Patent Application
20220066437
The present disclosure relates to diagnostic methods of machine tools.
One or more embodiments of the solution disclosed may relate to contexts of pattern recognition and/or predictive maintenance.
In the framework of diagnostic methods for machine tools, predictive maintenance seeks to predict the time to failure and, consequently, to identify the right moment to carry out the operation required for preventing failure. In this procedure, it is possible to identify one or more parameters of a system—which may, for example, be a machine tool with a plurality of movement axes, in particular a machine tool for laser cutting or laser welding—that are to be measured and processed using appropriate mathematical models in order to predict the time to failure.
In order to measure one or more parameters of a system, various methodologies may be used, such as measurement of vibrations, thermography, absorbed-current analysis, and detection of unusual vibrations. A variation of the quantities measured and monitored with respect to the status of normal operation may hence indicate a degradation of the performance of parts of the system, enabling warning of the usefullness of maintenance in order to prevent an imminent failure.
The quantities monitored may be analysed through a further variety of procedures, amongst which procedures of modal analysis. Modal analysis is the study of the dynamic behaviour of a structure when it is subjected to vibration.
In traditional diagnostic methods comprising modal analysis, the structure of the system that is to be monitored is excited with an external component, for example a calibrated impact hammer or a shaker, which applies dynamic impulse(s) to a part of the machine tool. Via a sensor it is possible to record the reaction of the structure to the above impulse. By analysing the signal acquired by the sensor, it is possible to extrapolate information on the features of the machine tool.
For instance, it is possible to use modal analysis for a method of acquisition of modal parameters of a numerical-control machine tool. Multiple accelerations and decelerations can be applied to the work bench of the numerical-control machine tool in order to generate excitation, so as to facilitate acquisition of the modal parameters of the numerical-control machine tool.
The use of external excitation elements involves the disadvantage of introducing perturbations in repeatability of the excitation: in fact, the impulse applied, for example, via the hammer or shaker, depends upon the generator, the user, and the point of application. The shaker, in fact, exerts a dynamic excitation, for example a driving force, the characteristics of which vary as the resting point of the table varies.
Moreover, machine tools of one and the same type installed in different environmental conditions, for example set on bases made of different materials, present behaviours that are similar but that differ from one machine tool to another as a result of external factors that introduce noise into the response, which is hard to filter out without having information on the external environment.
Modal-analysis methods for acquisition of modal parameters of machine tools can be used during calibration, or dry run, of machine tools: for example, it is possible to acquire modal parameters regarding the velocity of rotation of a main shaft. For instance, it is possible to use methods of this type for calibrating a machine tool with spindle in idle running conditions, via impulsive commands on rotation of the spindle.
A disadvantage of these known techniques is a poor compatibility with normal operation of the machine tool so that calibration has to be performed in a dedicated step (such as dry running) that involves machine-tool down times.
To use modal analysis for purposes of diagnostics and/or predictive maintenance, notwithstanding the vast activity in this area, further improved solutions are desirable.
An object of one or more embodiments is to contribute to providing such an improved solution.
According to one or more embodiments, the above object can be achieved by a method having the characteristics set forth in the ensuing claims.
A method for diagnosis of operation of a machine tool comprising one or more axes of movement may provide an example of the aforesaid method.
In one or more embodiments, the above method may acquire, analyse, and process an impulsive response of a structure of machine tool, it being possible to analyse the above impulsive response for providing information on an operating profile, which indicates the possibility of presence of faults in the structure and in the machine tool and/or the type of such faults.
One or more embodiments may regard a corresponding machine tool. A machine tool comprising control modules configured for implementing the diagnostic method and facilitating predictive maintenance may be an example of such a machine tool.
One or more embodiments may include a computer program product that can be loaded into the memory of at least one processing circuit (for example, a computer) and includes portions of software code for executing the steps of the method when the product is run on at least one processing circuit. As used in the present document, reference to such a computer program product is intended as being equivalent to reference to a computer-readable medium containing instructions for controlling the processing system in order to coordinate implementation of the method according to one or more embodiments. Reference to “at least one computer” is intended to point out the possibility of one or more embodiments being implemented in modular and/or distributed form.
One or more embodiments present the advantage of making it easier to achieve improved functions; for example, a mode of stimulation of a CNC chain makes it possible to carry out modal analysis during normal operations carried out by the tool, thus providing a flexibility such as to enable system testing in different positions and with different configurations of the axes.
One or more embodiments moreover present the advantage of making it possible to supply a neural network with homogeneous samples in so far as they derive from a stimulus with characteristics of high repeatability.
One or more embodiments may render the diagnosis advantageously independent of the external conditions in which the machine tool is used.
One or more embodiments may likewise favour remote training of neural networks, for example via data gathered in field.
The claims form an integral part of the technical teaching provided herein with reference to the embodiments.
One or more embodiments will now be described, purely by way of example, with reference to the annexed drawings, wherein:
In the ensuing description, one or more specific details are illustrated in order to provide an in-depth understanding of examples of embodiments of the description. The embodiments may be obtained without one or more of the specific details or with other methods, components, materials, etc. In other cases, known operations, materials, or structures are not illustrated or described in detail so that certain aspects of the embodiments will not be obscured.
Reference to “an embodiment” or “one embodiment” in the framework of the present description is intended to indicate that a particular configuration, structure, or characteristic described with reference to the embodiment is comprised in at least one embodiment. Consequently, phrases such as “in an embodiment” or “in one embodiment” that may be present in one or more points of the present description do not necessarily refer to one and the same embodiment.
In addition, particular conformations, structures, or characteristics may be combined adequately in one or more embodiments.
The references used herein are provided merely for convenience and hence do not define the sphere of protection or the scope of the embodiments.
The above machine-tool system 100 may comprise:
In what follows, reference will be made, for simplicity, to a system 100 comprising a machine tool 10 with mobile structure of a cantilever type with three axes (designated by the letters X, Y, Z), which is also referred to as “cartesian machine tool”. It is to be noted that the type of structure described is not binding; in fact, the solution discussed can be adapted to structures of some other type, e.g., with six degrees of freedom (i.e., with redundant axes).
As has been said, in various embodiments, the aforesaid cartesian machine tool 10 can be used for moving an end effector 101 comprising, for example, a laser-welding or laser-cutting head. In variant embodiments, the machine tool 10 may also use end effectors 101 that perform operations of additive manufacturing, or in general also other machining heads, or end effectors, including machining heads that operate without the aid of laser sources (e.g., arc-welding heads, punches, presses, benders, etc.).
The aforesaid machine tool 10 of the system 100 comprises the axes X, Y, Z, in
The movement of the first arm 102 and of the second arm 104 of the machine tool 10 in the working envelope 10, as well as of the end effector 101, is determined by a plurality of actuators and/or motors driven by the numerical-control unit 20.
The control module 20, as has been said, is hence configured for being coupled to the machine tool 10, in particular to the motors, for implementing movement of one or more arms 102, 104 and/or of the end effector 101 in the working envelope 10, as described in greater detail in what follows.
It is to be noted that for simplicity in what follows the expression “movement of the axes X, Y, Z” will be used as referring to the operation of driving the motors and/or actuators of one or more arms 102, 104 and/or of the end effector 101 so as to move the one or more arms 102, 104 and/or the end effector 101 according to the aforesaid one or more axes X, Y, Z.
The sensor 30 is coupled to the machine tool 10 and represents a set of sensors 30 for detecting measurements of quantities indicative of operation of one or more parts of the machine tool 10, each sensor of the set of sensors 30 producing corresponding read-out signals or read-out data S. For instance, the set of sensors may comprise at least one of the following:
For simplicity, in what follows the term “sensor 30” is used in the singular, it being understood that what has been said for the sensor 30 may be extended, for example to:
The sensor 30 is hence configured for supplying the above read-out data S to the control module 20. The control module 20 can use the read-out data S, for example for:
The control module 20, as exemplified in
The CNC unit 200 of the control module 20 comprises, for example (represented as dashed boxes within the stage 200):
The first processor 2002 operates as user interface for sending instructions and commands to the second personal computer 2004, which, for example, comprises an operating system of a Linux type associated to extensions of a real-time type for management of the machine tool 10. The second processor 2004 hence supplies trajectories to be executed to the servo-drive board 2006, for example of a PCI DSP type for control of one or more actuators or motors. In the second processor 2004 and in the servo-drive board 2006 a procedure of management of the mobile structure 10 is implemented.
As mentioned, the CNC unit 200 of the control module 20 controls operation of motors and actuators for movement of the axes of the mobile structure 10 according to programs or sequences of programming instructions P predetermined on the basis of the requirements of machining of the workpiece and in a co-ordinated way. The above programs P are prearranged for moving the end effector 101 within the envelope 10 represented in
In the CNC unit 200, the first processor 2002, according to one aspect of the solution described herein, can then be configured for generating a programming sequence P of movement of the axes X, Y, Z of the machine tool 10, where the programming sequence P comprises instructions that are such as to apply impulsive variations (for example, variations according to an excitation sequence) of a kinematic quantity (for instance, acceleration, velocity, position) that regards one of the above actuators.
The sequence of programming instructions P, once interpolated in the second processor 2004 of the CNC unit 200, is then supplied to at least one servo drive 2006, configured for controlling a movement, performed via at least one motor or one actuator of the axes X, Y, Z of the machine tool 10 according to the above sequence of programming instructions P.
The interface 202 of the control module 20 may comprise an input/output device, for example a display with touchscreen of a video display terminal for an operator, with which a user can, for example, modify instructions or parameters of instructions of the part program that represents the sequence of programming instructions P.
The intermediate stage 204 of the control module 20 may be configured for operating as manager of the artificial-neural-network processing stage 206, as discussed in what follows. The intermediate stage 204 can pre-process and/or post-process data during an exchange of data, for example read-out data S of the sensor 30 (or other data on the status of the machine tool 10) exchanged between the interface 202 and the artificial-neural-network processing stage 206 to which the intermediate stage applies an operation of transformation into the frequency domain (e.g., a Fourier transform).
The artificial-neural-network processing stage 206 of the control module 20 may comprise a set of artificial neural networks 2060, 2070 configured for being used, for example on the basis of instructions and signals received from the intermediate stage 204, for supplying one or more operating profiles of the machine tool, specifically one or more signals indicative of the status of the machine tool W, associated to which is, for example, information indicating anomalous operation, such as an alarm string warning of the risk of failure of a belt, or else information indicating proper operation.
The operating profile, specifically the aforesaid signal indicative of the status of the machine tool W, can hence be supplied by the artificial-neural-network processing stage 206 to the intermediate stage 204 and/or to other stages, for instance to:
The server SV can communicate with all the stages of the control module 20 to facilitate downloading of updates of software implementations of operations of the method, such as new versions of the software of the ANN processing stage 206, comprising, for example, the set of re-trained neural networks 2060, 2070. Likewise, the ANN processing stage 206 can send, for example via the intermediate stage 204 or the interface 202 (or else directly), data acquired in field to be added to a remote database on the server SV that contains data to be used for training the networks themselves in order to render subsequent data-processing operations more robust or facilitate analysis of new operating profiles of the machine tool 10, as discussed in what follows.
The control module 20 can hence be configured for exchanging instructions and data P, T, S, W, at input and output, with networks, for example the Internet, with communication modalities in themselves known.
The machine tool 100 may be configured, for example according to a control chain of the control module 20, for implementing operations of a method 1000 for diagnosis of a machine tool 10, as discussed in what follows in relation to
The method 1000 for diagnosis of a machine tool 10 and/or at least one machine tool 10 as described herein may be used in a system for predictive maintenance of a machine tool 10, for example enabling easier automation of provision of maintenance services, automatic planning of such services (type, duration, or date of the intervention), etc. An advantage of carrying out predictive maintenance is to enable reduction of direct maintenance costs, as well as reduction in production loss, which are the natural consequence of non-optimal maintenance interventions like the ones provided, for example, by corrective maintenance.
In one embodiment, the diagnostic method 1000 may, for example:
In particular, the operation of controlling 1210 the movement of the axes may comprise applying a trajectory T of the arms 102, 104 and/or of the end effector 101 that comprises a step-like deceleration of the movement along an axis, for example the axis X, of the machine tool 10, for example by “instantaneously” stopping movement of the arms 102, 104 and of the end effector 101 along the aforesaid axis X, as described in detail in what follows with reference to
The operation of receiving 1220 a read-out signal S of at least one sensor 30 follows, downstream, the operation of controlling 1210 the movement of the axes, thus facilitating acquisition, for example measurement via sensor reading S, of the response of the machine tool 10 to the sequence of instructions of movement regarding one of the actuators of the axes X, Y, Z driven by the CNC unit 200.
It is noted that, unlike traditional methods of modal analysis in which the excitation sequence is generated outside the system 100, for example with a hammer or a shaker as discussed previously, in the present solution the system is “self-exciting”, via self-generation of a sequence of programming instructions P, which propagates through the control chain of the diagnostic method 1000.
Yet a further advantage of the above self-generation of the variations of excitation, which may be comprised in a time sequence of excitations, in the CNC unit 200 of the control module 20 of the system 100 is to facilitate production of impulsive variations of excitation in the trajectory T that are extremely similar from one repetition of the method 1000 to another, for example each having one and the same “flat” harmonic content, i.e., a constant harmonic content in the range of frequencies present in the impulsive temporal variation of excitation in the trajectory T.
Consequently, for example with the same operating conditions of the machine tool 10, the sensor 30 records, downstream of application of the impulsive variations of excitation in the trajectory T, a signal S indicating the response of the system to the excitation and generated each time by application of an impulsive temporal variation in the trajectory T that presents characteristics that are substantially constant from one repetition to another. This improved repeatability in the impulsive variation of excitation in the trajectory T facilitates analysis of the read-out signal S acquired by the sensor 30 used for detecting any possible presence of faults, as well as the type of such faults, in operating profiles of the system 100, in particular of each component 10, 20, 30 of the machine-tool system 100 involved in generation, application, and/or reception of the excitation sequence that ultimately results in the trajectory T.
In fact, since instructions or signals corresponding to the impulsive variations of excitation in the trajectory T stimulate the system 100 in a forcing way (in fact, the aforesaid variation of excitation is also referred to as “driving force”), they traverse all the stages of the system 100, propagating through them. Hence, modal analysis of the response S read by the sensor 30 can bring out faults in operating profiles of the machine tool 10 also at levels higher and/or lower than that of the mechanical structure of the machine tool 10 of the system 100, such as malfunctioning of an actuator of one or more axes X, Y, Z, or of the CNC unit 200 itself, for example in a servo drive 2006.
As has been said, with traditional methods the excitation enters the system 100, directly in the mechanical structure of the machine tool 10, hence without the possibility of providing information on the status of the control module 20 or of the sensor 30.
Thanks to the repeatability of the excitation, with the method 1000 it is also advantageously possible to:
The above procedure of comparing the read-out datum S with the reference read-out datum S0, for example to filter out the noise due to the external environment, is particularly advantageous because it enables analysis of problems that are hard to detect in so far as in the modal analysis are at frequencies close to the frequencies of “free” vibration (i.e., of proper operation) of the machine tool 10.
The aforesaid reference read-out datum S0 can be stored in a memory accessible by at least one from among the interface 202, the intermediate stage 204, and the ANN processing stage 206 so as to enable its subsequent use to refine neural-network processing 206, as discussed in what follows.
In one embodiment, the operation of generating 1200 a sequence of programming instructions P of movement of the axes X, Y, Z of the machine tool 10, which comprises instructions that are such as to apply a trajectory T comprising temporal variations of impulsive excitation of a kinematic quantity (for example, the velocity v and/or the acceleration ac) that regards one of the actuators of the axes X, Y, Z may comprise a number of steps, one for each set of instructions for one or more of the axes X, Y, Z, as exemplified in
Specifically,
For instance, in
It is noted that, in the example represented in
The impulsive variation D may hence be an instantaneous stop, i.e., a stop in the shortest time allowed by the movement of the axis X and by the kinematic work point, i.e., an instantaneous acceleration ac from a first velocity (for example, a negative constant velocity v for instants t<t1) to a second velocity (for example, a velocity v equal to zero at the instant t=t1).
Once again illustrated by way of example in
It is noted that supplying a sequence of instructions of impulsive excitation in the sequence of programming instructions P comprises, for example, supplying a sequence of excitation instructions such that the corresponding variation of kinematic quantity D has a brief duration, which can be approximated to an instantaneous (impulsive) excitation.
As may be noted from the diagram of
Hence, the diagnostic method 1000 implemented in the system 100 facilitates propagation of the sequence of excitation instructions P, P′, P″ that determines the excitation D during the normal operating activities of the machine tool 10, in particular, in masked time, for instance:
What has been discussed above involves the advantage that, from the standpoint of the end user of the machine tool 10 and/or of the system, no machine-tool down times are required for applying the diagnostics method 1000, unlike what typically occurs in the case where the impulsive variation of excitation D in the trajectory T is applied by means of a hammer or with a shaker, as discussed previously.
The parameters of the sequence of instructions P for generation of the excitation D (for example, amplitude, frequency, period, number of waveforms in frequency, etc.) can be selected and chosen in a flexible manner, both in a manual mode and in an automatic mode via the part program corresponding to the sequence of programming instructions P and the interface of the first processor 2002 of the CNC unit. This flexibility of selection of the variation D in the trajectory T facilitates execution of the method 1000 on the system 100, for example with one and the same excitation D, applied at moments when the machine tool assumes different positions and with different configurations of the arms 102, 104 and/or of the end effector 101 with respect to the axes X, Y, Z. This leads to the advantage that the method facilitates detection of the type of criticality, for example increasing the precision and accuracy of the diagnostic method 1000.
Finally, the flexibility afforded by the programmability of the part program P in the generation of the excitation D means that the harmonic content of the impulsive variation D is substantially homogeneous over the band of interest, unlike traditional methods, as discussed previously in particular with reference to
It is noted that what has been discussed previously as regards the axis X of the example of
In step 1220, the response, specifically the frequency response, of the system to the excitation is detected by the sensor 30. As has been said, the sensor 30 is representative of a set of one or more sensors 30 for making measurements of quantities indicative of operation of one or more parts of the machine tool 10, to produce corresponding read-out signals S in the sensor 30.
In one or more embodiments, the sensor 30 comprises a type of sensor chosen according to:
Illustrated in this regard in
As represented, for example, in
It is possible to use for this purpose also a sensor element 30 that is already provided for normal operating functions. In particular, it is possible to use as sensor 30 in the diagnostic method 1000 the signal S that is supplied by an encoder, typically installed on the end effector 101, and that, after possible processing thereof, is then used both for correcting the position error of the axes moved X, Y, Z with respect to a programmed position (i.e., for implementing a feedback in control of movement of the axes X, Y, Z) and for recording the read-out signal S that indicates the impulsive response of the machine tool 10, specifically the frequency response. In fact, during propagation of the impulsive sequence of excitation T, the control axes X, Y, Z themselves are excited, for example vibrating as a function of the excitation T. Consequently, if the position of the axes X, Y, Z is recorded, for example by the CNC unit 20, during this vibration of reaction to the excitation T, a position error is caused in so far as the position of the axes X, Y, Z is not the same as the one set as target in the control 200.
The signal of the position error of the axes X, Y, Z with respect to a programmed position of the encoder on the end effector 101, which operates as virtual position sensor, can thus be used as read-out signal S of the sensor 30. The method 1000 hence presents an advantage in not requiring installation of further sensors in the machine tool 10 beyond the virtual sensor 30 already present.
In a further example of embodiment, the sensor 30 comprises a sensor of a capacitive type, for example a capacitive sensor, provided in a way in itself known, set in the proximity of an output point of a laser beam in a laser machine tool 10. Such a capacitive sensor 30, for example, supplies the measurement of the distance between the aforesaid output point of the laser and a sheet of metal being machined in so far as, to carry out proper cutting and prevent collision of parts of the machine tool 10, in particular of the end effector 101, with the aforementioned workpiece, it is useful to maintain a constant gap between the latter and the end effector. The read-out signal S supplied by the capacitive sensor 30 is useful for feedback control, for example via the control module 200, of an axis that manages this gap.
In a variant embodiment, it is possible to get the final portion of the impulse D to correspond to a positioning of the end effector 101 such that the sensor 30 is at a very short distance from the metal sheet. In this way, the read-out signal S coming from the sensor 30, which corresponds to the distance between the metal sheet and the end effector 101, assumes an oscillating pattern, the oscillations of which can be analysed in a way similar to those of the read-out signals of the triaxial accelerometer.
Also in this case, the advantage consists in not having to add further external devices for measuring the impulse response in the diagnostic method 1000, specifically the frequency response. In addition, the use of this capacitive sensor may be particularly indicated for analysing the set of faults linked to operation of the sensor 30 itself integrated in the machine tool 10.
It is noted that the sensor 30 is preferably set as downstream as possible in a mechanical transmission chain of the machine tool 10. With reference to a structure 10, as exemplified in
In a similar way, the sensor 30 located in a second position 302, fixed with respect to the first arm 102, can, for example, detect with higher sensitivity operating profiles that comprise faults linked to the mechanical transmission along the axis X.
Likewise, the sensor 30, located in a first position 300 fixed with respect to the reference system in which the machine tool 10 is installed, can, for example, detect with higher sensitivity operating profiles that comprise faults in relation to fixing of the machine tool to the ground.
The above sensor 30 may further comprise, for example, one or more of the following functions:
In one embodiment, the sensor 30 comprises the triaxial accelerometer, which acquires, following upon application of the impulsive variation D in the trajectory T, a read-out signal S having components ax, ay, az that represent the measurements of the acceleration time plots along each of the axes, as illustrated in
It is noted that processing of the sensor read-out signal S in the stages 204, 206 is described in what follows, for simplicity, in relation to the acceleration signals ax, ay, az, it being moreover understood that the present discussion is provided purely by way of non-limiting example in so far as such processing can extend to other types of signal coming from other types of sensor (position error, distance from metal sheet, etc.).
In the particular example of
Each curve in the graph of
It is noted that it is possible to choose to analyse even just one of the components of the acceleration ax, ay, az. For instance, if the excitation impulse T excites movement of the axis X it is possible to choose to analyse just the component az along the axis Z of the read-out signal S. This choice may be motivated by considerations of orientation between a position 300, 302, 304 of the sensor 30 and the axes X, Y, Z of the machine tool 10. This choice may be set, for example, as automatic selection or else may be made by a user.
As anticipated in relation to
Once acquisition of the read-out signal S of the sensor 30 that represents the response of the machine tool 10 to the variation D has been completed (e.g., with a total duration of the operation of 1 s), for example in step 1210 of the method 1000, the intermediate stage 204 of the control module 20 is configured for carrying out steps of data pre-processing, in order to supply the ANN processing stage 206 with the data to be analysed, as represented by way of example in
As exemplified in
An operation designated by 2023, which can be executed in parallel or in series with the operations already illustrated, is the operation of acquiring from the CNC unit 200, for example via the interface 202, data regarding:
In a further step (designated by 2025), which again can be executed in series or in parallel with the steps illustrated so far, the list of data acquired in step 2023, for one or more axes X, Y, Z, is sent to/stored in an array Vacc.
Finally the method may comprise the operation, designated by 2028, of converting the arrays of data acquired or processed, such as the set of arrays AD, FT, FT′, Vacc, into a data format compatible with a data format of the artificial-neural-network stage 206, to obtain a set of data Sf that comprise the frequency spectrum FT, FT′ of the read-out signals S. This data set Sf, which comprises data indicating the response of the machine tool to the impulsive variation, is hence supplied to the ANN processing stage 206.
It is noted that the operations 2022 to 2028 can also be executed in a different order and/or iteratively during operations performed in the course of ANN processing 206, or as discussed in what follows.
The ANN processing stage 206 may comprise a set of ANNs, which include, for example:
The aforesaid ANNs can operate in parallel on one and the same set of data Sf or on copies of one and the same set of data Sf received.
The different neural networks in the ANN processing stage 206 are arranged so as to present a number of layers and a number of neurons, and so as to be trained according to the types of anomalous operation or faults that are to be identified.
Some subsets of anomalous operations/faults sought in the analysis of the read-out signals S of the sensor 30, which are referred to as “contexts of analysis”, may be grouped into sets of faults/anomalies, for example, of the following kinds:
In embodiments of the solution discussed herein, the contexts/subsets of analysis can be selected as a function of the instructions contained in the part program P.
As has been said, processing, for example in step 1230 of the method 1000, of the set of data Sf that comprise the spectrum FT, FT′ of the read-out signal S in the ANN processing stage 206 is managed in multi-diagnostic or multi-context mode, as represented, for example, in
By “multi-context mode” is meant a mode of operation of the neural network that is defined in a flexible manner according to the requests. For instance, it may change according to the sub-set of events (in terms of faults) that are to be analysed, the type of data, in particular the frequency range fmin, fmax considered in the set of data Sf that is sent to the set of networks 2060, 2070 of the stage 206.
It is noted that the two neural networks 2060, 2070 may have one and the same architecture and differ as regards training, for example as regards the frequency range fmin, fmax in which they are trained to analyse anomalous operating profiles of the machine tool 10.
For instance, the first neural network 2060 can be supplied already trained to analyse operating profiles in a first, relatively wide, frequency range (for example, from 0 Hz to 100 Hz), whereas the second neural network 2062 may be supplied already trained to analyse operating profiles in a second, relatively limited, frequency range (for example, with frequencies f comprised between 5 Hz and 50 Hz). In one embodiment, the first neural network 2060 may comprise a network supplied already trained, for example, with training function according to the conjugate-gradient method and with data falling within a frequency range from 5 Hz to 50 Hz.
In other cases, the neural networks may be completely different in every parameter or have some parameters in common and others different.
As illustrated in the example of
The first processing layer 2064 may comprise a set of n neurons 2064n, where, for example, n=20, which receive at input the set of data Sf that comprise, for example, the frequency spectra FT′ of read-out signals S in closed frequency ranges fmin, fmax which are weighted by weights 2064w and added to bias data 2064b. The neurons 2064n can have one and the same transfer function, for example of a sigmoid type.
The second processing layer 2066 may comprise a second set of m neurons 2066m, where, for example, m=2, which receive at input data that are weighted by weights 2066w and added to bias data 2066b. The neurons 2066m may have one and the same transfer function, for example an activation function of a softmax type. The respective weights 2064w, 2066w of the first processing stage 2064 and of the second processing stage 2066 may be initialised at random values according, for example, to a Gaussian distribution.
To return to the discussion of
In the case where, in step 800, generic faults are found in the set of data Sf, it is possible to proceed to the step designated by 810, which comprises selecting one or more of the following operations:
As has been said, the spectrum FT (and/or the portion of spectrum FT′) may correspond either to the spectrum (and/or portion of spectrum) of the read-out signal S or to the spectrum (and/or portion of spectrum) of a signal obtained from comparison of the read-out signal S with the reference read-out signal, for example the signal indicating the absolute value of the differences between the spectrum of the read-out signal S and the spectrum of the reference read-out signal S0, in particular if the latter is already stored and/or when the intermediate stage 204 is configured for supplying this signal for the type of analysis to be carried out (and in the corresponding frequency range fmin, fmax).
Following upon processing via the aforesaid first ANN 2060 in step 814, the neural-network stage 206 may supply an indication, for example collected in the output stage 2068, of the presence or otherwise of the above fault or anomalous operation. In a step 820, it is possible to select, on the basis of the aforesaid indication provided by the first neural network 2060 one or more of the following operations:
As has been said, if for example in step 820 the first neural network identifies, on the basis of the analysis carried out in step 814, a fault on the mechanical transmission of the horizontal axis X, in step 822 the interface 200 receives the signal W indicative of the status of the machine tool. The interface 200 may be configured so as to issue a video message, displaying, for example, a message containing information processed on the basis of the operating profile analysed, for instance a string associated to the signal W indicative of the status of the machine tool such as:
“X-Transmission needs maintenance, please contact customer service . . . ”
In an embodiment, the above message may be personalised and vary according to the type of analysis made, the type of machine tool, the type of human-machine interface, etc.
It is noted that what has been discussed previously regarding a set of neural networks 206 that comprises two neural networks 2060, 2070 may be applied to any number of neural networks in the set of neural networks 206, hence supplying in series the set of data Sf each time to one neural network in the set of neural networks 206, for example to the neural network trained to recognize a specific fault or a specific type of anomalous operation in order to analyse operating profiles in the set of data Sf processed as a function of the sensor read-out signal S.
In an embodiment, the method 1000 may comprise a number of iterations of the sequence of operations detailed in step 1230, it being possible for these operations to be carried out also in a sequential order different from the one presented in the previous paragraph. In particular, as mentioned, successive iterations may comprise execution of one and the same sequence of operations but on subsets of portions of spectrum FT, FT′ in the set of data Sf indexed via different identifier codes, for example labelled with identifier numbers.
Specifically, the operations of the method for management of the neural network 206 may be applied in sequence to different data subsets FT, FT′, for example, in increasing or decreasing order of identifier number or in some other order, until a set or sub-set of types of anomalous operation and/or faults that are to be analysed (mechanical, electrical, software, etc.) has been covered.
Downstream of the sequence of processing operations 1230 and of the possible iterations of the method 1000, the various operating profiles analysed and results of the analyses of the individual neural networks 2060, 2070, may be merged in a merged operating profile, which comprises, for example, a global signal indicative of the overall status of the machine tool W′.
In an embodiment, the neural networks of the ANN processing stage 206 can be trained on data S acquired by the sensor during movement of the axes X, Y, Z. The step of training of the neural networks may be carried out, for example:
As an alternative or in addition to this, training input data can be supplied to the ANN processing stage 206 either in real time, during operation of the machine tool itself, or off-line, for example following upon downloading from a virtual repository accessible via Internet connection.
Training of the neural networks can be carried out more than once on one and the same network or on different networks with the same data or with different data so as to guarantee maximum flexibility of the system in order to be able to identify the largest number of possible types of faults and/or anomalous operation and supply the user stages with detailed information on the operating profile of the machine tool, as a signal indicative of the status of the machine tool W.
It is noted that in general the control module 20 may comprise one or more memory areas or memory devices (for example, databases) in which to store one or more sets of data, which comprise, for example, the read-out data S of the sensor, the set of data Sf, the reference read-out signal S0, and other data (for example, the weights 2064w, 2066w) processed during application of the method 1000.
As has been said, the neural networks of the set of neural networks 206 can be supplied already trained and can be updated so as to include new neural networks trained with new data, in the case where new types of faults or anomalous operation are found and new training or insertion of new trained neural networks is hence required: events previously not recognised by the system (for example, following upon recognition of new types of faults or anomalous operation) may be used for updating the remote training database. It is hence possible to download a new updated version, understood as implementing version, of the method 1000 that includes the re-trained neural networks (for example re-trained in off-line mode on data present on a remote server).
In general, the neural networks are configured for carrying out analysis of operating profiles in the response of the machine tool, measured by the sensor, to the impulsive variation, such as recognition of a correct operating profile (for example, by comparing the spectrum of the measured signal FT with that of the reference signal S0) or of a faulty operating profile (for example, by detecting the presence of peaks in the response of the machine tool, in particular in the frequency response).
Without prejudice to the underlying principles, the details and embodiments may vary, even appreciably, with respect to what has been described herein purely by way of example, without thereby departing from the sphere of protection, as defined in the annexed claims.
| Number | Date | Country | Kind |
|---|---|---|---|
| 102018000020143 | Dec 2018 | IT | national |
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/IB2019/060894 | 12/17/2019 | WO | 00 |
