Inspection system for controlling the joint ends of a sheet product

Abstract

Inspection system for checking joint ends of a sheet product An inspection system 100 for checking a distance between joint ends 102 of a sheet product 104 having predetermined length, width and thickness parameters is provided. A marking unit prints transverse marks (T1, T2) on a first end 106 and a second end 108 at a predetermined distance (D1, D2) from first edge portions (110, 116), second edge portions (112, 118) and center portions (114, 120). A detection unit acquires images formed from pixels associated with the distance between the transverse marks (T1, T2).A processing unit applies a geometric mesh to surfaces of the sheet product 104 representing nodes identifying the first end 106, the second end 108, the first edge portions (110, 116), the second edge portions (112, 118) and the central portions (114, 120). A junction is defined based on calculated distances (C1, C2, C3, C4) between nodal points (T1.1, T1.2, T1.3, T1.4) on the transverse mark T1 and nodal points (T2.1, T2.2, T2.3, T2.4) on the transverse mark T2. Figure for abstract: Fig. 1.

Description

Title of the invention: Inspection system for controlling the joint ends of a sheet product Technical field

[0001] The invention relates to an inspection system for inspecting joint ends of a sheet product and, in particular, for inspecting a distance between the joint ends of the sheet product in order to identify joint defects in association with material characteristics. Context

[0002] There are several means known in the current state of the art for controlling joint ends of a sheet product, for example, by measuring the available rubber between the ends of a tread before joining the ends. The available rubber defines a gap, an overlap or a perfect join between the ends. Control can also be carried out by measuring differences in height in thickness between joint ends of the tread to avoid a staggered join. Therefore, for a perfect join of the tread, it is important to have symmetrical ends with geometric characteristics including a match in thickness, width and profile to avoid gaps, overlaps or offsets between the ends of the tread.

[0003] During tire manufacturing, characterizing the geometry of rubber products (such as the tread in a green or unvulcanized state) can pose difficulties. Indeed, its plastic properties in the green state do not easily allow for the prediction of the material characteristics based on precise geometric characteristics. The difficulty in accurately characterizing the junction ends arises from the plastic and irregular deformation of the ends during cutting. This characterization also differs from one tire size to another since the tread profile (in particular its thickness and width) varies considerably. It is, therefore, difficult to automate the characterization of the junction with gap, overlap and offset defects.Therefore, characterizing the expansion or compression of the tread around the junction region is essential to measure the gap or overlap of offset defects. The expansion or compression of green rubber results mainly from compounding properties including the processing time to prepare a mixture, environmental factors including temperature and humidity, and various mechanical processing steps. during the definition of the tread shape.

[0004] Transverse marks or marking incisions are sometimes added to tread joint ends to determine whether the joint is acceptable after the tread has been wound onto a tire building drum (see, e.g., US10913229, JP4537624, and KR10-0746109). Furthermore, detection of an accurate tread position may be performed prior to joining the ends by using position sensors on an outer circumferential surface of a tire building drum (see, e.g., US2017 / 0074645).

[0005] According to the present invention, the conformity of a junction is improved by means of an inspection system which controls a distance between the transverse marks and the junction ends of a tread by taking into account material properties, in particular the expansion and compression of the rubber in a green (i.e. unvulcanized) state. The variation of these material properties, in particular in a plastic state due to irregular deformations, is detected using a machine vision system employing neural networks and / or artificial intelligence tools. Summary of the invention

[0006] An inspection system for monitoring a distance between joining ends of a sheet product having a profile defined by parameters of a predetermined length, a predetermined width and a predetermined thickness, the joining ends including a first end, a second end, first edge portions, second edge portions and central portions defining end surfaces of the joining ends along a longitudinal axis, the inspection system comprising:

[0007] - at least one marking unit configured to print trans marks versales (Ti, T2) on the first end and the second end at a predetermined distance (Di, D2) from the first edge portions, the second edge portions and the central portions;

[0008] - at least one detection unit configured to acquire one or more images in order to identify the transverse marks (Tb T2) on the sheet product, the images being formed from pixels associated with the distance of the transverse marks (Tb T2) relative to the first edge portions, the second edge portions and the central portions;

[0009] - at least one processing unit comprising one or more software or al algorithms or transformation models configured to process the product images in sheet form with a geometric mesh allowing to define one or more points nodal points (Tu, TL2, TL3, TL4) on the transverse mark Ti on the first end and one or more nodal points (Eu, EL2, EL3, EL4) on the first edge part, the second edge part and the central part of the first end;

[0010] - at least one processing unit comprising one or more software or al transformation algorithms or models configured with a geometric mesh for defining one or more nodal points (T2 b T2 2, T23, T24) on the transverse mark T2 on the second end and one or more nodal points (E2 b E2 2, E2 3, E2 4) on the first edge portion, the second edge portion and the central portion of the second end; and

[0011] - the processing unit being configured to define a junction on the basis of calculated distances (Ci, C2, C3, C4) between the nodal points (Tu, TL2, TL3, TL4) on the transverse mark Ti of the first end and the nodal points (T2 b T22, T23, T24 ) on the transverse mark T2 of the second end.

[0012] In some embodiments of the inspection system, the defined junction is a perfect junction (P) without defects, if the distances calculated (Ci, C2, C3, C4) by the processing unit between the nodal points (Tu, TL2, TL3, Ti,4) on the transverse mark Ti of the first end and the nodal points (T2.b T2 2, T2 3, T24) on the transverse mark T2 of the second end are equal to a sum of the predetermined distance (Di+D2).

[0013] In some embodiments of the inspection system, the defined junction is an imperfect junction due to a gap defect (G), if the distances calculated (Ci, C2, C3, C4) by the processing unit between the nodal points (Tu, TL2, TL3, Ti 4) on the transverse mark Ti of the first end and the nodal points (T2 b T22, T23, T24 ) on the transverse mark T2 of the second end are greater than a sum of the predetermined distance (Di+D2).

[0014] In some embodiments of the inspection system, the defined junction is an imperfect junction due to an overlap defect (O), if the distances calculated (Cb C2, C3, C4) by the processing unit between the nodal points (Tu, TL2, TL3, Ti 4) on the transverse mark Ti of the first end and the nodal points (T2 b T2.2, T23, T24) on the transverse mark T2 of the second end are less than a sum of the predetermined distance (Di+D2).

[0015] In some embodiments of the inspection system, the defined junction is an imperfect junction due to an offset defect (J), if the nodal points (Tu, Ti.2, Ti.3, TL4) on the transverse mark Ti and the nodal points (Eu, EL2, Eu, EL4) on the edge portions of the first end are not aligned with the nodal points (T2.b T2.2, T2.3, T24) on the transverse mark T2 and the nodal points (E2.i, E2 2, E2 3, E 24) on the edge portions of the second end.

[0016] In some embodiments of the inspection system, the defined junction is an imperfect junction due to a combination of at least one of gap defects (G), overlap defects (O), perfect junction (P), or offset defects (J).

[0017] In some embodiments of the inspection system, the processing unit comprising one or more software or algorithms or transformation models is trained to predict a quality of subsequent junctions or anticipate defects based on identified defects comprising at least one of gap defects (G), overlap defects (O), offset defects (J) or the perfect junction (P).

[0018] In some embodiments of the inspection system, the processing unit comprises an image processing module implementing one or more transformation models defined as point object detection based on the CenterNet neural network architecture.

[0019] In some embodiments of the inspection system, the processing unit comprises an image processing module implementing one or more transformation models comprising an encoder-decoder configured to adapt the acquired images to a lower or higher resolution.

[0020] In some embodiments of the inspection system, the processing unit comprises an image processing module configured to apply a geometric correction to compensate for non-planarity of a measurement region.

[0021] In some embodiments of the inspection system, a distance sensor is placed at a level of the detection unit to measure a distance between the detection unit and the surface of the sheet product to convert a pixel area into a distance with adaptation to the variation in diameter of the surfaces of the sheet product. Brief description of the drawings

[0022] The nature and various advantages of the invention will become more apparent upon reading the following detailed description in conjunction with the accompanying drawings, in which like reference numerals denote like parts throughout the drawings, and in which:

[0023] [Fig.l] represents a schematic view of an embodiment of an inspection system, according to the present invention.

[0024] [Fig.2] shows a schematic view of one embodiment of joint ends in a perfect joint, according to the present invention.

[0025] [Fig. 3] shows a perspective view of an embodiment of joint ends in a perfectly joined state of [Fig. 2], according to the present invention.

[0026] [Fig.4] shows a schematic view of an embodiment of joint ends in an imperfect joint due to a gap defect, according to the present invention.

[0027] [Fig.5] shows a perspective view of an embodiment of joint ends in an imperfect joint due to a gap defect of [Fig.4], according to the present invention.

[0028] [Fig.6] shows a schematic view of an embodiment of joint ends in an imperfect joint due to an overlap defect, according to the present invention.

[0029] [Fig.7] shows a perspective view of an embodiment of joint ends in an imperfect joint due to an overlap defect of [Fig.6], according to the present invention.

[0030] [Fig.8] shows a schematic view of an embodiment of joint ends in an imperfect joint due to a combination of gap defect, overlap defect and perfect joint, according to the present invention.

[0031] [Fig.9] shows a schematic view of an embodiment of junction ends in an imperfect junction due to an offset defect, according to the present invention.

[0032] [Fig. 10] shows a schematic view of an embodiment of a tire used to illustrate a geometric correction in order to compensate for non-flatness of a measurement region, according to the present invention. Detailed description

[0033] The present invention relates to an inspection system which monitors a distance between joint ends of a tread. The inspection system is capable of monitoring various profiles of products having different thicknesses and widths, while taking into account material properties including expansion and compression of the rubber, particularly arising from plastic properties or uneven deformations of the rubber in the green or unvulcanized state. The plastic properties of the rubber in the green state do not easily allow for the prediction of material characteristics by considering only the geometry of the joint ends or only the tread profile.The expansion or compression of rubber in the green state results primarily from compounding properties including processing time to prepare a mix, environmental factors (including temperature and humidity), and various mechanical processing steps during tread shape formation. This also differs from one tire size to another since the tread profile (including its thickness and width) varies considerably during production. In several examples, the sheet product may be a tire tread, rubber tracks, or . sheet products of a similar nature. For the purposes of the present invention, the sheet product may be referred to as a tread, however it will be understood that the invention is not limited to tire treads. The inspection method of the invention makes it possible to join tread ends efficiently and with satisfactory quality.

[0034] The disclosed inspection system inspects two joining ends of a tread forming a joined tread. Generally, treads are longitudinal portions of a tread surface for application to a tire carcass, the tread having a predetermined length, a predetermined width, and a predetermined thickness. The tread may include a tread having a tread pattern on a ground-contacting side of the tread. The tread pattern may include any known tread features, for example, including blocks and / or ribs separated by grooves and / or sipes. The joining ends of a single tread may be joined to form a continuous tread. As a further example, separate treads (i.e., segments, sections or lengths of tread) may be joined to form a single, integral tread comprising multiple joined tread segments. Although the present invention relates to a tread in the green state, it might also be possible to apply it to a tread that is at least partially vulcanized.

[0035] Each joining end of the tread includes an end surface (or "joining end surface") that forms the end of the tread. The joining end surface includes a cross-sectional surface of the tread extending laterally across the width of the tread and having a height that extends across the thickness of the tread. The joining end surface may extend across the width of the tread in a direction normal to the longitudinal direction of the tread or at any other oblique angle to the longitudinal direction. Further, the joining end surface may extend laterally along a linear or non-linear path.Similarly, the height of the joining end surface may extend across the tread thickness in any direction, including for example a direction normal to the longitudinal direction of the tread, and along any linear or non-linear path. Since the tread may include a tread pattern extending through the tread thickness, the tread end surface may include voids disposed within the tread thickness. rolling and arranged inwardly relative to an outer profile in cross section.

[0036] The longitudinal location where the joining end is formed along each tread portion is chosen to provide a joining end having a profile that generally matches the profile of the other joining end to which it is to be joined. For example, it is possible to select and form, or otherwise provide, joining ends that generally match (i.e., their cross-sections, profiles, and / or perimeters generally match). This provides a tread join that is consistent with adjacent portions or details of the joined tread, and with the overall tread pattern of the joined tread.In other words, by forming the joining ends such that they generally match, the joined or assembled tread may include a tread pattern that is not substantially disturbed at the join where the tread details of opposing joining ends are generally aligned with respect to a height (i.e., thickness) and width of the tread.

[0037] The arrangement of the joining ends results in a joined or assembled tread in which the joined tread portions extend generally in the same longitudinal direction, which may extend linearly as a tread ribbon or annularly as the tread. Alternatively, determining the location along any length of tread at which to form the joining end may be selected not only to generally match the joining end of the tread to be joined, but also to select a location that allows the tread to properly join.

[0038] Referring to [Fig.l], an inspection system 100 according to the present invention is shown. The inspection system 100 is capable of monitoring a distance between joint ends 102 of a tread 104. The joint ends 102 of the tread 104 include a first end 106 and a second end 108 of predetermined profile parameters including, but not limited to, length, width, and thickness. Further, each of the first end 106 and the second end 108 of the tread 104 includes end surfaces defining first edge portions (110, 116), second edge portions (112, 118), and center portions (114, 120).The inspection system 100 comprises at least one marking unit configured to print transverse marks (Tb T2) on the first end 106 and the second end 108 at a predetermined distance (Db D2) from the first edge portions (110, 116), the second . edge portions (112, 118) and center portions (114, 120) after manufacturing and cutting the tread 104 (see [Fig. 2]). Further, the tread 104 may include one or more tread segments.

[0039] The transverse marks (Tb T2) may be printed along the entire width of the tread 104 as shown in [Fig. 1]. The predetermined distance (Db D2) between the transverse marks (Tb T2) and the joining ends 102 are set by a cutting blade and the marking unit. The marking unit may include one or more marking rollers configured to apply transverse marks (Tb T2) at the predetermined distance (Db D2) between the transverse marks (Tb T2) prior to cutting the tread 104 into segments by the cutting blade. Although two transverse marks (Tb T2) appear on the tread 104, those skilled in the art understand that multiple transverse or linear marks may be formed for the purpose of interpreting the distance between the marks and the joining ends (102).

[0040] The inspection system 100 further includes a detection unit 122 that employs one or more sensors (not shown) that acquire information regarding the physical environment around the tread. In the following description, the terms “sensor,” “photographic equipment,” “camera,” and “optical sensor” may be used interchangeably and may refer to one or more equipment configured to detect two-dimensional (2D) and / or three-dimensional (3D) images, to perform 3D relief perception, and / or other types of physical environment sensing. In one example, the sensors of the detection unit 122 may be any commercially available RGB-D camera with a frame resolution of 1920 x 1080 achieving a frame rate of 30 fps or frames per second. In another example, the one or more sensors may be a laser profilometer.The sensors of the detection unit 122 may be incorporated remotely from the tread 104, attached either to a stationary support or to a movable support covering the field of view of the inspection system 100. Further, the sensors of the detection unit 122 may be aligned with a center of a tire building drum and / or with the center of the junction ends 102.

[0041] The sensor(s) of the detection unit 122 of the inspection system 100 detect the presence of an arrangement of the tread 104 in the field of view of the sensor, which triggers the acquisition by the sensor of the image of the tread 104 as well as its profile parameters. In certain embodiments of the inspection system 100, the sensor is triggered when the tread 104 enters the field of view against the background of the acquired image. The acquired images are formed of pixels associated with the distance of the transverse marks (Tb T2) rela tively to the first edge portions (110, 116), to the second edge portions (112, 118) and to the central portions (114, 120). The sensors of the detection unit 122 send the acquired profile parameters of the tread 104 shown in [Fig.l].

[0042] In one embodiment of the present invention, the inspection system 100 comprises a distance sensor suitable for being placed at the lens of the camera of the detection unit 122 to measure a distance between the camera and a surface of the tread 104. The processing unit can then adapt a conversion rate from a pixel to a distance, for example in mm. This principle of the distance sensors, the focal length of the camera of the detection unit 122, and the height of a pixel allows the inspection system 100 to adapt to a variation in the diameter of the surfaces of the tread 104 resulting from a manufacturing dispersion or a dimensional variation.

[0043] The sensors of the detection unit 122 are configured to acquire one or more images of the tread 104 comprising the transverse marks (Ti, T2) and the distances (Dh D2) relative to the junction ends 102. The acquired image or images are transferred and recorded as acquired images in the memory of a processing unit. The processing unit, which executes the instructions of an image processing module of the processing unit, analyzes the image in order to determine one or more parameters of the tread 104 subject to imaging.The term "processor" (or alternatively the term "programmable logic circuit") refers to one or more devices capable of processing and analyzing data and comprising one or more software packages for processing them (e.g., one or more integrated circuits known to those skilled in the art as being included in a computer, one or more controllers, one or more microcontrollers, one or more microcomputers, one or more programmable logic controllers (or "PLCs"), one or more application-specific integrated circuits, one or more neural networks and / or one or more other known equivalent programmable circuits). The processing unit comprises software for processing the data acquired by the subsystems associated with the inspection system 100 (and the corresponding data obtained) as well as software for identifying and locating variances and identifying their sources for correction..

[0044] The processing unit comprising one or more software, algorithms and / or models is configured to process the images of the tread 104 by starting with a geometric mesh of the acquired images of the tread 104 using state-of-the-art techniques. In one embodiment of the present invention (see Figures 2 and 3), the geometric mesh makes it possible to define nodal points (Th, TL2, Ti.3, TL4) on the transverse mark Tisurla first end 102. Further, nodal points (Eu, EL2, EL3, Ei 4) are defined on the edge portions of the first end 106. Similarly, the geometric mesh makes it possible to define nodal points (T2.i, T2.2, T23, T24) on the transverse mark T2 on the second end 108. Further, nodal points (E21, E22, E23, E24) are defined on the edge portions of the second end 108. Although the processing unit defines 16 nodal points according to an example of the present invention, those skilled in the art may understand that a number N of nodal points may be defined based on the profile of the tread 104 or using processing techniques using various software or algorithms or models.

[0045] In an embodiment of the present invention as shown in Figures 2 and 3, the junction between the first end 106 and the second end 108 is called a perfect junction P if the distances calculated (Ci, C2, C3, C4) by the processing unit between the nodal points (Tu, TL2, TL3, TL4) on the transverse mark Ti of the first end 106 and the nodal points (T2.b T2 2, T2 3, T2 4) on the transverse mark T2 of the second end 108 are equal to a sum of the predetermined distance (Di+D2) defined after manufacturing and cutting into segments of the tread 104. Furthermore, the nodal points (Tu, TL2, T[ 3, Ti.4) on the transverse mark Ti and the nodal points (Eu, EL2, EL3, Ei.4) on the edge portions of the first end 106 are aligned with the nodal points (T2.b T22, T23, T24) on the transverse mark Tl and the nodal points (E2.i, E22, E23, E24) on the edge parts of the second end 108.In this perfect junction P, there is an adequate amount of rubber available between the first end and the second end without expansion or compression of the rubber so that there are no gap defects, overlap defects or offset defects. In an example, if the predetermined distance Di and D2 of the transverse marks (Tb T2) is 75 mm relative, respectively, to the first end 106 and the second end 108, then the calculated distance (Ci C2C3, C4) between the nodal points on the transverse marks Ti and T2 is 150 mm, so that there is no gap or overlap between the junction ends 102.

[0046] In an embodiment of the present invention as shown in Figures 4 and 5, the junction between the first end 106 and the second end 108 is called an imperfect junction due to a gap defect G if the distances calculated (Ci.C2C3, C4) by the processing unit between the nodal points (Tu, TL2, TL3, Ti.4) on the transverse mark Ti of the first end 106 and the nodal points (T2.i, T2.2, T23, T24) on the transverse mark T2 of the second end 108 are greater than a sum of the predetermined distance (Di+D2) defined after manufacturing and cutting into segments of the tread 104. Furthermore, the nodal points (Tu, TL2, TL3, Ti,4) on the transverse mark Ti and the nodal points (Eu, EL2, EL3, EL4) on the edge portions of the first end 106 are aligned with the nodal points (T2.b T2.2, T2.3, T24) on the transverse mark Ti and the nodal points (E2 b E22, E23, E24 ) on the edge portions of the second end 108. In this imperfect junction, there is not enough rubber available between the first end 106 and the second end 108, which results in a gap defect G. The gap defect G arises mainly from plastic properties of the rubber resulting in expansion of the rubber and, consequently, a gap between the first end 106 and the second end 108.In one example, if the predetermined distance Di and D2 of the transverse marks Ti and T2 is 75 mm relative, respectively, to the first end 106 and the second end 108, then the calculated distances (Ci, C2, C3, C4) between the nodal points on the transverse marks Ti and T2 is 154 mm, so that there is no gap between the joining ends 102.

[0047] In an embodiment of the present invention as shown in Figures 6 and 7, the junction between the first end 106 and the second end 108 is called an imperfect junction due to an overlap defect O if the distances calculated (Ci, C2, C3, C4) by the processing unit between the nodal points (Ti.i, TL2, Ti.3, TL4) on the transverse mark T of the first end 106 and the nodal points T2.b T22, T23, T24 on the transverse mark T2 of the second end 108 are less than a sum of the predetermined distance (Di+D2) defined after manufacturing and cutting into segments of the tread 104.Furthermore, the nodal points (Tu, Ti 2, Ti 3, T[ 4) on the transverse mark Ti and the nodal points (Eu, EL2, EL3, Ei 4) on the edge portions of the first end 106 are aligned with the nodal points (T2 b T2 2, T2 3, T24) on the transverse mark Ti and the nodal points (E 21, E22, E23, E24) on the edge portions of the second end 108. In this imperfect junction, there is an excess of rubber available between the first end 106 and the second end 108, which results in an overlap defect O. The overlap defect O arises mainly from plastic properties of the rubber resulting in compression of the rubber and, consequently, an overlap between the first end 106 and the second end 108.In one example, if the predetermined distance Di and D2 of the transverse marks (Ti, T2) is 75 mm relative to, respectively, the first end 106 and the second end 108, then the calculated distance (Ci C2C3, C4) between the nodal points on the transverse marks Ti and T2 is 148 mm due to an overlapping junction between the junction ends 102.

[0048] In an embodiment of the present invention as shown in [Fig.8], the junction between the first end 106 and the second end 108 is also referred to as an imperfect junction due to a combination of at least one of the gap defect G, the overlap defect O or the perfect junction P. It is determined that a junction is this imperfect junction if the distances calculated (Ci, C2, C3, C4) by the processing unit between the nodal points (Tu, TL2, TL3, Ti 4) on the transverse mark T of the first end 106 and the nodal points (T2 b T22, T23, T24) on the transverse mark T2 of the second end 108 are greater than, less than or equal to the sum of the predetermined distance (Di + D2) due to unequal expression or compression of the rubber at several nodal points on the transverse marks (Tb T2).

[0049] In this specific example of the present embodiment as shown in [Fig.8] illustrating a combination of defects, if the calculated distance Ci between the nodal points Tu and T2 is greater than the sum of the predetermined distance (Di+D2), a gap defect G may result. Furthermore, if the calculated distances (C2C3) between the nodal points (Ti 2, TL3) and (T2 2, T2 3) are equal to the sum of the predetermined distance (Di+D2), the perfect junction P without defects is obtained. In the similar example, if the calculated distance C4 between the nodal points Ti4 and T24 is less than the sum of the predetermined distance (Di+D2), it may result in an overlap defect O. In this imperfect junction due to a combination of defects, there is excess rubber, insufficient rubber or adequate rubber available between few nodal points (Ei.i, EL2, EL3, Ei.4) on the edge parts of the first end 106 on the transverse mark Ti and the nodal points (E2.b E2 2, E2 3, E24) on the edge parts of the second end 108. .

[0050] In an embodiment of the present invention as shown in [Fig.9], the junction between the first end 106 and the second end 108 is also called an imperfect junction due to a misalignment of the nodal points on the junction ends 102. If the nodal points (Tu, TL2, TL3, TL4) on the transverse mark Ti and the nodal points (Eu, EL2, Ei 3, EL4) on the edge portions of the first end 106 are not aligned with the nodal points (T21, T22, T23, T24) on the transverse mark T2 and the nodal points (E2 b E22, E23, E24) on the edge portions of the second end 108. The imperfect junction between the first end 106 and the second end 108 results in an offset junction J. This imperfect junction can result from a variation in thickness of the first end 106 and the second end 108 due to plastic properties of the rubber causing expansion or expression.

[0051] In one embodiment of the present invention, in the event of a non-compliant or imperfect splice due to at least one of the gap defect G, the overlap defect O or the offset defect J, the splicing operation is interrupted and a technician can be alerted. Depending on the defect, the technician can repair the imperfect splice if possible or the defective tread 104 can be replaced with a new tread 104. The processing unit comprising One or more software or algorithms or transformation models can be trained based on the identified defects including G-gap defect, O-overlap defect, J-shift defect or perfect splice to predict quality of subsequent splices or anticipate defects. This learning technique can develop optimization or regulation for the algorithm to change parameters of a splicing machine to anticipate or correct performance.

[0052] The image processing module may implement one or more machine learning models using the acquired parameters of the tread 104 from the sensors to identify junction ends 102 of the tread 104. Although the embodiments are described herein with respect to the use of neural networks (and specifically convolutional neural networks (CNNs)) as the machine learning model, other types of machine learning models may be used.These include, but are not limited to, models employing linear regression, logistic regression, decision trees, support vector machines, naive Bayesian models, K-nearest neighbors (kNN), where K stands for clustering, random forest, dimensionality reduction algorithms, gradient algorithms, neural networks (e.g., autoencoders, CNNs, RNNs, perceptrons, logarithmic short-term memory (LSTM), Hopfield, Boltzmann, deep belief networks, deconvolution, generative adversarial networks (GANs), etc.) as well as their complements and equivalents.

[0053] In one embodiment of the present invention, the image processing module of the processing unit may include transformation models configured to identify transverse marks (Tb T2) on the tread 104. In one example, a Hough transform model may be applied to detect lines or marks in acquired images of the tread 104.

[0054] In the preferred embodiment of the present invention, the image processing module may implement one or more machine learning models based on a neural network architecture defined as point object detection, proposed as CenterNet in 2014. The “CenterNet: Point Objects” model was one of the turning points in anchor-free object detection algorithms. Anchor-free object detection is a technique for directly predicting a bounding box of an image relative to a fixed reference in the image. An object is modeled as the center point of its bounding box. The size of the bounding box and other object properties are obtained by inference or regression from a point of interest at the center (see https: / / arxiv.org / pdf / 1904.07850.pdf).

[0055] In one aspect of the present invention, the CenterNet model includes an encoder-decoder for processing input images of the tread 104. The encoder is configured to iteratively reduce a resolution of the acquired images and may increase a depth of the image. In a specific example, if the resolution of the acquired image is 1024x1024x3 (RGB Image => 3), the encoder may iteratively reduce a spatial dimension to 16x16 but increase the depth to 1024 using a convolution operation. The convolution operation is defined as a filter application to retain features of interest in the acquired image. This special lower resolution "image" is referred to as an "interest point map". A decoder is configured to increase the format of the interest point map with a transposed convolution.In order to reduce the computational cost, the size of the point of interest map can be increased to half the resolution (512 x 512). In addition, the geometric mesh can be applied with a resolution of half the size of the acquired input image.

[0056] In one embodiment according to the present invention, since there is only one object per image, center detection heads, and associated detection heads may not be required. The center detection head is defined as an interest point identification at a center of each of multiple objects in the image. The associated detection head is defined as N interest points that are related to the center interest point. Additionally, the CenterNet model can contain two heads, namely a depth head for 16 nodal interest points (T ii» Ti.2, Us, Ti.4, Ei.i, Ei.2» E13, E14, T2.i, T2.2, T2.3, T2.4, E2.i, E2.2, E2.3, E2.4), and an offset prediction head for 32 points, double the number of interest points, allowing for offset prediction associated with reduced resolution due to undersampling within the neural network, and may be necessary due to memory constraints.

[0057] The processing unit may configure the inspection system 100 to one or more parameters of the junction ends 102 of the tread 104 that are calculated by the image processing module. The processing unit may also refer to a reference (e.g., a size chart for various treads) to make a final determination of the target tread parameter(s). The reference may include known tread parameters corresponding to a plurality of known commercially available treads. For example, after the image processing module calculates one or more tread parameters, the processing unit may compare the calculated tread parameters to the known tread parameters recorded in the reference. The image processing module may also be trained by a data augmentation technique of artificially augmenting a training set by creating modified copies of a dataset using existing data. This includes applying minor changes to the dataset or using deep learning to generate new data points. In the present invention, the images of the tread 104 may be transformed by geometric transformations including flipping, cropping, rotating, translating, and Kernel filters to increase or decrease the sharpness of an image and color space transformations including changes to the RGB color channels, intensification of any color, changing the brightness. Employing data augmentation techniques helps train the algorithm and make it more robust to environmental variations.

[0058] The identification of the junction ends 102 of the tread 104 is relevant to the representation and may be determined by post-processing a previously generated segmentation of the tread 104. For example, a method may be used to determine whether a pixel is a candidate for the areas of the junction ends 102 including the first end 106, the second end 108, the first edge portions (110, 116), the second edge portions (112, 118), and the center portions (114, 120). For example, active contour models may be applied, along with path planning and distance transformations, to extract the portions of the tread 104.A morphology-based level set model may be used to perform tread area extraction 104 by learning the structural patterns of a target tread-like object 104 and estimating the junction ends 102 of the object as a path. The invention therefore takes advantage of artificial intelligence (or "AI")-based methods and tools to supplement partial information provided by perception.

[0059] In an embodiment of the present invention as shown in [Fig. 10], the image processing module is configured to apply a geometric correction to compensate for non-flatness of a measurement region. The calculated distance measured between the transverse marks (Tb T2) on the first end 106 and the second end 108 may be curved due to the radius of the tread 104. Since the acquired images detected on the image taken by the camera are flat, it is necessary to apply the geometric correction.

[0060] As shown in [Fig. 10], let us take 3 surfaces Ao A3 and A4 having the relation Ao — A3 + A4

[0061] A0=(D + R).C

[0062] A3=R.Cos (Alpha).C

[0063] A4 = (D + r). C

[0064] C = Rope / 2

[0065] Therefore, with the relation Ao= A3+ A4

[0066] (D + R).C = R.Cos (Alpha).C + (D + r). C

[0067] r = R. (1 - R.Cos (Alpha)) - (Equation 1)

[0068] Constraints relating to the Rope (C):

[0069] (D + r). Tan (Theta) = R.Sin (Alpha) - (Equation 2)

[0070] Substituting r from Equation 1 into Equation 2

[0071] (D + R. (1 - R.Cos (Alpha))). Tan (Theta) = R.Sin (Alpha)

[0072] (D + r). Tan (Theta) = R.Sin (Alpha) + R.Cos (Alpha). Tan (Theta) - (Equation 3)

[0073] Multiplying by Cos (Theta) in Equation 3

[0074] (D + r). Sin (Theta) = R.Sin (Alpha). Cos (Theta) + R.Cos (Alpha). Sin (Theta) -(Equation 4)

[0075] With the formula Sin (a+b) in the right term of Equation 4

[0076] (D + r). Sin (Theta) = R.Sin (Alpha + Theta) - (Equation 5)

[0077] Therefore,

[0078] Alpha = Arcsin(((D+R) / R).Sin(Theta)) - Theta - (Equation 6)

[0079] Ylmage= D. Tan (Theta) - (Equation 7)

[0080] Y Pneumatic R.[(((D+R) / R).Sin(Theta)) - Theta] - (Equation 7)

[0081] Therefore, final equation

[0082] Ypneumatic-R*[Arcsin(((D+R) / R).Sin(Arctan(YImage / D)) - Arctan(YImage / D))] - (Equation 8)

[0083] Although embodiments of the disclosed apparatus have been illustrated and described, it will be understood that various changes, additions, and modifications may be made without departing from the spirit or scope of the present disclosure. Accordingly, no limitations should be imposed on the scope of the disclosed invention except those set forth in the appended claims.

Claims

Claims

1. An inspection system (100) for monitoring a distance between joining ends (102) of a sheet product (104) having a profile defined by parameters of a predetermined length, a predetermined width and a predetermined thickness, the joining ends (102) including a first end (106), a second end (108), first edge portions (110, 116), second edge portions (112, 118) and central portions (114, 120) defining end surfaces of the joining ends (102) along a longitudinal axis, the inspection system (100) comprising: - at least one marking unit configured to print transverse marks (Tb T2) on the first end (106) and the second end (108) at a predetermined distance (Db D2) from the first edge portions (110, 116), second edge portions (112, 118) and central portions (114, 120); - at least one detection unit (122) configured to acquire one or more images in order to identify the transverse marks (Tb T2) on the sheet product (104), the images being formed of pixels associated with the distance of the transverse marks (Tb T2) relative to the first edge portions (110, 116), to the second edge portions (112, 118) and to the central portions (114, 120); - at least one processing unit comprising one or more software or algorithms or transformation models configured to process the images of the sheet product (104) with a geometric mesh making it possible to define one or more nodal points (Tbb Tb2, Tb3, Ti,4) on the transverse mark (Tl) on the first end (106) and one or more nodal points (Ebb Eb2, Eb3, Ei,4) on the first edge portion (110), the second edge portion (112) and the central portion (114) of the first end (106); - at least one processing unit comprising one or more software or algorithms or transformation models configured with a geometric mesh for defining one or more nodal points (T2b T22, T23, T24) on the transverse mark (T2) on the second end (106) and one or more nodal points (E2b E2 2, E23, E24) on the first edge portion (116), the second edge portion (118) and the central portion (120) of the second end (108); and - the processing unit being configured to define a junction on the based on calculated distances (Cb C2, C3, C4) between the nodal points (Tu, TL2, Ti 3, Ti 4) on the transverse mark (TJ of the first end (106) and the nodal points (T2.i, T22, T2 3, T24) on the transverse mark T2 of the second end (108).

2. Inspection system (100) according to claim 1, wherein the defined junction is a perfect junction (P) without defects, if the distances calculated (Cb C2, C3, C4) by the processing unit between the nodal points (Tu, TL2, TL3, Ti 4) on the transverse mark (TJ of the first end (106) and the nodal points (T21, T2 2, T2 3, T24) on the transverse mark (T2) of the second end (108) are equal to a sum of the predetermined distance (Di+D2).

3. An inspection system (100) according to claim 1, wherein the defined junction is an imperfect junction due to a gap defect (G), if the distances calculated (Ci, C2, C3, C4) by the processing unit between the nodal points (Tu, TL2, TL3, TL4) on the transverse mark (T 1) of the first end (106) and the nodal points (T2.b T2 2, T2 3, T24) on the transverse mark (T2) of the second end (108) are greater than a sum of the predetermined distance (Di+D2).

4. An inspection system (100) according to claim 1, wherein the defined junction is an imperfect junction due to an overlap defect (O), if the distances calculated (Cb C2, C3, C4) by the processing unit between the nodal points (Tu, TL2, TL3, T14) on the transverse mark (TJ of the first end (106) and the nodal points (T 2.1, T22, T23, T24) on the transverse mark (T2) of the second end (108) are less than a sum of the predetermined distance (Di+D2

5. )■ Inspection system (100) according to claims 1 to 4, wherein the defined junction is an imperfect junction due to an offset defect (J), if the nodal points (Tu, TL2, TL3, TL4) on the transverse mark (TJ and the nodal points (Eu, EL2, EL3, Ei4) on the edge portions (110, 112, 114) of the first end (106) are not aligned with the nodal points (T21, T2 2, T2 3, T24) on the transverse mark (T2) and the nodal points (E2 b E2 2, E2 3, E24) on the edge portions (110, 112, 114) of the second end (108).

6. An inspection system (100) according to claims 1 to 5, wherein the defined junction is an imperfect junction due to a combination of at least one of gap defects (G), overlap defects (O), offset defects (J) or perfect junction (P).

7. An inspection system (100) according to claims 1 to 6, the processing unit comprising one or more software or algorithms or transformation models and being trained to predict a quality of subsequent junctions or anticipate defects based on identified defects comprising at least one of gap defects (G), overlap defects (O), perfect junction (P) or offset defects (J).

8. An inspection system (100) according to claim 1 or 6, wherein the processing unit comprises an image processing module implementing one or more transformation models defined as point object detection based on the CenterNet neural network architecture.

9. An inspection system (100) according to claims 1 to 7, wherein the processing unit comprises an image processing module implementing one or more transformation models comprising an encoder-decoder configured to adapt the acquired images to a lower or higher resolution.

10. An inspection system (100) according to claims 1 to 8, wherein the processing unit comprises an image processing module configured to apply a geometric correction to compensate for non-planarity of a measurement region.

11. An inspection system (100) according to claims 1 to 9, wherein a distance sensor is provided at a level of the detection unit (122) for measuring a distance between the detection unit (122) and the surface of the sheet product (104) to convert a pixel area into a distance with adaptation to the diameter variation of the surfaces of the sheet product (104).