Shearography system for detecting bloat inside a tire

Abstract

The invention relates to a shearography system (100) incorporating a means for the automatic segmentation of anomalies detected by learning in images of tires obtained by shearography. The invention also relates to a tire manufacturing plant comprising the disclosed shearography system. Figure for the abstract: Fig. 5

Description

Title of the invention: Shearography system for detecting blowholes inside a tire. Technical field

[0001] The invention relates to a shearography system for detecting blowholes inside a tire. More particularly, the invention relates to a system that performs automatic diagnosis of anomalies such as "blowholes" or "bubbles" detected by machine learning in tire images obtained by shearography.

[0002] In the field of tire manufacturing, the internal components of tires intended for use on vehicles are difficult to inspect. These tires are generally black due to the black rubber compound they are made of (resulting from the use of carbon to reinforce the elastomeric compounds). Furthermore, blowholes (or "bubbles") inside tires are very difficult to produce. The presence of a blowhole may be linked to a known anomaly (which includes, but is not limited to, dents, scratches, cuts, chips, smooth teeth, material deficiencies or excesses, and cracks). As used here, the term "anomaly" refers to a detected difference (represented, for example, by a deviation from a standard value) without further information (therefore, an "anomaly" is not automatically interpreted as a "defect").

[0003] In the workshops for checking manufactured and retreaded tires, there are shearography machines with high inspection rates. Shearography refers to non-destructive testing (NDT) methods using interferometry in the visible light spectrum (see Dubos, Camille et al., Shearography: From Digital Mock-up to Robotic Inspection, COFREND 2023 Conference, https: / / doi.org / 10;58286:28523 (June 6-8, 2023) ("the Dubos reference"). Shearography is an optical technique used to check the porosity of composite materials (thus enabling the detection of air bubbles inside a manufactured or retreaded tire). Shearography is known for its robustness against environmental disturbances and vibrations (see the Dubos reference). Therefore, automating these machines is a significant opportunity.

[0004] However, a problem encountered by these machines concerns the production of extremely noisy images, with speckle noise inherent to their acquisition principle, which uses interferometry. In industry, historical criteria The characteristics that allow us to distinguish an anomaly are its area and amplitude. The industry must therefore detect different amplitude classes and also segment anomalies to assess their areas. As an example, [Fig. 1] shows a flank image (image (a) in [Fig. 1]) and a vertex image (image (b) in [Fig. 1]), both typical images obtained with a commercially available shearography machine.

[0005] Regardless of the type of tire, the steps in a verification procedure are essentially the same. The verification procedure includes a preparation step, during which the operation of the sensors is checked, the sensors are calibrated, and the tire is positioned for inspection. After the preparation process, the verification procedure includes a data acquisition step, during which the sensors collect and record data corresponding to the tire's position and its physical measurement. It is understood that the sensor(s) and / or the tire could be moved during this process. After the data acquisition process, the verification procedure includes a data analysis step, during which the acquired data is analyzed and a report of the results is generated.The verification process includes a final step during which the results are saved and the tire is returned to the inspection line of the verification workshop.

[0006] There are ongoing efforts to automate at least part of the tire inspection process. For example, one proposed solution uses a hybrid set of convolutional neural networks (or "CNNs") with faster, high-performance region-based convolutional neural networks (or "Faster Region-based CNNs"). In this solution, region proposal generation and object detection are performed using CNNs to detect bubble anomalies in tire shearography images. Anomaly detection is carried out using an artificial intelligence-based recognition algorithm for deep learning (namely, "Faster R-CNN") that detects limiting anomalies. The proposed model primarily uses sliding-window CNNs to divide the shearography images into regions in order to identify bubble anomalies.Subsequently, this step is followed by the implementation of faster region-based CNNs to identify bubble anomalies in tire shearography images. A faster region-based convo-lutional neural network (R-CNN) is chosen to minimize the false positive rate (see Chang, Chang, Chuan-Yu et al., Quality Assessment of Tire Shearography Images via Ensemble Faster Region-Based ConvNets, Electronics, doi.10.3390 / electronics9010045 (2020) (“Chang reference”)).

[0007] Other solutions propose the application of convolutional neural networks (or "fully convolutional network" or "FCN") in the industrial field. For example, one solution proposes a method based on an FCN to accurately locate and segment anomalies in radiographic images of tires (see Wang, Ren et al, "Tire Defect Detection Using Fully Convolutional Network", IEEE Access, doi:10.1109 / ACCESS.2019.2908483 (April 13, 2019) ("the Wang reference").

[0008] Another solution applicable to tire defect inspection proposes an end-to-end residual structure integrated U-net with a hybrid loss function and a coordinated attention module (or "HLU2-Net"). In the HLU2-Net, a new residual U-structure is used to replace the U-Net's encoding-decoding block to merge multi-scale and multi-level functions. Furthermore, a coordinated attention module is introduced to highlight useful features and weaken irrelevant ones (see Zheng, Zhouzhou, Zhang, Yan and Yu, Bin, "HLU2-Net: A Residual U-Structure Embedded U-Net with Hybrid Loss for Tire Defect Inspection", IEEE Transactions on Instrumentation and Measurement, doi:10.1109 / TIM.2021.3126847 (February 2023) ("the Zheng reference").

[0009] There are still major difficulties in detecting anomalies related to blowholes in tires. Poor visual quality makes acquiring tire images difficult, and the various machines used for automatic detection often exhibit certain undesirable characteristics (such as low contrast and low brightness). Furthermore, tire images incorporate images of the tread and sidewalls, which consist of different textures (the tread is made of thick rubber, and the sidewall is largely made of rubber reinforced with fabric and / or steel cables). It is also understood that some anomalies exist only in the tread and / or sidewall images.

[0010] In the tire manufacturing process, diagnosing "blowing" type anomalies in tire images obtained by shearography is an important task. Therefore, the disclosed invention relates to the automatic verification of "blowing" or "bubbles" type anomalies detected through machine learning by building a model based on labeled shearographic images. Summary of the invention

[0011] The invention relates to a shearography system incorporating a means for the automatic segmentation of anomalies detected by learning in images of pneumatics obtained by shearography, characterized in that the shearography system comprises: - a diagnostic device incorporating a shearography device, the diagnostic device comprising: - an inlet where tires intended for diagnostic use are introduced into the device; - a processing area where the device's shearography system takes images of each tire positioned within it; and - an outlet where the tires exit the device to undergo further processing; - a means of transfer which transports to the device the tires intended for diagnosis by the device; - at least one conveyor in scrolling which introduces the tires towards the entrance of the device to position them in the processing space and to remove them from the exit of the device; - a detection system that gathers information on the physical environment around the shearography system, the detection system comprising one or more sensors that detect the presence of tires in their field of vision; and - a communication network that manages the data entering the shearography system from the device, the communication network comprising at least one communication server for executing programmed instructions stored in a memory of one or more processors of the shearography system that employs (a) an image processing module of the processor, which analyzes images obtained from tires positioned in the processing space of the device; and (b) a module for executing an algorithm for detecting anomalies from the data obtained from the shearographic images and video images of the tires; in which the images taken are 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, performs a diagnostic procedure comprising the following steps: - a detection step of a tire positioned in the processing space of the device during which the device detects the presence of an arrangement of anomalies inside the tire within the detection field of the device; - an image acquisition stage during which the device takes as input shearographic images and video images containing a visible label that allows for the detection of a tire anomaly; and - a preprocessing step for the raw data of the images taken by the device, during which the pre-processed images are used as input to a trained neural network based on the Atrous Spatial Pyramid Pooling (ASPP) structure to obtain the results of segmentation of anomalies detected by learning in the images of tires taken.

[0012] In certain embodiments of the shearography system of the invention, the ASPP structure is arranged between an encoder and a decoder.

[0013] In certain embodiments of the shearography system of the invention, the trained neural network incorporates a stratified division performed on a dataset obtained from the device and a balancing of the training set; so that stratified division allows training data, validation data and test data to be distributed while maintaining a distribution of anomalies across all divisions.

[0014] In certain embodiments of the shearography system of the invention, the shearography system is configured to determine: - a class probability of each tire image taken from a training sample; and - a defined loss function which is calculated by the error between each tire image and a corresponding label in the training sample.

[0015] In certain embodiments of the shearography system of the invention, the defined loss function comprises a combination of IoULoss and CrossEntropy loss.

[0016] In certain embodiments of the shearography system of the invention, the data entering the shearography system includes general information concerning an identified tire.

[0017] In certain embodiments of the shearography system of the invention, the sensor(s) of the detection system are arranged in relation to the transfer means and the conveyor to manage the rate of introduction of the tires into the device.

[0018] In certain embodiments of the shearography system of the invention, at least one sensor of the detection system includes one or more learning programming modes to feed and train at least one neural network.

[0019] The invention also relates to a tire manufacturing installation comprising the disclosed shearography system.

[0020] Other aspects of the invention will become evident from the following detailed description. Brief description of the drawings

[0021] The nature and various advantages of the invention will become more evident upon reading the following detailed description, together with the accompanying drawings, in which The same reference numbers designate identical parts everywhere, and in which: [Fig.l] The [Fig.l] represents a side image and a top image obtained by a known shearography machine. [Fig.2] Fig.2 represents a schematic cross-sectional view of one embodiment of a known tire. [Fig.3] Fig.3 represents an embodiment of a diagnostic system of the invention which implements a method enabling the automatic diagnosis of anomalies inside one or more tires. [Fig.4] The [Fig.4] represents exemplary shearographic and video images taken of a tire during a diagnostic procedure implemented by the system of the [Fig.3]. [Fig.5] The [Fig.5] represents the architecture of a model of a neural network trained by the system of the [Fig.3]. [Fig. 6] Fig. 6 represents an improved pixel detection accuracy achieved by including ASPP in the model of Fig. 5. [Fig. 7] Fig. 7 represents a training implementation of the neural network shown in Fig. 5. Detailed description

[0022] When considering the characteristics of a tire intended for a diagnostic procedure, its geometry must be taken into account. A tire is an object with a known geometry, generally comprising several superimposed layers of rubber (or "layers"), as well as a metallic or textile fiber structure constituting a carcass that reinforces the tire's structure. The type of rubber and the type of reinforcement are chosen according to the desired final characteristics. Figure 2 includes a schematic representation of a tire 10, which typically includes two circumferential beads designed to allow the tire to be attached to a rim. Each bead includes an annular reinforcing bead. The construction of a tire is typically described by a representation of its components in a meridian plane, that is, a plane containing the tire's axis of rotation.The radial, axial, and circumferential directions respectively refer to the directions perpendicular to the tire's axis of rotation, parallel to the tire's axis of rotation, and perpendicular to any meridian plane. The terms "radially," "axially," and "circumferentially" mean, respectively, "along a radial direction," "along the axial direction," and "along a circumferential direction" of the tire. The terms "radially inside" and "radially outside" mean "closer, respectively." "further away" from the axis of rotation of the tire, in a radial direction.

[0023] The tire 10 also includes a tread 12, comprising a tread surface 12a and an inner surface 12b. The tread 12 is intended to come into contact with the ground via a tread surface 12a. The inner surface 12b ensures pressure maintenance within the tire.

[0024] The tire 10 further comprises a crown reinforcement including a working reinforcement 14 and a shrink-fit reinforcement 16, the working reinforcement 14 having working layers represented by layers 14a and 14b. The tire 10 also comprises two sidewalls (a sidewall 18 being shown in [Fig. 1]) and two reinforced blocks 20 with a bead 22. A radial carcass layer 24 extends from one bead to the other, surrounding the bead in a known manner. The tread 12 has reinforcements consisting, for example, of superimposed layers having known reinforcing threads. In some embodiments, the tire may include a rubber compound 26 that dissipates static electricity produced during rolling.

[0025] The tread 12 is bounded, in the radial direction, by two circumferential surfaces, the outermost of which is the tread surface 12a and the innermost of which is called the tread base surface. The tread base surface (or "bottom surface") is defined as the surface of the tread surface translated radially inward by a radial distance equal to the tread depth. It is common for this depth to decrease over the outermost axially circumferential portions (called "shoulders") of the tread 12.

[0026] Furthermore, the tread of a tire is delimited, along the axial direction, by two lateral surfaces. The tread is further constituted by one or more rubber compounds.

[0027] To achieve good grip on wet surfaces, cutouts are arranged in the tread 12. A cutout is defined as either a well, a groove, an incision, or a circumferential groove, and forms a space opening onto the tread surface 12a. The performance of a tread pattern must remain sufficiently consistent despite wear of the tread to ensure the longevity of the tire. Consequently, it is necessary to maintain a certain thickness of rubber material between the bottom face of the cutouts (grooves or grooves) and the reinforcing elements to guarantee the tire's durability. For example, grooves must be wide enough to allow the evacuation of liquid present on the ground surface regardless of the stage of wear of the tread (12).

[0028] With reference to the figures, in which the same numbers identify identical elements, [Fig. 3] represents an embodiment of a shearography system (or "system") 100 of the invention. The system 100 implements a diagnostic method (or "method") enabling the automatic diagnosis of anomalies such as "blowing" or "bubbles" inside one or more tires P (it is understood that the term "anomaly" is used here to refer to blowing, bubbles, and the corresponding anomalies). The method incorporates a machine learning method based on data corresponding to the tire images obtained by shearography. It is understood that the system 100 can be part of a tire manufacturing plant.

[0029] As used herein, the term “method” or “process” may include one or more steps performed by at least one computer system comprising one or more processors to execute instructions that perform the steps. Unless otherwise specified, any sequence of steps is given by way of example and does not limit the described methods to any particular sequence. It is understood that the system 100 can implement the diagnostic method in any physical environment without prior knowledge of the configuration of anomalies to be detected on the inner surfaces of the imaged tires. By performing the method, the system 100 achieves continuous improvement in the recognition of anomalies and their relative positioning along the inner surface of a tire.

[0030] The system 100 includes a diagnostic device (or "device") 102 which comprises a shearography device as known to those skilled in the art. Tires intended for diagnosis by the device 102 arrive by a known transfer means (for example, one or more tires P could arrive at the system 100 by means of one or more conveyors 104 as shown in [Fig. 3]). The tires P arrive one at a time on a conveyor 106 which ensures the movement of each tire towards an inlet 102a of the device 102 (see arrow A in [Fig. 3]).

[0031] After being introduced into the apparatus 102, the introduced tire is positioned in a processing space where the apparatus 102 will image it (see [Fig. 3] where tire P* represents such a positioned tire). During this step, the tire P* is positioned in the apparatus 102 so that the shearography device of the apparatus 102 can take images of the tire. It is understood that the identified tire could be positioned by the conveyor 106 onto a worktable or equivalent support so that the apparatus 102 can image it. The support can be configured to move rotationally, alternately vertically and / or alternately horizontally, thus allowing images to be taken of a variety of tires.

[0032] When the device 102 detects the tire P* positioned within the device 102, the device 102 detects the presence of an arrangement of anomalies inside the tire P* within the device's detection field, which triggers it to capture an image of a surface of the tire P*. The device 102 (and particularly its shearography device) takes images of the tire P* detected within the device 102, including taking one or more videos of the tire P*. This substantially simultaneous acquisition of images and videos allows for the assignment of labels to the anomalies, thus facilitating their subsequent detection. The system 100 "searches" the obtained images for the presence of the anomalies "seen" by the shearography device. If no anomalies are detected, the shearography device continues to acquire images until the search for the tire is exhausted.

[0033] At the end of a diagnostic process carried out by the apparatus 102, the conveyor 106 transports each imaged tire P to an outlet 102b of the apparatus for further processing (for example, storage of the checked tires, subsequent analysis for tires which require it, and / or removal of tires bearing one or more anomalies).

[0034] The system 100 also includes a detection system (not shown) for gathering information about the physical environment around the system. 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 system 100 (it is understood that the terms "sensor" and "camera" are used interchangeably). In embodiments of the system 100 shown in [Fig. 3], the sensor(s) of the detection system could be fixed above, beside, and / or around the system 100 (and particularly arranged with respect to the conveyors 104 and 106 to manage the transfer rate and the movement of the tires P, respectively).In embodiments of the system of the invention, these sensors could be part of an overall detection system that employs these sensors together with one or more sensors positioned in the physical environment in which the system 100 is installed. It is understood that one or more sensors (including cameras) may include one or more programming methods, including learning, to feed, modify, and train at least one neural network.

[0035] To properly manage the movement of the tires P to the device 102, which ensures clear image capture of each identified tire, it is necessary to identify the tire being diagnosed and detect the positioning of the relevant anomalies. Thus, the detection data refers to a plurality of recordings representative of the positioning of the anomalies of at least one A known tire or a portion of a known tire tracked over time. For example, sensing data may include one or more positions from records of the positions of a reference point on a portion of the tire (e.g., the inner surface 12b) over time or at defined time intervals; sensor data taken over time; a video stream that has been processed using a computer vision technique; and / or data indicating the operating status of device 102 over time. In some cases, sensing data may include representative data of one or more continuous movements of conveyor 106 before it stops so that device 102 can take one or more images of a tire.

[0036] Referring again to [Fig. 3], to implement the computer-based diagnostic method, the system 100 includes a communication network (or "network") 108 that manages the data entering the system 100 from various sources (for example, from the device 102 and its associated shearography device). The communication network 108 incorporates one or more communication servers (or "servers") 108a, each comprising one or more processors operationally connected to a memory. The memory is configured to store an application for analyzing representative data of the imaged tires (and more specifically, the images taken of the inner surfaces of these imaged tires).The processor(s) include an analysis application execution module that performs image processing, and the processor(s) are capable of executing programmed instructions stored in memory to carry out the steps of the diagnostic process (as described below).

[0037] Data entering system 100 may include general information about an identified tire. General information includes stored data concerning the identification of the identified tire (including, without limitation, its place of production, distribution and / or storage, production date, retreading history if applicable, and its position and mounting history). General information may also include the retreading rank (if applicable) of the identified tire. Data corresponding to a retreading rank of an identified tire is usually managed by the entity that manages the use of one or more identified tires (e.g., one or more persons and / or one or more companies) and / or the manufacturer of such tires.

[0038] 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 programs for their processing (for example, 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 several microcomputers, one or more programmable logic controllers (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 the data captured by device 102 and the detection system of system 100, as well as one or more software programs for identifying and locating variances and identifying their sources in order to correct them.

[0039] In System 100, memory may include both volatile and non-volatile memory devices. Non-volatile memory may include solid-state memories, such as NAND flash memory, keep-alive memory (or 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 data when System 100 (or a part of System 100) is powered off or loses its power supply. Volatile memory may include static and dynamic RAM that stores program instructions and data, including a learning application.

[0040] In embodiments of the system 100, the processor can configure the device 102 (and in particular its shearography device) according to one or more parameters calculated by an image processing module incorporated in the processor's memory. The image processing module analyzes the images of the tires and, more specifically, the images of the anomalies seen by the shearography device of the device 102.

[0041] The processor can also refer to a reference (for example, a table of various tire sizes) to perform a final determination of one or more parameters of a tire being imaged. The reference can include parameters from a plurality of commercially available, known tires. For example, after the image processing module has calculated one or more parameters of the imaged tire (see, for example, tire P* being imaged by device 102 in [Fig. 3]), the processor can compare the calculated parameters with the known parameters stored in the reference. The processor can retrieve the parameters of commercially available, known tires that most closely match the calculated parameters to configure device 102.The tire reference may include measurements corresponding to a plurality of commercially available tires. For example, for a tire size 225 / 50R17, the number "225" identifies the tire's cross-sectional area in millimeters, the number "50" indicates the sidewall aspect ratio, and the measurement "R17" represents the diameter of the rim. rim in inches (being approximately 43.18 centimeters).

[0042] It is understood that one or more sensors (including cameras) may include one or more programming modes, including learning, to feed, modify, and train at least one neural network. Referring again to [Fig. 3], and further to [Fig. 4], the neural network that takes as input the stereographic images obtained by the device will also process a video image obtained by the detection system of system 100. The system takes both types of images as input in order to take into account the presence of labels, which will be more visible on the video image. By way of example, [Fig. 4] represents a shearographic image (image (a) of [Fig. 4]) and a video image (image (b) of [Fig. 4]). The video image contains a visible label that allows for the detection of an anomaly, thus providing information to avoid detecting the "traces" around the detected anomaly.

[0043] With further reference to Figures 3 and 4, and moreover to Figures 5 and 6, [Fig. 5] represents an architecture of the model trained by the system 100. The architecture of the model represented in [Fig. 5] is based on a neural network incorporating the Atrous Spatial Pyramid Pooling (or “ASPP”) structure together with an encoder 200 and a decoder 202. To classify a pixel, the ASPP exploits multi-scale features by employing several parallel filters with different rates. The advantage is that it allows us to take into account a context that is more spatially distant from the pixel to be classified (for example, instead of taking 9 contiguous pixels in the "feature map", we will look for pixels a little further away) (see Chen, Liang-Chieh et al., DeepLah: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs, https: / / doi.org / 10.48550 / arXiv.1606.00915 (May 12, 2017) ("the Chen reference"). The [Fig.6] represents the results of this ASPP structure which improves the accuracy on pixel detection, particularly the comparison of the loss rate of the training set and the validation set (image (a) of [Fig.4]) and the overlap rate of the training set and the validation set (image (b) of [Fig.4]). .

[0044] Thus, the system 100 offers a means of automatically segmenting anomalies such as "blowing" or "bubbles" detected by learning in tire images obtained by shearography. During a diagnostic process implemented by the system 100, raw image data collected by the device 102 (the shearographic images) are obtained and preprocessed by the network 108. The preprocessing is performed to create training samples, which are divided into three parts: a training set, a validation set, and a test set to verify the performance of the trained network. The preprocessed images are taken as input to the model architecture based on the ASPP structure to obtain anomaly segmentation results. Based on this, the system determines the class probability of each tire image in the training sample; the defined loss function calculates the error between each tire image and the corresponding label in the training sample. Depending on the error and any deviation from the predefined conditions, the model parameters are updated.

[0045] The ASPP structure is defined in the network to capture multi-scale information from anomalies. The original data can therefore be directly extracted efficiently to obtain fast and accurate segmentation results: a supervised method based on deep learning primarily trains the model and label images through a neural network, automatically extracts the relevant features of the anomalies, and automatically completes the segmentation.

[0046] With reference to Figures 3 to 6, and further to [Fig. 7], [Fig. 7] represents a training embodiment of the neural network shown in [Fig. 5]. To improve the results, a stratified split is performed on a dataset obtained from device 102 and a balancing of the training set. This consists of (1) homogeneously distributing the shearographic images obtained from device 102 between "training," "validation," and "test"; and (2) repeating the rare indices in the training set to present the different classes equally. The stratified split allows the training, validation, and test data to be distributed while maintaining the distribution of anomalies across all splits (i.e., the same proportion is observed in the training / validation / test split). During the training process of the model shown in the [Fig.[5] A balanced split-train means that for training, and only training, the balance is artificially maintained by recalling "rare" examples at regular intervals. It is possible to experiment with several strategies to gain more leverage with additional hyperparameters that allow for a better definition of the balance between false and correct detection.

[0047] The loss function is a combination of IoULoss, which penalizes incorrect intersections over unions, and CrossEntropy loss. The various parameters are precisely chosen from PRCurves (with the threshold on the x-axis and the accuracies for the classes in question on the y-axis); the kernel size, in order to eliminate singular points that could lead to false detection; and the trigger threshold (in mm²) for each anomaly class, allowing for the final classification of the pneumatic.

[0048] Compared to the prior art, the disclosed system 100 has the following beneficial effects: - the disclosed system 100 uses an architecture based on the ASPP structure to directly extract features from original data information, and directly segment the target based on the obtained features, which has high efficiency and does not depend on manual intervention; - The constructed model has an ASPP structure that can capture multi-scale information from anomalies of the type "blowings" or "bubbles" and can efficiently extract the multi-scale receptive field from the high-level feature map characteristic, which improves the model's ability to effectively reduce the loss of information on anomaly characteristics during transmission processes.

[0049] Thus, the advantage of segmentation is that it would be possible to detect both the surface characteristic of the tire and also to classify the different types of amplitudes into different classes.

[0050] A diagnostic process can be performed by PLC control and can include pre-programmed management information. For example, a process setting can be associated with the parameters of the tire being imaged, and / or the properties of tires already imaged in a tire manufacturing plant incorporating System 100. System 100 (and / or a plant incorporating System 100) can easily repeat one or more steps of the process in a predetermined order to ensure the training of the disclosed model.

[0051] System 100 (and / or a tire manufacturing plant incorporating System 100) may include pre-programmed management information. For example, a process setting may be associated with the parameters of typical physical environments in which System 100 operates. In embodiments of the invention, System 100 (and / or a tire manufacturing plant incorporating System 100) may receive voice commands or other audio data representing, for example, a step or a stop in image capture (either shearographic or video images). A request may be made that includes a request for the current status of an ongoing diagnostic process cycle. A generated response may be represented audibly, visually, tactilely (for example, using a haptic interface), and / or virtually and / or augmented.This response, along with the corresponding data, can be recorded in a neural network.

[0052] For all embodiments of system 100, a monitoring system could be implemented. At least part of the monitoring system can be provided in a portable device such as a mobile network device (for example, a telephone). mobile, a laptop, one or more network-connected wearable devices (including "augmented reality" and / or "virtual reality" devices), network-connected wearable clothing and / or any combination thereof and / or any equivalent). It is conceivable that detection and comparison steps could be performed iteratively.

[0053] The terms "at least one" and "one or more" are used interchangeably. The ranges presented as being "between a and b" encompass the values ​​"a" and "b".

[0054] Although particular embodiments of the disclosed apparatus have been illustrated and described, it will be understood that various changes, additions, and modifications can be made without departing from the spirit or scope of this disclosure. Therefore, no limitations should be imposed on the scope of the invention described except those set forth in the appended claims.

Claims

Demands

1. A shearography system (100) incorporating a means for the automatic segmentation of anomalies detected by learning in images of tires obtained by shearography, characterized in that the shearography system (100) comprises: - a diagnostic device (102) incorporating a shearography device, the diagnostic device (102) comprising: - an inlet (102a) where pneumatics (P) intended for diagnostic use are introduced into the device; - a processing area where the shearography device of the apparatus (102) takes images of each tire (P) positioned within it; and - an outlet (102b) where the pneumatics (P) exit the device (102) to perform further treatment; - a transfer means (104) which transports to the device (102) the tires (P) intended for diagnosis by the device; - at least one conveyor (106) in scrolling which introduces the tires (P) towards the inlet (102a) of the device (102) to position them in the processing space and to bring them out of the outlet (102b) of the device; - a detection system that gathers information on the physical environment around the shearography system (102), the detection system comprising one or more sensors that detect the presence of tires in their field of vision; and - a communication network (108) which manages the data entering the shearography system (100) from the device (102), the communication network comprising at least one communication server (108a) for executing programmed instructions stored in a memory of one or more processors of the shearography system which employs (a) an image processing module of the processor, which analyzes images obtained from the tires (P) positioned in the processing space of the device (102); and (b) a module for executing an algorithm for detecting anomalies from the data obtained from the shearographic images and the video images of the tires (P); in which the images taken are 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, performs a diagnostic process comprising the following steps: - a step of detecting a tire (P*) positioned in the processing space of the device (102) during which the device (102) detects the presence of an arrangement of anomalies inside the tire (P*) in the detection field of the device; - an image acquisition step during which the device (102) takes as input shearographic images and video images containing a visible label which allows the detection of an anomaly of the tire (P*);and - a preprocessing step of the raw data of the images taken by the device (102), during which the preprocessed images are used as input to a trained neural network based on the Atrous Spatial Pyramid Pooling (ASPP) structure to obtain the results of the segmentation of anomalies detected by learning in the images of tires taken.;

2. The shearography system (100) of claim 1, wherein the ASPP structure is arranged between an encoder (200) and a decoder (202).

3. The shearography system (100) of claim 2, wherein the trained neural network incorporates a stratified split performed on a dataset obtained from the device (102) and a balancing of the training set; such that the stratified split allows for the distribution of training data, validation data and test data while maintaining a distribution of anomalies in all the splits.

4. The shearography system (100) of claim 3, wherein the shearography system (100) is configured to determine: - a class probability of each tire image taken in a training sample; and - a defined loss function that is calculated by the error between each tire image (P) and a corresponding label in the training sample.

5. The shearography system (100) of claim 4, wherein the defined loss function comprises a combination of IoULoss and CrossEntropy loss.

6. The shearography system (100) of any one of the claims indications 1 to 5, in which the data entering the shearography system includes general information concerning an identified tire.

7. The shearography system (100) of any one of claims 1 to 6, wherein the sensor(s) of the detection system are arranged relative to the transfer means (104) and the conveyor (106) to manage the rate of introduction of the tires (P) into the device (102).

8. The shearography system (100) of any one of claims 1 to 7, wherein at least one sensor of the sensing system includes one or more learning programming modes for feeding and training at least one neural network.

9. A tire manufacturing plant comprising the shearography system of any one of claims 1 to 8.

Images