Removal of defects from the buffed tire casing surface.

Abstract

A method for determining whether to apply a remediation process to remove defects from the surface of a buffed casing of a pneumatic tire, the method comprising: receiving tire-specific usage data; determining a defect threshold based on the received tire-specific usage data; acquiring three-dimensional profile data of at least one portion of the casing surface using a surface profile measuring device; processing the profile data to determine the total number of defects on at least one portion of the surface; determining whether the total number of defects is less than a defect threshold; determining, on the condition that the total number of defects is less than a defect threshold, to apply a remediation process to remove defects from the surface; and applying the remediation process to the tire.

Description

Technical Field

[0001] The present disclosure relates to methods and systems for detecting defects on the surface of a buffed casing of a pneumatic tire, and processes for removing those defects.

Background Art

[0002] When the tread of a vehicle tire (i.e., the part of the tire that contacts the road) wears out, it is often economical and environmentally friendly to replace the worn tread with a new tread rather than installing a completely new tire on the vehicle. The process of replacing the tread is often referred to as "retreading." Retreading a tire typically involves three steps. First, the worn tread is removed, leaving the tire's "buffed casing" exposed. Next, the surface of the buffed casing is inspected for local defects. To remove the identified defects, a treatment (also known as "skiving") is applied to the tire. Finally, a new tread is applied to the "skived" casing, and the retreaded tire is installed on the vehicle for further use.

[0003] In order to reduce (or completely eliminate) the need for human operator intervention, it is generally desirable to provide an automated station for performing retreading. Even when automated defect inspection and skiving are performed, defects may not be correctly identified, or inappropriate surface treatment may be applied to the identified defects. Thus, these stations often still require undesirable manual input by an operator.

Summary of the Invention

[0004] An object of the present invention is to provide a novel and useful method and system for detecting surface defects of a tire and selecting a treatment process for removing the surface defects of the tire.

[0005] A first aspect of the present invention relates to a method for detecting defects on the surface of a buffed casing of a pneumatic tire. The method can be performed by an automated tire inspection and skiving station as part of a retreading process applied to a tire (for example, after a worn tread has been removed and before a new tread has been applied). The tire is mounted on a rotating device configured to rotate the tire in steps around its axis of rotation. The rotating device includes an angle encoder for controlling the angular position of the tire. The method involves rotating the tire in steps. Using a camera having a single pixel line (also known as a “linear camera”), a one-dimensional array of pixel intensity values ​​(“line image”) for the corresponding image portion of the equatorial plane of the casing is captured at each stop in the stepwise rotation. The method further includes associating each one-dimensional array of pixel intensity values ​​with the corresponding angular position of the tire, synthesizing multiple one-dimensional arrays to form two-dimensional image data of the equatorial plane of the casing based on the associated angular positions of the tire, and processing the two-dimensional image data of the equatorial plane of the casing to detect defects.

[0006] This aspect of the present invention enables faster detection of surface defects on tire casings than conventional systems by, for example, using a linear camera to image relevant portions of the casing surface. Conventional systems may rely on cameras that capture two-dimensional images. Such cameras are typically slower than linear cameras. Furthermore, two-dimensional images captured by cameras may contain important areas that cannot be used for defect detection because these portions of the tire surface are out of focus (because the equatorial plane, which is the most relevant part of the tire surface, is curved). Consequently, the time spent acquiring and processing these areas, and the memory used to store the corresponding image data, is usually wasted. In contrast, since all portions captured by a linear camera are in focus, the efficiency of the defect detection process can be improved. The present invention can thereby improve the productivity of automated tire inspection stations.

[0007] In one embodiment, the tire may have first and second shoulders, and each imaging portion may extend substantially from the first shoulder to the second shoulder of the tire. Thus, defects can be detected at any position between the shoulders.

[0008] In one embodiment, the method may further include acquiring three-dimensional profile data of the equatorial plane of a casing using a surface shape measuring device, and identifying one or more regions of interest on the equatorial plane from the profile data. Regions of interest can be identified in many ways (for example, by thresholding the profile data or by using a neural network). In one example, regions of interest can be identified as regions having anomalous distance values ​​in the profile data, i.e., profile data values ​​greater than a threshold, such as a predetermined threshold. The method may further include identifying portions of image data corresponding to one or more regions of interest on the equatorial plane, and processing the identified portions of the image data or each identified portion in order to detect defects.

[0009] Therefore, three-dimensional profile data can be used to identify regions of interest and further process only the image data corresponding to these regions. In other words, profile data can be used to filter image data and extract portions relevant to further image processing. In this way, less image data is processed, and the overall time required to detect defects on the tire surface can be reduced.

[0010] In the above, processing the identified portion of the image data or each identified portion in order to detect defects is, The process involves using a neural network to process identified portions of image data or each identified portion in order to detect defects and, for each detected defect, generate a score distribution across a corresponding set of possible tire defect types. This includes selecting a treatment process from a set of treatment processes for a region of interest on the equatorial plane or for each region of interest, based on the score distribution of defects detected in the corresponding portion of the image data.

[0011] In this way, the appropriate corresponding remediation process can be automatically selected for each of several types of defects. This means that the appropriate remediation process for each defect can be selected more quickly and reliably than when the remediation process is selected by a skilled human operator. The method may further include applying each selected remediation process to a region of interest on the equatorial plane or to each region of interest.

[0012] Neural network models can include convolutional neural networks, i.e., neural networks that include at least one convolutional layer applying a trained kernel to a two-dimensional dataset. Experimentally, it has been shown that such networks can process patches of image data corresponding to the regions of interest identified as described above to generate defect labels that match those generated by human experts.

[0013] This concept provides a second independent embodiment of the present invention, which is a method for selecting a treatment process for removing defects from the surface of a buffed casing of a pneumatic tire using a neural network model. The method can be implemented on one or more computing devices. One or more computing devices can be connected to an automated tire inspection and skiving station. The method includes receiving image data of a portion of the surface of the buffed casing. The image data can be generated by an inspection and skiving station connected to a computing device. For example, the images can be captured using a linear camera (as in the first embodiment of the present invention) or using a two-dimensional camera. The method further includes detecting at least one defect and, for each detected defect, processing the image data using a neural network to generate a score distribution across a set of defect labels (e.g., "clean skive", "rust", "frayed metal", "no belt", "unknown", etc.). The set of defect labels includes defect labels associated with corresponding possible tire defect types and, optionally, at least one defect label ("unknown") that is not associated with any defect type in the set of defect types (e.g., can be assigned to a defect not included in the set of defect types). The neural network may be any known architecture suitable for performing object detection and classification on input images. For example, the neural network model may include a convolutional neural network.

[0014] This aspect of the present invention can enable more informed decision-making regarding which treatment process to select. This is because neural networks can recognize (and distinguish) more defect types than conventional methods typically used to process images of surface defects on buffed casings. This makes it possible to select the most appropriate treatment for a particular part of the tire without requiring a skilled human operator to make the decision, thus increasing the efficiency of inspection and skiving stations.

[0015] In either a first or second aspect of the present invention, the neural network can be trained using a known supervised learning method. For example, the neural network can be trained by tuning its network parameters, thereby prompting the neural network to process a training dataset that has been (manually) labeled for defect detection and to generate a score distribution that matches each label in the training dataset. The method further includes selecting a treatment process for a portion of the surface from a set of treatment processes based on the score distribution. In one embodiment, the method may further include applying the selected treatment process to the portion of the surface using a repair device controlled by one or more computer devices (e.g., by using a tool attached to a positioning device such as a robotic arm).

[0016] A third aspect of the present invention relates to a method for selecting a treatment process for removing defects from the surface of a buffed casing of a pneumatic tire. The method includes: acquiring three-dimensional profile data of at least a portion of the surface of the casing (e.g., the equatorial plane or a portion of the sidewall of the casing) by using a surface shape measuring device; processing the profile data to determine the total number of defects on at least a portion of the surface; and determining whether the total number of defects is less than a defect threshold. The method further includes, provided that it is determined that the total number of defects is less than a defect threshold, applying a first treatment process to the surface of the buffed casing; acquiring image data of the equatorial plane of the casing; processing the image data of the equatorial plane of the casing to detect defects; and selecting a further treatment process from a set of treatment processes based on the detected defects on the equatorial plane of the casing.

[0017] By determining whether the number of defects falls below a threshold, if the expected future lifespan of the tire after the remediation process is deemed unacceptably short, the remediation process will not be performed on the tire. Thus, uneconomical repairs are avoided, and the risk to occupants of vehicles equipped with rapidly failing tires is reduced.

[0018] The first treatment process may include applying a wire brush to the tire casing.

[0019] In one embodiment, the method may further include applying a selected further treatment process to the buffed surface of the casing (for example, by using a tool attached to a robotic arm).

[0020] This aspect of the present invention enables faster inspection and skiving of tire casings. This is because, by applying the first treatment process, some defects can already be removed (to identify defect-specific treatments) before the relatively slow steps of image acquisition and image processing are performed. This can reduce the time required to perform these steps and increase the productivity of defect inspection and skiving stations.

[0021] A fourth aspect of the present invention relates to a method for determining whether (or not) to apply a remediation process to remove defects from the surface of a buffed casing of a pneumatic tire. The method includes receiving tire-specific usage data. The method further includes determining a defect threshold based on the received tire-specific usage data, acquiring three-dimensional profile data of at least a portion of the casing surface by using a surface shape measuring device, processing the profile data to determine the total number of defects on at least a portion of the surface, and determining whether the total number of defects is less than the defect threshold. The method further includes determining to apply a remediation process to remove defects from the surface, provided that the total number of defects is determined to be less than the defect threshold, and applying the remediation process to the tire. If the total number of defects is greater than or equal to the defect threshold, the tire may be considered unsuitable for retreading.

[0022] This aspect of the present invention enables the accurate assessment of a tire's suitability for retreading using available information about its history. This allows for the implementation of detailed safety considerations and reduces costs (for example, by avoiding further processing steps if the tire is found to be unsuitable).

[0023] Tire-specific usage data can, for example, identify the tire's previous use. This can be done by identifying the previous use (e.g., truck tire, bus tire, etc.) from a predetermined set of possible uses for the tire. Additionally or alternatively, tire-specific usage data can identify (at least) previous measurements of the tire's physical quantities (e.g., results of previous defect inspections) or the tire's service history. Alternatively, tire-specific usage data may identify the tire's intended use (e.g., by identifying one of a predetermined set of uses for the tire), independently of the tire's previous use.

[0024] Tire-specific usage data can be stored (and read from) devices mounted inside or on the tire, or otherwise attached to the tire (e.g., tire-mounted sensor (TMS) units, or radio transmission devices such as radio frequency identification (RFID) tags mounted inside or on the tire). Additionally or alternatively, tire-specific usage data may be stored (and read from) a database. The database may be stored on a remotely located cloud server or a local server. If tire-specific usage data is stored in a database, the correct tire-specific usage data can be read from the database using an identification code that uniquely identifies the tire (the identification code may be on the tire, or stored in a device mounted inside or on the tire). Generally, tire-specific usage data can be updated throughout the entire lifecycle of a particular tire. Additionally or alternatively, tire-specific usage data may include the date of manufacture, manufacturer, model number, batch or serial number, and / or further details that identify the tire's lifecycle. In one embodiment, tire-specific usage data can further identify defects (e.g., number of defects, type of defects, etc.) detected during previous retreading processes applied to the tire.

[0025] In a further example, the tire-specific usage data can indicate the type of past users of the tire (i.e., can include a label indicating one of a plurality of predetermined user classes). This is important because the past usage of the tire can be related to its future usage, and as a result, the tire-specific usage data can indicate its future usage. For example, if the past usage of the tire was on a passenger vehicle (e.g., a school bus), a higher level of safety may be desired, and thus, a lower defect number threshold than that applicable to a truck may be desired. More generally, the defect number threshold may depend on the user class (user type) indicated by the user class.

[0026] In one embodiment, applying a treatment process to the surface of the buffed casing can include obtaining image data of the equatorial plane of the casing and processing the image data to (i) detect surface defects, (ii) associate at least one defect label from a set of defect labels with each detected defect, and (iii) determine the total number of defects associated with each label in the set of defect labels. The method further includes determining a set of additional defect number thresholds, each additional defect number threshold being associated with one label in the set of defect labels, based on the received tire-specific usage data, and for each label in the set of defect labels, determining whether the total number of defects associated with the respective label is below the corresponding additional defect number threshold, and applying a treatment process to the surface of the buffed casing, conditional on determining that the total number of defects associated with each label in the set of defect labels is below the corresponding additional defect number threshold.

[0027] By considering individual thresholds for various defect types, more sophisticated safety considerations can be implemented. For example, the threshold for "rust" defects (i.e., the maximum allowable number of "rust" defects before the tire is considered unsuitable for retreading) may be higher for truck tires than for bus tires. Note that some types of defects may be more critical depending on the use of the tire. For example, certain types of defects are more likely to cause future problems in the case of tires to be used on vehicles carrying heavy loads or vehicles exposed to extreme weather conditions.

[0028] There is benefit in using a combination of multiple aspects of the first to fourth aspects of the present invention, but each of these aspects is independent and can be used without the other aspects. For example, the image data used in the second to fourth aspects of the present invention can be captured by an imaging device configured to capture a two-dimensional array of intensity values.

[0029] The fifth to eighth aspects of the present invention relate to a system configured to execute the methods according to the first to fourth aspects, respectively.

Brief Description of the Drawings

[0030] Embodiments of the present invention will be described by way of example only, with reference to the accompanying drawings. [Figure 1] An exemplary system for detecting defects on the surface of a buffed casing of a pneumatic tire is schematically shown. [Figure 2] It is a flowchart of an exemplary method for detecting defects on the surface of a buffed casing. [Figure 3] An exemplary system for defect classification is schematically shown. [Figure 4A] An exemplary image of a buffed tire casing is shown. [Figure 4B] A table of experimentally obtained defect classification results is shown. [Figure 5]This is a flowchart illustrating an exemplary method for selecting a treatment process to remove defects from the surface of a buffed casing using a neural network model. [Figure 6] A schematic diagram illustrates an exemplary system for implementing the method shown in Figure 5. [Figure 7] The following shows an exemplary score distribution for each of the three defects. [Figure 8] An exemplary scheme for assigning treatment scores is shown. [Figure 9] An example of mapping data for selecting treatments based on treatment scores is shown. [Figure 10] This is a flowchart illustrating a further exemplary method for selecting a treatment process to remove defects from the surface of a buffed casing. [Figure 11] This is a flowchart illustrating an exemplary method for determining whether to apply a corrective process to remove a defect. [Figure 12] This is a flowchart illustrating an exemplary method for performing automated skiving of buffed casings. [Modes for carrying out the invention]

[0031] Figure 1 shows a system 1 for inspecting pneumatic tires, such as truck tires and bus tires. System 1 can be used on tires undergoing a retreading process. In particular, system 1 is suitable for detecting defects on the surface of tires from which the worn tread has been removed in the initial steps of the retreading process (such tires are also known as buffed tire casings or buffed tire carcasses). System 1 may operate automatically, i.e., requiring only minimal human intervention (if necessary). In some embodiments, system 1 may be part of an automated defect inspection and skiving station that performs defect detection and defect removal procedures on the buffed casing. The buffed tire casing inspected and treated by such a defect inspection and skiving station may then be subject to further steps of the retreading process (e.g., application of a new tread). In some embodiments, system 1 is part of an automated retreading station.

[0032] System 1 includes an electric hub 7 for releasably mounting the tire 3 and for rotating the tire 3 around its axis of rotation (in the embodiment of Figure 1, the axis of rotation of the tire 3 is parallel to the z-axis). The electric hub 7 may include a servo motor having an angle encoder for controlling the angular position of the tire 3. In the embodiment of Figure 1, the tire 3 is a buffed tire casing. In a broad sense, the outer exposed surface of a buffed tire casing 3 may have surface defects (commonly referred to as scratches). Such defects can affect whether the tire casing 3 is suitable for retreading. These defects often need to be removed before a new tread can be applied to the tire casing 3.

[0033] The outer surface of the tire casing 3 may include a substantially toroidal curved surface 5 (also known as the equatorial plane or "crown" portion of the casing) and two opposing side wall portions 4 (one side wall portion 4 can be seen in Figure 1, while the other side wall portion is located on the opposite side of the tire casing 3 and cannot be seen in this figure). The intersection of the side wall 4 and the equatorial plane 5 forms the shoulder portions 6 and 8 of the tire casing 3.

[0034] System 1 further comprises a camera 9 configured to image an elongated portion of the outer surface of the tire casing 3. In particular, camera 9 has only a single pixel line (such a camera is also known as a “linear camera,” and the resulting image is known as a “line image” or “one-dimensional image”). Thus, for each “line image” acquired, camera 9 can output a one-dimensional array of pixel intensity values ​​(i.e., a one-dimensional array containing numerical values ​​that identify the pixel intensity measured at each position spaced apart in the z direction). In the embodiment of Figure 1, camera 9 may be a color camera (for example, the camera may provide RGB values ​​for each pixel intensity value). In other embodiments, camera 9 may be a monochrome camera.

[0035] In the embodiment shown in Figure 1, the camera 9 is configured to image a portion 10 of the equatorial plane 5. In particular, the camera 9 may be positioned such that the imaging portion 10 extends substantially over the entire length of the equatorial plane 5 in the z direction, i.e., substantially from shoulder 6 to shoulder 8. The imaging portion 10 may be narrower in the z direction than in the circumferential direction of the equatorial plane 5 (i.e., the rotational direction around the axis of the tire). In particular, the imaging portion 10 only needs to extend to a length corresponding to the imaging length of one pixel of the camera 9 in the circumferential direction of the equatorial plane 5. For example, the imaging portion 10 may extend less than 1 mm along the circumferential direction of the equatorial plane 5.

[0036] In one embodiment, the camera 9 may be mounted on a robotic arm configured to adjust the position of the camera 9 so that the imaging portion 10 is in focus of the camera 9 (i.e., so that, for example, the spatial resolution, sharpness, and / or contrast of the resulting image meet sufficiency criteria). Additionally or alternatively, the camera 9 may include an adjustable focus system configured to focus on the imaging portion 10.

[0037] System 1 further comprises a control unit 15 configured to control the rotation of the tire casing 3 by the electric hub 7 and the acquisition of images by the camera 9. As described below with reference to Figure 2, an image of the entire equatorial plane 5 can be generated by combining multiple line images acquired by the camera 9. For this purpose, the control unit 15 can synchronize the acquisition of line images with the stepwise rotation of the tire casing 3 (by the electric hub 7) such that one of the multiple line images is captured at each stop in the stepwise rotation.

[0038] System 1 may further include a surface shape measuring device camera 13 configured to capture three-dimensional profile data of a portion 14 of the equatorial plane 5. The surface shape measuring device 13 may be a laser surface shape measuring device. The surface shape measuring device 13 may be positioned at a fixed location on the tire 4 facing the equatorial plane 5 in order to capture the three-dimensional profile of the equatorial plane 5. In other embodiments, the surface shape measuring device 13 may be mounted on a robot arm.

[0039] The surface shape measuring device 13 is configured to measure the distance between the surface shape measuring device 13 and a plurality of points contained in a portion 14 of the equatorial plane 5. In particular, the plurality of points measured by the surface shape measuring device 13 are located on a line parallel to the z-axis (i.e., parallel to the rotation axis of the tire). The surface shape measuring device 13 can emit a laser beam that moves from shoulder 6 to shoulder 8 to scan the portion 14 of the equatorial plane 5. The surface shape measuring device 13 can output a one-dimensional array containing numerical values ​​indicating the distance of corresponding points on the equatorial plane 5 from the rotation axis of the tire, for example, each numerical value specifying the distance between the surface shape measuring device 13 and the corresponding point on the equatorial plane 5.

[0040] The surface shape measuring device 13 can be controlled by the control unit 15. As described below with reference to Figure 2, the three-dimensional profile data of the entire equatorial plane 5 can be generated by combining multiple profile data measurements. For this purpose, the control unit 15 can synchronize the acquisition of profile data with the stepwise rotation of the tire casing 3 (by the electric hub 7) so that one of the multiple profile data is captured at each stop in the stepwise rotation.

[0041] In one embodiment, System 1 may include means (e.g., a computing device) for processing line images acquired by the camera 9 and profile data acquired by the surface shape measuring device 13 in order to perform steps S103 and / or S105 of the method described below with reference to Figure 2. In another embodiment, System 1 may be connected to a remote computing device or cloud computing service (e.g., via the Internet) to process line images acquired by the camera 9 and / or profile data acquired by the surface shape measuring device 13 in order to perform steps S103 and / or S105 of Figure 2.

[0042] System 1 may be further configured to acquire three-dimensional profile data of the outer surface of the side wall 4 of the tire casing 3. In this case, the surface shape measuring device 13 can be mounted on a positioning device (e.g., a robotic arm) configured to move the surface shape measuring device 13 between the respective positions to acquire profile data of the side wall 4 or the equatorial plane 5. Alternatively, System 1 may include a second surface shape measuring device configured to acquire three-dimensional profile data of the outer surface (at least a portion thereof) of the side wall 4.

[0043] System 1 may further comprise several tools (not shown in Figure 1) for applying a treatment process to a selected portion of the equatorial plane 5. These tools may include polishing tools, such as (wire) brushes, carbide wheels, and router cutters. System 1 may further comprise a robotic arm (or other positioning device) for operating the tools. The tools may be configured to be operated by the robotic arm. The treatment process can be applied to remove defects on the equatorial plane 5 (defects may be identified according to the methods described below with reference to Figure 2). The treatment process can be performed by the robotic arm based on a set of treatment parameters. The treatment parameters may specify the tools used for a particular treatment, and the operating conditions of the tools, such as the angular velocity of the tool (typically defined in revolutions per second), the angle of the tool, and so on. For example, to perform a treatment process, the robotic arm can be (releasably) mounted on the tool specified by the treatment parameters and can operate the tool according to the operating conditions specified by the treatment parameters.

[0044] System 1 in Figure 1 includes a surface shape measuring device 13, but in other embodiments, the surface shape measuring device 13 may not be present.

[0045] Referring to Figures 2 and 3, a method for detecting defects on the surface of a buffed casing is described. Schematically, defects are detected by image processing of image data representing the equatorial plane 5 (or the relevant portion thereof) of the casing 3, which is generated by synthesizing multiple line images. This method can be implemented using System 1 in Figure 1. In the first step S101, multiple line images of corresponding portions of the equatorial plane 5 are acquired using a linear camera 9. The acquisition of multiple line images can be synchronized with the stepwise rotation of the tire casing 3 by the hub 7. In particular, one of the multiple line images can be acquired at each stop during the stepwise rotation.

[0046] For example, the control unit 15 can control the hub 7 to rotate the tire casing 3 by one revolution in steps (i.e., the rotation phase alternates with a stationary phase to "stop"). In each step of the stepwise rotation, the tire casing 3 is rotated by a predetermined angle (e.g., 0.1 degrees, 0.5 degrees, or 1 degree). Between each stationary phase (i.e., between any two steps of the stepwise rotation), one of several line images can be captured. Each acquired line image can be associated with the corresponding angular position of the tire casing 3 (provided by the angular encoder of the hub 7). For each line image, a tuple (φ, Z) may be formed, where φ specifies the angular position of the tire casing 3 when each image was captured, and Z = [z1, z2, ..., z N ] is the corresponding one-dimensional array provided as output by camera 9 (z1~z N (where represents the numerical value of N pixels in camera 9).

[0047] In one embodiment, in step S101, profile data of the equatorial plane 5 can also be acquired using the surface shape measuring device 13. For this purpose, the acquisition of profile data by the surface shape measuring device 13 can be synchronized with the stepwise rotation of the tire casing 3 such that one of a plurality of profiles is acquired at each stop in the stepwise rotation. Each acquired profile can be associated with the corresponding angular position of the tire casing 3 (provided by the angular encoder of the hub 7). For each acquired profile, a tuple (φ,H) may be formed, where φ specifies the angular position of the tire casing 3 when each profile was acquired, and H=[h1,h2,...,h M ] is a corresponding one-dimensional array provided as output by the surface shape measuring device 13 (h1~h M (where represents the numerical value of the distance between the surface shape measuring device 13 and M points on the corresponding portion 14 of the equatorial plane 5). In this embodiment, the number of line images acquired and the number of three-dimensional profiles captured may be the same.

[0048] In step S103, multiple line images are combined to form two-dimensional image data of the equatorial plane 5. In particular, the multiple line images may be combined based on the associated angular positions of the tire casing 3. For example, each row of the two-dimensional image data can be input based on its respective tuple (φ, Z). The pixel value Z can form the numerical value of each row, and the angular position φ can identify the row index of each row having two-dimensional image data. In a broad sense, the effect of using the linear camera 9 is that a complete two-dimensional image of the tire surface can be generated faster than conventional methods that rely on acquiring several two-dimensional images. This is because linear cameras are generally significantly faster than two-dimensional cameras. Furthermore, because the equatorial plane 5 is a curved surface, important areas of any acquired two-dimensional image are often unusable for defect detection because the tire surface is out of focus (and therefore the time spent acquiring these areas is usually wasted). Thus, an image segmentation process is required to select useful information from the two-dimensional image, and the rest of the image is removed. This process consumes time and computational resources. When image segmentation is performed after multiple two-dimensional images have been captured, a considerable amount of memory is required to store the images.

[0049] In an embodiment in which multiple profiles (one-dimensional arrays of distance values) are acquired in step S101, in step S103, the multiple acquired profiles can be combined to form composite profile data (i.e., a two-dimensional array) representing the three-dimensional profile of the equatorial plane 5. In this sense, this (composite) profile data can be said to be "three-dimensional". For example, each row of the composite profile data can be input based on its respective tuple (φ, H). The distance value H can form the numerical value of each row, and the angular position φ can identify the row index of each row having composite profile data.

[0050] Although the generation of image data and composite profile data (by performing steps S101 and S103) has been described as being based on synchronizing the stepwise rotation of the tire casing 3 with the acquisition of each data, in another embodiment, it is understood that the tire casing 3 can be rotated continuously while the camera 9 and the surface shape measuring device 13 acquire their respective data. In this case, the acquired data may be adversely affected by motion blur. By appropriately adjusting the rotation speed and the respective acquisition times of the camera 9 and the surface shape measuring device 13, the effects of motion blur can be mitigated, and image data and composite profile data suitable for defect detection can be obtained.

[0051] Referring back to Figure 2, in step S105, two-dimensional image data of the equatorial plane 5 is processed for defect detection. Defect detection may include determining the presence of surface defects in specific identified portions of the equatorial plane 5. In general, defect detection can be performed more efficiently and reliably by processing two-dimensional image data rather than processing line images individually. This is because defects are identified (at least partially) by their shape, and defects commonly found on the surface of a buffed casing may extend (circumferentially) across several line images. Thus, the shape of any detected object can be determined more reliably by processing two-dimensional image data. The image data can be processed using any suitable image processing technique (e.g., known object detection algorithms, neural networks such as neural network model 40 described below with reference to Figure 3).

[0052] In one embodiment, processing two-dimensional image data may include identifying candidate objects and then classifying the candidate objects as one of a set of possible tire defect types (defect classes) by comparing the determined characteristics of the objects (e.g., size, shape, and / or color) with the characteristics (manually defined or learned) of (commonly found) defects.

[0053] In a broad sense, processing the entire two-dimensional image data can be computationally expensive (because the two-dimensional image data can be large). This can hinder (near) real-time defect detection and slow down the (mostly automated) process of inspecting the buffed tire casing 3 (and thus reduce the productivity of the automated station). Therefore, it is desirable to reduce the computational resources required to process the two-dimensional image data. This can be achieved by identifying one or more regions of interest (ROIs) on the equatorial plane 5 that are likely to contain defects, and by processing only the portion of the image data corresponding to the ROIs. The buffed tire casing typically has a small number of small defects (typically fewer than 100 defects, or even fewer than 10 defects, each of which is typically 100 mm) 2 This can be beneficial because it only has (and may extend over less than) a certain period. Therefore, the number and size of ROIs can be reduced, and the overall processing time can be shortened.

[0054] In embodiments where composite profile data of the equatorial plane 5 is obtained in steps S101 and S103, the composite profile data can be used to identify one or more ROIs on the equatorial plane 5. In one embodiment, an ROI is identified by identifying anomalous distance values ​​in the profile data, i.e., distance values ​​greater than a predetermined threshold. Such anomalous distance values ​​may indicate craters (or "skives") on the equatorial plane 5. An ROI can be defined as a region of a predetermined size centered on each of the anomalous distance values. In other embodiments, ROIs can be identified as a segmentation process by processing the composite profile data with a neural network model (e.g., a neural network including a convolutional neural network). The neural network can be trained to generate these ROIs by supervised learning based on a database of the set of profile data and the corresponding ground truth ROIs.

[0055] Figure 3 illustrates the process described above for identifying ROIs and processing only the image data corresponding to the ROIs. Exemplary profile data 20 can be processed to identify an exemplary ROI 24 centered on an unusual distance value indicating a potential defect 22. Next, ROI 24 in profile data 20 is mapped to identify the corresponding ROI 28 in exemplary image data 26, i.e., ROI 24 and ROI 28 correspond to substantially the same portion of the equatorial plane 5. Such mapping can be understood to take into account that (at each step of the stepwise rotation of the tire casing 3) the camera 9 and surface shape measuring device 13 are typically inspecting spatially separated portions 10, 14 of the equatorial plane 5 (as shown in Figure 1), and as a result the row indices of ROI 24, 28 are offset from each other by a (fixed) offset. Image data within ROI 28 is processed to detect defects.

[0056] Generally, the neural network model 40 can be configured to receive an input image of the equatorial plane 5 (e.g., the entire two-dimensional image data, or a separate portion of the image data corresponding to an identified ROI). The neural network model 40 can process the input image to identify one or more defects and classify the detected defects. For example, for each detected defect, the neural network 40 can generate a score distribution across a set of corresponding possible tire defect types. The neural network 40 can then provide, as output, the classification results (e.g., the identified defects, the corresponding score distributions, and a list of the defect locations in the input image). Based on the score distribution of each defect, one of a set of corresponding defect labels can be assigned to each defect. In one embodiment, the set of defect labels may include "clean skive" (associated with clean areas with embedded and exposed belt wires), "rust" (associated with rust on or near belt wires), "frayed metal" (associated with loose belt wires), "no belt" (associated with unexposed belts), and "unknown" (associated with defects other than those mentioned above).

[0057] The neural network model 40 may be a known architecture suitable for performing object detection and classification on input images. For example, the neural network model 40 may include a convolutional neural network. The neural network model 40 can be implemented using one or more computer systems located in one or more locations.

[0058] In one embodiment, the neural network 40 may include a conventional object detection model that is initially pre-trained (e.g., by supervised learning) on ​​a large image dataset drawn from an readily available library, or approximately annotated by an automated process, for example. The neural network 40 may be alternatively trained, or further trained, on a training dataset generated by taking and annotating images of surface defects in tire casings buffed by experts. Known techniques for data augmentation can be used to increase the size of the training dataset, thereby improving the performance of the neural network model 40.

[0059] Figure 4A illustrates exemplary defect detection and classification performed by the neural network 40. Figure 4A shows two images annotated according to the corresponding classification results generated by the neural network 40. In this case, the images are conventional two-dimensional images (i.e., not image data generated by combining line images) showing two different parts of a buffed tire surface. In the upper image of Figure 4A, the neural network 40 detected three defects (labeled by the neural network 40 as "clean skive," "rust," and "frayed metal"). In the lower image of Figure 4A, the neural network 40 detected four defects (one defect labeled as "clean skive," one defect labeled as "rust," and two defects labeled as "frayed metal"). The labels shown in the example of Figure 4A correspond to the defect labels with the highest scores in their respective score distributions.

[0060] Figure 4B shows a table of "confusion matrices" experimentally obtained for the neural network 40 (in this example, the set of defect labels consists of "clean skive," "rust," "frayed metal," and "unknown"). The confusion matrix in Figure 4B compares the labels assigned by the neural network 40 with those assigned by human experts. In particular, the diagonal elements of the confusion matrix indicate the probability that the neural network 40 will label the corresponding defect with the same label as a human expert. In this example, the average value of the diagonal elements corresponding to the defect types "clean skive," "rust," and "frayed metal" is greater than 0.7, indicating good classification performance by the neural network 40.

[0061] The neural network 40 can typically recognize (and distinguish) a greater number of defect types than conventional methods for processing images of surface defects in buffed casings. As will be described in more detail below, this provides additional information that can be used to improve the selection of further processing steps for buffed tire casings, such as whether the casing is suitable for retreading and, if so, how to best treat the surface defects of the casing.

[0062] Generally, buffed tire casings with surface defects may need to be treated before the tire can be further processed, i.e., the defects may need to be removed. As described above with reference to System 1, the treatment process can be performed by a tool supported on a positioning device (e.g., a robotic arm) based on a set of treatment parameters. The treatment parameters can specify the tool used for a particular treatment, and the operating conditions of the tool, e.g., the angular velocity of the tool (typically defined in revolutions per second), the angle of the tool, etc. The treatment process for a particular part of the tire can be selected from several available treatments. It is desirable to provide a method that enables the automatic selection of the most appropriate treatment so as to further reduce the need for human intervention. Such a method for selecting a treatment process to remove defects from the surface of a buffed casing is described below with reference to Figure 5. Figure 6 schematically shows a system for implementing the method of Figure 5. The data processing steps of the method of Figure 5 can be implemented using one or more computers located in one or more locations.

[0063] In step S201, image data of a portion of the buffed surface of the casing is received. The portion of the surface may have at least one surface defect. For example, the image data may represent a portion of the equatorial plane 5 of the casing 3. The received image data is typically a two-dimensional array of numbers.

[0064] Generally, the received image data can be acquired in any preferred manner, for example, by combining multiple line images acquired using a conventional (two-dimensional) camera or by using a linear camera. In one embodiment, the received image data can be generated by System 1 as described above. The received input image data can correspond to the ROI of the identified surface as described above, with reference to Figure 3.

[0065] In step S203, a neural network model is used to process the received image data to detect defects and, for each detected defect, generate a score distribution across a set of defect types. Step S203 can be performed using the neural network model 40, as described above with reference to Figure 3.

[0066] In step S205, the output of the neural network 40 (i.e., the score distribution) is provided to the treatment selection module 42, which selects a treatment process (for the portion of the surface represented in the received image data) from a set of treatment processes based on the score distribution. In other words, the treatment process is selected based on the detected defects. Generally, if only one defect (or multiple defects of the same type) is detected, the selection of the treatment process can be performed based on a predetermined one-to-one mapping between the defect type and the treatment process. Such a one-to-one mapping is no longer appropriate when multiple defects of different types are detected in the same portion of the surface (i.e., when multiple defects are close to each other, as shown in the example in Figure 4, for example). In this case, the treatment process can be selected based on the combination of defect types detected in the input image. A subset of the set of possible defect labels is assigned to the portion of the surface shown in the image, and the treatment for the surface is selected based on that subset of defect labels.

[0067] The treatment selection module 42 can process the received score distribution and assign one defect label to each of the detected defects. For example, the treatment selection module 42 can assign the defect label with the highest corresponding score within each score distribution to each detected defect. To illustrate this, Figure 7 shows an exemplary score distribution table of an input image in which three defects were detected. For each defect, the corresponding row in Figure 7 gives five numerical values ​​(scores) that collectively form a score distribution across a set of four defect types, plus an optional score ("unknown") indicating that the defect is not one of the four defect types. The treatment selection module 42 can assign the label "Frayed Metal" to "Defect 1" because this label has a higher score than the other labels. Similarly, the treatment selection module 42 can assign the label "Clean Skive" to "Defect 2" and the label "Rust" to "Defect 3".

[0068] Optionally, if the score distribution satisfies the uncertainty criterion (for example, none of the scores in the score distribution of the set of defect types exceed the threshold, and / or the score for "Unknown" is the highest), the action selection module 42 may assign the label "Unknown" to the defect (instead of the defect type whose score distribution happens to be the highest). Thus, "Unknown" is one of the set of possible defect labels, but it is not associated with any particular defect type within the set of defect types.

[0069] Next, the treatment selection module 42 can process the assigned labels to determine a treatment score. In one embodiment, an integer (1 to 9) is determined as the treatment score, depending on the assigned label of the defect detected for a particular input image. Figure 8 shows a Venn diagram illustrating an exemplary method for determining a treatment score. In Figure 8, circles 60, 62, 64, 66, and 68 represent the defect labels "clean skive," "rust," "frayed metal," "no belt," and "unknown," respectively. This is a predetermined set of defect labels, optionally including "unknown," which indicates that the defect has a defect type that cannot be identified (e.g., it is not one of the (four) set of defect types on which the score distribution is based). Therefore, points within a first subset of circles (e.g., circles 60 and 62) but outside of other circles (e.g., circles 64, 66, and 68) represent images that have one or more defects of the type corresponding to the subset of circles, but not one or more defects of the type corresponding to the other circles (e.g., clean skives and rust defects, but no other defects).

[0070] In Figure 8, each logical set represented in the Venn diagram (i.e., each subset of defect labels) is assigned a corresponding remediation score. The remediation scores may differ for each subset of defect labels. At least one subset of defect labels contains multiple corresponding defect labels. The remediation scores are shown in Figure 8 as numbers assigned to the overlapping regions representing those defect labels. In other words, the remediation scores represent each subset of defect labels. For example, for an input image in which three defects are detected with the score distribution in Figure 7 (i.e., at least one frayed metal defect (defect 1), at least one clean skive defect (defect 2), and at least one rust skive (defect 3)), the remediation selection module 42 can determine 7 remediation scores based on the Venn diagram in Figure 8. This is because the table in Figure 7 contains defects of the types "clean skive," "rust," and "frayed metal," and the corresponding logical set (i.e., the overlapping area of ​​circles 60, 62, and 64, which corresponds to the subset of defect labels consisting of "clean skive," "rust," and "frayed metal") is assigned to the treatment score value in 7.

[0071] It should be noted that the remediation score is typically not defined for all possible corresponding subsets of defect labels. For example, the defect "missing belt" can be very serious, so it may not be necessary to consider situations where another type of defect is detected in the same part of the tire. Therefore, if at least one defect of type "missing belt" is detected in an image, the image is assigned a remediation score of 9 (corresponding to a single defect label "missing belt"), regardless of whether other types of defects are also present.

[0072] The treatment selection module 42 can select a treatment process based on the determined treatment score. For example, each subset of defect labels can be mapped by mapping the data to a specific treatment process in a set of treatment processes. Figure 9 shows exemplary mapping data. In this example, the set of treatment processes includes four treatments. In each row (i.e., for each treatment score), one treatment column is marked as an "on" entry, indicating that this treatment is selected for each treatment score. For example, "Treatment 2" can be selected for a treatment score of 5.

[0073] It should be noted that the actions identified by the mapping data for a subset of defect labels containing at least two defect labels may differ from at least one of the actions that the mapping data would perform to individually map each of the defect labels in the subset. For example, if only defects of the type “frayed metal” are detected, the corresponding action score is 2, which maps to action 1 under the mapping data in Figure 9. If only defects of the type “rust” are found, the corresponding action score is 4, which maps to action 3 under the mapping data in Figure 9. However, if both “frayed metal” and “rust” are found in the image, i.e., the subset of defect labels is labeled with action score 6, which maps to action 2 under the mapping data in Figure 9, i.e., to an action different from the respective actions corresponding to action scores 2 and 4.

[0074] The mapping data in Figure 9 does not map treatment scores 8 and 9 to any arbitrary treatment. Treatment score 8 indicates the defect label "unknown," making it impossible to determine an appropriate treatment. Treatment score 9 indicates the defect label "no belt," suggesting that there may be no suitable treatment for the corresponding type of defect. If a given image of a tire is assigned one of these two treatment scores (a subset of defect labels), the tire may be considered processed.

[0075] Otherwise, the selected treatment process can be applied to the relevant portion of the surface, i.e., the portion of the surface represented in the input image. In the embodiment, the selected treatment process can be applied locally only to the portion of the surface that has the detected defect. For example, the output of the neural network 40 may include, for each detected defect (in addition to the score distribution described above), coordinates that identify the location of the defect in the input image and information that identifies the extent of the defect. Thus, the region of the surface with defects can be determined from the output of the neural network 40 (e.g., the region inside the rectangle shown in Figure 4), and the selected treatment can be applied mainly to these regions.

[0076] As described above, an embodiment of System 1 may include one or more computing devices for implementing steps S201 to S205. In some of these embodiments, System 1 may include a tool and a positioning device (e.g., a robotic arm) for operating the tool, and a selected treatment process may be applied using System 1.

[0077] In a broad sense, acquiring and processing image data of the equatorial plane of a casing can be a relatively slow process (even when the image data is captured by a linear camera as described above). Therefore, to increase productivity (of automated defect inspection and skiving stations), it may be advantageous to provide a method in which a first treatment process is applied to the equatorial plane of the casing before acquiring and processing the image data (i.e., more time- and resource-efficient). Such a method is illustrated with reference to Figure 10. This method is described as being implemented using an embodiment of System 1 in Figure 1, which further comprises one or more computing devices, tools, and a robotic arm for operating the tools.

[0078] In step 301, three-dimensional profile data of at least a portion of the surface of the casing 3 is acquired by using the surface shape measuring device 13. For example, three-dimensional profile data of the equatorial plane 5 can be acquired (and combined) as described above with reference to Figures 1 and 2. Additionally or alternatively, three-dimensional profile data of at least a portion of the sidewall surfaces 4 of the casing 3 is acquired. In step 303, the profile data is processed (e.g., by one or more computing devices) to determine the total number of defects on at least one portion of the surface. Several methods are known for doing this. For example, defects can be detected from the profile data by identifying anomalous distance values ​​in the profile data, i.e., distance values ​​greater than a predetermined threshold. The predetermined threshold can be experimentally selected based on a desired trade-off between the number of false positive defect identifications and the number of false negative defect identifications. In other embodiments, the profile data can be processed using a neural network model (e.g., a neural network including a convolutional neural network) and defects can be detected based on it.

[0079] In step S305, it is determined whether the total number of defects falls below a predetermined defect threshold. If the total number of defects is equal to or greater than the defect threshold, the buffed tire casing 3 may be determined to be unsuitable for retreading. This is because such a tire, once retreaded, may not be considered (sufficiently) safe to deploy for use (even if these defects are removed by the treatment). The defect threshold may be fixed or selected based on the type of tire (e.g., truck tires, bus tires). Defect thresholds for different tire types may differ because different safety considerations and / or regulations apply. Generally, it is advantageous to determine whether a particular tire is suitable for retreading before further (time-consuming) steps (such as taking and processing surface images or applying treatments) are performed.

[0080] If the total number of defects is less than the defect threshold number, the first remediation process is applied to the buffed surface of the casing (step S307). The first remediation process can be applied to the entire outer surface of the casing 3, or only to a portion of the outer surface of the casing 3, such as the equatorial surface 5, the side walls 4, or localized areas of the surface around defects detected in step S303. The first remediation process may include applying a (wire) brush to the outer surface of the casing 3. In general, the first remediation process (i.e., the tools and operating conditions) can be selected to be effective against widespread and frequently found tire defects so that the total number of defects is reduced before further (defect type-specific) remediation is selected (i.e., steps S309-S313).

[0081] In step S309, image data of the equatorial plane 5 of the casing 3 is acquired. For example, the image data can be acquired using a linear camera 9 as described above with reference to steps S101 and S103 in Figure 2. Alternatively, the image data can be acquired using a conventional (two-dimensional) camera. In step S311, the image data of the equatorial plane 5 of the casing 3 is processed to detect defects. For example, the image data of the equatorial plane 5 of the casing 3 can be processed as described above with reference to steps S105 in Figure 2 and / or step S203 in Figure 5. In step S313, a further treatment process is selected from a set of treatment processes based on the detected defects on the equatorial plane of the casing. The further treatment process can be selected as described above with reference to step S205 in Figure 5. Furthermore, in step S313, the selected further treatment process can be applied to the buffed equatorial plane 5 of the casing 3 (for example, as described above with reference to Figure 5).

[0082] As described above, when evaluating whether a buffed tire casing is suitable for retreading, it may be advantageous to consider the type of tire (e.g., truck tires, bus tires, etc.). In addition to the type of tire, further information about the tire can be used in such evaluation. This allows for a more accurate and well-informed decision, and therefore contributes to the implementation of safety considerations and cost reductions (for example, by avoiding further processing steps if the tire is found to be unsuitable). It is desirable to provide a method in which information about a particular tire, such as the tire's history, is considered when evaluating the tire's suitability for retreading. Such a method is illustrated with reference to Figure 11. This method can be implemented in an embodiment of System 1.

[0083] In step S401, tire-specific usage data for tire 3 is received. Generally, tire-specific usage data includes information that identifies tire 3. Tire-specific usage data may identify the previous use of tire 3 (e.g., truck tire, bus tire, etc.). Additionally or alternatively, tire-specific usage data may identify other details of tire 3, such as the date of manufacture, manufacturer, model number, batch or serial number, material properties, and / or further details of the tire that identify the tire's lifecycle (e.g., the service history of tire 3). In one embodiment, tire-specific usage data may identify previous measurements of physical quantities of tire 3, such as information about defects detected during a previous retreading process applied to the tire (e.g., the number of defects, the type of defects, etc.). Tire-specific usage data can be stored in a device mounted on or otherwise attached to tire 3, such as a received (or read) radio frequency identification (RFID) tag or tire mounting sensor (TMS) unit. Such a device can be read (by an appropriate reader) to retrieve the tire-specific usage data. Additionally or alternatively, tire-specific usage data can be stored in a (remote) database (for example, on a cloud server). Such a database can store tire-specific usage data for multiple tires. In this case, tire 3 may have an associated identification code that uniquely identifies tire 3 among multiple tires. Tire-specific usage data may be read from the database based on the identification code. The identification code may be provided directly on tire 3 or stored in a device mounted on tire 3 (for example, an RFID tag or TMS).

[0084] In step S403, a primary defect threshold is determined based on the received tire-specific usage data. In other words, the primary defect threshold for a particular tire is determined by considering the specific tire-specific usage data for that tire. In one embodiment, a corresponding set of further defect thresholds is determined based on the received tire-specific usage data. Each of the further defect thresholds can be associated with one label from a set of defect labels. For example, the defect thresholds for each of the defect labels "clean skive," "rust," "frayed metal," "no belt," and "unknown" can be determined based on the received tire-specific usage data.

[0085] In step S405, three-dimensional profile data of at least a portion of the surface of the tire casing 3 is acquired using a surface shape measuring device. For example, three-dimensional profile data of the equatorial plane 5 can be acquired (and combined) as described above with reference to Figures 1 and 2. Additionally or alternatively, three-dimensional profile data of at least a portion of the sidewall surfaces 4 of the casing 3 is acquired. In step S407, the profile data is processed to determine the total number of defects (for example, as described above with reference to step S303 in Figure 10). In step S409 (similar to S305 in Figure 10), it is determined whether the total number of defects is below a primary defect threshold. If the total number of defects is equal to or greater than the primary defect threshold, the buffed tire casing 3 can be determined to be unsuitable for retreading. If the total number of defects is below the primary defect threshold, the method proceeds to step 411. In step S411, if no further defect threshold has been determined in step S403, a process for removing defects can be applied to the surface of the buffed casing 3.

[0086] Alternatively, in embodiments where a further defect threshold is determined in step S403, step S411 may include acquiring image data of the equatorial plane 5 of the casing 3 (for example, as described above with reference to step S309 in Figure 10). Step S411 may further include processing the image data to detect surface defects and associating each detected defect with at least one defect label from a set of defect labels (for example, as described above with reference to step S203 in Figure 5). The total number of defects associated with each label in the set of defect labels can be determined. Next, for each label in the set of defect labels, the total number of defects associated with each label is compared with the corresponding further defect threshold. For each label in the set of defect labels, if it is determined that the total number of defects associated with each label is less than or equal to the corresponding further defect threshold, the treatment is applied to the buffed surface of the casing 3. For one of the labels in the set of defect labels, if it is determined that the total number of defects associated with each label is higher than the corresponding further defect threshold, the tire casing 3 can be determined to be unsuitable for retreading. By considering individual thresholds for various defect types, it becomes possible to make more accurate and informed decisions regarding further processing and subsequent use of the tire casing 3.

[0087] Figure 12 shows a method for performing automated skiving of buffed casings. This method combines several aspects of the methods described above with reference to Figures 2, 5, 10, and 11. This method can perform reliable, efficient, and highly automated defect detection, detection identification, treatment selection, and treatment application. This method can be implemented using the embodiment of System 1 described above with reference to Figure 1.

[0088] In the first step S501, the tire casing 3 is loaded and an initial inspection is performed. For this purpose, tire-specific usage data is retrieved as described in step S401 of Figure 11 (for example, by scanning the RFID tag on the tire 3 to read the tire-specific usage data, or by reading the identification code and looking up the corresponding tire-specific usage data in a database that stores tire-specific usage data). The primary defect count thresholds for the equatorial plane 5 and sidewall 4 are determined based on the retrieved tire-specific usage data. Furthermore, three-dimensional profile data of the equatorial plane 5 and sidewall 4 of the tire casing 3 is acquired. The profile data is processed (for example, by using a neural network, as described in S303 of Figure 10) to detect defects and determine the total number of defects on the equatorial plane 5 and the total number of defects on the sidewall 4. In step S503, the total number of defects on the equatorial plane 5 and sidewall 4 are compared with their respective primary defect count thresholds. If either the total number of defects on the equatorial plane 5 or the total number of defects on the sidewall 4 is greater than the respective primary defect threshold, the tire casing 3 may be determined to be unsuitable for retreading, and the tire may be unloaded.

[0089] If the total number of defects on the equatorial plane 5 and the total number of defects on the sidewalls 4 are less than or equal to their respective primary defect thresholds, the first treatment (or a set of predetermined treatments) is applied to the tire casing 3 (as described above with reference to step S307 in Figure 10).

[0090] Next, a series of defect detection and defect correction steps can be repeatedly applied to the equatorial plane 5. This can be done by a loop counter (S507) that tracks the number of iterations performed and compares this number to a threshold. If the loop counter determines that the number of iterations performed is greater than the threshold, the method in Figure 12 is completed. If the number of iterations performed is less than or equal to the threshold, step S509 is performed.

[0091] In step S509, image data of the equatorial plane 5 is generated by synthesizing line images acquired using a linear camera 9 (as described with reference to steps S101 and S103 in Figure 2). The ROI of the image data (identified based on profile data as described with reference to Figure 3) is processed by a neural network 40 to detect defects within the ROI and assign defect labels. Tire-specific usage data (stored in a device mounted on the tire 3 or in a database) may be updated based on the output of the neural network 40. The total number of defects associated with each label in the set of defect labels is determined.

[0092] A set of further defect thresholds is determined based on the received tire-specific usage data, and each of these further defect thresholds can be associated with one of the labels in the set of defect labels (as described with reference to step S403 in Figure 11). If, for one of the labels in the set of defect labels, the total number of defects associated with that label is higher than the corresponding further defect threshold, the tire casing 3 can be determined to be unsuitable for retreading. If, for each of the labels in the set of defect labels, the total number of defects associated with that label is less than or equal to the corresponding further defect threshold, the action for each identified ROI is selected based on the corresponding output of the neural network 40 (as described with reference to S205). In step S511, the selected action is applied to each portion of the equatorial plane 5, and a counter for the number of iterations performed is incremented. Next, step S507 is performed again, and the loop counter checks whether the number of iterations performed exceeds a threshold. If it exceeds the threshold, the method is completed. If it does not exceed the threshold, steps S509 and S511 are performed again.

[0093] Those skilled in the art should understand that various modifications can be made to the embodiments described above without departing from the scope of the present invention.

Claims

1. A method for determining whether to apply a treatment process to remove defects from the surface of a buffed casing of a pneumatic tire, Receiving tire-specific usage data, The defect threshold is determined based on the tire-specific usage data received, Using a surface shape measuring device, acquire three-dimensional profile data of at least one portion of the surface of the casing, Processing the profile data to determine the total number of defects on at least one portion of the surface, Determining whether the total number of defects is less than the defect threshold, On the condition that the total number of defects is less than the defect threshold, it is decided to apply the treatment process for removing defects from the surface, A method comprising applying the treatment process to the tire.

2. The method according to claim 1, wherein the tire-specific usage data identifies at least the previous use of the tire and / or previous measurements of the physical quantities of the tire.

3. The method according to claim 1 or 2, wherein the tire-specific usage data identifies at least the maintenance history of the tire.

4. The method according to any one of claims 1 to 3, wherein receiving tire-specific usage data includes receiving the tire-specific usage data from a wireless transmitting device mounted inside or on the tire, and / or from a tire mounting sensor unit mounted on the tire.

5. The method according to any one of claims 1 to 4, wherein receiving tire-specific usage data includes reading the tire-specific usage data from a database.

6. The method according to claim 5, wherein the database is stored on a cloud server.

7. Applying the treatment process to the surface of the buffed casing is To acquire image data of the equatorial plane of the casing, The aforementioned image data, (i) Detect surface defects, (ii) Associate each detected defect with at least one defect label from the set of defect labels, (iii) Processing to determine the total number of defects associated with each label in the set of defect labels, Based on the received tire-specific usage data, determine a corresponding set of further defect thresholds, wherein each further defect threshold is associated with a set of further defect thresholds that correspond to one of the defect labels. For each label in the set of defect labels, determine whether the total number of defects associated with each label is less than the corresponding further defect threshold. The method according to any one of claims 1 to 6, further comprising applying the treatment process to the surface of the buffed casing, provided that for each label in the set of defect labels, it is determined that the total number of defects associated with each label is less than the corresponding further defect threshold.

8. A system for determining whether to apply a remediation process to remove defects from the surface of a buffed casing of a pneumatic tire, the system comprising a surface shape measuring device, a computing device, and a robotic arm configured to support a tool, the system The tire receives its own tire-specific usage data. Using the computing device, a defect threshold is determined based on the received tire-specific usage data. Using the surface shape measuring device, acquire three-dimensional profile data of at least a portion of the surface of the casing. Using the computing device, process the profile data to determine the total number of defects on at least one portion of the surface. Using the computing device, determine whether the total number of defects is less than the defect threshold. On the condition that the total number of defects is determined to be less than the defect threshold, it is decided to use the computing device to apply the treatment process for removing defects from the surface. A system configured to apply the treatment process to the tire using the aforementioned tool.

9. The system according to claim 8, wherein the tire-specific usage data identifies at least the previous use of the tire and / or previous measurements of the physical quantities of the tire.

10. The system according to claim 8 or 9, wherein the tire-specific usage data identifies at least the maintenance history of the tire.

11. The system according to any one of claims 8 to 10, wherein the system is configured to receive tire-specific usage data from a wireless transmitting device mounted inside or on the tire, and / or from a tire mounting sensor unit mounted on the tire.

12. The system according to any one of claims 8 to 11, wherein the system is configured to read tire-specific usage data of the tire from a database.

13. The system according to claim 12, wherein the database is stored on a remotely located cloud server.

14. The system is further equipped with a camera, Using the aforementioned camera, image data of the equatorial plane of the casing is acquired. Using the computing device, the image data is processed (i) Detect surface defects, (ii) Associate each detected defect with at least one defect label from the set of defect labels, (iii) Process to determine the total number of defects associated with each label in the set of defect labels, Using the computing device, determine a corresponding set of further defect thresholds based on the received tire-specific usage data, wherein each further defect threshold determines a corresponding set of further defect thresholds associated with one label from the set of defect labels. Using the computing device, determine whether the total number of defects associated with each label in the set of defect labels is less than the corresponding further defect threshold. The system according to any one of claims 8 to 13, configured to apply the treatment process to the surface of the buffed casing by using the tool, provided that, for each label in the set of defect labels, it is determined that the total number of defects associated with each label is less than the corresponding further defect threshold.

15. A computer program that, when executed by a computer, includes instructions causing the computer to perform the method described in any one of claims 1 to 7.

Images