipment. In the illustrated embodiment caramel is being swirled into the ice cream composition ICO before the ice cream composition with the caramel is finally formed at the ice cream former outlet ICFO. The ice cream composition ICOis divided into ice cream items by a cutter CUT placed just below the ice cream former outlet ICFO. The ice cream item ICI is formed by shaping and dividing the ice cream composition in the ice cream former ICF. One or more of the ice cream item properties of the ice cream item ICI is being measured by a sensor SENS when the ice cream item ICI gets in contact with the transportation surface TSU. The sensor SENS is communicatively coupled with a controller (not shown) wherein the controller may indicate if an adjustment of one or more ice cream item properties of the ice cream is required. The adjustment may be done automatically or by an operator.

[0103] Fig. lb illustrates three smart freezers SF connected to an ice cream former ICF. The three smart freezers SF are connected to the same ice cream former ICF. The three smart freezers SF may typically be used when an ice cream product comprising more than one ice cream type is made. Each smart freezer SF is connected to the ice cream former ICF where the ice cream composition ICO from each smart freezer first gets into contact with the other ice cream compositions shortly before the ice cream former outlet ICFO. The ice cream former ICO may in another embodiment of the invention be facilitated with the same type of ice cream composition from the different smart freezer SF. One smart freezer may feed multiple ice cream formers ICF (e.g. in relation to fillers) but for typically other types of ice cream formers ICF than fillers, each ice cream former will be fed by its own separate smart freezer SF.

[0104] Three types of ice cream compositions may be fed from the smart freezers SF in fig. lb and the ice cream compositions being connected shortly before the ice cream former outlet. By connecting the difference types of ice cream composition, it may facilitate to make an ice cream item with more than one type of ice cream composition. The ice cream former ICF may be connected to any number of smart freezer SF and the ice cream items ICI being made may be made of any kind of number of different ice cream composition types.

[0105] Fig. lb illustrates an embodiment of the invention where three smart freezers SF are cooling three different types of ice cream compositions. The three different types of ice cream compositions could be according to color, flavors, density, or any other relevant ice cream type. The three different ice cream types are guided to the icecream former ICF where the ice creams compositions are being shaped into the respective form for an ice cream item being processed. The three types of ice cream may be formed like e.g., a traffic cone with different types of ice cream for the button, middle and top. The ice cream may also be formed as a happy face with different types of ice cream for face, mouth and eyes. When the ice cream has been formed the mass of ice cream is guided towards the ice cream former outlet ICFO where a cutter CUT is placed just outside of the ice cream former outlet ICFO. The cutter CUT is used to divide the mass of ice cream into ice cream items ICI. The ice cream items ICI may fall to a transportation surface TSU where a sensor SENS is placed to measure one or more ice cream item properties ICIP. The sensor SENS is communicatively coupled with a controller (not shown) wherein the controller is configured for notifying an operator if any adjustments is to be made to adjust the one or more ice cream item properties. The controller may also be configured for automatically adjusting the one or more ice cream item properties by adjusting any kind of processing units along the production line e.g., freezer, ice cream former, hardening tunnel, mixer or any other relevant processing unit along the production line.

[0106] Fig. 1c illustrates a part of the production line PL with an ice cream former ICF, where the ice cream former is an ice cream filler. Fig. 1c further illustrates a transportation surface TSU configured for holding ice cream items ICI, e.g., a cone ICI. The ice cream former ICF is illustrated with 8 tubes from which an ice cream composition (not shown) is being conveyed through from a smart freezer (not shown). The ice cream composition is conveyed through the ice cream former outlet ICFO to the transportation surface TSU, where the transportation surface is illustrated with 8 pockets holes for ice cream items ICI. The ice cream former ICF and the transportation surface TSU are not limited to 8 tubes / ice cream former outlets ICFO and pockets, but may have any number of tubes / ice cream former outlets ICFO and pockets according to production line PL. The number of tubes / ice cream former outlets ICFO and pockets will typically correspond to the number of manufacturing lanes MAL in the production line PL. One of the tubes of the ice cream filler FILL is illustrated with an ice cream former outlet ICFO which is placed inside an ice cream item ICI, here illustrated as an ice cream cone ICI. The ice cream former outlet ICFO may also beused for e.g., biscuits, boats, or any other ice cream item ICI, where a mass of ice cream is to be added. The positioning of the ice cream items ICI may be measured at a measuring location LOC, where the measuring location LOC is located at the same place as the ice cream former ICF. The position of the ice cream items ICI may also be measured both upstream and downstream at measuring location LOC placed on either side of the ice cream former ICF in a conveying direction COND. The conveying direction COND is indicated by an arrow in the figure.

[0107] Fig. 2a illustrates a smart freezer SF according to an embodiment of the invention. The smart freezer SF may be combined with any of the elements of the embodiments of the invention illustrated on figs, la-lc, 3, 4a-4b, and 5a-5c.

[0108] The smart freezer SF comprises a mix inlet 1 connected to a mixing pump 2 and further to mix piping 3. A flow meter 4 is connected to the mix piping 3. A dasher motor 5 is configured to rotate a dasher DAS (not visible) via belts. A cylinder pressure sensor 6 measures the pressure at the inlet of the dasher DAS. An air inlet 7 provides for an inflow of air into the dasher DAS, said air flow being controlled by an air controller 8. A temperature sensor 9 is configured to measure the temperature of the ice cream composition ICO in a freezing cylinder 10. The dasher DAS rotates inside the freezing cylinder 10.

[0109] Ice cream ingredients ICN are provided as an input at the mix inlet 1. The ice cream ingredients ICN may comprise a mixture of water and other ingredients such as sugar, cream and flavoring. As the ice cream ingredients ICN enter the freezing cylinder 10, they constitute an ice cream composition ICO. The flow meter 4 measures the flow of the ice cream ingredients ICN and regulates the mix pump 2 according to e.g. desired volume of ice cream composition ICO per unit time. The dasher motor 5 is configured to rotate the dasher according to a specific number of rotations per minute, ensuring air is well distributed in the freezing cylinder 10 and that the desired ice cream quality FICQ is achieved. The inflow of air at air inlet 7 is typically quantified as an overrun parameter which parameterizes the amount of air in the ice cream composition ICO as a volume percentage relative to the amount of fluid. The air in the ice cream composition ICO is also considered as an ice cream ingredientICN, the inflow of which is indicated on the right side of fig. 2a. The air controller 8 controls the inflow of air according to a desired overrun.

[0110] The desired viscosity of the ice cream composition ICO is controlled by adjusting a viscosity setting. This is not the real fluid viscosity of the ice cream composition ICO, pumped out of the smart freezer SF, but rather a representation. The viscosity setting is typically calculated as the actual power of the dasher motor 5, relative to its max power. The actual power is mainly affected by the resistance of the ice cream composition ICO, when turning the dasher DAS, due to the varying fluid viscosity of ice cream composition ICO inside the freezing cylinder 10.

[0111] The refrigeration system RSYS is typically adjusted, in order to achieve the desired viscosity setting. By applying more or less cooling power to the freezing cylinder 10, the real fluid viscosity of ice cream composition ICO inside the freezing cylinder 10 is changed until the calculated viscosity representation is equal or close to the viscosity setting.

[0112] The ice cream composition ICO is pumped out of the freezing cylinder 10 using a product pump 11. The temperature sensor 9 measures the temperature of the ice cream composition ICO, pumped out. This is typically used as a monitoring measurement, but it can also be used as an input for the regulation of the refrigeration system RSYS, if the smart freezer SF is temperature regulated.

[0113] A user interface UI is illustrated, through which a desired ice cream quality FICQ (not shown) may be set. The smart freezer SF in turn provides a smart freezer setting SFS which comprises smart freezer parameters such as dasher rotations per minute, overrun, viscosity, and cylinder pressure. The smart freezer setting also depends on product flow e.g. the volume of ice cream composition per unit time.

[0114] Fig. 2b illustrates inside of a smart freezer and the principles of making and crystallizing an ice cream composition ICO. The smart freezer SF comprises an open dasher DAS with scraper blades SCB located on the outside of the dasher DAS. The dasher DAS is located within a freezing cylinder comprising a freezing cylinder wall 10, where the freezing cylinder 10 is located in the smart freezer SF. An open dasherDAS is the most common type used, but a closed dasher may also be used. Ice cream ingredients ICN (not shown) are being fed to the smart freezer SF at one end where the ice cream ingredients ICN typically is a few degrees Celsius. At the moment just after the ice cream ingredients ICN enters the smart freezer SF it may be understood as an ice cream composition ICO. Air is added to the smart freezer SF by a compressor (not shown) at the same end as the other ice cream ingredients ICN. The air is typically warmer than the ice cream ingredients. The ice cream composition ICO - which is now understood as ice cream ingredients including air mixed into it while the temperature is lowered - ends up at the circumference of the freezing cylinder as a mixed slushy substance composition where the cooling of the smart freezer begins. As soon as the ice cream ingredients ICN and injected air from air compressor (not shown) has been fed to the smart freezer SF, it should be understood as an ice cream composition ICO. The smart freezer SF continuously cools the ice cream composition ICO and mixes the ice cream composition ICO, so air is distributed uniformly. The scraper blades SCB of the dasher DAS scrapes the ice cream composition ICO from the smart freezer wall to the centre of the dasher DAS illustrated in the fig 2b. This process crystallizes the ice cream composition ICO. The air component constitute what is known in the industry as overrun. For some types of Ice cream, Overrun is small or even zero..

[0115] Fig. 3 system illustrates the principles of an optional layout of a production line for ice cream products having an ice cream hardening tunnel HT applied according to an embodiment within the scope of the invention. An illustrated production line PL comprising a transportation surface TSU extending from four smart freezers SF and four ice cream formers ICF through the hardening tunnel HT to a coating station COA where the ice cream items can be coated and further to a packing station (wrapping foil station - not shown). Each ice cream former ICF makes a manufacturing lane (not shown) of ice cream items and therefore production line PL is illustrated in the figure comprises four manufacturing lanes of ice cream items (not shown) but may not by limited to four manufacturing lanes. The transportation surface being movable in a direction indicated by associated arrows by an automatic adjustable drive system (not shown) under the control of a cooling control system CON. Along the transportation surface a plurality of measuring locations LOC is provided for measuring and / orobserving the one or more ice cream item properties of ice cream items and communicatively send the measured one or more ice cream item properties to the cooling control system CCS, where the cooling control system CCS also is to be understood as a controller for the production line PL. The measuring locations LOC along the production line PL are not limited to these locations but may be placed at any location along the production line PL. The measuring location LOC may comprise one or more sensors or an operator.

[0116] One or more of the individual measuring location LOC may establish the relevant measurements and then the measured data may be applied as a basis for an upstream adjustment. In other words, a controller CCS of the production line controlling the adjustment of ice cream item properties may be fed with measurement data from one or more measuring locations and thereby be configured for the adjustment of ice cream item properties based on data from one or more measuring locations.

[0117] Moreover, data from one measuring location may be fed to not only one location, e.g. a smart freezer, but also to controllers relevant for other adjustment locations (i.e. devices to be controlled) of the process. The cooling control system CCS may also be referred to as a controller or control system.

[0118] Moreover, the ice cream hardening tunnel HT includes an adjustable cooling arrangement (not shown) also controlled by the cooling control system CON controlling cooling temperature and optionally also adjustably controlling air flow within the hardening tunnel HT.

[0119] It should be noted that the cooling control system CCS may be a singular arrangement or a number of co-functioning controllers. The illustrated cooling control system CCS is communicatively coupled with a user interface UI by means of which an operator has access to modify the quality of ice cream composition ICC in the smart freezer or modify properties of ice cream items along the production line PL. The operator may also modify ice cream composition / items any devices related to ice cream item properties of ice cream items at the production line PL for ice creamproducts on the basis of the properties of ice cream items. It is thus noted that many state of the art production line for ice cream products may be controlled according to the invention only with an addon measuring properties of the ice cream item along the production line, upstream, thereby making it possible for an operator, or the control system, making timely adjustment of the ice cream item properties by modifying the production line parameters.

[0120] Upstream the hardening tunnel HT ice cream items may be positioned on the transportation surface TSU by an ice cream item former ICF, here in the form of four individual stations connected to a mixer MIX, a smart freezer SF, an ingredients feeder typically between a smart freezer and ice cream former, ice cream formers, stick inserters and / or a cutter / valve, thereby facilitating a continuous and automatic placement of ice cream items (not shown) on the transportation surface TSU prior to being transported along the production line. One smart freezer may be connected to four ice cream formers (not shown) when producing an ice cream product with one type of ice cream. Two or more smart freezers may also be connected to one ice cream former (not shown) for making an ice cream product comprising two or more different types of ice cream, e.g., when making an ice cream product with vanilla and strawberry ice cream.

[0121] The mixer MIX mixes ingredients relevant for the recipe of the ice cream composition to be produced and the smart freezers SF provides the quality for the applied ice cream formers ICF of the ice cream item ICI. The desired ice cream quality for the ice cream composition prior to the ice cream former ICF from the smart freezer is applied according to e.g., keeping the desired flow of ice cream to the ice cream former in order to keep the correct volume or weight for the ice cream items leaving the ice cream former ICF.

[0122] Inside the ice cream hardening tunnel HT the transportation surface TSU extends through the hardening tunnel HT so as to facilitate a cooling of the ice cream items from a temperature the ice creams items may have upstream the tunnel, to a temperature of the ice cream items which is lower when the ice cream items leaves the ice cream hardening tunnel HT downstream the hardening tunnel HT.

[0123] The length of the transportation surface TSU, the cooling applied by the cooling system (not shown) including optional internal ventilation, movement of cool air within the hardening tunnel HT, the speed of the transportation surface TSU, etc will determine the resulting cooling from one temperature, e.g. minus 5 degrees Celsius to e.g. minus 18 degrees Celsius, measured as core temperature.

[0124] Some of these parameters are referred to as adjustable tunnel parameter, and these adjustable tunnel parameters may be adjusted manually and / or automatically.

[0125] Fig. 3 illustrates an output of a hardening tunnel HT controlled within the scope of the invention. An ice cream hardening tunnel HT has an exit of the hardening tunnel HT through which a transportation surface TSU extends towards an ice cream item transferring system via an optional ice cream loosener LOS. In the present embodiments, the transportation surface TSU is implemented to transport ice cream items ICI on conveyor plates or trays. The transportations surface is moving in the direction of the arrows COND during operation. If a reference to a transportation surface TSU is made, the reference will be made with respect to a / the surface of the conveyor plates or trays if such plates are applied. If, the conveyor transports the ice cream items directly on the conveyor elements, a transportation surface TSU is to be understood as the surface upon which the ice cream items are conveyed. Other implementations of the conveyor may thus of course be applicable within the scope of the invention, with or without “loose” plates or trays positioned on the top of the underlying conveyor, although easy removal plates / trays / etc are advantageous as these may easily be positioned and removed on the conveyor and easy to clean in a run-time environment. Furthermore, it will be easier to make format changes, if for instance the removal plates / trays / are specifically designed / formed to carry or keep specific ice cream item types (e.g. if ice cream items are carried in “pockets”). The illustrated embodiment includes a sensor for measuring ice cream item properties, where the sensor is a core temperature measuring system CMS, here placed just outside the hardening tunnel HT.

[0126] Fig. 4a illustrates a block diagram of the process along the production line of producing ice cream products. The production line PL comprises a plurality of workingstations according to the location along the production line PL. At first, different ice cream ingredients are fed o a mixer MIX (technically optional) where ingredients (a mixture of ice cream compounds) are being mixed. The substance from the mixer MIX is then transferred to a smart freezer SF where the substance is being processed into an ice cream composition. The mix is being added to the smart freezer SF and at the same end of the freezer air is added. Air may also be understood as an ice cream ingredient typically fed into the smart freezer through a separate dedicated inlet. In the step in the smart freezer SF the input ice cream ingredients (mixture and air) is processed while being cooled and partly frozen, and thereby crystallizing parts of the ice cream composition and / or changing the ice cream composition. The ice cream composition is cooled to a temperature below zero degrees Celsius through the smart freezer SF to get the desired ice cream quality with the right size and shape of the ice cream foam-structure. At the input end of the smart freezer SF air is continuously fed into the smart freezer SF. Within the smart freezer, the ice cream composition is subject to shear e.g. by a dasher in a freezing cylinder (e.g. the freezing cylinder 10 of fig. 2a) to make the ice cream composition more soft and less cold to eat.

[0127] In the next step of the process the ice cream composition is guided to an ice cream former ICF where the ice cream composition is being shaped and divided into ice cream items. The ice cream former ICF may be an ice cream cutter, where a flow of the ice cream composition is being guided to an outlet shaped as the desired ice cream product. A stick inserter may typically be placed at the end of the outlet of the ice cream former ICF when the ice cream item type is an ice cream on a stick. At the end of the ice cream cutter a metal wire is placed to divide the ice cream composition into ice cream items and let the ice cream items drop to a conveyor. The ice cream former ICF may also be an ice cream filler, where a valve divides the stream of the ice cream composition. After the ice cream composition has been divided by the valve the ice cream may be push by a piston into an ice cream container like e.g., a waffle, biscuit or non-edible container.

[0128] The ice cream items are being conveyed from the ice cream former ICF to a hardening tunnel HT where the ice cream items are cooled down through the hardeningtunnel HT. The hardening tunnel may vary in type from each production line according to what type of ice cream item types are being made. A exemplary hardening tunnel HT and some of its upstream and downstream production line equipment is illustrated in fig. 3. These components may e.g. be included in the present fig. 4a and fig. 4b systems. Ice cream items are being conveyed on a transportation surface TSU. That type of hardening tunnel HT typically comprises air ventilation, vaporisers, heat exchanger and other things (not shown) related to controlling the temperature in a uniform way within the hardening tunnel HT. Other types of hardening tunnels may be hardening tunnels where the conveyor comprises a casting form. The hardening tunnels with conveyors with casting forms may be circular and rotating in an approximately horizontal plan around a centre. Ice cream items are shaped and divided into the casting forms of the conveyor and after that conveyed in a rotational motion. At the bottom side of the casting form a cooling liquid is applied to hardening the ice cream items in the casting forms. Another type of hardening tunnel with casting forms may also be used. The casting form are conveyed in a linear direction and the casting forms are connected as slits in a belt. The slits with the casting forms are filled with ice cream items in one end and conveyed in a linear direction with the casting form on top. After releasing the ice cream items from the casting forms the casting forms turns 180 degrees at the end of the belt and is being conveyed back to the ice cream former. At the ice cream former the casting forms may once again be filled with ice cream items. The number of casting forms perpendicular to the conveying direction may vary from e.g., 2-12 or even more according to the specific production of ice cream products. For both hardening tunnels with casting forms cooling fluid may be applied from below to cool the ice cream items by spraying it on the top and let the cooling fluid run down along the outside of the casting forms as in a so-called cascade system. More often, the system is a bath, where the cold liquid is coming in from below and the slightly heated fluid is flowing over the outsides of the casting forms. All three types of hardening tunnels is used for hardening and cooling ice cream items after the ice cream former. The ice cream items may be conveyed through the hardening tunnel HT in multiple minutes or even hours to ensure a uniform frozen temperature through the ice cream item. Through the hardening tunnel HT an adhesion between the ice cream items and the surface where the ice cream items are being conveyed on may beestablished. The adhesion may also evolve through the hardening tunnel due to the cooled conditions. In some production lines PL a hardening tunnel is optional and the ice cream items are cooled prior to the ice cream former in the smart freezer. This is typically a very expensive process to cool the smart freezer enough and further very complicated to process the ice cream composition through the ice cream former due to a stiffer ice cream composition.

[0129] The next step along the production line of the ice cream product is a packing station PACK, where the ice cream items are being wrapped in typically a foil. The ice cream items may be placed in a longitudinal foil which are being welded together to enclose typically one ice cream item. Afterwards the foil is being cutted and divided so one ice cream item is packed and enclosed in a foil. The packing station PACK may also be a packing station PACK where ice cream items are being packed directly in boxes of cardboard or paper without any foil. The packing station PACK may also comprise a secondary packing station PACK where the individually foil-enclosed ice cream items are being packed into boxes. The ice cream items may be packed in the boxes or containers by hand or by a robot.

[0130] Along the entire production line PL one or more sensor (not shown) may be placed at different locations to measure different properties, parameters and / or characteristics of the ice cream composition, ice cream characteristics and / or different stations (ice cream former, hardening tunnel, smart freezer, mixer, coating station, packing station, ingredients feeder, etc.). The sensor may further be connected to one or more controllers (not shown) with a wire or wireless to send the observed or measured properties, parameters and / or characteristics to the one or more controllers. The one and more controllers may further be connected to the different stations along the production line PL to adjust the different working stations based on measurements and / or observations. The location of the one or more sensors SENS may be used for any upstream adjustments of the ice cream composition, ice cream item or any parameters related to upstream working stations by the controller.

[0131] The one or more sensors may be a weight sensor, vision, camera, flow sensor, pressure sensor, temperature sensor, distance sensor or any other sensor for observingany kind of ice cream characteristics, parameters or properties. The sensors may also be used to measure the parameters of any of the working stations.

[0132] Through the different steps along the production line PL illustrated in fig. 4 an ice cream composition ICO is made in the smart freezer SF from ingredients ICN in the mixer MIX. The ice cream composition ICO is being shaped and divided into an ice cream item ICI at the ice cream former ICF. The ice cream item ICI are being processed through the production line PL where the hardening tunnel and packaging station refines the ice cream item. At the end of the production line PL a ice cream product is made which is ready for sale. This may be an individually ice cream product in a foil which typically is bought in small shop. The ice cream product may also be a box with multiple foiled ice cream products which may be bought in a super market.

[0133] Fig. 4b illustrates the same embodiment as illustrated in fig. 4a with one or more optionally parts added to the production line PL of ice cream products. Before the ice cream former ICF an optional ingredients feeder INF may be placed to mix the ice cream composition from the smart freezer SF with small pieces of chocolate, berries, cake, caramel or any other edible things.

[0134] Fig. 4b further illustrates that after the ice cream items have left the hardening tunnel HT one or more optional coating stations COA is placed along the production line PL. The coating station COA may typically be a station where the ice cream item is being dipped into a bath of warm chocolate. The ice cream items are then being lifted with the coated chocolate dripping off the ice cream items, but with a new layer added to the ice cream item. The ice cream item may also be coated with other edible things or different types of chocolate. After the first dip of coating an additional coating may be provided in form of another layer of chocolate or solid small pieces of chocolate, berries or any other edible solid stuff.

[0135] Additional optional working stations along the production line PL which is not shown may be any of the following: Smart Cutter, Coating check station, serialization, packaging check station, a manual station, a cleaning station, or any other working station related to the production of ice cream products.

[0136] Fig. 5a illustrates a method for adjusting an ice cream composition ICO that may be implemented by automatic process control of a smart freezer SF. Automatic process control is implemented by a controller CON for providing a smart freezing setting SFS based on analysed observed ice cream characteristics ICC of said ice cream composition and an input desired ice cream quality FICQ comprising one or more desired ice cream characteristics ICC.

[0137] Fig. 5a illustrates automatic process control of a smart freezer wherein process control is implemented as a closed-loop negative feedback system utilizing a controller CON. Ice cream composition ICO is outputted from the smart freezer SF, where one or more sensors or an operator observes the ice cream composition ICO from the smart freezer SF and inputs observed ice cream characteristics ICC e.g. by a user interface UI to be compared with a desired ice cream quality. Ice cream characteristics may relate to different aspects of the smart freezer and its output ice cream composition ICO such as temperature, viscosity, or foam structure of ice cream composition ICC.

[0138] The observed ice cream characteristics are fed to a summing point at which the error between desired ice cream characteristics and the observed ice cream characteristics is computed. The error E is supplied as an input to the controller CON which provides a smart freezing setting SFS comprising a plurality of adjustable smart freezer parameters SFP .

[0139] The smart freezing setting SFS is provided as an input to the smart freezer SF to adapt the ice cream composition such that a desired ice cream quality FICQ of the ice cream composition ICO may be obtained or the deviation minimized e.g. through achieving or maintaining appropriate observed ice cream characteristics ICC.

[0140] The desired ice cream quality FICQ is representation of the desired state of the system i.e. it forms a representation of the desired ice cream composition ICO as provided by the smart freezer SF. The desired ice cream quality FICQ can be parameterized according to one or more variables which may or may not correspond to observable ice cream characteristics ICC. The desired ice cream quality FICQ canalso be parameterized in terms that are not directly observable or measurable as ice cream characteristics ICC. Suitable conversion of the desired ice cream quality FICQ must then be handled by the controller CON.

[0141] The automatic process control of a smart freezer as illustrated on fig. 5a may be implemented using different design methodologies.

[0142] In an example, a design methodology may require a model estimate of the smart freezer SF, i.e. the dynamics of the smart freezer and associated sensors may be formalized in a model which expresses the dynamics of the system. This model may e.g. be a set of differential equations which can be further modified to form a statespace model. A model of the smart freezer may e.g. be obtained via methods for system identification, or it may be based on known physical laws and relationships such as modelling the behaviour of the ice cream composition using computational fluid dynamic modelling. Such modelling may also be used to improve the performance of machine learning control models according to various embodiments of the invention. For example including initializing a machine learning control model based on e.g. a simulation of fluid dynamics.

[0143] According to the present example, a controller CON may be designed based on the estimated dynamical model of the smart freezer SF. A design methodology which may be employed is a lead-lag compensator designed with pole-zero placement by which the controller CON is modelled as a transfer function in cascade with the obtained model of the smart freezer SF. The desired transfer function of the controller is obtained by analysing the open-loop system consisting of at least the controller CON and the smart freezer SF. The lead-lag compensator is then analytically derived by adapting the poles and zeros of the open-loop system based on the required dynamic and static parameters of the closed-loop system i.e. when the smart freezer SF is being regulated by the controller CON in a negative feedback loop as indicated on fig. 5a.

[0144] Dynamic and static parameters may include design goals such as desired overshoot, system bandwidth, stability requirements, time delay, or steady-state errorwhich may be determined approximately by placing the poles and zeros of the openloop system by deriving a lead and a lag compensator.

[0145] In an example, the dynamical model of the smart freezer is given by a transfer function in the frequency domain. The controller CON may equivalently be modelled as a transfer function in the frequency domain. The lead-lag compensator is then designed according to derived dynamic and static behaviour of the open-loop system which is analytically determined. Based on such design constraints, parameters of the lead and lag compensator may be analytically derived, or they may be derived based on experimentation or measurements or a combination of the preceding methods.

[0146] In an example, the controller CON is a proportional-integral-derivative PID controller implemented in software or hardware.

[0147] A hardware implementation of the PID controller may follow a design approach substantially similar to the one described in the preceding example.

[0148] Design of a PID controller for implementation in software can be achieved by using parts of the design approach as described in the preceding example. Given a dynamical model of the smart freezer SF and the transfer function of a PID controller which includes a proportional, integral, and derivative term in the time domain, the parameters of the PID controller may be derived by experimentation or modelling of the frequency response of the open-loop system comprising at least the PID controller and the model of the smart freezer SF. The continuous time transfer function of the PID controller can then be approximated as a discrete time model suitable for implementation in software by utilizing e.g. a Pade approximation.

[0149] In an example, the observed ice cream characteristics of the automatic process control are ice cream temperature, ice cream viscosity, ice cream shape, and ice cream weight. The desired ice cream quality FICQ formed as an input to the system may correspondingly be parameterized in terms of desired ice cream characteristics such as ice cream temperature, desired ice cream viscosity, desired ice cream shape, and desired ice cream weight.

[0150] The observed ice cream characteristics are provided as sensor outputs from the smart freezer i.e. the smart freezer comprises at least a temperature sensor for measuring ice cream temperature, a viscosity sensor for measuring ice cream viscosity, a camera for measuring ice cream shape, a weight sensor for measuring ice cream weight or an operator for observing the mouthfeel of the ice cream e.g. according to foam-structure.

[0151] The sensor outputs are provided in a feedback loop to the input of the controller CON whereby the error E between the desired ice cream quality FICQ and the observed ice cream characteristics is computed.

[0152] The error E is provided to the controller CON as input, and the controller CON then provides a smart freezer setting SFS comprising smart freezer parameters. In the present example, the smart freezer parameters comprise at least dasher-rotation, smart freezer temperature, smart freezer pressure, overrun, time in freezer, and volume-flow.

[0153] It should be noted that other embodiments may apply fewer smart freezers settings.

[0154] The smart freezer setting SFS provided by the controller CON is provided to the smart freezer SF which adapts the output of its smart freezer parameters such that they correspond to the supplied smart freezer setting SFS. Adaptation of smart freezer parameters may involve some or all of the parameters of the smart freezer SF such as compression or decompression of refrigerant, increase or decrease of dasher-rotation, or adjustment of the overrun.

[0155] The controller CON is designed such that the supplied smart freezer setting SFS drives the output of the smart freezer SF towards the desired ice cream quality FICQ by adapting the parameters of the smart freezer SF via the negative feedback loop as indicated on fig. 5a.

[0156] In an example, the observed ice cream characteristics ICC measured at the output of the smart freezer SF may comprise physical inputs to the smart freezer SFsuch as measurements related to the mix of ingredients that are supplied as a physical input to the smart freezer SF. Such ice cream characteristics ICC may be relevant to the automatic process control of a smart freezer SF because of variation in the composition of the ingredients.

[0157] Fig. 5b illustrates an embodiment of the invention related to fig. 5a wherein measured smart freezer parameters MSFP are incorporated into the closed-loop negative feedback control system along with ice cream characteristics ICC. Measured smart freezer parameters MSFP are distinguished from ice cream characteristics ICC in that they are not related to measurements performed directly on the ice cream composition ICO. The measured smart freezer parameters are provided to the summing point where they are compared with the desired ice cream quality FICQ to produce an error E which is provided as an input to the controller CON. Note that the controller CON of fig. 5b is typically not identical to the controller illustrated on fig. 5a, because the controller CON must be adapted to account for additional error E inputs.

[0158] The embodiment of the invention illustrated on fig. 5b utilizes additional measurements that are available from the smart freezer SF, but that do not necessarily relate directly to the ice cream composition ICO. Examples of measured smart freezer parameters are temperature of freezer, rotation speed of dasher, flow of refrigerant, temperature of refrigerant, power consumption and so on. Incorporating such measurements into the closed-loop negative feedback control system may allow for a better design of the controller CON and consequently allow the system to more effectively achieve a desired ice cream quality FICQ.

[0159] Fig. 5c illustrates an embodiment of the invention related to fig. 5a with an additional input parameter to the controller CON. The input parameter is an external operational parameter EOP. The external operational parameter EOP provides a measure of conditions that necessitate that the controller CON is adapted to effectively continue to control the smart freezer SF to achieve a desired ice cream quality FICQ. The external operational parameter EOP may be one or more measurements of ambient conditions that affect the dynamics of the system such as ambient temperature or pressure. The one or more measurements of ambient conditions are provided as anexternal operational parameter EOP to the controller CON which may in turn be adapted accordingly.

[0160] The external operational parameter EOP may be a parameter which indicates changes to the dynamics of e.g. the smart freezer or to sensors providing ice cream characteristics or measured smart freezer parameters MSFP. The external operational parameter EOP can be related to measurements performed on the smart freezer SF which indicate dynamical changes that necessitate an adaptation of the controller CON.

[0161] The controller CON illustrated on fig. 5c is typically not identical to the controllers of fig. 5a or 5b because the controller CON of fig. 5c is an adaptive controller. The adaptive controller of fig. 5c can be implemented through various means e.g. through providing for the adaptation of its internal parameters such as the locations or poles or zeros based on the external operating parameter EOP that is provided to the controller CON as an input. The adaptive controller is therefore able to function even if the smart freezer cannot be modelled as a time-invariant system because it can adapt to dynamics that change with time whether these are external or internal to the system as a whole.

[0162] It is to be understood that the embodiments illustrated on figs. 5a-5c may be combined by incorporating aspects of each example embodiment into other embodiments. As an example, closed-loop negative feedback control can be implemented by combining the embodiment of fig. 5b with that of fig. 5c such that the closed-loop negative feedback control system utilizes both ice cream characteristics ICC and measured smart freezer parameter MSFP and additionally comprises an adaptive controller CON which takes as an input an external operational parameter EOP that provides for e.g. adaptation of the controller’s internal parameters.

[0163] In an optional embodiment of the invention, the controller, including the controller illustrated in, e.g., fig. 5a includes a trained machine learning control model. The trained machine learning control model provide smart freezing setting to the smart freezer SF. In this embodiment of the invention, the machine learning control modelis trained to learn how inputs to the smart freezer in the form of a smart freezing setting are mapped to a corresponding ice cream quality provided by the freezer (output) given the smart freezing setting. Once this knowledge of the freezer is learned by the model, the trained machine learning model may be applied to control smart freezing setting SFS based on a desired ice cream quality characterized by one or more desired ice cream characteristics and further based on ice cream characteristics ICC associated with the ice cream made by the smart freezer SF. Advantageously, the control provided by the trained machine learning control model is data driven and hence, advantageously, providing an analytical model of the system may not be required. The machine learning control model may be trained based on training data collected during making of ice cream with the smart freezer. The training data may at least comprise a desired ice cream quality, a corresponding smart freezing setting, corresponding achieved ice cream quality (output), and preferably ice cream ingredients. The achieved ice cream quality may as already mentioned, e.g., be quantised by obtaining ice cream characteristics ICC of the ice cream being produced by the smart freezer utilizing the smart freezing setting. Based on the training data, the machine learning model thereby may learn to determine the smart freezing setting needed to achieve a desired ice cream quality. When the machine learning control model is implemented to provide smart freezing setting, the machine learning control model receives the ice cream characteristics and may thus adjust the smart freezing settings based on the received ice cream characteristics in order to achieve the desired ice cream characteristics. As illustrated in fig. 5a, the model may receive the error between the desired ice cream quality and the actual ice cream quality. E.g., the actual ice cream quality may be characterized by way of ice cream characteristics, and the desired ice cream characteristics may be characterised by desired ice cream characteristics. The ice cream characteristics and the desired ice cream characteristics may include the same parameters. The error may then be calculated by a comparison between the obtained one or more ice cream characteristics and the one or more desired ice cream characteristics.

[0164] Optionally, the machine learning control model may in addition to the desired ice cream quality and the ice cream quality (sometimes referred to as actual ice creamquality) be trained based on further input training data, including, e.g., one or more external operational parameter(s) and / or one or more parameter(s) related to the ice cream ingredients. By training the machine learning control model on data collected in various conditions, including various ambience conditions, the machine learning model may provide adaptive control, e.g., adaptive in the sense that the model learns to provide smart freezing setting to the smart freezer to achieve an ice cream quality specified by the desired ice cream quality under various conditions in which the model has been trained.

[0165] The machine learning control model may be trained on different datasets, to enable the machine learning model to be able to control the smart freezer and achieve desired ice cream quality when producing different types of ice cream, which may, e.g., be based on different ingredients, and thereby require a different control to achieve a desired ice cream quality.

[0166] Fig 6 schematically illustrates training of a machine learning control model based on reinforcement learning according to an embodiment of the invention.

[0167] The reinforcement learning control model illustrated in fig. 6 may be considered a relatively simple-to-implement reinforcement learning control model, and other embodiments of the invention may include more advanced reinforcement learning models, as described elsewhere in this disclosure.

[0168] The reinforcement learning control model or system illustrated in fig. 6 includes an agent AGT and an environment ENV comprising at least a smart freezer SF for producing ice cream composition ICO based on ice cream ingredients ICN. The agent AGT utilizes a policy, which may be considered an algorithm or a a mathematical framework, to determine actions (smart freezing settings SFS) based on the current state ST of the environment ENV. The environment ENV responds or changes as a result of the actions (smart freezing settings SFS) determined by the agent (policy) into a new state ST and feedback associated with the new state ST is provided in the form of rewards RW that enable the agent AGT to refine its strategy over time. In this example, the state of the environment (the smart freezer) may be based onmeasured or otherwise obtained ice cream characteristics, external operating parameters or other data associated with the environment (smart freezer, ice cream composition etc.). Again, the environment ENV may include the smart freezer, the produced ice cream, the external operating parameters etc. During training of the reinforcement learning control model, the model is applied to control the smart freezer, and the reinforcement learning control model, the agent determines actions based on a policy that improves during learning and in return, receives a state and a corresponding reward. The training of the reinforcement learning control model may be considered an optimization problem, wherein the model is trained to optimize the rewards or the accumulated rewards. In this way, the reinforcement learning control model learns over time a policy, that provides the actions for a given states that optimizes the reward, and thereby the model learns to output the optimal smart freezing setting (action) for given states of the environment including the quality of the ice cream in relation to the desired ice cream quality.

[0169] The agent AG is responsible for deciding the smart freezing setting (action) based on the learned policy. In other words, the agent determines one or more actions that influences the environment ENV, and in this case, by providing smart freezing settings to the smart freezer. The agent AGT incorporates a policy. The policy is implemented using machine learning algorithms capable of learning and adapting from the environment and based on the state ST of the environment ENV which changes based on the action. The learning is further based on rewards given when the action leads to a desired state, e.g., when the action provides a desired ice cream quality etc. A reward system provides feedback to the agent in terms of rewards. The reward is based on the outcomes of the agent's actions and guides the agent in improving its decision-making strategy, e.g., determining the smart freezing setting that optimizes the reward.

[0170] The environment may be considered a system. The system behaviour may optionally modelled or be learned based on machine learning. Such knowledge of the behaviour of the system may improve the performance of the machine learning control model, and / or minimize the time required to train the machine learning control model.

[0171] Optionally, various types of sensors may be utilized to measure external operating parameters or ice cream characteristics.

[0172] Optionally, the system may comprise a quality control interface configured to provide input about the ice cream quality. The quality control interface may, e.g., enable a quality control person to input information about the ice cream quality, e.g., after testing the ice cream quality.

[0173] Rewards may be given continuously based on ice cream characteristics that may be measured continuously. However, some rewards may also be provided less frequently. E.g., rewards related to ice cream quality that requires a test person to taste test the ice cream, or rewards provided based on quality control of the ice cream that requires measurement or testing of samples of ice cream, e.g., using equipment that is not implemented directly in the ice cream production line. Some parameters related to the ice cream quality may be given higher rewards, e.g., the outcome of a taste test by an expert quality control person may be weighted higher than, e.g., temperature measurements of the ice cream.

[0174] During training of the reinforcement learning control model, the model may be applied to control the smart freezer, and the reinforcement learning control model, the policy, determines actions and in return, the receives a state and a corresponding reward. The training of the reinforcement learning control model may be considered an optimization problem, wherein the model is trained to optimize the rewards or the accumulated rewards. In this way, the reinforcement learning control model learns over time, which actions for a given states that optimizes the reward, and thereby the model learns to output the optimal smart freezing setting (action) for given states of the environment including the quality of the ice cream in relation to the desired ice cream quality.

[0175] The following comprises a description of embodiments of the invention without references to a specific figure. These embodiments include different types of machine learning control models, which may be implemented to control smart freezing settings. Most of these models may be considered more complex variants of thereinforcement learning control model described in relation to fig. 5a. The models may be implemented in a control system, including the control systems illustrated in fig. 5a - fig. 6c. Machine learning based control models are advantageous, e.g., because the models learn the system behaviour based on data, and hence, the models may be optimized over time by retraining the models as more and more data are collected. The models are also adaptive because the models may be adapted by training the model based on different training data. Hence the model may be adapted to control the smart freezer in production of various types of ice cream and adapted to various production line locations that may be characterized by differing in external operating parameters.

[0176] Different types of machine learning models may be utilized for the machine learning control model, including e.g., various types of reinforcement learning models, including deep reinforcement learning, and Q-leaming, etc. Further models that may be utilized as the machine learning control model includes one or more of the following reinforcement learning control models: deep deterministic policy gradient (DDPG) algorithm, Proximal Policy Optimization (PPO), actor critic algorithms including soft actor critic (SAC), deep Q-network (DQN). We note that also genetic programming may be utilized according to an embodiment of the invention.

[0177] The deep deterministic policy gradient (DDPG) algorithm is a model -free, off-policy reinforcement learning method, which may advantageously be implemented as a machine learning control model, according to an embodiment of the invention. A DDPG agent is an actor-critic reinforcement learning agent that searches for an optimal policy that maximizes the expected cumulative long-term reward, concurrently learns a Q-function and a policy. The algorithm uses off-policy data and the Bellman equation to learn the Q-function and uses the Q-function to learn the policy.

[0178] Proximal policy optimization (PPO) may be utilized as a machine learning control model according to an embodiment of the invention. The PPO may be classified as a policy gradient method for training an agent’s policy network. The policy network is the function that the agent uses to make decisions. To train the right policy network, PPO takes a small policy update (step size), so that the agent may reliably reach the optimal solution. A too-big step may direct policy in the falsedirection, thus having little possibility of recovery; a too-small step lowers overall efficiency. Consequently, PPO implements a clip function that constrains the policy update of an agent from being too large or too small. Advantageously, PPO strikes a balance between performance and comprehension.

[0179] The Deep Q-network (DQN) algorithm is a model -free, off-policy reinforcement learning method, which may advantageously be implemented as a machine learning control model, according to an embodiment of the invention. Deep q-network agent is a value-based reinforcement learning agent that trains a critic to estimate the expected discounted cumulative long-term reward when following the optimal policy. DQN may be considered a variant of Q-learning that features a target critic and an experience buffer. DQN may be considered a relatively simple and effective model. Moreover, DQN may mitigate data correlation. In essence, DQN combines principles of deep neural networks with Q-leaming, enabling agents to learn optimal policies in the complex control of the smart freezer. DQN may utilize experience replay, which may advantageously help in decorrelating the sequential experiences by storing them in a replay memory buffer. This memory buffer is randomly sampled during the network update to break the temporal dependencies and stabilize learning.

[0180] Soft Actor Critic (SAC) is an algorithm that optimizes a stochastic policy in an off-policy way, forming a bridge between stochastic policy optimization and DDPG-style approaches. While the SAC algorithm may be best suited for continuous action spaces, it may be implemented as a machine learning control model according to an embodiment of the invention.

[0181] Advantageously, reinforcement learning focuses on the interaction between an agent and its environment, the smart freezer, where the agent learns to maximize rewards by selecting optimal control actions, e.g., selecting optimal smart freezing setting. E.g., the agent receives a reward when the determined smart freezing setting results in an ice cream quality similar to the desired ice cream quality.

[0182] When training the reinforcement learning model, rewards may be given when ice cream characteristics become closer or substantially matches the desired ice cream characteristics. E.g., when the error between the ice cream characteristics and the desired ice cream characteristics is minimized. This may be considered dense reward. Rewards may also be given when the ice cream quality becomes closer to the desired ice cream quality or when it substantially matches the desired ice cream quality, as determined, e.g., by a quality control scheme. This quality control scheme may involve various parameters including ice cream characteristics, such as foam structure, temperature, pressure etc., however, the quality control scheme may also involve a quality control person tasting an ice cream produced by the smart freezer, and based on the taste test, the quality control person will give a feedback as to whether the quality of ice cream substantially matches the desired ice cream quality. Optionally, the feedback from the quality control person may be given more weight, e.g., a higher reward, compared to other rewards.

[0183] A machine learning control model based on reinforcement learning may take long time to train, in turn, resulting in lots of waisted ice cream, as the reinforcement learning model trains by testing different smart freezing settings and learning from the corresponding feedback (rewards) given in response to these actions.

[0184] Optionally, reward shaping and / or imitation learning may be utilized to train the reinforcement learning control model. Advantageously, in imitation learning, an expert in ice cream making is determining the smart freezing setting to achieve a given desired ice cream quality, and furthermore, the expert is evaluating the ice cream quality of the produced ice cream, and giving feedback as to whether the ice cream quality of the produced ice cream matches the desired ice cream quality. The expert may during production adjust the ice cream setting and when adjusting the settings continuing to evaluate the ice cream quality. The reinforcement learning control model is then capable of learning from the expert, which smart freezing settings that maximizes the reward (provides the desired ice cream quality). Advantageously, imitation learning may vastly minimize the amount of training iterations it takes before the reinforcement learning control model becomes capable of performing at a levelthat provides the desired ice cream quality. Optionally, the training of the reinforcement learning control model may be continued without the imitation learning, to improve the performance of the reinforcement learning model.

[0185] Notice that the concept of imitating the smart freezing setting determined by a human expert in ice cream making may be implemented using various types of machine learning, including different types of supervised learning and thereby not only reinforcement learning. However, while reinforcement learning may sometimes be time consuming to train, reinforcement learning control models may typically outperform traditional supervised learning algorithm in controlling the smart freezer by controlling smart freezing settings.

[0186] The reinforcement learning models may be trained and operated on various data related to the smart freezer, including, e.g., external operational parameters. The type of data, including, e.g., external operational parameters, that may be used to train machine learning control models, have already been described previously, and we note that these data may also be applied to train other machine learning models according to the invention, including reinforcement learning models such as those described above.

[0187] Optionally, the training of machine learning control models according to the invention may include using a model of the smart freezer as a starting point. E.g., an analytical model or an empirically determined model. Optimally, the machine learning model may first be trained to copy or approximate the model of the smart freezer and then the machine learning control model having approximated the model may be trained based on training data. This may advantageously minimize the iterations required to train the machine learning control model to control smart freezing settings.

[0188] In an optional embodiment of the invention, the control of the smart freezing settings may be based on a genetic algorithm. A genetic algorithm approach may further advantageously be combined with machine learning control approaches, including those already described in this disclosure. Advantageously, genetic algorithms may be said to breed the solution to the control problem using an interactiveprocess involving probabilistic selection of the fittest solutions by means of a set of genetic operators.

[0189] Fig. 7 illustrates an embodiment of the invention in which the smart freezer setting SFS provided to the smart freezer SF (not shown) varies with time. For illustrative purposes, three different smart freezer settings SFS at three different times are shown. These correspond to smart freezer setting tl SFStl at time tl, smart freezer setting t2 SFSt2 at time t2, and smart freezer setting t3 SFSt3 at time t3. Each of the three smart freezer settings SFS comprise smart freezer parameters SFP. With reference to smart freezer setting tl SFStl, four smart freezer parameters are shown and these correspond to smart freezer parameter 1 SFP1, smart freezer parameter 2 SFP2, smart freezer parameter 3 SFP3, and smart freezer parameter n SFPn. A smart freezer setting SFS may of course comprise fewer or more smart freezer parameters SFP than the four illustrated on fig. 7. The smart freezer parameters SFP1, SFP2, SFP3, SFPn may typically refer to parameters a user may control manually when using some kind of a conventional ice cream freezer user interface. It should however also be noted that the smart freezer parameters may also be automatically deducted from each other or automatically controlled within the scope of the invention on the basis of an automatically provided smart freezer setting according to the provisions of the invention. Please refer to fig 8a-c and e.g. fig. 5a-c for details on how these setting are provided and how these settings are applied as assistance / guiding to an operator of the smart freezer as a response to a desired ice cream quality manually input via the user interface UI or as a response to desired ice cream quality e.g. automatically associated to the ice cream recipe, type of ice cream, the specific smart freezer and / or the specific production line.

[0190] Moreover, in the above referred figures different more automatic implementations are indicated e.g. automatically provided new and updated adjustable smart freezer parameters SFP which may be at least partly automatically applied in the control loop of the smart freezer.

[0191] In the present example, the smart freezer parameters may e.g. comprise dasher-rotation, dasher power consumption (also referred to as viscosity or dasherresistance within the art), smart freezer temperature, smart freezer pressure, overrun, time in freezer, and volume-flow.

[0192] The parameter values of each of the four smart freezer parameters SFP1, SFP2, SFP3, and SFPn at time tl are indicated on fig. 7 as respectively Xtl, Ytl, Ztl, and Wtl. The indicated parameter values Xtl, Ytl, Ztl, and Wtl may for example respectively correspond to particular values of dasher revolutions per minute, overrun, viscosity, and cylinder pressure. The four parameter Xtl, Ytl, Ztl, and Wtl may be adapted as a function of time, i.e. at time t2 a new smart freezer setting t2 SFSt2 may be obtained which has new values of the four smart freezer parameters SFP1, SFP2, SFP3, and SFPn these values being indicated on the figure as Xt2, Yt2, Zt2, and Wt2.

[0193] The smart freezer setting tn SFStn at time tn illustrates that the smart freezer parameters SFP may be further adapted at a different time tn. The parameter values of the smart freezer parameters SFP1, SFP2, SFP3, and SFPn are indicated at time tn as Xtn, Ytn, Ztn, and Wtn. The three different smart freezer settings SFStl, SFSt2, and SFStn thus each comprise unique smart freezer parameters with particular parameter values that are adapted such that the smart freezer provides a desired ice cream quality at each of the different times tl, t2, and tn.

[0194] Fig. 8a illustrates an embodiment of the invention in which the smart freezer SF is controlled in a closed-loop feedback by a controller CON providing smart freezer settings SFS to a user interface UI. For illustrative purposes, the observed ice cream characteristics ICC are shown as being used directly as an input to the controller CON, but it must be understood that the embodiment on fig. 8a can also implement a closed- loop negative feedback control system i.e. according to embodiments as illustrated on figs. 5a-5c where the difference between a desired ice cream quality FICQ (not shown) and a observed ice cream characteristics ICC is the input to the controller CON.

[0195] The user interface UI provides an interface through which smart freezer settings SFS provided by the controller CON may be selected or evaluated by an operator (not shown). The operator may evaluate a smart freezer setting SFS via theuser interface UI and select a smart freezer setting SFS via the user interface UI which is transmitted to the smart freezer SF as a user confirmed smart freezer setting UCSFS.

[0196] The user interface may comprise a processor, communication means, power supply, and means for visualizing different smart freezer settings SFS. The user interface UI may also be implemented in hardware i.e. via buttons, switches, and levers through which the operator can e.g. select or approve a smart freezer setting SFS.

[0197] Fig. 8b illustrates an embodiment of the invention in which the smart freezer SF is controlled in a closed-loop feedback by a controller CON providing smart freezer settings SFS to a user interface UI. The smart freezer settings SFS are provided to the user interface UI but the operator (not shown) is not able to forward these smart freezer setting SFS to the smart freezer SF. The user interface UI thus allows for the operator to inspect the smart freezer settings that are provided by the controller CON.

[0198] The controller CON provides the smart freezer setting SFS directly to the smart freezer SF i.e. the operator does not approve or select an appropriate smart freezer setting SFS as in the embodiment illustrated on fig. 8b.

[0199] Fig. 8c illustrates an embodiment of the invention in which the smart freezer SF is controlled in a closed-loop feedback by a controller CON providing smart freezer settings SFS to a user interface UI. The embodiment of fig. 8c is different from those of figs. 8a-8b in that a smart freezer setting SFS may be provided directly by the controller CON to the smart freezer SF or via the user interface as selected by an operator (not shown) and comprising a user confirmed smart freezer setting UCSFS.

[0200] The operator can provide a desired ice cream quality FICQ via the user interface UI to the controller CON in any of the embodiments of the invention illustrated on figs. 8a-8c. The desired ice cream quality FICQ may for example be selected according to observations made by the operator such as texture or stickiness of the ice cream composition. The desired ice cream quality FICQ may correspond to a number of parameters which the operator can adapt accordingly. The user interface UI may aid in obtaining the desired ice cream quality FICQ in that the desired ice cream quality is parameterized in a way that is easy for an operator to comprehendthus making the smart freezer SF more user-friendly than if the operator were to control individual parameters such as temperature, pressure or dasher rotation directly. As already mentioned, an important object of the invention is to facilitate that an operator may use the smart freezer without little or no deep knowledge about the mutal complexities related to adjustable smart freezer parameters when those are set or to be set.

[0201] Thus instead of focusing on these complexities, the operator may simply select a desired ice cream quality FICQ and / or observed ice cream characteristics ICC in terms he or she understands (which is not immediately relatable to adjustable smart freezer parameters: The operator may thus e.g. for example select:• an ice cream ingredient recipe (e.g. selected one from a list of recipes)• ice cream composition out of freezer (volume flow per time)• overrun• estimated ice cream quality o estimated real fluid Viscosity o sensory features like creaminess, coldness, etc., o functional features like stickiness, consistency, etc ... o ice cream composition temperature

[0202] The freezer will then estimate the setting (a setting of a combination of adjustable parameters), e.g., at least three, needed to achieve this specified FICQ.

[0203] In assisted mode, this can be a suggestion, and the operator can follow it, decide to modify or even completely overrule. See e.g. fig. 8a.

[0204] In automated mode, the settings are used, and might even be hidden for the operator, se e.g. fig. 8b.

[0205] The used freezer mode can be depending on different user groups, so for instance an expert can use assisted and an operator with less experience and skills can use automated mode. This may e.g. be obtained in relation to fig. 8c, implementing both assisting and automated mode.

[0206] There can be many different ice cream ingredient ICN recipes, so a smart freezer SF might not have any or only few data on the selected one.

[0207] Also, there can be time dependent variations on same type of ice cream ingredient ICN recipe, due to for instance maturing conditions which might be somewhat different from time to time, when prepared before used in the Smart Freezer. Also, there can be some variations in some of the ice cream ingredient ICN, for instance vegan ice cream ingredients ICN can have varying properties depending on where they are grown.

[0208] The operator can adjust the ice cream composition ICO, by describing the observed ice cream composition ICO pumped out of the freezer, for instance inputting that the ice cream composition ICO too sticky or not consistent, have air void in the flow etc. These inputs may typically be prepared and therefore be avail to a user for selection e.g. from a menu and then the controller CON will translate the relatively request into a concrete setting of adjustable parameters which may be automatically applied or visualized as a suggestion to the operator.

[0209] It is noted that the dialog between an operator which may not be very skilled on how to set the adjustable smart freezer parameters but skilled enough to notice whether ice cream is functionally not as intended or e.g. in terms of sensory performance. Therefore, such an ongoing dialog between an operator good enough to express desired ice cream quality (FICQ) as “soft” parameters and when something needs to be adjusted (while relating this to the current and updated setting of the adjustable smart freezer parameters) will be a perfect date set for machine learning purposes therefore on the long perfecting the automatic providing of adjustable smart freezer parameters. This is e.g. elaborated on in relation to fig. 5a-c and fig. 6, e.g. also in combination with fig. 8a-c.

[0210] In other words, the smart freezer may correct the ice cream composition, and at the same time learn how the particular current ice cream ingredient ICN recipe needs to be processed.

[0211] The smart freezer may in this way also be able to learn from an experienced and skilled expert user.

[0212] List of reference signs:ACP Adaptive controller parameters,AGT AgentANL Automatically analyzingDAS DasherDICC Desired ice cream characteristicsCON ControllerCOA Coating stationCOP CompressorE ErrorENV EnvironmentEOP External operating parameter(s)FICQ Desired ice cream qualityICC Ice cream characteristicsICF Ice cream former,ICN Ice cream ingredientsICO Ice cream compositionINF Ingredients feederMIX Mixer,PACK Packing station,PL Production line,RSYS Refrigeration systemRW RewardSCB Scraper blade(s),SF Smart freezerSFP Adjustable smart freezer parametersSFS Smart freezing settingST State1 Mix inlet2 Mix pump3 Mix piping4 Flow meter5 Dasher motor6 Cylinder pressure sensor 7 Air inlet8 Air controller9 Temperature sensor10 Freezing cylinder11 Product pump

Claims

Claims1. A method for adjusting an ice cream composition (ICO) processed by a smart freezer (SF) comprising: providing ice cream ingredients (ICN) to said smart freezer (SF), freezing and processing said ice cream ingredients (ICN) into an ice cream composition (ICO), observing one or more ice cream characteristics (ICC) of said ice cream composition (ICO), automatically analyzing said one or more ice cream characteristics (ICC), automatically providing a smart freezing setting (SFS) comprising a plurality of adjustable smart freezer parameters (SFP) based on said analysed observed ice cream characteristics (ICC) and a desired ice cream quality (FICQ) of said ice cream composition (ICO).

2. The method according to claim 1, wherein adjusting of said ice cream composition (ICO) is performed by using the provided smart freezer setting (SFS) as a basis for adjusting the smart freezer parameters defined by said provided smart freezer setting.

3. The method according to any one of the preceding claims, wherein said desired ice cream quality (FICQ) of said ice cream composition is defined prior to said step of providing said ice cream ingredients (ICN) to said smart freezer (SF).

4. The method according to any one of the preceding claims, wherein said desired ice cream quality (FICQ) is based on an ice cream composition type (ICCT).

5. The method according to any one of the preceding claims, wherein said observing is a measuring performed by one or more sensors.

6. The method according to any one of the preceding claims, wherein said method comprises an additional step of measuring the smart freezer parameters (SFP) as theyare set prior to automatically analysing said one or more ice cream characteristics (ICC).

7. The method according to any one of the preceding claims, wherein said observing is done by one or more sensor(s).

8. The method according to any one of the preceding claims, wherein said ice cream characteristics (ICC) are macroscopic characteristics.

9. The method according to any one of the preceding claims, wherein said ice cream characteristics (ICC) are microscopic characteristics.

10. The method according to any one of the preceding claims, wherein said automatically analysing said one or more ice cream characteristics is based on said ice cream type being made.

11. The method according to any one of the preceding claims, wherein said smart freezer setting (SFS) relates to a desired ice cream composition type.

12. The method according to any one of the preceding claims, wherein said ice cream ingredients are continuously added to the smart freezer.

13. The method according to any one of the preceding claims, wherein said ice cream composition is based on said ice cream ingredients (ICN) being added to said smart freezer (SF).

14. The method according to any one of the preceding claims, wherein multiple types of ice cream ingredients (ICN) are added to said smart freezer (SF).

15. The method according to any one of the preceding claims, wherein said ice cream composition is adapted to an ice cream type.

16. The method according to any one of the preceding claims, wherein said ice cream composition is made from one or more types of ice cream types with different flavours.

17. The method according to any one of the preceding claims, wherein said provided smart freezing setting (SFS) comprising a plurality of set adjustable smart freezer parameters (SFP) based on said analysed observed ice cream characteristics (ICC) and a desired ice cream quality (FICQ) of said ice cream composition (ICO) is correlated to power consumption of the smart freezer.

18. The method according to any one of the preceding claims, wherein said provided smart freezing setting (SFS) comprising a plurality of set adjustable smart freezer parameters (SFP) based on said analysed observed ice cream characteristics (ICC) and a desired ice cream quality (FICQ) of said ice cream composition (ICO) is correlated to power consumption of a dasher of the smart freezer.

19. The method according to any one of the preceding claims, wherein said automatically providing a smart freezing setting (SFS) is performed by a controller (CON) configured to implement a closed-loop feedback control comprising the steps of: establishing an error (E) as the difference between said desired ice cream quality (FICQ) and said observed ice cream characteristics (ICC), processing said error (E) to obtain said smart freezing setting (SFS), communicating said smart freezing setting (SFS) to a user interface (UI).

20. The method according to any one of the preceding claims, wherein said automatically providing a smart freezing setting (SFS) is performed by a controller (CON) configured to implement a closed-loop feedback control comprising the steps of: establishing an error (E) as the difference between said desired ice cream quality (FICQ) and said observed ice cream characteristics (ICC), processing said error (E) to obtain said smart freezing setting (SFS), providing said smart freezing setting (SFS) as an input to said smart freezer (SF).

21. The method according to any one of the preceding claims, wherein said automatically providing a smart freezing setting (SFS) is performed by a controller (CON) configured to implement a closed-loop feedback control comprising the steps of: establishing an error (E) as the difference between said desired ice cream quality (FICQ) and said observed ice cream characteristics (ICC), processing said error (E) to obtain said smart freezing setting (SFS), providing said smart freezing setting (SFS) as an input to said smart freezer (SF) for automatic adjustment of a present setting of the smart freezer.

22. The method according to any one of the preceding claims, wherein said error (E) further comprises a difference between said desired ice cream quality (FICQ) and measured smart freezer parameters (MSFP).

23. The method according to any one of the preceding claims, wherein said controller (CON) is a proportional-integral-derivative PID controller.

24. The method according to any one of the preceding claims, wherein said controller (CON) is a lead-lag compensator.

25. The method according to any one of the preceding claims, wherein said controller (CON) is configured for training based on said observed ice cream characteristics.

26. The method according to any one of the preceding claims, wherein said controller (CON) comprises adaptive controller parameters (ACP).

27. The method according to any one of the preceding claims, wherein said adaptive controller parameters (ACP) are configured for manual adjustments.

28. The method according to any one of the preceding claims, wherein said automatically analysing is analysed based on machine learning.

29. The method according to any one of the preceding claims, wherein said automatically analysing is analysed based on a machine learning control model.

30. The method according to any one of the preceding claims, wherein said machine learning control model is a reinforcement learning model.

31. The method according to any one of the preceding claims, wherein said machine learning control model is trained at least based on ice cream characteristics.

32. The method according to any one of the preceding claims, wherein said machine learning control model is trained and wherein said training includes reward shaping.

33. The method according to any one of the preceding claims, wherein said machine learning control model is trained and wherein said training of said machine learning control model includes imitation learning.

34. The method according to any one of the preceding claims, wherein said smart freezer (SF) is controlled by a controller comprising a machine learning control model.

35. The method according to any one of the preceding claims, wherein said an additional step of measuring is done prior to defining said desired ice cream quality (DICQ).

36. The method according to any one of the preceding claims, wherein the controller (CON) is configured for running the smart freezer in an assistance mode via the user interface, using input from the user interface in the form of observed ice cream characteristics (ICC) and / or a desired ice cream quality (FICQ) and outputting at least one smart freezer setting (SFS) which the user may set by manually setting the adjustable smart freezer parameters (SFP) on the smart freezer (SF).

37. The method according to any one of the preceding claims, wherein the controller (CON) is configured for running the smart freezer in an automatic mode via the user interface, using input from the user interface in the form observed ice cream characteristics (ICC) and / or a desired ice cream quality (FICQ)) and automaticallyapplying a smart freezer setting (SFS) by automatically setting the adjustable smart freezer parameters (SFP) on the smart freezer (SF).

38. The method according to any one of the preceding claims, wherein said smart freezer setting (SFS) is also depending on the power consumption of the smart freezer.

39. A smart freezer (SF) comprising: one or more sensor(s) (SENS) for observing ice cream characteristics (ICC), a controller (CON) configured for receiving ice cream characteristics (ICC), a freezer and a mixer for freezing and processing ice cream ingredients (ICN), a user interface (UI) for setting a plurality of adjustable smart freezer parameters (SFP), wherein said plurality of adjustable smart freezer parameters (SFP) defines a smart freezer setting (SFS) for said smart freezer (SF).

40. The smart freezer (SF) according to claim 39, wherein said smart freezer (SF) is upstream mechanically connected to a mixer (MIX).

41. The smart freezer (SF) according to any one of the claims 39-40, wherein said smart freezer (SF) is downstream mechanically connected to one or more an ice cream formers (ICF).

42. The smart freezer (SF) according to any one of the claims 39-41, wherein said smart freezer comprises a communication module.

43. The smart freezer (SF) according to any one of the claims 39-42, wherein the smart freezer comprises a user interface (UI) configured for visualizing e.g. by means of a display, the current smart freezer setting (SFS) and / or an updated smart freezer setting in response to a received observed ice cream characteristics (ICC) and / or a desired ice cream quality (FICQ).

44. The smart freezer (SF) according to any one of the claims 39-43, wherein said smart freezer is connected to a server / cloud.

45. The smart freezer (SF) according to any one of the claims 39-44, wherein the controller (CON) is configured for running the smart freezer in an assistance mode via the user interface, using input from the user interface in the form of observed ice cream characteristics (ICC) and / or a desired ice cream quality (FICQ) and outputting at least one smart freezer setting (SFS) which the user may set by manually setting the adjustable smart freezer parameters (SFP) on the smart freezer (SF).

46. The smart freezer (SF) according to any one of the claims 39-45, wherein the controller (CON) is configured for running the smart freezer in an automatic mode via the user interface, using input from the user interface in the form observed ice cream characteristics (ICC) and / or a desired ice cream quality (FICQ)) and automatically applying a smart freezer setting (SFS) by automatically setting the adjustable smart freezer parameters (SFP) on the smart freezer (SF).

47. The smart freezer (SF) according to any one of the claims 39-46, wherein said smart freezer is operated according to any one of the claims 1-38.

Images