Robot unit, method for controlling metal products moving in a device, and method for manufacturing metal reels

Abstract

A robotic unit (10) for automatically controlling a moving metal product (P) and an associated computerized method are configured for manipulating or cutting parts of said metal product (P). The invention also relates to an apparatus (100) and a method for manufacturing metal reels (200).

Description

[Technical field]

[0001] The present invention relates to a robotic unit and related methods for controlling moving metal products, in particular moving metal products such as rods and wires, in industrial processes of metal reel manufacturing, starting from, but not limited to, hot rolling processes, the robotic unit being configured to identify, control, manipulate and possibly remove parts of the product that may cause problems or cobbles, or be used for simple sampling.

[0002] The invention also relates to an apparatus including the robot unit described above, and an associated method for the manufacture of metal reels. [Background technology]

[0003] Apparatus is known for producing metal reels in which hot rolled metal products, such as smooth or ribbed rod or wire, are fed continuously or semi-continuously (e.g., in a "billet-to-billet" feed) to a winding unit where a laying or coil-forming head forms a continuous series of coils which are staggered lengthwise on a continuous transport conveyor (usually a rollerway) where the coils are cooled.

[0004] The product thus arranged finishes its run in the reel forming unit and is dropped into a well containing a trolley with a centering rod at its centre and on which the coils are stacked one above the other to form a skein / spool.

[0005] It is known that the first and / or last part of the product may not be formed correctly. It is therefore necessary to check at least the condition of the last coil, or more generally the condition of the last part of the product, while the product is progressing on the roller ways, to ensure that the automatic reel forming unit can operate without the risk of being blocked on the roller ways along the continuous transport conveyor.

[0006] Coil jamming along the roller ways or inside the forming unit is one of the most common problems and must be avoided, as it can lead to deformation of the skeins, plant stoppages, damage to structures, increased maintenance interventions and, in the presence of operators, a high risk of accidents.

[0007] In conventional rolling plants the last portion of the advancing product is manually controlled by one or more operators monitoring the terminal zone of the roller ways through which the product passes at a temperature of about 500°C. When manually cutting off the tail portion of a moving product, difficulties can be presented by the high relative speed of the roller ways to the operator, the temperatures involved and the weight of the product portion being removed from the roller ways, as well as difficulties caused by the diameter of the product and other factors, particularly when sampling and quality control are taken into account.

[0008] Automated systems have also been proposed to perform these tasks in order to recognize, cut, and remove the final part of the product being processed. However, these systems cannot reliably distinguish the final part of the product from the previous homogenous part, nor can they accurately distinguish whether this final part has defects or a position / shape that will cause subsequent problems during reeling. As a result, the final part is often removed, and more of it is removed than is actually needed, resulting in reduced plant productivity and increased processing waste. Alternatively, the final part is not recognized and removed, resulting in the creation of cobbles or substandard reels.

[0009] WO 2017 / 082908 relates to rolling mills in general for producing hot rolled products, and in particular to the trimming and removal of the head and tail ends of such products, and the correct positioning of the new ends after trimming. WO 2017 / 082908 does not provide any measures to identify the presence of defects in the ends, heads or tails of such products. WO 2017 / 082908 therefore does not provide a system that allows the process to be optimized, taking into account that, as mentioned above, the removed portion of the product is often larger than the amount actually required.

[0010] Another example of a method and device for cutting the leading and trailing ends of a metal wire that is subsequently wound on a reel is described in US Patent No. 4,995,251 and US Patent No. 3,756,289. Neither US Patent No. 4,995,251 nor US Patent No. 3,756,289 mentions the detection of defects, and therefore suffer from the same drawbacks as WO 2017 / 082908.

[0011] Korean Patent Publication No. 2019-0032908 and the publication by BLUG A. et al., "On potential of current CNN cameras for industrial surface inspection," relate to the detection of micro surface defects, rather than macro shape defects or cobble defects, in the manufacturing of metal products using artificial intelligence.

[0012] JP 07-98217 A relates to an apparatus for identifying defects in three-dimensional objects, which can be installed in a production line for the three-dimensional objects.

[0013] Thus, none of these documents allows optimizing the operation of the apparatus and method for the manufacture of metal reels in its function of selectively removing the head or tail of the metal product being worked.

[0014] Therefore, there is a need to achieve a robotic unit and method for controlling moving metal products, and an apparatus and method for manufacturing metal reels, which overcomes at least one of the shortcomings of the prior art.

[0015] One object of the present invention is to provide a robotic unit capable of performing quality control on at least one of the products being processed, in particular on the last part thereof, thereby reducing the false identification of defects and therefore waste due to excessive removal of material.

[0016] Another object of the present invention is to provide a computerized method for managing robotic units, preferably according to a self-learning logic.

[0017] Another object of the present invention is to create an apparatus for manufacturing metal reels which is more productive and requires fewer plant stops than apparatuses known in the prior art.

[0018] Another object of the invention is to provide a method for manufacturing metal reels that is particularly efficient and allows automated, on-the-fly control of the product being processed.

[0019] Applicant has invented, tested and embodied the present invention to overcome the shortcomings of the existing art and to obtain these and other objects and advantages. Summary of the Invention

[0020] The invention is set forth and characterized in the independent claims. The dependent claims describe further features of the invention or variants of the main inventive idea.

[0021] Corresponding to the above mentioned objectives, a robotic unit for controlling a moving metal product, preferably a rod or a hot rolled wire, comprises at least one robotized articulated arm and an acquisition and control unit including a video monitoring system and a processing system configured to acquire and process a sequence of frames of said metal product and to send actuation signals to said robotized articulated arm in order to manipulate or cut a part of said metal product.

[0022] According to one aspect of the invention described above, the processing system implements artificial intelligence configured to at least a) recognize a terminal portion of the metal product, and b) identify whether the terminal has a macroscopic defect, such as a shape defect or a cobble defect.

[0023] The processing system can further selectively manage movement of the robotic articulated arm (manipulation, repositioning, or cutting with possible subsequent repositioning) in a substantially linear movement along the feed direction of the metal product based on the identified defects.

[0024] This allows, in an automated manner, to manipulate the workpiece in a targeted manner only if there is a defect in the tail, and in this case to perform a specific operation based on the type of defect, thereby optimizing the robot unit, the equipment to which it is attached, and the corresponding method, thereby reducing the waste during processing and improving the quality of the final product.

[0025] According to another aspect of the invention, the robotic articulated arm can be moved in at least six degrees of freedom, more preferably seven degrees of freedom, allowing the manipulation of metal products, particularly but not limited to ends, regardless of their position, shape and structure.

[0026] According to another aspect of the invention, the video monitoring system comprises at least one video camera located on a fixed support, on the robotic articulated arm, or on the manipulative head of the arm, thereby allowing the product to be imaged on the fly, i.e. tracked, improving image quality and accuracy.

[0027] According to another aspect of the invention, the robot unit further comprises an interception member, which is vertically movable between a rest position below the plane of movement of the metal product and not interfering with the metal product, and an interception position incident on the plane of movement to temporarily stop a portion of the metal product intended to be manipulated / cut by the robotized articulated arm, which allows stretching of the metal product coil to facilitate grasping, manipulation, possible cutting and extraction of the portion of the product.

[0028] The invention also relates to a computerized method for controlling a moving metal product, preferably a rod or a hot rolled wire, comprising acquiring, by means of a video monitoring system of an acquisition and control unit, a sequence of frames of said metal product processed by a processing system, transmitting actuation signals to a robotized articulated arm and driving said robotized articulated arm on the basis of said actuation signals.

[0029] According to one aspect of the invention, the method provides for artificial intelligence implemented in the processing system to at least a) recognize an edge of the metal product, and b) identify whether the edge has a macroscopic defect, e.g. a shape defect or a cobble defect.

[0030] According to another aspect of the invention, based on the type of defect detected, the robotic articulated arm is moved to reposition the tail portion on the fly by means of its gripping tool or moved to cut the tail portion on the fly by means of its cutting tool.

[0031] According to another aspect of the invention, the artificial intelligence has a recognition algorithm for identifying a first point and a second point for each frame of the sequence of frames, the first point and the second point defining the area between them as a control zone, and based on the width and position of the control zone, identifying the frame as either a center frame or a tail frame of the metal product, or an empty frame, thus obtaining a macroscopic identification of the entire length of the product.

[0032] According to another aspect of the invention, the artificial intelligence comprises a convolutional neural network comprising an input level or layer receiving each time one frame of the sequence of frames, one or more convolutional levels or layers capable of identifying macroscopic geometric and visual characteristics of the metal product, at least one fully connected level configured to classify the frames based on information received from at least one of the one or more convolutional levels or layers, and an output level or layer defined by a binary class vector comprising a first class and a second class, both classes being characterized by respective confidence indices for the absence of defects and the presence of defects.

[0033] According to another aspect of the invention, if the confidence index associated with the second class exceeds a predetermined threshold, the convolutional network establishes that the analyzed frame is associated with a metal product having a defect and generates a specific actuation signal to send to the robotized articulated arm for correction of the defect.

[0034] According to another aspect of the invention, the at least one fully connected level classifies each of the frames into a class of belonging comprising tail frames with shape defects and tail frames with cobble defects based on the information received from at least one of the one or more convolutional levels or layers. The present invention also relates to an apparatus for the production of metal reels, the apparatus comprising a winding unit having a laying head and configured to receive a metal product in a linear form and form it into a coil, an automatic reel-forming unit capable of receiving the metal product formed into a coil and forming it into a reel, and a roller way configured to move the metal product formed into a coil from an initial end corresponding to a position where the winding unit is located to a terminal end corresponding to a position where the automatic reel-forming unit is located.

[0035] According to one aspect of the invention, the apparatus further includes a robotic articulated arm disposed at a position on the roller way that is included between the starting end and the terminal end.

[0036] One aspect of the invention relates to a method for producing metal reels, whereby a winding unit receives a metal product in wire form, a laying head forms it into a coil, an automatic reel forming unit receives the coiled metal product and forms a reel, and a roller way moves the coiled metal product from a start corresponding to a position where the winding unit is located to a finish corresponding to a position where the automatic reel forming unit is located.

[0037] According to one aspect of the invention, in the method, a robotic articulated arm moves laterally to a position included between the starting and ending ends of the roller way to manipulate, reposition, or cut the metal workpiece. These and other aspects, features and advantages of the present invention are explained, by way of non-limiting example, in the following illustrative embodiments taken in conjunction with the accompanying drawings, in which: [Brief description of the drawings]

[0038] [Figure 1] FIG. 1 is a perspective view of a robot unit for controlling moving metal products according to the present invention; [Diagram 2] FIG. 2 is an enlarged detail view showing an interception member capable of intercepting a moving metal product. [Diagram 3] FIG. 1 is a schematic diagram illustrating the operation of artificial intelligence to identify defects that may be present in a metal product and its tail. [Figure 4] 1 is a top view of an apparatus for manufacturing metal reels according to the present invention; [Diagram 5] 1 is a side view of an apparatus for manufacturing metal reels according to the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0039] It must be made clear that the expressions and terminology used in this specification, as well as the figures of the accompanying drawings similarly described, have the sole function of better illustrating and explaining the invention, the scope of which is defined by the claims, and that their function is to provide a non-limiting example.

[0040] To facilitate understanding, the same reference numbers have been used wherever possible in the drawings to identify identical common elements. It will be understood that elements and features of one embodiment may be suitably combined or incorporated into other embodiments without further description.

[0041] With reference to FIG. 1, this shows a robot unit 10 for controlling a metal product P according to the invention, which may be a smooth or ribbed rod or a metal wire moving in a substantially straight feed direction X.

[0042] The metal product P is laid down in a feed direction X on the basis of a series of coils arranged in an offset configuration, using a machine of known type, for example a coil forming head.

[0043] The metal products P are moved on a movement plane Z which is substantially horizontal but which may possibly be inclined upwards or downwards, for example in the feed direction X.

[0044] After its movement on the plane of movement Z, the metal products P can be packed into reels or skeins 200 .

[0045] The unit 10 comprises at least one robotic articulated arm 11 having an operating head 12 each equipped with a gripping tool 13 configured to manipulate the metal product P and a cutting tool 14 for cutting the metal product P, i.e. for separating a particular portion of the metal product P.

[0046] This particular portion may be the end or tail portion P1 (see FIG. 3) or, for example, a sample portion located at a position included between the tail portion P1 and the beginning or head portion P2 of the metal product P. Between the head portion P2 and the tail portion P1 is a central portion P3 (see FIG. 3) having a uniform, laterally aligned, coiled, substantially planar structure.

[0047] The robot articulated arm 11 can be moved according to preferably six degrees of freedom, but additionally seven degrees of freedom. For this purpose, the robot unit 10 can be equipped with a slider or rail 15, for example parallel to the feed direction X, on which the arm 11 is mounted to slide, so that the arm 11 can follow the product P to be processed.

[0048] The robot unit 10 may also be provided with an interception member 23 (see Figure 2) which is vertically movable between a rest position located below the plane of movement Z of the metal product P and not interfering with the metal product P, and an interception position incident on the plane of movement Z to hold a coil of metal product P which is intended to be manipulated / cut by means of the robot articulated arm 11.

[0049] The robot unit 10 may also comprise a depositing unit 24 comprising an arm 24a on which the portion of the metal product P to be removed is deposited.

[0050] The robot unit 10 has an acquisition and control unit equipped with a video monitoring system 18 for acquiring video sequences and / or sequences of digital images or frames, in particular sequences of frames 19 of the metal product P being processed (see FIG. 3).

[0051] The sequence of frames 19 may be continuous, or may be acquired at regular or irregular intervals.

[0052] The video monitoring system 18 may comprise at least one video camera 18a, preferably located on the robotic articulated arm 11 (see Figures 1 and 2).

[0053] A video camera 18a may be positioned on the scanning head 12 and point in a direction substantially perpendicular to the feed direction X.

[0054] In this case, the sequence of frames 19 may be acquired "with tracking" through both the inherent mobility of the robot articulated arm 11 and the translational movement of the arm 11 along rails 15, if provided.

[0055] Alternatively, the video camera 18a can be positioned along or on the plane of movement Z to guide the robotic articulated arm 11 at a fixed position relative to the metal workpiece P being processed.

[0056] The acquisition and control unit also includes a processing system 21 which is connected to the video monitoring system 18 (see FIG. 3).

[0057] The processing system 21 processes the sequence of frames 19 and sends actuation signals S to the arm 11 to manipulate, reposition and / or sever the tail portion P1 or the sample portion.

[0058] The processing system 21 implements artificial intelligence configured to at least (a) recognize the termination P1 and (b) identify whether the termination P1 is defective or not.

[0059] The defects in question may be of the macro type, i.e. independent of microscopic surface defects, but may be shape, geometric or cobble defects, where a portion of the metal product P may have a configuration that "exits" from the moving surface Z.

[0060] The artificial intelligence is therefore able to identify a particular portion of a metal product P along its length in the feed direction X in accordance with the presence of a defect at its end P1.

[0061] This artificial intelligence is therefore particularly optimized since it is able to very quickly first identify the portion of interest, for example the sequence of frames 19 which includes the end portion P1, and subsequently recognise the presence of possible macroscopic defects.

[0062] In comparison to prior art solutions, these operations need to be performed almost instantaneously, since the detection of a defect is followed by actuation of the arm 11 (manipulation, repositioning, or cutting with possible subsequent repositioning) advantageously without interruption or slowdown.

[0063] Furthermore, the macroscopic analysis performed by the artificial intelligence of the present invention presupposes a "global" comparison along the entire length of the metal product P and therefore cannot be assimilated to a microscopic surface inspection.

[0064] According to a possible implementation, the artificial intelligence as described above may also be configured to c) classify the type of defect, for example a shape defect or a cobble defect.

[0065] In the particular case of shape defects, the processing system 21 is configured to identify the position and geometrical characteristics of the terminal portion P1 relative to the homogenous portion P3, and to recognise, for example, possible variations in the position relative to the homogenous portion P3.

[0066] The processing system 21 may include a CPU 21a and a storage module 21b that may be connected to the programmable CPU 21a and may be, for example, a commercially available random access memory (RAM), a read only memory (ROM), a floppy disk, a hard disk (HARD DISK), non-volatile memory (NOR FLASH and NAND FLASH), mass memory, or other form of digital storage or electronic database.

[0067] The storage module 21b can store one or more artificial intelligence based recognition algorithms, such as, for example, neural networks, SVM (Support Vector Machines), neuro-fuzzy networks, genetic algorithms, and the like.

[0068] The CPU 21a may be any type of microprocessor or processor capable of executing the recognition algorithms as described above.

[0069] In a possible embodiment, in order to take advantage of the level of evolution and sophistication reached by image recognition neural networks, algorithms based on state-of-the-art deep convolutional networks (CNN or DCNN) designed for image classification can be stored in the storage module 21b. These architectures are usually composed of one or more convolutional layers and can end with a module of fully connected layers. The cascaded application of convolutional layers allows the extraction of local features of the acquired image at different scale levels. The convolutional neural network then processes the starting image to provide an alternative representation of it. Finally, this output is processed and can be used in several tasks such as image classification and object recognition.

[0070] Examples of this type of architecture are e.g. EfficientNet, MobileNet, ResNet, or VGG16.

[0071] In the embodiment described here, the artificial intelligence is implemented on a storage module 21b comprising a convolutional neural network 25 comprising an input level or layer 26 which receives each time a frame 19a, 19b, 19c of the sequence of frames 19, represented as a matrix of pixels, that the convolutional neural network 25 has to analyze.

[0072] The convolutional neural network 25 also includes one or more convolution levels 27 that are capable of identifying visual features, such as curves, lines, edges, etc. depicted in the frames 19a, 19b, 19c.

[0073] Frames 19a, 19b, 19c are then processed sequentially by the layers of a convolutional network: each filter of the nth convolutional level 27, which can be identified as a scalar matrix containing numerical values ​​or weights, is applied to all the sectors that make up the output of the previous level according to a predetermined step, resulting in a filtered image, also known to those skilled in the art as an activation map.

[0074] Between the nth convolution level 27 and the nth+1 convolution level 27, there may be non-linear levels or layers whose function is to introduce non-linearity into the system that is computing substantially linear operations during the convolution level 27. After some non-linear levels, there may be so-called "pooling" levels or layers. There may be additional regularization layers, such as so-called "batch normalization" layers.

[0075] The possibility of composing multiple functional components of a convolutional network allows for a wide variety of possible architectures. It is common practice to use architectures studied in the literature that have demonstrated good predictive capabilities on similar tasks (e.g., EfficientNet-type architectures).

[0076] The convolutional neural network 25 comprises at least one fully connected level or layer 28 downstream of the convolutional layer 27 and is configured to classify the filtered image based on the information received from the last previous layer. The fully connected layer 28 comprises a set of nodes trained to recognize whether a metal product P identified in the filtered image is defective or not.

[0077] Downstream of the fully connected level 28 is an output level 29 defined by a binary class vector containing a first class ("defective products") and a second class ("good products"), each characterized by a reliability index PD1 and PD2, respectively.

[0078] If the confidence index PD2 for the second class exceeds a predefined threshold PD, the convolutional neural network 25 establishes that the just analyzed frame 19a, 19b, 19c contains a defective metal product P and generates and sends a specific actuation signal S to the robotic articulated arm 11 in order to correct the defect.

[0079] The convolutional neural network 25 is first trained in a manner known per se, using a "training set" comprising sample images showing different configurations of the coils characterizing the metal product P. In this way, the weights of the convolutional network 25 will be appropriately trained. If the convolutional network 25 even partially replicates an architecture known in the literature (e.g. ResNet, EfficientNet), it is possible to train the weights of the network starting from a configuration of a network trained on a different dataset, usually trained on a larger dataset, by a technique known as "transfer learning".

[0080] According to one embodiment, before transmitting the frames 19a, 19b, 19c to the input level 26 of the convolutional neural network 25, a recognition algorithm 30 can be applied in advance, the recognition algorithm 30 being capable of identifying one or more frames 19b of the series of frames in which the tail, i.e. the end, of the metal product P is visible.

[0081] In particular, the recognition algorithm 30 is capable of recognizing a first point A and a second point B on each frame 19a, 19b, 19c of the series of frames 19, which define a control strip therebetween, and where, according to the width and position of the control strip, it is possible to identify whether the frame in question is a central frame 19a or a tail frame 19b or an empty frame 19c of the metal product P.

[0082] The central frame 19a is a frame in which the central portion P3 is visible, the tail frame 19b is a frame in which the end portion P1 is visible, and the empty frame 19c is a frame in which the metal product P is already completed and is therefore invisible.

[0083] A reference system (x,y) can be associated with each frame, where the abscissa x is parallel to the feed direction X of the metal product P and the ordinate y is orthogonal to the feed direction X. Points A and B, shown as dotted lines in FIG. 3, are projected onto the abscissa x and define a control strip between them.

[0084] Only tail frames 19b can be analyzed by the convolutional neural network 25. Alternatively, the convolutional neural network 25 can be configured to accept any frame 19a, 19b, 19c at input, and recognition is performed directly within it.

[0085] According to one possible embodiment, the convolutional neural network 25 may comprise an additional fully connected level configured to classify the filtered images into their respective classes including tail frames 19b with shape defects and tail frames 19b with cobble defects based on the information received from the last previous level.

[0086] According to some embodiments, both during the training step of the convolutional neural network 25, and possibly thereafter, a verification process is provided in which an operator records: a) the type of frames 19a, 19b, 19c, b) for tail frames 19b, whether there are defects, and c) for tail frames 19b that have defects, the type of defect.

[0087] According to some embodiments, the processing system 21 may comprise an automation module 21c associated with the arm 11 and configured to receive the information processed by the convolutional neural network 25 and manage the movement of the robotic articulated arm 11 as required.

[0088] In the event that the convolutional neural network 25 identifies a defect in the metal product P being processed, the movement of the robot articulated arm 11 repositions the tail portion P1 to align with the central portion P3, either by means of the gripping tool 13 in the case of a shape defect, or by means of the cutting tool 14 in the case of a cobble defect.

[0089] The acquisition and control unit can also be automatically or manually managed to command the movements of the robotic articulated arm 11 to perform the sampling of the metal product P. In this case, the recognition algorithm 30 can be used to identify the central frame 19a corresponding to the central portion P3 along which the sample portion is to be taken.

[0090] 4-5, in some embodiments, an apparatus 100 for producing a metal reel 200 from, for example, a hot rolled metal product P is shown.

[0091] The apparatus 100 comprises a winding unit 110 having a winding head 111 and configured to receive linear metal products P and form them into a coil, and an automatic reel forming unit 113 capable of receiving the metal products P formed into coils and grouping them into a reel 200.

[0092] The apparatus 100 comprises a roller way 112 or cooling belt on which the metal product P formed into a coil is placed. The roller way 112 has a starting end 112b corresponding to where the winding unit 110 is located and a terminal end 112b corresponding to where the automatic reel forming unit 113 is located.

[0093] The roller way 112 defines the movement plane Z of the metal product P and may integrate a cooling system to bring the temperature of the metal product P from approximately 1000° C. at the exit from the winding unit 110 to approximately 500° C. at the end of the roller way 112.

[0094] The apparatus 100 includes a robot unit 10 disposed at a position included between a start end 112 a and a finish end 112 b of a roller way 112 .

[0095] The robot unit 10 is disposed to the side of the roller way 112 so that the robot articulated arm 11 can cooperate with the processing of the metal product P.

[0096] The rails 15 of the robot unit 10 can be arranged parallel to the roller ways 112 .

[0097] The interception member 23 may be configured to be inserted between the rollers of the roller way 112 to intercept one or more coils that the robotic articulated arm 11 must manipulate or cut.

[0098] Some embodiments also relate to a method for manufacturing a metal reel 200, in which a winding unit 110 receives a linear metal product P, a laying head 111 forms it into a coil, an automatic reel forming unit 113 receives the coiled metal product P and forms the metal reel 200, and a roller way 112 moves the coiled metal product P from a starting end 112a to a terminal end.

[0099] The robot unit 10 operates at positions included between the starting end 112a and the ending end 112b to manipulate, reposition or cut the metal product P as previously described.

[0100] It will be apparent that modifications and / or additions of components can be made to the robotic unit and method for controlling a moving metal product, and the apparatus and method for manufacturing metal reels described hereinbefore, without departing from the field and scope of the present invention.

[0101] It is also clear that although the invention has been described with reference to some particular examples, a person skilled in the art can certainly realise many other equivalent forms of robotic units and methods for controlling moving metal products, and many other equivalent forms of devices and methods for manufacturing metal reels, having the features set out in the claims and therefore all falling within the field of protection defined by the claims.

[0102] In the following claims, references in parentheses have the sole purpose of improving readability and shall not be considered as limiting elements with regard to the field of protection defined by the same claims.

Claims

1. A robotic unit (10) for controlling a moving metal product (P), preferably a rod or a hot rolled wire, comprising: At least one robotic articulated arm (11); an acquisition and control unit, a video monitoring system (18); a processing system (21) configured to acquire and process a sequence of frames (19) of the metal product (P) and to send actuation signals (S) to the robotized articulated arm (11) in order to manipulate or cut a portion of the metal product (P); an acquisition and control unit including: The processing system (21) includes at least a) recognizing the tail portion (P1) of the metal product (P); b) implementing an artificial intelligence configured to identify whether said tail portion (P1) has macroscopic defects; the processing system (21) is configured to selectively manage movement of the robotized articulated arm (11) based on the identified defects; A robot unit (10) characterized in that, based on the type of defect detected, the robotized articulated arm is moved to reposition the tail part (P1) on the fly by means of its gripping tool (13) or to cut the tail part (P1) on the fly by means of its cutting tool (14).

2. 2. The robot unit (10) according to claim 1, characterized in that the robotized articulated arm (11) can be moved in at least six degrees of freedom.

3. 2. The robot unit (10) according to claim 1, characterized in that the video monitoring system (18) comprises at least one video camera (18a) located on the operating head (12) of the robotized articulated arm (11).

4. 2. The robot unit (10) according to claim 1, characterized in that the video monitoring system (18) comprises at least one video camera (18a) located on a fixed support.

5. Further comprising an interception member (23), 2. The robot unit (10) according to claim 1, characterized in that the interception member is vertically movable between a rest position, located below the plane of movement (Z) of the metal product (P) and not interfering with the metal product, and an interception position, where it enters the plane of movement (Z) to temporarily stop a part of the metal product (P) intended to be manipulated / cut by the robotized articulated arm (11).

6. 1. A computerized method for controlling a moving metal product (P), preferably a rod or a hot rolled wire, comprising: acquiring, by means of a video monitoring system (18) of an acquisition and control unit, a sequence (19) of frames of said metal product (P) which are processed by a processing system (21); Sending an actuation signal (S) to a robotic articulated arm (11) and driving the robotic articulated arm (11) based on the actuation signal (S); The method comprises at least a) recognizing the tail portion (P1) of the metal product (P); b) identifying whether the tail portion (P1) has macroscopic defects; 10. A method according to claim 9, wherein said robotic articulated arm is provided with an artificial intelligence implemented in said processing system (21) such that, based on the type of defect detected, said robotic articulated arm is moved to reposition said tail part (P1) on the fly by means of its gripping tool (13) or to cut said tail part (P1) on the fly by means of its cutting tool (14).

7. 7. The method according to claim 6, wherein the artificial intelligence has a recognition algorithm (30) for identifying a first point (A) and a second point (B) for each frame (19a, 19b, 19c) of the sequence of frames (19), the first point (A) and the second point (B) defining the area between them as a control zone, and identifying the frame as either a central frame (19a) or a tail frame (19b) of the metal product (P), or an empty frame (19c), based on the width and position of the control zone.

8. The artificial intelligence comprises a convolutional neural network (25); The convolutional neural network (25) an input level or layer (26) receiving each one of the frames (19a, 19b, 19c) of said sequence of frames (19); one or more convolution levels or layers (27) that function to identify macroscopic geometric and visual characteristics of said tail portion (P1); at least one fully connected level (28) configured to classify the frame (19a, 19b, 19c) based on information received from at least one of the one or more convolutional levels or layers (27); an output level or layer (29) defined by a binary class vector including a first class and a second class, both classes being characterized by respective confidence indices (PD1, PD2) for the absence and presence of a defect; 8. The method of claim 6 or claim 7, comprising:

9. The method according to claim 8, characterized in that the confidence index (PD2) associated with the second class for the frame (19a, 19b, 19c) is analyzed, and if the confidence index (PD2) exceeds a predetermined threshold (PD), the convolutional network (25) establishes that the analyzed frame (19a, 19b, 19c) is associated with a metal product (P) having a defect and generates a specific actuation signal (S) to be sent to the robotized articulated arm (11) for correcting the defect.

10. 9. The method of claim 8, wherein the at least one fully connected level (28) is configured to classify each of the frames (19a, 19b, 19c) into a class of belonging comprising tail frames (19b) with shape defects and tail frames (19b) with cobble or clogging defects based on the information received from at least one of the one or more convolutional levels or layers (27).

11. 10. The method of claim 9, wherein the at least one fully connected level (28) is configured to classify each of the frames (19a, 19b, 19c) into a class of belonging comprising tail frames (19b) with shape defects and tail frames (19b) with cobble or clogging defects based on the information received from at least one of the one or more convolutional levels or layers (27).

12. An apparatus (100) for the production of metal reels (200), comprising: a winding unit (110) having a winding head (111) and configured to receive the metal product (P) in linear form and form it into a coil; an automatic reel-forming unit (113) capable of receiving the metal product (P) formed into a coil shape and forming it into a reel (200); a roller way (112) configured to move the metal product (P) formed into a coil shape from a start end (112a) corresponding to a position where the winding unit (110) is disposed to a finish end (112b) corresponding to a position where the automatic reel forming unit (113) is disposed; Equipped with An apparatus (100) characterized in that it is arranged at a position included between the starting end (112a) and the ending end (112b) of the roller way (112), and further comprises the robot unit (10) according to any one of claims 1 to 5.

13. A method for producing a metal reel (200), comprising the steps of: The winding unit (110) receives the wire-shaped metal product (P), and the laying head (111) forms it into a coil; An automatic reel forming unit (113) receives the metal product (P) formed into a coil shape and forms a reel (200); A roller way (112) moves the metal product (P) formed into a coil shape from a start end (112a) corresponding to a position where the winding unit (110) is disposed to a finish end (112b) corresponding to a position where the automatic reel forming unit (113) is disposed; 6. A method according to claim 1, wherein the robot unit (10) according to any one of claims 1 to 5 is moved laterally to a position included between the starting end (112a) and the ending end (112b) of the roller way (112) in order to manipulate, reposition or cut the metal product (P).

Images