Method for predictive maintenance of automatic machines for manufacturing or packaging consumer goods

Abstract

A method of predictive maintenance of an automatic machine (1) for manufacturing or packaging consumer goods, comprising the steps of: detecting and recording, by means of at least one respective local control unit (11), at least one sample sequence (SS) related to at least one motorized measure (MM) of at least one electric actuator (4); transmitting the recorded sample sequence (SS) to a data processing unit (5); defining, on the basis of the detected at least one sample sequence (SS) and at least with respect to the detected motorized measure (MM), at least one multidimensional tolerance range (TH) within an anomaly matrix (AM) having at least two statistical features (STF) as dimensions; calculating the two statistical features (STF) so as to define a position of an actual condition (AC) within the anomaly matrix (AM); determining the urgency of the necessary maintenance on the basis of the position of the actual condition (AC) in the anomaly matrix (AM) and of the multidimensional tolerance range (TH).

Description

[0001] Cross-reference to related applications

[0002] This patent application claims priority to Italian Patent Application No. 102020000014944, filed on June 23, 2020, the entire contents of which are incorporated herein by reference. Technical Field

[0003] This invention relates to a method for predictive maintenance of automated machines used in the manufacture or packaging of consumer products.

[0004] The present invention finds an advantageous but non-exclusive application in predictive maintenance of automated packaging machines for manufacturing cigarette boxes, and the following disclosure will explicitly relate to this application without loss of generality. Background Technology

[0005] In manufacturing plants that process consumer goods, various systems for predictive maintenance have recently been proposed, namely, systems that can determine in advance when maintenance interventions (such as adjustments, cleaning, or replacement of parts) are needed for automated machines.

[0006] By predicting in advance when maintenance interventions will be required, these interventions can be programmed in a coordinated and reasonable manner; in this way, maintenance interventions, machine downtime, and the amount of discarded products are optimized, i.e., reduced to a minimum.

[0007] Typically, these predictions are used in systems that are particularly expensive in terms of time and design. In particular, during the construction of automated machines, special sensors are usually installed on the most risky machine parts based on experience or after advanced simulations.

[0008] In some cases, particularly for detecting wear on components, these sensors are embedded inside the components and generate warnings or alarms once the wear becomes noticeable. In other cases, accelerometers, cameras, and / or thermometers are mounted near the component to be monitored in order to detect any excessive changes in a single analyzed feature.

[0009] However, the aforementioned changes in local variables may depend on a variety of factors that are not always detectable by appropriate sensors. For example, rapid blade wear may be due to dirt in the cutting area, loose screws, vibration, overheating of the surrounding area, changes in the angle of cutting or material entry, or a combination of these characteristics.

[0010] In traditional systems, which merely focus on detecting local one-dimensional features and defining the current status of the component with respect to a reference value (usually a scalar) on the basis of the latter, the increased risk of component failure can be overlooked due to a combination of more than one factor. According to some of these systems, the oscillations of a predetermined signal are compared with the oscillations detected by the appropriate sensors, a one-dimensional threshold (at one or both endpoints) is established, beyond which a warning of the necessary maintenance is generated.

[0011] Due to the huge amount of data to be managed and transmitted in real time, some known systems are generally unable to perform high-frequency sampling. In other known systems, the problem has been attempted to be solved by locally averaging the values detected at high frequency and sending only the average values to a central data processing unit, which significantly reduces the amount of data to be managed, but also reduces the accuracy of the data, since the central processing unit does not take into account the individual values and therefore cannot take into account any peak values that can indicate the approach of a failure.

[0012] Furthermore, it often happens that, by comparing the oscillations of a predetermined reference signal with the oscillations of the current signal detected by the respective sensors, it is not possible to effectively perform maintenance prediction, since the current signal can have oscillations indicative of a failure, but not exceeding the upper threshold or the lower threshold set starting from the reference signal.

[0013] The presence of all the sensors necessary to perform effective and efficient predictive maintenance determines a huge increase in the manufacturing costs of the automatic machines; furthermore, the use of said sensors does not allow to predict some failures caused by the synergy of several factors.

[0014] Patent US 5852351 describes a local unit for acquiring data from the sensors of a machine for the predictive maintenance of the machine itself. The local acquisition unit is installed on the machine to detect the signals from the sensors and periodically stores the values of said signals in a memory. At predetermined times, an operator equipped with a portable electronic device (for example a portable computer) approaches the machine to transfer (preferably by infrared transfer) the contents of the memory of the local acquisition unit to the memory of the portable electronic device. The acquisition method described in patent US 5852351 is simple and cheap to implement, but on the other hand has a high management cost, since it often requires the intervention of an operator who reads the data stored in the memory of the local acquisition unit; furthermore, if the reading of the data stored in the memory of the local acquisition unit is not performed at a high temporal frequency, the predictive maintenance system cannot predict in good margin when it is necessary to perform a maintenance intervention.

[0015] Patent application US 2003046382 describes a method for the remote diagnosis of automatic machines, according to which a local acquisition and control unit is coupled to the automatic machine, which is connected to a series of sensors arranged on the automatic machine. The local acquisition and control unit periodically reads the signals provided by the sensors and compares them with a model of the automatic machine stored in the local acquisition and control unit; if the local acquisition and control unit detects a significant anomaly between the signals provided by the sensors and the model of the automatic machine, the local acquisition and control unit transmits information relating to the anomaly to a remote diagnosis system, which formulates a diagnosis of the anomaly and then sends a request for technical intervention to a service centre, which can perform maintenance operations on the automatic machine. According to the preferred embodiment described in patent application US 2003046382, the remote diagnosis system comprises a first remote diagnosis station (computer or network of computers) to formulate a first diagnosis, a further second remote diagnosis station (computer or network of computers) to formulate a second diagnosis if the first remote diagnosis station is unable to formulate a diagnosis, and a team of technicians to formulate a third diagnosis if even the second remote diagnosis station is unable to formulate a diagnosis. SUMMARY

[0016] The aim of the present application is to provide a method for the predictive maintenance of automatic machines for the manufacture or packaging of consumer goods, which is at least partially free of the aforementioned drawbacks and at the same time is simple and inexpensive to implement.

[0017] According to the present application, a method for the predictive maintenance of automatic machines for the manufacture or packaging of consumer goods is provided, according to what is claimed in the attached claims. An automatic machine for the manufacture or packaging of consumer goods configured to perform the method described above is also provided.

[0018] The claims describe preferred embodiments of the present application, which form an integral part of the present description. BRIEF DESCRIPTION OF DRAWINGS

[0019] The present application will now be described with reference to the accompanying drawings, which illustrate some non-limiting embodiments thereof, in which:

[0020] Figure 1 is a perspective schematic view of an automatic machine for the manufacture of tobacco industry products;

[0021] Figure 2 shows an anomaly matrix with two statistical features as dimensions as a function of the degree of motorization;

[0022] Figure 3 shows possible diagrams relating to the general steps of the method and how they are connected to each other;

[0023] Figure 4 is a graph showing the comparison between the correct measure and the measure determining the necessary subsequent maintenance warning; and

[0024] Figure 5 is a graph showing the comparison between a series of statistical features in a suitable configuration and the same series of statistical features in a configuration that is unsuitable, resulting in a subsequent maintenance warning. DETAILED DESCRIPTION

[0025] Figure 1 An automatic machine 1 for manufacturing tobacco industry products is shown, in particular an automatic packaging machine 1 for applying transparent overwraps to packets of cigarettes.

[0026] The automatic machine 1 comprises various elements designed to work on articles (in the illustrated embodiment, packets of cigarettes 2). Figure 1 In particular, the automatic machine 1 comprises one or more electric drives 3 configured to control at least one electric actuator 4.

[0027] According to some preferred but not limiting embodiments, the electric actuator 4 comprises an electric motor, in particular of the brushless type. According to other not illustrated embodiments, the actuator 4 also comprises a drive type different from an electric motor (for example an electric actuation cylinder, etc.).

[0028] In some non-limiting cases, the electric drives 3 are grouped in a dedicated area of the automatic machine 1 (for example a general or dedicated electrical panel). Alternatively or additionally, some electric drives 3 are provided at the respective electric actuators 4. For example, in the case of electric motors, the respective drives can also be provided on the stator of the electric motor itself. In other words, in some non-limiting cases, the electric drives 3 are provided on a machine control cabinet (which can be the same as or different from the control cabinet in which the data processing unit 5 is also located). Alternatively or additionally, some electric drives 3 can be provided on the respective electric actuators 4 to which they are connected.

[0029] In particular, the electric drives 3 are also configured to periodically detect and record (for example in a local storage unit within each electric drive 3) a sample sequence SS (for example Figure 4 each point of the graph shown) related to at least one motorized measure MM of the at least one electric actuator 4 at a sampling frequency SF. In some non-limiting cases, the motorized measure MM comprises (is) the speed error of the electric actuator 4. In other non-limiting cases, for example Figure 4In the case shown, the motorization measure MM comprises (is) the torque (or the required current) error of the electric actuator. In other non-limiting cases not shown, the motorization measure MM comprises any difference between a reference value and an actual value (related to the electric actuator 4 and detected by the corresponding electric driver 3). Obviously, the same electric driver 3 can detect and record different measures related to the same electric actuator 4, and / or different electric drivers 3 can record mutually different measures related to different electric actuators 4.

[0030] Moreover, the automatic machine 1 comprises a data processing unit 5 (in particular a processor or a dedicated industrial PC) configured to periodically receive the sequence of samples SS previously detected at the sampling frequency SF at a transmission frequency TF equal to or lower than the sampling frequency SF.

[0031] Moreover, as Figure 1 shown in non-limiting embodiments, the automatic machine 1 comprises a local storage unit 6 configured to contain (i.e. store internally for reading and / or writing) the anomaly matrix AM (for example as shown in Figure 2 Particularly, the anomaly matrix AM has at least two statistical features STF based on the detected motorization measures MM as dimensions. More accurately, the storage unit 6 comprises an area dedicated to a database DB used by the data processing unit 5 for processing and updating the model of the automatic machine 1.

[0032] The term "statistical feature STF" refers to all functions (functionality) applicable to a set of data, which can be defined and calculated by statistical analysis, in particular any scalar value that can be defined by performing statistical operations on the sequence of samples SS related to (at least) the motorization measure MM. Examples of statistical features can be: mean, median, mode, shape coefficient and shape index (kurtosis and skewness). Examples of signal measures can be: margin (or clearance) coefficient, peak coefficient, pulse factor, peak value, root mean square or RMS value, signal-to-noise and distortion ratio SINAD, signal-to-noise ratio SNR, standard deviation STD, total harmonic distortion THD.

[0033] Advantageously but not necessarily, the automatic machine 1 comprises at least one local acquisition unit 7 connected to (or determined) a node of a bidirectional, digital and local industrial network (for example of the I / O Link ® type). In Figure 1 non-limiting embodiments, in order to allow high-speed and high-quality data transmission, the industrial network is a local wired network (i.e. with a cable connection) on the automatic machine 1.

[0034] In Figure 1In a non-limiting embodiment, multiple local acquisition units 7 with different characteristics are provided. Specifically, the acquisition unit 7 can be any type of sensor configured to detect values ​​(preferably analog values) of local state metrics (LSMs) such as temperature and vibration. The local acquisition unit 7 is also configured to transmit the detected local state metrics (LSMs) to the data processing unit 5.

[0035] Advantageously, but not necessarily, the local acquisition units 7 are each set on different mechanical groups 10 installed on the automated machine 1. In this way, the status of each mechanical group can be monitored, and manufacturing of only one part of the machine associated with the group 10 to be maintained can be stopped.

[0036] In detail, at least one local acquisition unit 7 includes smart tags and / or IoT (Internet of Things) sensors. In this way, the data processing unit 5 can be notified of the status of a single mechanical group 10 (including a mobile mechanical group, such as a group of units moving on a direct drive system) identified by information sent from a corresponding smart tag, or by an IoT sensor mounted on the group 10, or from a single component of the automated machine 1.

[0037] Advantageously but not necessarily, the automatic machine 1 also includes a communication interface 8 ( Figure 1 The communication interface 8 is configured to connect to the data processing unit 5 and allow the maintenance program 9 to be transferred to maintenance resources, such as... Figure 1 The operator O (or maintenance robot) is shown. In Figure 1 In a non-limiting embodiment, communication interface 8 is a (tactile) screen configured to alert operator O about an upcoming maintenance operation. After the maintenance procedure 9 is transmitted to the appropriate maintenance resource, the resource (or...) Figure 1 In the case of the operator O), maintenance operations are performed in the order and time phases indicated in maintenance procedure 9.

[0038] According to another aspect of the present invention, a method for predictive maintenance of an automated machine 1 for manufacturing or packaging consumer products is provided.

[0039] The method includes the following steps: periodically detecting and recording at least one sampling sequence SS associated with the motorization metric MM of at least one electric actuator 4 by means of a corresponding local control unit 11 at a sampling frequency SF. Specifically, the local control unit 11 includes at least one electric drive 3 or a local acquisition unit 7, the at least one electric drive 3 being configured to drive at least one motor of the automated machine 1, and the local acquisition unit 7 being configured to periodically acquire the sampling sequence SS (i.e., values) of the local state metric LSM and periodically transmit them to the data processing unit 5.

[0040] Advantageously, the method also comprises the step of periodically transmitting the sequence of samples SS recorded at the data processing unit 5 at a transmission frequency TF equal to or lower than, preferably lower than, the sampling frequency SF. In detail, the sampling frequency SF is a particularly high frequency compared to the transmission frequency TS, since the accuracy of the detection also depends on the sampling speed, which is precisely defined by the frequency SF. On the other hand, the transmission frequency TF determines the speed with which the data processing unit 5 can update the database DB and therefore the model of the automatic machine 1.

[0041] Advantageously, but not necessarily, the sampling frequency SF is greater than or equal to 2 kHz (i.e. the corresponding sampling time is less than or equal to 500 microseconds), equal to or greater than 4 kHz (i.e. the sampling time is less than or equal to 250 microseconds). In this way, it is possible to carry out a dense sampling, thus greatly reducing the risk of losing certain information indicative of future anomalies, thus reducing the need for maintenance.

[0042] In particular, the sampling frequency SF corresponds to the so-called cycle time of the local control unit 11, i.e. the refresh time of the sensors in the case of local acquisition units or the closing time of the speed loop by means of the electric drive 3.

[0043] Advantageously, but not necessarily, the transmission frequency TF is lower than or equal to 0.2 Hz (i.e. the time between one transmission and the next of the sequence of samples SS is greater than or equal to 5 seconds), in particular less than or equal to 0.1 Hz (i.e. the time between one transmission and the next of the sequence of samples SS is greater than or equal to 10 seconds), more particularly less than or equal to 0.067 Hz (i.e. the transmission time is greater than or equal to 15 seconds). In this way, it is possible to avoid sending all the detected data in real time to the data processing unit 5 continuously and therefore to reduce the continuous traffic of information, since the same data (sequence of samples SS) are sent in packets.

[0044] Advantageously, but not necessarily, during the recording of the motorized measures MM and / or of the local state measures LSM, the plurality of local control units 11 (i.e. the electric drive 3 and the local acquisition units 7) receive from the data processing unit 5, at a synchronization frequency SCF, a synchronization signal MS to be included in the recording of the sequence of samples SS. In particular, the synchronization signal MS is included for every “n” recorded sequence of samples SS. More precisely, the synchronization frequency SCF is lower than the sampling frequency SF, but higher than the transmission frequency TF.

[0045] In particular, the synchronization frequency SCF corresponds to the so-called cycle time of the data processing unit 5. In detail, the data processing unit 5 is a PLC or an industrial PC and the synchronization frequency SCF is greater than or equal to 200 Hz, in particular greater than or equal to 500 Hz, more particularly greater than or equal to 1 kHz (kilohertz).

[0046] Advantageously but not necessarily, the synchronization signal MS is an analog signal (i.e. not digital, with the possibility of assuming a plurality of different values). In this way, it is possible to synchronize each sequence of samples SS even after transmission (in the data group, given that the transmission frequency TF is significantly lower than the sampling frequency SF). In other words, knowing the digital (analog) value of the synchronization signal MS and the instant at which the transmission takes place, it is possible to re-phase the sequences of samples SS over time, despite the fact that they are transmitted in blocks (groups).

[0047] According to some non-limiting embodiments, the synchronization signal MS (for example from the PLC - unit 5 - to the driver 3) is the position of the physical or virtual main shaft of the automatic machine 1. In particular, the instantaneous value of the so-called sawtooth wave of the (virtual) main shaft of the automatic machine 1 is considered the synchronization signal MS. In this way, the position of the main shaft is used as a reference for the re-phasing over time of the sequences of samples SS transmitted from the local control unit 11 to the data processing unit 5. Since, thanks to the re-phasing by means of the synchronization signal MS, the amount of data to be sent is greatly reduced, instead of sending the data and the corresponding instant of recording (as occurs in the systems of the prior art), it is only the value of the sample SS that is sent, as well as the value of the main shaft position for the next synchronization of the sequence of samples SS sent for all "n" samples.

[0048] In other non-limiting cases, the synchronization signal MS is a suitable counter (incremental or decremental) which is used as a main reference according to what described in the foregoing.

[0049] Advantageously but not necessarily, the method comprises the further step of using the synchronization signal MS as a reference to synchronize the samples SS sent to the data processing unit 5 to know which sample SS corresponds to a given instant or a given phase of the automatic machine 1. In particular, the data processing unit 5 pre-processes the sequences of samples SS sent by synchronizing each sequence over time.

[0050] In particular, the method also comprises the further step of defining a multidimensional tolerance range TH (in particular by training a model by means of an unsupervised classifier, as described below) within at least the anomaly matrix AM (based on at least one detected sequence of samples SS and involving at least the detected motorized measure MM (and / or the local state measure LSM), the anomaly matrix AM having at least two statistical features STF as dimensions (for example selected from the group formed by those described previously). In other words, the statistical features STF defining the dimensions of the anomaly matrix AM are calculated as a function of the detected motorized measure MM. Figure 2

[0051] ​Advantageously but not necessarily, in particular, the sequence of recorded samples SS involves, in addition to the motorization measure MM, a local state measure LSM which involves the condition of one or more mechanical groups 10 (including at least one element), in particular detecting the value of the local state measure by means of at least one local acquisition unit 7, connected to a node of a bidirectional, digital and local, point-to-point and wired (or wireless) industrial network.

[0052] In particular, the local state measure LSM comprises vibrations detected more accurately in multiple dimensions, and / or temperatures and / or accelerations.

[0053] In Figure 2 In a non-limiting embodiment, the anomaly matrix AM comprises two dimensions defined by two respective statistical features STF and STF' calculated with respect to the motorization measure MM (also with the local state measure LSM); in particular, the abscissa represents a statistical feature STF (function of the motorization measure MM) called kurtosis, while the ordinate represents a statistical feature STF' (also function of the same motorization measure MM). In this non-limiting case, the motorization measure MM is the torque error.

[0054] According to some preferred but non-limiting embodiments, the motorization measure MM is the speed error of an electric motor (for example brushless) and is detected in particular by means of a respective driver. In detail, it is surprising that by using this motorization measure MM it is possible to more easily detect anomalies in the behavior of the electric motor. In particular, it is found that using the speed error as motorization measure MM allows to highlight behaviors caused by friction. In particular cases, the variation of friction in the kinematic movement allows to improve the assessment of the wear of the components of the automatic machine 1, improving the estimate of the predictive maintenance.

[0055] Advantageously, the method comprises a further step of: for each detected sampling sequence SS, calculating at least two statistical features STF (to define at least one multidimensional matrix) in order to define the position of the actual condition AC within the anomaly matrix AM.

[0056] In some non-limiting cases, the condition AC corresponds to a single sample SS. In particular, a cloud of consecutive actual conditions AC is defined for the sampling sequence SS.

[0057] In other non-limiting cases, the position of the actual condition AC is calculated from a plurality of samples SS. In other non-limiting cases, the position of the actual condition AC within the anomaly matrix AM is determined from the entire sampling sequence SS detected between one transmission and another between the local control unit 11 and the data processing unit 5.

[0058] Advantageously but not necessarily, as Figure 2In non-limiting embodiments of the application, the multidimensional tolerance range TH, TH', TH" is defined via an unsupervised classifier, in particular a K-means algorithm.

[0059] In Figure 2 In non-limiting embodiments of the application, the unsupervised classifier used for computing (defining) the tolerance range TH, TH', TH" is the so-called K-means algorithm for partitioning analysis of groups. In particular, using this algorithm, it is possible to first compute the center C, C', C" of the groups, i.e. of the sampling sequences SS received by the data processing unit, and subsequently, based on the distribution of the actual condition AC, i.e. of the samples SS, determine the range TH, TH', TH". In detail, in the center portion, three repetitions of the above method are shown, which involve the correct operating condition determined according to three different (consecutive) sampling sequences SS. Figure 2

[0060] In addition, the method comprises the step of determining the urgency of the necessary maintenance based on the position of the actual condition AC in the anomaly matrix AM and the multidimensional tolerance range TH, in particular by verifying the presence of a dangerous condition DC in the proximity of or outside the tolerance range TH. Figure 2

[0061] According to non-limiting embodiments of the application, the tolerance range TH, TH', TH" is configured to have a non-linear shape, in particular elliptical or circular. Figure 2

[0062] In some non-illustrated non-limiting cases, the tolerance range TH has a different (complex) shape based on the type of anomaly to be detected.

[0063] According to some non-illustrated non-limiting embodiments, the metric MM, LSM used for computing the statistical features STF, STF' varies depending on the anomaly to be detected.

[0064] Advantageously but not necessarily, the tolerance range TH is periodically updated (see the presence of ranges TH' and TH" in Figure 2 including the values of the most recent sampling sequences SS detected.

[0065] In some non-limiting cases, the tolerance range TH is updated based only on the values of the most recent sampling sequences SS detected.

[0066] In other non-limiting cases, the tolerance range TH is updated based on the values of the most recent sampling sequences SS detected and the values of some (or all) previous sampling sequences SS detected.

[0067] ​​​According to some preferred non-limiting embodiments, the method comprises the further step of training a model of the automatic machine by means of an unsupervised classifier, in particular a K-means algorithm, using as input a plurality of statistical features STF, STF' (for example some of the statistical features listed above) generated by known faults.

[0068] According to some non-limiting cases, the anomaly matrix MA comprises a plurality of groups GR, each group corresponding to the state of a different mechanical element of the automatic machine 1 or of mechanical elements (or groups 10) having similar structural characteristics.

[0069] In Figure 2 non-limiting embodiments, the groups GR are shown as being processed during possible known anomalous conditions and are simulated or empirically tested to understand how the statistical features STF, STF' (or some of the statistical features listed previously) determine the deviation on the anomaly matrix AM of the actual condition AC. In particular, the anomalies F1, F2 and F3 are determined by varying (increasing / decreasing) the motion friction in a specific mechanical group 10 and by calculating the statistical features STF, STF' based on the torque error (metric MM) detected by the respective driver. On the other hand, the anomaly F4 is generated by simulating an increase in the play in the same mechanical group 10. In these first four anomalies, the variation in terms of features STF (in this case kurtosis) is evident. Furthermore, the anomalies F5 and F6 are generated by simulating a known torque disturbance on the mechanical group 10 described above from the outside. Furthermore, the anomalies F7 and F8 indicate the weighting of the group 10 with different masses. Finally, the cloud HS of the actual condition AC indicates the simulation of correct operation neglecting (from the virtual laboratory) environmental conditions such as humidity, temperature, some friction, etc. These conditions and all the other known potential anomalies can be used to improve the model of the automatic machine 1 and define a plurality of anomaly matrices AM as a function of different statistical features STF (for example some of the statistical features listed previously) in order to effectively detect different types of possible anomalies.

[0070] According to other non-limiting cases, or in addition, for each mechanical element, or for each group, a specific anomaly matrix AM is defined, with the statistical features STF as dimensions, which best detects the deviation from the expected value of the specific element or group 10.

[0071] Advantageously, but not necessarily, the model of the automatic machine 1 is periodically updated to include the most recent sampling sequences SS detected. In particular, the model is also updated in the case of unexpected anomalies (or unexpected faults), defining a fault area DA (for example a cloud) on the anomaly matrix AM of the actual condition AC. Figure 1 .

[0072] Advantageously but not necessarily, the method comprises the further step of calculating the speed at which the successive actual conditions AC move within the anomaly matrix AM, in particular the speed at which the most recent actual condition moves towards the tolerance range TH. The higher said speed, the sooner preventive maintenance needs to be performed.

[0073] Advantageously but not necessarily, the method comprises the further step of periodically scheduling maintenance procedures 9 based on the position or speed of the most recent actual condition AC within the anomaly matrix AM. In particular, the maintenance procedures 9 are transmitted to the maintenance resource (operator O) through the communication interface 8 (which can be a mobile device, such as a PC, a tablet or a smartphone, in addition to the HMI).

[0074] According to some preferred non-limiting embodiments, the method further comprises the step of periodically sending (and updating at a frequency equal to or lower than the transmission frequency) the updated maintenance procedures 9 to the maintenance resource, for example to Figure 3 The operator O shown, in turn, performs the preventive maintenance operations in the order determined in the (periodic) schedule detailed by the maintenance procedures 9.

[0075] Advantageously but not necessarily, the motorization metrics MM comprise the torque / current provided by the electric motor and / or the electric motor follow-up error and / or the load percentage and / or the RMS and / or the torque error. All these motorization metrics MM are detected, in particular by means of an oscilloscope inside the electric drive 3.

[0076] Advantageously but not necessarily, the method described so far can be applied locally to the automatic machine 1, i.e. without the need to use a distributed data sharing system (cloud) and / or without the need for an essential internet connection.

[0077] In Figure 2In a non-limiting embodiment, possible connections between some general steps of the method are illustrated. Specifically, in this non-limiting embodiment, one or more electrical drivers 3 and / or one or more local acquisition units communicate bidirectionally with the data processing unit 5, transmitting sequences of recorded and detected sample sequences SS and periodically receiving a synchronization signal MS (at a synchronization frequency). Two separate sub-steps 20 and 30 are provided within the data processing unit 5. In step 20, the data processing unit 5 processes the transfer of data to the database DB. Specifically, in block 21, the received sample SS is collected; in block 22, the received sample SS is preprocessed to synchronize it using the synchronization signal MS. Subsequently, in block 23, the statistical features STF required to assess the presence of any anomalies are extracted (processed / calculated). The extracted statistical features STF (i.e., the actual state AC within the anomaly matrix AM) are then stored in the database DB (particularly in a unidirectional manner, as indicated by arrow 19). However, in step 30, the data processing unit 5 processes the detection of any anomalies. In block 31, by determining the tolerance range TH (after center C) and verifying the possible existence of the hazardous condition DC, the cloud of the actual condition AC (e.g.) is considered. Figure 4 The information is classified (specifically, by the K-means algorithm or any type of unsupervised classifier). In any case, after classifying the received information, in box 32, training of the database is performed, including the information just classified in the model of the automated machine 1. In this case, communication 18 is bidirectional because data is received from the database DB during classification and sent to the database DB during training. Communication 17 between the database DB and communication interface 8 is also bidirectional because the maintenance resource can transmit any maintenance performed in addition to receiving the maintenance program 9, thereby allowing the data processing unit 5 to update the program 9.

[0078] exist Figure 5 and Figure 4 In a non-limiting implementation, a comparison between correct and abnormal operating conditions is shown (which thus determines maintenance predictions). Specifically, Figure 5 The value S1 represents the torque error (eNm) over time under correct operating conditions, while value S2 represents the torque error (eNm) over time under abnormal operating conditions. Using the method described above, anomalies can be detected when determining deviations in the condition AC (i.e., characteristic STF, STF' calculated as a function of sample SS) within the anomaly matrix AM. In solutions of known techniques, this type of anomaly (which essentially follows the trend of correct conditions, exhibiting some slight inaccuracies and tensions in the signal) is difficult to detect. Specifically, Figure 1The trend of a plurality of statistical features STF (of the type listed above, for example) related to the motorization measure MM is shown, indicating correct operating conditions in the left part of the graph (i.e. from 40 to 51 features) and abnormal operating conditions in the right part of the graph (i.e. from 40" to 51" features). Using the methods described so far, it is possible to train the model of the automatic machine 1 so that the data processing unit 5 can determine whether the actual condition AC is in the correct area of the anomaly matrix AM or in the abnormal area through a multi-factor evaluation (the deviation of a single value does not necessarily lead to an anomaly).

[0079] Advantageously but not necessarily, the automatic machine 1 is configured to perform the method described above.

[0080] In ​ In the preferred and non-limiting embodiment shown, the tobacco industry products processed by the automatic machine 1 are cigarette packets 2. According to different embodiments not shown, the automatic machine 1 is of a different type (for example, a packer, a cellophane wrapping machine or a packaging machine, a food machine, a machine for hygienic absorbent articles, etc.) and therefore the products are cigarettes, filter tips, tobacco packets, cigars, nappies, chocolates, etc.

[0081] Although the present application described above makes particular reference to a very precise embodiment, it should not be considered limited to this embodiment, since all the variations, modifications or simplifications apparent to the person skilled in the art fall within the scope of the present application, for example: the addition of further actuators, use on another type of machine of the tobacco industry other than a packaging machine, anomalies other than those described (but which in any case can affect production, causing so-called "warnings"), use of different data transmission systems or devices, algorithms other than those mentioned, statistical features other than those mentioned, etc.

[0082] The present application has a plurality of advantages.

[0083] First of all, it allows to increase the efficiency of the automatic machine to which it is applied, since the predictive nature of the faults it determines allows to significantly reduce the number of unexpected and non-optimized interruptions in terms of time (for example, damage to a component in the absence of a relevant spare part available). All this involves a significant reduction in the production recovery time, thus increasing the productivity of the automatic machine.

[0084] Furthermore, this reduction in time obviously allows to proportionally reduce the costs due to periodic maintenance, unlike the case of preventive maintenance (by estimating the average wear of a component and replacing it even in the absence of evident signs of malfunction), allowing to replace a component only when it is really necessary, thus obviously saving costs, making it unnecessary to stock unnecessary spare parts, or in any case to evaluate the actual need.

[0085] Moreover, thanks to the difference between the synchronization signal and the sampling frequency and the transmission frequency, the present application allows to perform a very high frequency sampling, effectively managing the amount of data, without having to transmit them in real time to the data processing unit. In addition, the present application allows to periodically recompute a new tolerance range by updating the model of the machine, thus continuously improving the knowledge and adaptability of the automatic machine.

[0086] Another advantage of the present application lies in the definition of a multidimensional control, which allows to also consider anomalies that would not be possible to detect by individually monitoring single values. Moreover, the present application also determines a reduction of costs, since the possibility of exploiting what has already been detected by means of components present on the machine in any case (such as, for example, the drives) at least partially eliminates the need to increase the appropriate sensors, which would otherwise be required for carrying out predictive maintenance.

[0087] Finally, by continuously performing the above method, it is possible to perform a predictive maintenance of the automatic machine, in order to reduce (even cancel) the number of semi-finished products that are discarded due to uncompleted processing cycles, usually due to sudden failures. The result, from an economic and environmental point of view, is a further increase in productivity and a significant reduction in waste.

Claims

1. A method of predictive maintenance of an automatic machine (1) for manufacturing or packaging consumer goods; the method comprising the steps of: - detecting and recording, by means of at least one respective local control unit (11), at least one sample sequence (SS) relating to at least one motorized measure (MM) of at least one electric actuator (4) periodically and with a sampling frequency (SF); - transmitting the recorded sample sequence (SS) to a data processing unit (5) periodically and with a transmission frequency (TF) equal to or lower than the sampling frequency (SF); - defining, based on the detected at least one sample sequence (SS) and at least with respect to the detected motorized measure (MM), at least one multidimensional tolerance range (TH) within an anomaly matrix (AM) having at least two statistical features (STF) as dimensions; - calculating, for each detected sample sequence (SS), the at least two statistical features (STF) so as to define the position of the actual condition (AC) within the anomaly matrix (AM); - determining the urgency of the necessary maintenance based on the position of the actual condition (AC) in the anomaly matrix (AM) and the multidimensional tolerance range (TH); wherein the motorized measure (MM) is a speed error of an electric motor and the motorized measure (MM) is used to detect anomalies of behavior of the electric motor caused by friction in the behavior thereof.

2. The method of claim 1, wherein, The motorized measure (MM) is the speed error detected by the respective driver.

3. The method of claim 1, wherein, During recording, each local control unit (11) receives a synchronization signal to be included in the recording of the sample sequence (SS) at a synchronization frequency (SFC); the synchronization signal is included in all n samples (SS); the synchronization frequency (SFC) is lower than the sampling frequency (SF) but higher than the transmission frequency (TF).

4. The method of claim 3, wherein, The synchronization signal is the position of a physical or virtual main shaft of the automatic machine (1).

5. The method according to claim 3, further comprising the further step of synchronizing the samples (SS) transmitted to the data processing unit (5) using the synchronization signal as a reference to know which sample corresponds to a given instant or to a given time phase of the automatic machine (1).

6. The method of claim 1, wherein, The sequence of recorded samples (SS) also relates to local state measures (LSM) relating to the condition of one or more devices installed on the automatic machine (1), the local state measure (LSM) values being detected by means of at least one local acquisition unit (7) connected to a node of a bidirectional, digital and local industrial network.

7. The method of claim 6, wherein, The local state measures (LSM) comprise vibrations and / or temperatures and / or accelerations.

8. The method of claim 7, wherein, The local state measures (LSM) comprise vibrations detected in several dimensions.

9. The method of claim 1, wherein, The sampling frequency (SF) is greater than or equal to 2 kHz.

10. The method of claim 9, wherein, The sampling frequency (SF) is greater than or equal to 4 kHz.

11. The method of claim 1, wherein, The transmission frequency (TF) is less than or equal to 0.2 Hz.

12. The method of claim 11, wherein, The transmission frequency (TF) is less than or equal to 0.1 Hz.

13. The method of claim 1, wherein, Said multidimensional tolerance range (TH) is defined via an unsupervised classifier; the tolerance range (TH) is configured to have a non-linear shape; said tolerance range (TH) is periodically updated, including the values of the detected last sampling sequence (SS).

14. The method of claim 13, wherein, Said multidimensional tolerance range (TH) is defined via a K-means algorithm.

15. The method of claim 13, wherein, Said tolerance range (TH) is configured to have an elliptical or circular shape.

16. The method according to claim 1, further comprising the step of: training a model of said automatic machine (1) by means of an unsupervised classifier, using as input a plurality of statistical features (STF) generated by known faults; said model is periodically updated, including the detected last sampling sequence (SS); in case of unexpected fault, said model is also updated, defining on said anomaly matrix (AM) an area (AC) of fault.

17. The method according to claim 16, further comprising the step of: training a model of said automatic machine (1) by means of a K-means algorithm, using as input a plurality of statistical features (STF) generated by known faults.

18. The method according to claim 1, further comprising the step of: calculating the speed at which a continuous actual condition (AC) moves within said anomaly matrix (AM).

19. The method according to claim 18, further comprising the step of: calculating the speed at which the last actual condition (AC) moves towards said tolerance range (TH).

20. The method according to claim 1, further comprising the step of: periodically scheduling a maintenance procedure (9) based on the position or speed of the last actual condition (AC) within said anomaly matrix (AM).

21. The method according to claim 20, further comprising the step of: periodically transmitting the updated maintenance procedure (9) to a maintenance resource.

22. The method of claim 1, wherein, Said anomaly matrix (AM) comprises a plurality of groups, each group corresponding to the state of a different mechanical element of said automatic machine (1) or of mechanical elements having structural features in common; for each group a multidimensional tolerance range (TH) is defined.

23. The method of claim 1, wherein, Said motorized measure (MM) comprises torque / current supplied by the motor and / or motor follow-up error and / or load percentage and / or RMS value.

24. An automatic machine (1) for manufacturing or packaging consumer goods; said automatic machine (1) comprises: - one or more electric drives (3) configured to control at least one electric actuator (4) and to periodically detect and record, at a sampling frequency (SF), a sampling sequence (SS) related to at least one motorized measure (MM) of said at least one electric actuator (4); - a data processing unit (5) configured to periodically receive, at a transmission frequency (TF) equal to or lower than said sampling frequency (SF), said sampling sequence (SS) recorded at said sampling frequency (SF); - a local storage unit (6) configured to contain an anomaly matrix (AM) having at least two statistical features (STF) based on at least one detected motorized measure (MM). Said automatic machine (1) is configured to perform the method according to claim 1.

25. The automatic machine (1) according to claim 24, further comprising at least one local acquisition unit (7) connected to a node of a bidirectional, digital and local industrial network; said automatic machine (1) further comprises a communication interface (8) configured to be connected to said data processing unit (5) and to allow said data processing unit (5) to transmit maintenance programs (9) to maintenance resources; said at least one local acquisition unit (7) comprises smart tags and / or loT sensors; the electric drives (3) are arranged on a machine control cabinet or on the respective electric actuators (4) to which said machine control cabinet is connected; said automatic machine (1) comprises a plurality of local acquisition units (7), each of said local acquisition units (7) being arranged on a different mechanical group installed on said automatic machine (1).

Images