Inspection system for assemblies of rubber products in visible layers, both in their entirety and at the edges of these products.

Abstract

The invention relates to an inspection system (100) for assemblies of layered rubber products, which performs a method for inspecting extruded, profiled, layered rubber products such that the inspection system (100) uses a rubber product detection algorithm to construct a segmentation model for detecting, within an assembly of rubber products, fully visible assembled products and partially visible product edges. The invention also relates to a tire production site comprising the disclosed inspection system (100). Figure for the abstract: Fig. 2

Description

Title of the invention: System for inspecting assemblies of rubber products in fully visible layers and the edges of these products Technical field

[0001] The invention relates to an inspection system based on machine vision for detecting rubber products that are fully visible and the edges of these products that are partially visible. The inspection system uses a segmentation model to detect these two types of products from a collection of images of rubber products assembled in layers. Context

[0002] In the field of tire manufacturing, profiled extruded rubber products must be assembled precisely to ensure anticipated subsequent performance. These rubber products are often extruded in the form of a sheet that allows the different products to be assembled on top of each other. During tire assembly, cameras of various types (including, without limitation, matrix type and profilometer type cameras) are often employed to control the position of the different products being laid. Referring to [Fig.l], an exemplary image is given of an assembly of layered products arranged on a drum that include one or more products fully visible (designated by lines IA) and the edges 1B of the rubber products partially visible. Obviously, the image in [Fig.l] presents a difficult contrast due to the tire being a black product.The positions of rubber products assembled in layers are therefore difficult to perceive in relation to each other. This is even more true when there is little 3D information: the contrast is very low and this generates a large number of parasitic edge detections.

[0003] This edge detection falls into the domain of edge detection, which presents a fundamental problem in computer vision. Edge detection has several applications, including, but not limited to, image segmentation, object detection, and video object segmentation. Edge detection is strongly related to the context and semantic cues in the image. Therefore, it is critical to obtain an appropriate representation to capture both high- and low-level visual cues (see Pu, Menyang et al. EDTER: Edge Detection with Transformer, https: / / arxiv.org / pdf / 2203.08566.pdf (16 March 2022) (the “referenced EDTER”).

[0004] The EDTER reference discloses a solution which proposes the use of trans trainer for generic edge detection with affordable computational cost. This solution incorporates feature learning that contains an image's global context and fine-grained local cues through a two-stage framework. In a first stage, an image is divided into coarse-grained patches, and a global transformer encoder is run on these patches to capture the long-range global context. Then, a novel bidirectional multi-level aggregation (or "BiMLA") decoder is developed to generate high-resolution representations for edge detection. In a second stage, the entire image is divided into multiple sequences by sampling with a non-overlapping sliding window. Then, a local transformer works on each sequence in turn to explore the short-range local cues.All local landmarks are integrated and fed into the local BiMLA decoder to obtain the pixel-level feature maps. Finally, the information from the two stages is fused by a feature fusion module (or “FFM”) and then fed into a decision head to predict the final contour map (see page 2 and [Fig.2] of the EDTER reference).

[0005] There is also the domain of object detection, the detection of which identifies objects as boxes aligned along the axis of an image. Most effective object detectors enumerate a nearly exhaustive list of potential object locations and classify each of them. The prior art proposes a detector that uses keypoint estimation to find center points and regresses on all other properties of an object (such as size, 3D location, orientation, and pose)(see Xingyi Zhou et al., Objects as Points, https: / / arxiv.org / pdf / 1904.07850.pdf (25 April 2019)(the “CenterNet” reference). This approach is based on modeling an object as a single point that includes the center point of its bounding box (see, for example, [Fig. 2] of the CenterNet reference).The "CenterNet" model is one of the milestones of the anchor-free object detection algorithm that can directly predict the bounding box of an image relative to a fixed reference in the image. The size of the bounding box and other object properties are inferred or regressed from a key point located in the center.

[0006] Improvements in the areas of edge detection and object detection could benefit the tire manufacturing field where layered rubber products are difficult to detect. Thus, the disclosed invention makes it possible to provide an inspection system using a rubber product detection algorithm to make it robust to inaccuracies in labeling these products and their edges. The ultimate goal is to correlate the uniformity criteria with the reality of what is manufactured by assembling the rubber products. Summary of the invention

[0007] The invention relates to a system for inspecting assemblies of layered rubber products which carries out a method for inspecting extruded rubber products profiled in layers, characterized in that the system comprises: - at least one external mobile equipment comprising one or more equipment which facilitates the manufacture of tires; - a detection system which collects information on the physical environment around the at least one external mobile equipment, the detection system comprising one or more sensors which detect the presence of an assembly of rubber products in the field of vision of the sensor; and - a communication network that manages incoming data to the inspection system from at least one external mobile equipment that is part of a tire production facility, the communication network comprising at least one communication server for executing programmed instructions stored in a memory of one or more processors of the inspection system that employs (a) an image processing module of the processor, which analyzes images obtained from the layered product assemblies positioned on the external mobile equipment to determine the presence and positioning of each fully visible product and each partially visible product; and (b) a module for executing a rubber product detection algorithm from the data obtained from the images of the product assemblies; wherein the captured images are formed of pixels associated with the parameters and positioning of the fully visible products and the partially visible product borders, these images being transferred and stored as captured images in the memory of the inspection system, and the memory, by executing the instructions of the image processing module in order to determine one or more parameters of each product of the imaged product assembly, executes an inspection process comprising the following steps: - a step of receiving an initial position signal of an assembly of layered rubber products; - a step of acquiring images of the product assembly, this step comprising a step of identifying the shape and number of layered rubber products in the collected image; - a data collection stage during which no processing is applied; - a data annotation step in order to identify the edges to be detected subsequently; - a segmentation model training step, during which the inspection system gathers the images obtained during the image acquisition step, and during which the inspection system identifies the edges or products; and - a use step, during which the identification of unchanged pixels by a maxpooling operation makes it possible to detect the most probable edge in a neighborhood; so that the inspection system uses the rubber product detection algorithm to build a segmentation model to detect in the rubber product assembly the fully visible assembled products and the edges of the partially visible products.

[0008] In some embodiments of the inspection system of the invention, the captured images are formed of pixels associated with the parameters and positioning of the fully visible products and the borders of partially visible products, these images being transferred and stored as captured images in the memory of the inspection system, and the memory, by executing the instructions of the image processing module in order to determine one or more parameters of each product of the imaged product assembly, executes an inspection process comprising the following steps: - a step of detecting the edges of the product assembly carried out after the identification step; and - a step of obtaining a coordinate system of the inspection system, this step comprising a step of correspondence between the coordinate system of the inspection system and a coordinate system of external mobile equipment; so that a set of coordinate points of the obtained correspondence is sent to a control unit to provide an expected coordinate position for each fully visible product and the borders of each partially identified visible product in the images.

[0009] In some embodiments of the inspection system of the invention, the segmentation model comprises an instance segmentation model or a semantic segmentation model.

[0010] In some embodiments of the inspection system of the invention, the edge detection performed by the inspection system includes semantic segmentation using a focal loss function to inform the model of the labeling uncertainty.

[0011] In certain embodiments of the inspection system of the invention, during the step of training a segmentation model, the inspection system performs a preprocessing step with a Gaussian kernel consisting of convolving the received initial position signal.

[0012] In certain embodiments of the inspection system of the invention, the sensor(s) of the detection system are selected from one or more cameras configured to perform two-dimensional (2D) and / or three-dimensional (3D) image detection, one or more video cameras, one or more infrared cameras, one or more ultraviolet cameras, one or more electromagnetic sensors and their equivalents.

[0013] In certain embodiments of the inspection system of the invention, the detection system comprises at least one illumination source capable of capturing the images under different coded and / or pre-coded illuminations.

[0014] In some embodiments of the inspection system of the invention, the at least one external mobile equipment comprises a drum serving as a support on which the rubber products are assembled during a tire manufacturing process.

[0015] In some embodiments of the inspection system of the invention, the image processing module comprises a machine learning model based on the detection of objects as points.

[0016] The invention also relates to a tire production site comprising the disclosed inspection system.

[0017] Other aspects of the invention will become apparent from the following detailed description. Brief description of the drawings

[0018] The nature and various advantages of the invention will become more apparent from the following detailed description, taken in conjunction with the accompanying drawings, in which like reference numerals designate like parts throughout, and in which: [Fig.l] [Fig.l] represents an exemplary image of an assembly of rubber products intended for incorporation into a tire during manufacture. [Fig.2] [Fig.2] shows a system for inspecting rubber product assemblies of the disclosed invention. [Fig.3] [Fig.3] represents a representative convolution operation between an image of a target product detected by a data processing module of the inspection system of the invention and a Gaussian kernel. [Fig.4] [Fig.5] [Fig.6] Figures 4, 5 and 6 represent examples of detection by which the inventive inspection system has carried out a method of detecting rubber products in layers superimposed and assembled along a tire manufacturing line. Detailed description

[0019] Referring now to the Figures, in which the same numbers identify identical elements, [Fig. 2] represents a system for inspecting rubber product assemblies (or "inspection system" or "system") 100 which carries out a method allowing the inspection of extruded rubber products profiled in layers (or "products"). The products likely to be detected include assemblies of products visible in their entirety and the edges of these products partially visible (for example, profiled extruded rubber products intended for incorporation into radial tires intended to equip a vehicle).

[0020] Generally, a tire comprises a tread, intended to come into contact with a ground via a rolling surface, the two axial ends of which are connected via two sidewalls with two beads (or "rods") ensuring the mechanical connection between the tire and the rim on which it is intended to be mounted. The sidewall of a tire comprises at least a portion of the carcass reinforcement connecting the beads at the crown and axially outermost to the axially outermost carcass layer of the sidewall mixtures. The beads are made up of at least two windings of elementary metal wires substantially parallel to at least one carcass layer. In a meridian section, each of these windings forms an alignment substantially parallel to the reinforcing elements of the carcass layer.

[0021] A radial tire further comprises a reinforcing reinforcement, consisting of a crown reinforcement, radially inside the tread, and a carcass reinforcement, radially inside the crown reinforcement. It generally comprises a rubber sealing layer (or "inner rubber") extending between the two beads, being located axially inside the carcass reinforcement and generally being in contact with the inflation gas. The carcass reinforcement of a radial tire usually comprises at least one carcass layer comprising reinforcements, coated with a polymeric material of the elastomer or elastomeric type, obtained by mixing. A carcass layer comprises a main part, connecting the two beads together and generally wrapping, in each bead, from the inside to the outside of the tire around a circumferential reinforcing element to form a turn-up.

[0022] The crown reinforcement of a radial tire comprises a superposition of crown layers extending circumferentially, radially outside the carcass reinforcement. Each crown layer is made up of reinforcements, parallel to each other and coated with a polymeric material of the elastomer type or coating mixture. For a given meridian plane, the radial, axial and circumferential directions respectively designate the directions perpendicular to the axis of rotation of the tire, parallel to the axis of rotation of the tire and perpendicular to the meridian plane.

[0023] Thus, in carrying out the manufacture of a tire, the correct assembly (including relative positioning) of the rubber products is essential to its desired performance.

[0024] The inspection system 100 uses a rubber product detection algorithm (or "detection algorithm") to make it robust to labeling inaccuracies of these products and their borders. More particularly, the inspection system 100 uses a segmentation model to detect assembled products that are fully visible (for example, the entire product IA detected in image 102 of [Fig. 2]) and to also detect the borders of the assembled and partially visible products (for example, the borders IB detected in image 104 of [Fig. 2]). It is understood that the segmentation model may be an instance segmentation model or a semantic segmentation model.

[0025] In one embodiment of the inspection system 100, the edge detection (e.g., the IB edges shown in [Fig. 2]) performed by the system includes semantic segmentation using a focal loss function to inform the model of labeling uncertainty. It is understood that a focal loss function can address class imbalance during training in tasks such as object detection. Thus, by labeling an edge (such as the IB edges shown in image 104 of [Fig. 2]), a succession and a finite number of points are also labeled, in no case all of the points on the line forming part of the edge. To avoid this problem, the inspection system 100 performs preprocessing consisting of convolving the signal of the product targeted for detection with a Gaussian kernel.This technique allows the model not to be penalized too much if it predicts a product at a position close to the labeled position, this allows for small variations.

[0026] Referring again to [Fig. 2], the inspection system 100 includes at least one communication network (or "network") that manages incoming data to the system from various sources (e.g., from at least a portion of a manufacturing line where layered rubber products are assembled and its associated detection system). The communication network incorporates one or more communication servers (or "servers") each comprising one or more processors operatively connected to a memory. To build the segmentation model employed in product detection, the communication network may include at least one segmentation network 106 that manages the capture and labeling data (see [Fig. 2]).

[0027] The memory is configured to store an image processing module of the processor, which analyzes the obtained images of the layered product assemblies to determine the presence and positioning of each fully visible product and each partially visible product. The memory is also configured to store a module for executing the detection algorithm from the data obtained from the images of the layered product assemblies. The detection algorithm allows for continuous improvement, ensuring that the inspection system 100 improves from the experience it acquires, particularly on the detection of products (either whole visible products or partially visible products) to achieve the desired properties of the tire being produced.

[0028] The term "processor" (or, alternatively, the term "programmable logic circuit") refers to one or more devices capable of processing and analyzing data and including one or more software programs 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 processor includes one or more software programs for processing data captured by a detection system of the inspection system 100 (and the corresponding data obtained) as well as one or more software programs for identifying and locating variances and identifying their sources to correct them..

[0029] In the inspection system 100 of the invention, the memory may include both volatile and non-volatile memory devices. The non-volatile memory may include solid-state memories, such as NAND flash memory, keep-alive memory (KAM) for saving various operating variables while the processor is powered off, magnetic and optical storage media, or any other suitable data storage device that retains the data (e.g., when the mixing line detection system is disabled or loses power). The volatile memory may include static and dynamic RAM that stores program instructions and data, including a training application.

[0030] In embodiments of an inspection method performed by the inspection system 100, the processor may configure the inspection system 100 (and in particular a tire manufacturing line incorporating the inspection system 100) on one or more parameters of one or more target products detected by a data processing module. In these embodiments, it is understood that one or more means of reinforcement learning could be employed (including deep reinforcement learning). Those skilled in the art in this field will recognize that many Data processing techniques can be used to select and determine the parameters of rubber products. Several commercially available data processing systems can be used.

[0031] The processor may also refer to a reference (e.g., a table of parameters of layered rubber products assembled on the production line) to make a final determination of one or more parameters of an identified assembled product. The reference may include known product parameters corresponding to a plurality of commercially available tires. For example, after the data processing module calculates one or more parameters of an identified product, the processor may compare the detected parameters with the known parameters recorded in the reference.

[0032] The inspection system 100 also includes a detection system (not shown) for collecting information about the physical environment around a tire manufacturing line (e.g., images of the products being assembled on a drum). The detection system includes one or more sensors (including one or more cameras) configured to perform two-dimensional (2D) and / or three-dimensional (3D) image detection, 3D depth detection, and / or other types of detection of the physical environment around the manufacturing line (it is understood that the terms "sensor" and "camera" are used interchangeably). In embodiments of the inspection system 100, the one or more sensors of the detection system are attached along the manufacturing line to capture images of rubber products being assembled.

[0033] In one embodiment of the inspection system 100, the detection system includes at least one camera that provides 2D and / or 3D images (where the 3D cameras capture the images represented as a set of 3D points with X, Y, Z coordinates, and sometimes red, green, blue color values ​​(the "RGB" or "RGB-D" format)) (referred to as an "RGB-D type camera"). Two or more cameras may be oriented to provide a predetermined overlap between the cameras' fields of view. As used herein, the term "camera" includes one or more cameras.

[0034] The detection system of the inspection system 100 could include at least one video and / or photographic camera, an infrared camera, an ultraviolet camera, a set of electromagnetic sensors capable of capturing an image, and their equivalents. The detection system of the inspection system 100 may incorporate an illumination source having one or more illuminations (e.g., known programmable LEDs) to serve as a source of light on the products assembled on the manufacturing line. The illuminations may be encoded into the detection system, or they may be pre-encoded during a training process of a neural network (e.g., example, using one or more computer programs incorporating data representative of the lighting and images of the layered rubber products). Changes in the lighting source, which are captured in images obtained by the detection system, are represented by pixels of different color intensity (for example, the images obtained may contain indications of reflection due to a change in lighting). Thus, the image captured by the detection system reveals the superimposed layers in the assembled product as well as the borders between them. The camera and the lighting source may move in an alternating or random manner to adjust, respectively, the lens and the lighting according to the parameters of the rubber product placed in a capture area of ​​the detection system (for example, according to the length and / or width of the imaged product).

[0035] The one or more sensors of the detection system of the inspection system 100 detect the presence of a disposition of a product assembly in the field of view of the sensor, which triggers the sensor to capture the image of the assembly and the profile parameters of the individual assembled products. In some embodiments of the inspection system 100, the sensor is triggered when the product assembly enters the field of view against the background of the captured image. The captured images are formed of pixels associated with the parameters and positioning of the fully visible products and the partially visible product borders. These images are transferred and stored as captured images in the memory of the inspection system 100.The memory, in executing the instructions of the image processing module to determine one or more parameters of each product in the imaged product assembly, performs an inspection process comprising the following steps: . - a step of receiving an initial position signal of an assembly of layered rubber products; - a step of acquiring images of the product assembly, this step comprising a step of identifying the shape and number of layered rubber products in the collected image; - a data collection step during which no processing is applied - a data annotation step in order to identify the edges to be detected subsequently. - a segmentation model training step, during which the inspection system (100) gathers the images obtained during the image acquisition step and identifies the borders or products. It is during this step that a convolution by a Gaussian kernel is carried out to allow deviations from the labeling. - a usage step, during which the identification of unchanged pixels by a maxpooling operation makes it possible to detect the most probable border in a neighborhood.

[0036] It is understood that external mobile equipment comprises one or more pieces of equipment that facilitate the manufacture of the tires for which the assembled products are intended. For example, external mobile equipment may comprise a drum that is known in the field of tire manufacturing to serve as an assembly support.

[0037] In one embodiment of the inspection system 100 of the invention, the image processing module may deploy one or more machine learning models based on a neural network architecture defined as the detection of objects as points (as proposed by the CenterNet reference).

[0038] Referring again to [Fig. 3], and furthermore to figures 4 and 5, the inspection system 100 starts an inspection process with the input of one or more images of at least one rubber product in superimposed layers. To train the model to be constructed, the inspection system 100 gathers the images taken and labels them as “product” (represented by the entire product 1A of [Fig. 2]) or “border” (represented by the borders 1B of [Fig. 2]). To facilitate the training of the model, a smoothing of the target is carried out, this to prevent the polyline traced by the labeling (being imprecise by nature because it is discrete) from disturbing the learning (see line 200 of [Fig. 2] representing the labeling imprecision). To do this, the inspection system 100 employs a convolution operation between the target and a Gaussian kernel.This operation has the advantage of preserving the maximum value and authorizing the pixels around this central pixel (see [Fig.3]).

[0039] In inference, to retrieve the correct pixel, the inspection system 100 uses a property of maxpooling. Maxpooling is an operation that calculates the maximum value of the patches of a feature map and uses it to create a downsampled (pooled) feature map. This technique is typically used after a convolutional layer to add a small translation invariance, meaning that a small translation of the image does not significantly affect the values ​​of most of the pooled outputs. Thus, by performing maxpooling, the inspection system 100 enhances on average all pixels in the images except the local maxima. By identifying unchanged pixels, the local maxima are retrieved. By then thresholding on a confidence level, the border is retrieved.

[0040] Referring now to Figures 4 to 6, examples of detection are shown in which the inspection system 100 has performed a method of detecting the rubber products in layers superimposed and assembled along a tire manufacturing line. Referring to [Fig. 4], the inspection system 100 performs a method of detecting the edges of the bead wires T incorporated in a rubber product. Referring to [Fig. 5], the inspection system 100 performs a method of detecting a product P that ensures the position of the bead wire (for example, the bead wire T shown in [Fig. 4]). Referring to [Fig. 6], the inspection system 100 performs a method of detecting a product incorporating whole products PiA together with the edges P1B. In this image, it is clear that there are two strips of rubber positioned on each side of a drum (where the rubber products are assembled) to stiffen the bead wire (for example, the bead wire T shown in [Fig. 4]).

[0041] The inspection system 100 of the invention can easily repeat one or more steps of a rubber product detection method in an order to properly achieve the performance parameters of the tires being produced.

[0042] In advance of launching the inspection process which is part of a rubber product manufacturing process, commands for the rubber products could be received by known means (for example, by one or more communication networks integrated into a tire manufacturing site incorporating the inspection system 100). The site allows the commands to be received up to a specific time to ensure that the desired properties of the tires are achieved.

[0043] The rubber product detection methods performed by the inspection system 100 may be done by the PLC control and may include preprogramming of the production information of a rubber product. For example, a setting of an assembly process may be associated with the parameters of an identified rubber product so that a manufacturing line is configured on one or more parameters calculated by a data processing module obtained by the inspection system 100 from each manufacturing line.

[0044] A setting of a rubber product detection method performed by the inspection system 100 may be associated with the parameters of the typical physical environments in which the inspection system 100 operates. In embodiments of the invention, the inspection system 100 (and / or a tire production site incorporating the inspection system 100) may receive voice commands or other audio data representing, for example, a step or a stop of the manufacturing line. The request may include a request for the current status of a tire production cycle. A generated response may be represented audibly, visually, tactilely (for example, using a haptic interface), and / or virtually and / or augmentedly. This response, associated with the corresponding data, may be recorded in one or more networks neuronal.

[0045] For all embodiments of the inspection system 100, a monitoring system could be implemented. At least part of the monitoring system may be provided in a portable device such as a mobile network device (e.g., a mobile phone, a laptop, one or more network-connected wearable devices (including “augmented reality” and / or “virtual reality” devices), network-connected wearables, and / or any combinations and / or equivalents). It is envisaged that detection and comparison steps may be performed iteratively.

[0046] The terms "at least one" and "one or more" are used interchangeably. Ranges that are presented as being "between a and b" encompass the values ​​"a" and "b".

[0047] Although particular embodiments of the disclosed system have been illustrated and described, it will be understood that various changes, additions, and modifications may be practiced without departing from the spirit and 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. A system (100) for inspecting layered rubber product assemblies that performs a method for inspecting layered profiled extruded rubber products, characterized in that the system (100) comprises: - at least one external mobile equipment comprising one or more equipment which facilitates the manufacture of tires; - a detection system which collects information on the physical environment around the at least one external mobile equipment, the detection system comprising one or more sensors which detect the presence of an assembly of rubber products in the field of vision of the sensor; and - a communication network that manages incoming data to the inspection system (100) from at least one external mobile equipment that is part of a tire production facility, the communication network comprising at least one communication server for executing programmed instructions stored in a memory of one or more processors of the inspection system (100) that employs (a) an image processing module of the processor, which analyzes images obtained from the layered product assemblies positioned on the external mobile equipment to determine the presence and positioning of each fully visible product and each partially visible product; and (b) a module for executing a rubber product detection algorithm from the data obtained from the images of the product assemblies; wherein the captured images are formed of pixels associated with the parameters and positioning of the fully visible products and the partially visible product borders, these images being transferred and stored as captured images in the memory of the inspection system (100), and the memory, by executing the instructions of the image processing module in order to determine one or more parameters of each product of the imaged product assembly, executes an inspection process comprising the following steps: - a step of receiving an initial position signal of an assembly of layered rubber products; - a step of acquiring images of the product assembly, this step including a step of identifying the shape and number of layered rubber products in the collected image; - a data collection stage during which no processing is applied; - a data annotation step in order to identify the edges to be detected subsequently; - a step of training a segmentation model, during which the inspection system (100) gathers the images obtained during the image acquisition step, and during which the inspection system (100) identifies the borders or the products; and - a usage step, during which the identification of unchanged pixels by a maxpooling operation makes it possible to detect the most probable border in a neighborhood; such that the inspection system (100) uses the rubber product detection algorithm to construct a segmentation model for detecting in the rubber product assembly the fully visible assembled products and the edges of the partially visible products.

2. The inspection system (100) of claim 1, wherein the captured images are formed of pixels associated with the parameters and positioning of the fully visible products and the partially visible product borders, these images being transferred and stored as captured images in the memory of the inspection system (100), and the memory, by executing the instructions of the image processing module in order to determine one or more parameters of each product of the imaged product assembly, executes an inspection process comprising the following steps: - a step of detecting the edges of the product assembly carried out after the identification step; and - a step of obtaining a coordinate system of the inspection system (100), this step comprising a step of correspondence between the coordinate system of the inspection system and a coordinate system of an external mobile equipment; so that a set of coordinate points of the obtained correspondence is sent to a control unit to provide an expected coordinate position for each fully visible product and the borders of each partially identified visible product in the images.

3. The inspection system (100) of claim 1 or re- dication 2, wherein the segmentation model comprises an instance segmentation model or a semantic segmentation model.

4. The inspection system (100) of claim 3, wherein the edge detection performed by the inspection system comprises semantic segmentation using a focal loss function to inform the model of labeling uncertainty.

5. The inspection system (100) of any one of claims 1 to 4, wherein, during the step of training a segmentation model, the inspection system (100) performs a preprocessing step with a Gaussian kernel consisting of convolving the received initial position signal.

6. The inspection system (100) of any one of claims 1 to 5, wherein the sensor(s) of the detection system are selected from one or more cameras configured to perform two-dimensional (2D) and / or three-dimensional (3D) image detection, one or more video cameras, one or more infrared cameras, one or more ultraviolet cameras, one or more electromagnetic sensors and their equivalents.

7. The inspection system (100) of any one of claims 1 to 6, wherein the detection system comprises at least one illumination source capable of capturing the images under different coded and / or pre-coded illuminations.

8. The inspection system (100) of any one of claims 1 to 7, wherein the at least one external movable equipment comprises a drum serving as a support on which the rubber products are assembled during a tire manufacturing process.

9. The inspection system (100) of any one of claims 1 to 8, wherein the image processing module comprises a machine learning model based on detecting objects as points.

10. A tire production site, comprising the inspection system (100) of any one of claims 1 to 9.

Images