Virtual sensing system for monitoring state of container packaging machine

Abstract

A virtual sensing system (1) for monitoring the status of a container packaging machine (2) for packaging containers filled with pourable food products, the system (1) reconstructing a target status monitoring signal based on input data from the container packaging machine (2), and having an input module (6) configured to receive the target status monitoring signal from the container packaging machine (2), the receiving unit is used for receiving input data indicating a target state monitoring signal to be reconstructed from the container packaging machine (2); and an artificial intelligence-AI-module (8) for implementing a machine learning algorithm to generate an output state monitoring signal based on the input data, the output state monitoring signal being a reconstruction of the target state monitoring signal; and an output module (9) for providing the output condition monitoring signal generated by the AI module (8) for further processing thereof by a condition monitoring module (9) intended to evaluate and / or predict the condition of the container packaging machine (2) based on the output condition monitoring signal.

Description

Technical Field

[0001] The invention relates to a virtual sensing system for monitoring the state of a container packaging machine, in particular for producing composite packages filled with pourable food. Background Art

[0002] As is known to all, many liquid, semi-liquid or pourable foods, such as juice, UHT (ultra-high temperature treated) milk, wine, ketchup, etc., are distributed and sold in composite packages made of multi-layer composite packaging materials.

[0003] A typical example is the parallelepiped-shaped package for pourable foods called Tetra BrikAseptic TM , which is made by sealing and folding laminated strip packaging materials. The packaging material has a multi-layer structure, which includes a cardboard and / or paper base layer, covered on both sides with a layer of heat-sealable plastic material, such as polyethylene. For aseptic packaging for long-term storage of products, the packaging material also includes a layer of oxygen barrier material, such as aluminum foil, which is superimposed on a layer of heat-sealable plastic material and covered with another layer of heat-sealable plastic material to form the inner surface of the package that ultimately contacts the food.

[0004] Such composite packages are usually produced in a fully automated packaging line (or plant), which forms the composite package starting from a web of multilayer composite packaging material and fills the composite package with the pourable food product.

[0005] A typical packaging line comprises at least one filling machine, which forms a composite package starting from a multilayer composite packaging material and fills the composite package with the pourable food product. In addition, the packaging line may also include other container packaging machines upstream and / or downstream. The downstream packaging machine may, for example, include one or more buffer units for temporarily buffering the composite package, application units for applying straws or other elements to the composite package, and grouping units, such as a palletizer unit, for grouping a plurality of composite packages together into a storage unit, such as a palletizer.

[0006] It is well known that it is important to monitor the status of container packaging machines and related components in order to evaluate their current operating conditions and possible abnormalities of related parts and components, and to predict the occurrence of failures or damages.

[0007] Condition monitoring is usually achieved by installing suitable monitoring sensors within the packaging machine, acquiring corresponding detection signals and processing the same using suitable evaluation and prediction algorithms, configured to provide an indication of the current and predicted operating condition of the same.

[0008] Specifically, vibration analysis is an important part of this type of condition monitoring. Vibration analysis aims to monitor the vibration signal level and pattern within the packaging machine to detect abnormal vibration events in key parts (such as bearings, gears, servo motors, etc.) and evaluate the overall condition of the same machine. Vibration analysis can, for example, achieve real-time reaction to state changes, remote condition monitoring and predictive maintenance.

[0009] Therefore, vibration sensors (especially accelerometer sensors) are coupled to key parts of the packaging machine to monitor the corresponding operation. For example, the vibration sensor is coupled to the servo motor through the support flange to provide a detection signal indicating the vibration of the same servo motor during operation.

[0010] However, installing and maintaining condition monitoring sensors in packaging machines is costly and time-consuming. In addition, there may be places in the packaging machine where it is difficult or impossible to install monitoring sensors (for example, some parts of the machine may have accessibility issues or may operate in harsh environments that are not suitable for installing monitoring sensors).

[0011] In such cases, conventional monitoring of the operating state of the corresponding parts of the packaging machine may not be possible or at any rate difficult to achieve with the desired reliability and accuracy. Summary of the invention

[0012] An object of the present disclosure is to provide an improved solution to at least partially overcome the above-mentioned problems of known condition monitoring solutions.

[0013] Thus, according to the present disclosure, there is provided a virtual sensing system for monitoring the status of a container packaging machine as defined in the accompanying claims.

[0014] According to a second aspect, there is provided a method for monitoring the status of a container packaging machine as defined in the accompanying claims.

[0015] According to a third aspect, a computer program product is provided, comprising instructions which, when the program is executed by a computing unit, cause the computing unit to perform the method. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Non-limiting embodiments of the present invention will be described by way of example with reference to the accompanying drawings, in which:

[0017] According to the implementation scheme of this solution, Figure 1 A schematic diagram showing a virtual sensing system for condition monitoring of a packaging machine in a packaging line Figure 2

[0018] Figure 2 is with Figure 1 Schematic diagram of the virtual sensing system;

[0019] Figure 3 It is a depiction Figure 1 Figure 1 shows the continuous learning architecture of the artificial intelligence module in the virtual sensing system;

[0020] Figure 4 It is a depiction Figure 1 A diagram of a training pipeline of an artificial intelligence module in a virtual sensing system; and

[0021] Figure 5 yes Figure 1 Schematic block diagram of possible implementations of artificial intelligence modules in a virtual sensing system. DETAILED DESCRIPTION

[0022] Figure 1 A schematic block diagram of a computer-implemented virtual sensing system, generally indicated at 1, is shown, which is used to monitor the operation of a container packaging machine 2, which is configured to produce packaging containers, in particular composite packaging formed from a multi-layer composite packaging material, filled with pourable products, in particular food, such as milk, cream, juice, wine, ketchup, sugar, salt, emulsions, solutions containing solid particles (such as beans), etc.

[0023] The container packaging machine 2 is part of a container packaging line 2', which also comprises a processing machine (not shown here). The container packaging machine 2 is, for example, a filling machine.

[0024] The multilayer composite packaging material may comprise at least one layer of fibrous material, such as a paper or paperboard layer, and at least two layers of heat-sealable plastic material, such as polyethylene, with the fibrous material layers interposed between each other. One of the two layers of heat-sealable plastic material defines the inner surface of the packaging container and ultimately contacts the pourable food packaged in the same packaging container.

[0025] The multilayer composite packaging material may further comprise a layer of gas-barrier and light-barrier material, such as an aluminium foil or ethylene vinyl alcohol (EVOH) film, in particular arranged between one of the heat-sealable plastic material layers and the fiber material layer. Preferably, the multilayer composite packaging material may further comprise a further layer of heat-sealable plastic material, the further layer of heat-sealable plastic material being between the gas-barrier and light-barrier material layer and the fiber material layer.

[0026] Each packaging container can extend along the longitudinal axis, have a longitudinal seam portion (extending along the respective longitudinal axis) and a pair of transverse sealing strips, in particular a transverse top sealing strip and a transverse bottom sealing strip. Specifically, each packaging container can have a substantially parallelepiped structure.

[0027] Furthermore, the packaging container may comprise at least two transverse walls (transverse to the respective longitudinal axis) and arranged at opposite sides of the packaging container, and a plurality of side walls extending between the transverse walls.

[0028] More specifically, a corresponding transverse wall of each packaging container can define a bottom wall, and another corresponding transverse wall can define a top wall. Specifically, the bottom wall can define a support surface suitable for being placed on a (horizontal) plane, such as a shelf in a distribution point, and the top wall is opposite to the bottom wall.

[0029] In a manner not shown in detail, the container packaging machine 2 can comprise at least: - a conveying device for advancing the web of multilayer packaging material along an advancement path; - a tube forming device for forming a tube from an advancing web of multilayer packaging material; - Sealing device for sealing the tube longitudinally; - a filling device, coupled to the inlet line, for filling the pipe 2 and the pipe 3 with the pourable food product - a package forming unit for forming the tube and transversely sealing the tube in order to form a composite package from the tube filled with the pourable food product.

[0030] In more detail, the package forming unit may include at least: - a forming device configured to form, transversely seal and transversely cut the tube to obtain composite packages, in particular pillow-shaped packages and - a final folder configured to receive the composite package from the forming device and to form a packaging container, in particular a pillow-shaped package, from the composite package.

[0031] The container packaging machine 2 further comprises a control device 4 for controlling the packaging operation; the control device 4 comprises, for example, a PLC (Programmable Logic Controller) unit or any suitable processing and computing unit, configured to execute a computer program designed to control the operation of the same container packaging machine 2.

[0032] According to one aspect of the present solution, the virtual sensing system 1 is configured to reconstruct a target signal, in particular a state monitoring target signal, such as a vibration signal, for state monitoring of the container packaging machine 2, by data fusion of other signals, measurements, parameters and / or information (typically represented as input data) obtained from the same environment of the target signal (i.e., from the same container packaging machine 2).

[0033] The virtual sensing system 1 is configured to learn how to interpret the relationships between the input data and combine them so as to be able to reconstruct the target signal with sufficient accuracy to replace a real physical sensor, in particular a vibration sensor (such as an accelerometer sensor), which may therefore not be present in the container packaging machine 2 for the purpose of condition monitoring.

[0034] In more detail, the virtual sensing system 1 is implemented in a processing computing unit 1' and it comprises: an input module 6 configured to receive input data from the container packaging machine 2 indicating (or associated with) a target state monitoring signal to be reconstructed; an artificial intelligence (AI) module 8 configured to implement a suitable neural network algorithm, in particular an algorithm of a recurrent nature, in order to generate an output state monitoring signal based on the input data received from the input module 6, the output state monitoring signal being an accurate approximation of the target state monitoring signal; and an output module 7 configured to provide the output state monitoring signals generated by the AI ​​module 8 for further processing by a state monitoring module 9 intended to evaluate the actual state of the container packaging machine 2 and / or predict the future state of the container packaging machine 2 based on the same output state monitoring signals. That is, the virtual sensing system 1 comprises a state monitoring unit 9.

[0035] The condition monitoring module 9 may be implemented in the same processing and computing unit 1 ′ and configured to implement condition monitoring of the container packaging machine 2 and, in general, of the container packaging line 2 ′. For example, the processing and computing unit 1 ′ may be configured to implement vibration analysis in case the output condition monitoring signal is configured to reconstruct a target vibration signal.

[0036] In a possible embodiment, the virtual sensing system 1 is remote from the container packaging machine 2 and is implemented in a central processing unit 100 (in this case comprising a processing computing unit 1′) located in a remote server (e.g. in the “cloud”), where operational data and information related to the container packaging line 2′ are received and processed to control and manage its operation.

[0037] In another possible embodiment, the virtual sensor system 1 can be implemented locally on the container packaging machine 2 , for example in a control device 4 of the same container packaging machine 2 .

[0038] Virtualization of real monitoring sensors by virtual sensing system 1 has many advantages, including: - Reduce the number of physical sensors on site, thereby reducing installation and maintenance costs; - Virtual sensing at critical points where installing real sensors is very difficult or even impossible, such as in very small components or moving parts, and the only other possible solution is to install expensive wireless sensors; -Improve the anomaly detection of the field fault monitoring solution by collecting data and adjusting the neural network algorithm implemented by AI module 8 according to those specific anomalies, thus being able to detect new anomalies that cannot be detected by current condition monitoring technologies.

[0039] In order to accurately replace the physical monitoring sensors in the container packaging machine 2, the AI ​​module 8 in the virtual sensing system 1 must be properly trained based on a suitable set of input data so that the neural network algorithm can be trained; for example, the weights of the neural network can be adjusted through the training process.

[0040] like Figure 1 As shown, such training is implemented by a training module 10, which may be included in the processing calculation unit 1' and / or the central processing unit 100 discussed above (or may be located outside thereof).

[0041] Training can be achieved with reference to multiple training container packaging machines 2, where (at least initially) physical monitoring sensors are retained for comparing corresponding detection signals with output state monitoring signals generated by the AI ​​module 8, and adjusting the weights (generally parameters) of the neural network implemented in the same AI module 8 accordingly.

[0042] like Figure 2 As shown schematically, during the training or "development" of the AI ​​module 8 of the virtual sensing system 1 (herein represented as a "virtual sensor"), the input module 6 receives both: a target state monitoring signal (in this case, an actual vibration signal provided by a physical, real sensor 15 (in this case, an accelerometer)) and the input data discussed above indicating the target state monitoring signal to be reconstructed.

[0043] In the example shown, the physical monitoring sensor to be removed from the container packaging machine 2 is a vibration sensor coupled to a machine servo motor (e.g., a rotary servo motor configured to drive jaws designed to cut tubes to obtain composite packages), and the input data may include operating signals related to the same servo motor, such as torque and speed signals, which are detected by suitable sensors coupled to the servo motor.

[0044] In order to have a larger input data pool, the first and second derivatives of the same torque and speed signals can also be considered.

[0045] During the development phase, the output state monitoring signal provided by the virtual sensing system 1 is compared with the target state monitoring signal, and then the parameters of the AI ​​module 8 (e.g., the weights of the corresponding neural network) are adjusted based on the comparison (e.g., the difference) between the same output state monitoring signal and the target state monitoring signal.

[0046] After the AI ​​module 8 is properly trained, the virtual sensing system 1 can be deployed for condition monitoring in the field of the container packaging machine 2, where the physical sensor 15 is no longer installed or used (in other words, the container packaging machine 2 has no physical sensor 15 for condition monitoring purposes). The vibration signal is now estimated by the virtual sensing system 1, which provides an output condition monitoring signal generated by the AI ​​module 8.

[0047] It should be noted that the development phase of the AI ​​module 8 can be performed with a selected number of training container packaging machines 2 in mind, for example, a number of machines at the company premises of the respective manufacturer. In this case, the container packaging machines 2 installed at the customer's premises can therefore be free of physical sensors 15 for condition monitoring purposes, at least during the above-mentioned deployment phase.

[0048] As an alternative, a selected number of container packing machines 2 provided to selected customers can also participate in the development of the AI ​​module 8 as training container packing machines.

[0049] According to one aspect of the present solution, the training module 10 implements a continuous learning method for training the AI ​​module 8 of the virtual sensing system 1 .

[0050] Continuous machine learning, also known as continuous or online learning, is a machine learning approach whereby an AI model continuously learns and evolves based on an ever-increasing amount of input data while retaining previously learned knowledge. This approach thus provides the AI ​​model with the ability to autonomously learn and adapt as new and different input data comes in.

[0051] In the context of the present solution, the continuous machine learning applied to the AI ​​module 8 provides for continuous retraining of the neural network, gradually adjusting the weights and parameters of the corresponding neural network algorithm with continuous and repeated training phases based on different input data. This approach allows to continuously improve the performance of the AI ​​module 8 until it is deemed ready for industrial deployment (at which point it will be possible to exploit all the advantages of the virtual sensing solution in the field).

[0052] like Figure 3 Schematically, the continuous learning method is used to collect input data from selected training container packaging machines 2, as shown in step 20. It should be noted that these training container packaging machines 2 can be training machines at the manufacturer and / or field training machines installed at the customer, in which case physical monitoring sensors 15 are installed to collect input data.

[0053] All collected data are received, for example at the central processing unit 100 , and stored in the data memory 19 .

[0054] The input data is then used to train the neural network algorithm in the AI ​​module 8 of the virtual sensing system 1 , in particular, to adjust relevant parameters (eg, weights), as shown in step 21 .

[0055] In this regard, it is to be noted that monitoring algorithms, such as predictive maintenance models, performance monitoring models, quality control models and process control models, which are typically used to monitor the operation and performance of the associated container packaging machine 2 in the same central processing unit 100, can be utilized to provide data labels for the training of the AI ​​module 8, essentially selecting the input data on which the virtual sensing system 1 must be retrained.

[0056] Then, any suitable metric is used to evaluate the performance of the AI ​​module 8 in reconstructing the target state monitoring signal, and the same target state monitoring signal is compared with the output state monitoring signal provided by the virtual sensing system 1, as shown in step 22.

[0057] If the performance is assessed as good or satisfactory (e.g., meeting or exceeding a certain quality threshold), the virtual sensing system 1 is considered ready for deployment in the field, as shown in step 24, so that the same virtual sensing system 1 is used for condition monitoring of an on-site container packaging machine 2 that has no on-board physical or actual condition monitoring sensors (or whose physical sensors are no longer used for this purpose).

[0058] like Figure 3 As shown, the training of the AI ​​module 8 continues, further input data is collected and further parameters of the same AI module 8 are adjusted. This continuous training can, for example, adapt to any possible changes or modifications that occur in the container packaging machine 2.

[0059] Reference now Figure 4 Another aspect of the present solution is discussed, which relates to a specific training pipeline of the AI ​​module 8 of the virtual sensing system 1 , which can be implemented by the training module 10 .

[0060] Specifically, it is first noted that when the AI ​​module 8 is deployed to reconstruct an approximation of a target state monitoring signal, the AI ​​module 8 is configured to implement a regression algorithm in order to output an actual estimate of the same signal.

[0061] However, the applicant has found that training the AI ​​module 8 configured as a regressor may not achieve the best training results.

[0062] Therefore, one aspect of the present solution envisages a training pipeline for the AI ​​module 8, according to which the training is divided into two consecutive epochs: a first training phase 26, in which the AI ​​module 8 is configured as a classifier, thereby assigning an output class (provided as an output of the AI ​​module 8) to the input data among a certain number of classes; and - A second training epoch 28, in which the same AI module 8 is in effect configured as a regressor, providing actual output values ​​for the reconstructed signal.

[0063] In more detail, Figure 4 As shown, in the first training phase 26 of the pipeline, the AI ​​module 8 is configured to implement a classification algorithm (which may be of any suitable nature, as appropriate) and provide a classification result, as shown in step 30 .

[0064] The classification results are then used to adjust the parameters of the AI ​​module 8, for example by measuring the performance of the neural network through a confusion matrix, as schematically shown in step 31, and in particular to set the weights of the neural network, as shown in step 32.

[0065] This constitutes a pre-training phase for the AI ​​module 8, which is followed by a second training epoch 28 of the pipeline, in which the AI ​​module 8 is configured to implement a regression algorithm (again of any suitable nature).

[0066] In this second training period, the AI ​​module 8 is fine-tuned to reconstruct the actual value of the target state monitoring signal with high accuracy, as shown in step 34. Thanks to the parameters (e.g., weights) previously set by the classifier in the first training period, the training of the AI ​​module 8 in this second training period becomes easy and provides more accurate results.

[0067] Specifically, as schematically shown in step 36 , the output state monitoring signal provided by the trained virtual sensing system 1 closely matches the target state monitoring signal.

[0068] According to a possible implementation, the AI ​​module 8 is configured to implement a recursive neural network algorithm, in particular a LSTM (Long Short-Term Memory) algorithm, for reconstructing the target state monitoring signal.

[0069] like Figure 5 As schematically shown in FIG. 8 , the AI ​​module 8 comprises a plurality of (three in the example) input stages 40 , each receiving respective input data (torque and speed signals and their first and second order derivatives in the example).

[0070] The AI ​​module 8 also includes a plurality of (three in the example) neural network units or stages 42, each of which receives respective input data from a corresponding input stage 40 via respective input weighting blocks 43, and also receives the output of the previous unit according to a recursive structure.

[0071] The outputs of the individual neural network units 42 are provided to an attention stage 45 via respective output weighting blocks 44, thereby implementing an appropriate attention algorithm to give less or more importance (i.e., attention) to one or more outputs received from the neural network units 42 in the recursive structure.

[0072] The AI ​​module 8 further comprises an output stage 46 , located after the attention stage 45 , configured to provide a reconstructed value of the output state monitoring signal.

[0073] Applicants have discovered that using a recursive neural network algorithm, and in particular an LSTM (Long Short Term Memory) algorithm, to reconstruct the target state monitoring signal is particularly advantageous and can provide more accurate and reliable results in signal reconstruction.

[0074] From the above description it is clear that the advantages of the solution discussed are present.

[0075] Anyway, once again, the discussed solutions can: reduce the number of real sensors deployed in the field, thereby reducing the associated installation and maintenance costs; perform virtual sensing at critical points where installation of real sensors is very difficult or not feasible; improve anomaly detection for field fault monitoring.

[0076] In particular, training and developing the AI ​​modules in a virtual sensing system using the discussed continual learning approach allows to continuously improve the performance of signal reconstruction thanks to a progressively larger pool of input data until the same virtual sensing system is ready for deployment in an industrial setting.

[0077] The same continuous learning approach, with repeated training over time, can adapt to changes in the container packaging machine and its operations, such as changes in materials, production parameters, etc.

[0078] The discussed training pipeline has two distinct and consecutive training periods, where the AI ​​module is first trained as a classifier and then as a regressor, which is shown to be beneficial for improving the performance of signal reconstruction.

[0079] In addition, the use of recurrent neural network structures can further improve the performance of signal reconstruction in virtual sensing systems.

[0080] Obviously, modifications may be made to what is described herein without departing from the scope of protection defined in the accompanying claims.

[0081] In particular, it is to be emphasized that the AI ​​module 8 in the virtual sensing system 1 can implement different types of neural network algorithms suitable for reconstructing target state monitoring signals.

[0082] The virtual sensing system 1 can also reconstruct other types of target state monitoring signals (in addition to the above-mentioned vibration signals); in addition, the input data can include other signals and / or other order derivatives of the same signal.

[0083] It is also to be emphasized that the discussed solution can be applied to any packaging line 2', including any packaging machine 2, and to any type of pourable food product.

[0084] The systems and methods disclosed herein can be implemented as software, firmware, hardware, or a combination thereof. In hardware implementation, the task division between the functional units or modules mentioned in the above description does not necessarily correspond to the division of physical units; on the contrary, a physical module can perform multiple functions, and a task can be performed by multiple physical modules in collaboration.

[0085] Some modules or all modules can be implemented as software executed by a digital signal processor or microprocessor, or as hardware or an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which may include a computer storage medium (or non-temporary medium) and a communication medium (or temporary medium). As is well known to those skilled in the art, the term "computer storage medium" includes volatile and non-volatile, removable and non-removable media for storing information (such as computer-readable instructions, data structures, program modules or other data) implemented in any method or technology. Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other storage technology, CD-ROM, digital versatile disc (DVD) or other optical disk storage, cassette, magnetic tape, disk storage or other magnetic storage device, or any other medium that can be used to store the desired information and can be accessed by a computer.

[0086] The target condition monitoring signal may be a vibration signal. Advantageously, the system may replace a physical vibration sensor which may not be present in the container packaging machine (2) for condition monitoring due to harsh environments or physical constraints.

[0087] Specifically, the target monitoring signal may be a vibration signal associated with a servo motor in the container packaging machine 2. It will be appreciated that the AI ​​module may be configured to generate a plurality of output state monitoring signals that are reconstructions of a plurality of target state monitoring signals corresponding to vibration signals associated with a plurality of servo motors.

[0088] The input data may include a plurality of operating signals associated with components of the packaging machine 2 that cause vibrations, such as servomotors. Such operating signals may include torque and / or speed signals, and optionally corresponding derivatives.

[0089] The machine learning algorithm may be trained based on a training data set that includes reference input data (indicating a reference target state monitoring signal to be reconstructed) and a reference target state monitoring signal. The input data may be collected simultaneously with respect to the reference target state monitoring signal. That is, the acquired input data may indicate the acquired target state monitoring signal.

[0090] The container packaging machine 2 may be operated during a training period to collect reference input data and reference target state monitoring signals.The container packaging machine 2 may be the same packaging machine 2 and / or a different packaging machine 2 previously disclosed.

[0091] One or more embodiments may relate to a method for monitoring the state of a container packaging machine 2, which is used to package containers filled with pourable food, the method comprising reconstructing a target state monitoring signal based on input data from the container packaging machine 2, such as a vibration signal associated with a servo motor in the container packaging machine, such as an operating signal associated with the servo motor in the container packaging machine 2, preferably, the operating signal includes torque and speed signals, and more preferably, the operating signal includes corresponding derivatives of the torque and speed signals.

[0092] The reconstruction steps include: - receiving input data from the container packaging machine 2 indicating a target state monitoring signal to be reconstructed and - implementing a machine learning algorithm to generate an output state monitoring signal based on the input data, the output state monitoring signal being a reconstruction of the target state monitoring signal and - providing output status monitoring signals generated by the AI ​​module 8, - processing the output status monitoring signal to evaluate and / or predict the status of the container packaging machine 2 based on the output status monitoring signal.

[0093] The method may include utilizing an algorithm designed to monitor the operation and / or performance of the training container packaging machine 2 based on input data, including one or more predictive maintenance models, performance monitoring models, quality control models, and process control models to provide data labels for training the AI ​​module 8.

[0094] One or more embodiments may involve a method of training the AI ​​module 8 and the previously described machine learning algorithm, the method comprising: - obtaining input data from a container packaging machine 2, for example a reference packaging machine 2 or the same packaging machine previously disclosed as being used during normal operation, - Obtaining a state monitoring signal from the physical state monitoring sensor 15 located at the container packaging machine 2 -Training AI module 8 and machine learning algorithms based on input data and condition monitoring signals.

[0095] The method may include implementing a training pipeline for training the AI ​​module 8, according to which the training is divided into two consecutive epochs: - a first training period, in which the AI ​​module 8 is configured as a classifier; and - A second training period, in which the AI ​​module 8 is configured as a regressor, providing actual output values ​​for the reconstructed signal.

[0096] The method may include: - During the first training period of the pipeline, the classification algorithm is implemented, providing classification results for adjusting the parameters of the machine learning algorithm during the pre-training phase of the AI ​​module 8, -During the second training epoch of the pipeline, a regression algorithm is implemented to fine-tune the parameters of the machine learning algorithm.

[0097] One or more embodiments may relate to a method of verifying the AI ​​module 8 and the machine learning algorithm as described above, the method comprising evaluating the performance of the machine learning algorithm by checking whether the state monitoring signal from the state monitoring sensor substantially corresponds to the output state monitoring signal generated by the AI ​​module 8.

[0098] During validation of the machine learning algorithm, the acquired physical signal (i.e., the condition monitoring signal from the condition monitoring sensor) can be compared with the output condition monitoring signal generated by the AI ​​module. If the output condition monitoring signal differs by more than a predetermined threshold within a predetermined time period, the machine learning algorithm can be retrained.

[0099] One or more embodiments may also relate to a computer program product comprising instructions, which, when executed by a computing unit, causes the computing unit to perform at least one of the following: a method for monitoring the state of a container packaging machine, a method for training an AI module 8 and a machine learning algorithm, and / or a method for validating an AI module 8 and a machine learning algorithm.

Claims

1. A virtual sensing system (1) for monitoring the state of a container packaging machine (2) for packaging containers filled with pourable food, the system (1) being configured to reconstruct a target state monitoring signal based on input data from the container packaging machine (2), and comprising: an input module (6) configured to receive the input data from the container packaging machine (2), the input data being indicative of the target state monitoring signal to be reconstructed; an artificial intelligence - AI - module (8) configured to implement a machine learning algorithm to generate an output state monitoring signal based on said input data, the output state monitoring signal being a reconstruction of said target state monitoring signal; an output module (7) configured to provide the output state monitoring signal generated by the AI ​​module (8), and A state monitoring module (9) is intended to evaluate and / or predict the state of the container packaging machine (2) based on the output state monitoring signal. 2 . The system of claim 1 , wherein the target state monitoring signal is a vibration signal.

3. A system according to claim 1 or claim 2, wherein the machine learning algorithm has been trained based on a training data set, which includes reference input data indicating a reference target state monitoring signal to be reconstructed and the reference target state monitoring signal.

4. The system according to any one of claims 1 to 3, further comprising a training module (10) operably coupled to the AI ​​module (8) and configured to train the AI ​​module (8) based on a continuous learning method.

5. The system according to any of the preceding claims, wherein the AI ​​module (8) is a neural network comprising: A plurality of neural network units (42), each neural network unit receiving respective input data and arranging them according to a recursive structure; and an attention stage (45) which implements an attention algorithm on the output received from the neural network unit (42) to provide a reconstructed value of the output state monitoring signal.

6. A system according to any one of the preceding claims, wherein: - the target monitoring signal is a vibration signal associated with a servo motor in the container packaging machine (2), and / or - the input data comprises a plurality of operating signals associated with a plurality of servomotors in the container packaging machine (2), preferably, the plurality of operating signals comprises torque and / or speed signals, even more preferably, the plurality of operating signals comprises respective derivatives of the torque and speed signals.

7. System according to any of the preceding claims, wherein the AI ​​module (8) is contained in a central processing unit (100) located in a remote server, which is remote from the container packaging machine (2).

8. A packaging line (2') comprising a virtual sensing system (1) according to any one of the preceding claims.

9. A method for monitoring the state of a container packaging machine (2), the container packaging machine (2) being used for packaging containers filled with pourable food, the method comprising reconstructing a target state monitoring signal based on input data from the container packaging machine (2), wherein the reconstruction comprises: receiving said input data from said container packaging machine (2) indicating said target state monitoring signal to be reconstructed; as well as implementing a machine learning algorithm to generate an output state monitoring signal based on the input data, the output state monitoring signal being a reconstruction of the target state monitoring signal; as well as providing the output state monitoring signal generated by the AI ​​module (8), The output status monitoring signal is processed to evaluate and / or predict the status of the container packaging machine (2) based on the output status monitoring signal.

10. The method according to claim 9, comprising: - obtaining input data from said container packaging machine (2), - obtaining a state monitoring signal from a physical state monitoring sensor (15) located at the container packaging machine (2); and - Training the AI ​​module (8) based on the input data and the state monitoring signal.

11. The method according to claim 9 or claim 10, comprising implementing a training pipeline for training the AI ​​module (8), according to which training is divided into two consecutive epochs: a first training period, wherein the AI ​​module (8) is configured as a classifier; and In the second training phase, the AI ​​module (8) is configured as a regressor to provide actual output values ​​for the reconstructed signal.

12. The method according to claim 11, comprising: - implementing a classification algorithm during the first training phase of the pipeline, providing classification results for adjusting the parameters of the machine learning algorithm during the pre-training phase of the AI ​​module (8), - During the second training phase of the pipeline, implementing a regression algorithm to fine-tune the parameters of the machine learning algorithm.

13. The method according to any one of claims 9 to 12 comprises evaluating the performance of the machine learning algorithm by checking whether the state monitoring signal from the state monitoring sensor substantially corresponds to the output state monitoring signal generated by the AI ​​module (8).

14. The method according to any one of claims 9 to 13, comprising utilizing an algorithm designed to monitor the operation and / or performance of the training container packaging machine (2) based on the input data, including one or more of a predictive maintenance model, a performance monitoring model, a quality control model and a process control model, to provide data labels for the training of the AI ​​module (8).

15. A computer program product comprising instructions which, when executed by a computing unit, cause the computing unit to perform the method according to any one of claims 9 to 14.

Images