METHOD FOR TESTING TIRES

Description

[0001] The invention relates to a method for testing tires. Tires are rotationally symmetrical objects with a diffusely scattering surface, such as vehicle or aircraft tires.

[0002] It is known that interferometric measuring methods such as ESPI (Electronic Speckle Pattern Interferometry) or the shearing method are used to determine the area displacements or strains of an object under test with a diffusely scattering surface. Which method is used for an object under test depends primarily on the size to be measured and the required measuring range. Another aspect when selecting the measuring method is the cost of the measuring system. For example, the shearing method has advantages over other interferometric measuring methods because it requires simple laser diodes due to the short required coherence length. Furthermore, as taught in DE 198 59 725 C2, for example, it can also be operated with multiple laser diodes, so that the illumination can be adapted to the object under test with comparatively little effort.The differometric method described in US 6,934,018 B2 represents an instrumental simplification compared to the shearing method, in which the use of an interferometer is not necessary.

[0003] In the aforementioned interferometric measurement methods, two states of the object under test are typically compared during static testing. Images of the object under two different loading conditions are taken and the interferograms of the two conditions are subtracted. This results in a differential interferogram, which represents the displacement or strain of the object between the two conditions in the form of interference lines. If the measuring head is equipped with a phase-shifting unit, an extended evaluation can be performed based on the principle of the phase-shift method (W. Osten, "Digital Processing and Evaluation of Interference Images," Chapter 6, Akademie Verlag ISBN 3-05-501294-1). This generates phase images that assign a specific phase angle to each pixel.The phase-shift method offers the advantage of a standardized image whose gray values ​​are proportional to the determined phase angle throughout the image. The readability of such images is significantly increased compared to simple interference line images.

[0004] In contrast, diffrometry subtracts speckle patterns from different loading conditions. Consequently, it is only possible to distinguish between points where the deformation is zero and those where the deformation is non-zero. A quantitative evaluation is not possible. Likewise, a phase-shift method is not feasible due to the lack of an interferometer.

[0005] Depending on the shape and size of the object to be tested, the interferometric measurement is performed sequentially or simultaneously in several sections. A typical application of this type is the interferometric testing of tires. For example, DE 42 31 578 C2 describes a sector-by-sector testing of tires. US Pat. No. 5,786,533 proposes a method for the targeted localization of defects in tires, particularly in the belt area. An evaluation of the entire belt area is only possible if the tire is tested in several sectors.

[0006] The positioning of the measuring head relative to the object to be tested is generally achieved by geometrically measuring the object to be tested and subsequently calculating a suitable position for the measuring head. This is described, for example, in DE 103 33 802 B4.

[0007] Typically, in measurement systems that test sections, the resulting images of the individual sections are displayed individually or arranged side by side on a visual system. Based on this, EP 1 284 409 A1 describes a method for transforming the resulting images, or the sections of the object to be tested depicted therein, into a common coordinate system. For this purpose, the object surface is topometrically measured in a planar manner, parallel to the recording of the object's deformation, and the appropriate spatial coordinates are assigned pixel by pixel to the resulting images depicting the object's deformation. This method uses a planar contour detection system, which can also be an integral component of the measuring head.

[0008] In practice, however, it has been shown that the equipment required for area-wide topometric recording of the object surface, particularly when using multiple measuring heads, entails considerable costs, and furthermore, the computing time required for contour detection is impractical for industrial use. However, EP 3 545 259 B1 discloses a method in which the metrological recording of the spatial coordinates of the surface points depicted in the individual resulting images is avoided and, instead, the spatial positions of the sections of the object surface depicted in the resulting images are calculated relative to the object. The known method also makes it possible to transfer the image points from different sections of the object into a common coordinate system and display them in an overall view.The overall view can be a development of the object surface, which is easier for the viewer to understand than the usual individual sector representations.

[0009] The state of the art also includes the use of artificial intelligence, particularly in the form of machine learning, for the automated evaluation of interferometric tire test results. Machine learning, which has been known for decades, involves algorithms generating a statistical model based on training data, including training images. Patterns and regularities are recognized from the training data, enabling, for example, automatic diagnosis or classification.

[0010] This is described in EP 2 065 674 B1, published on September 15, 2010, although at that time the effort required for this was still considerable and the use of automatic diagnosis or classification was therefore not very widespread in practice. However, with the increasing digitalization in recent years, artificial intelligence has found its way into many areas of life. The now banal realization of using artificial intelligence, particularly in the form of machine learning, to automate processes that previously required manual activities is also testified to by a large number of patent applications published in recent years in a wide variety of technical fields. One example of this in the field of tire testing is DE 10 2019 120 696 A1 or US 2021 / 0224971 A1.

[0011] To evaluate the results of a tire test, the images obtained from a tire, for example using an interferometric measurement method, are fed to a deep learning module for evaluation. This requires that the deep learning module has been trained beforehand. Training usually consists of (a) specifying training data with the (image) features typically to be identified, such as defect locations, (b) testing the parameters created by the deep learning module through learning processes using validation data, and (c) evaluating test data with a trainer checking the accuracy of the result. However, the data used to train the deep learning module—i.e., training, validation, and test data—must not overlap or be identical, as this would lead to distorted weighting.

[0012] Tire defects can appear in a variety of ways in the resulting images. This can be due, for example, to the distance of the defect from the inspected tire surface.

[0013] Furthermore, images obtained during interferometric tire testing often contain features caused by irrelevant deformations of the tire, but do not represent a defect. Such features, which frequently appear in images of the tire sidewall, especially at the bead where the transition to the surrounding area, i.e., an area outside the tire, occurs, can easily be misinterpreted as a defect in the tire.

[0014] Furthermore, deformations in the area surrounding the tire, for example a conveyor belt transporting the tire, can be misinterpreted as a defect in the tire.

[0015] DE 10 2019 120 696 A1 proposes the obvious solution to the problems described above: a deep learning module automatically detects image areas outside the tire and independently locates defects inside the tire. However, even detecting the area outside the tire requires a relatively high level of computing power, which must be further increased if different tire types and sizes are to be processed. The known approach requires nothing more than sufficient effort to train the artificial system and is therefore trivial.

[0016] The invention is based on the object of specifying a method and a device for the interferometric testing of a tire, in which the independent detection of relevant defect locations in the object to be tested is improved by an artificial system which is trained by means of at least one training data set and is particularly suitable for carrying out method steps for machine learning, in such a way that the evaluation of the result data can be carried out more reliably with a reduced effort for training the artificial system and with a lower computing effort.

[0017] This object is achieved by a method, a system, a computer program product, a computer-readable storage medium and a device according to the claims.

[0018] The invention is based on the discovery of systematically and automatically analyzing and categorizing images from an interferometric tire test. It has proven advantageous to convert the images of a tire into a standardized representation. In this process, the (size) representation of the object to be tested, which depends in particular on the position of the camera, especially the measuring head, relative to the object to be tested, is eliminated from the resulting images, overlap areas are removed, and image areas outside the object to be tested are eliminated.

[0019] The method according to the invention comprises the following method steps, wherein the processor and the storage medium are suitable for carrying out the method steps: a) Creating an image of at least a partial area of ​​the tire surface using a camera; b) Creating input data comprising an image consisting of at least one image; c) Analyzing the input data to determine and categorize a critical area comprising a relevant defect location and a non-critical area comprising a non-relevant anomaly; d) Saving the analyzed image as result data adjusted for the non-critical area; and e) Saving the adjusted result data as training data in the training data set.

[0020] The method according to the invention is characterized by the fact that the required labor and computing effort are significantly reduced by the automated process. In particular, the fact that the artificial system performs the analysis also reduces the time required by an operator to interpret the defect locations. Furthermore, the fact that the training data set is continuously expanded with further training data means that the analysis of the artificial system is continuously improved. Accordingly, over time, even an operator with less experience can carry out the interpretation of defect locations. This is because the images can be adapted in such a way that they are particularly suitable for the subsequent process steps. Finally, the trained system can perform the analysis and categorization independently, so that, for example, the sorting of defective tires can be carried out fully automatically and without an operator.

[0021] To carry out the method according to the invention, a testing device as described below can be used. The processor and the storage medium can be an integral part of the testing device, but can also be a centrally located computer that is only connected to the testing device via a radio link to transmit data. A central computer has the advantage that multiple testing devices can transmit data, such as the recordings, to the central computer, which can thus more easily build a large data set, which is also stored, in particular, as a training data set. Furthermore, this central computer can be equipped with particularly high computing power.

[0022] Result data can be result images, but also result images with additional information about the result image.

[0023] A tire is a tire of a vehicle, such as an automobile, a work vehicle or an aircraft, or a similar rotationally symmetrical object.

[0024] An image, for example, is a measurement result created from individual images taken by a camera. The camera is preferably suitable for conducting an interferometric test. For example, several individual images are taken to generate a phase image for an interferometric test.

[0025] In an advantageous embodiment, the recording is a phase image and / or a video image. A phase image is usually acquired in combination with a coherent light source during an interferometric inspection. The phase image is particularly suitable for identifying and categorizing a defect location. A video image is preferably acquired in combination with non-coherent illumination to avoid the speckle formation that occurs with coherent illumination. The video image is therefore particularly suitable for identifying and categorizing a non-relevant anomaly.

[0026] Furthermore, in an advantageous development, a contrast image corresponding to the phase image can be calculated. The contrast image indicates the quality, in particular the contrast, of the calculated phase value for each pixel of the phase image. It is advantageously obtained from the individual images used to generate the phase image. High contrast is characterized by a large amplitude of the phase modulation. If the contrast is too small, the relevant pixels are expediently eliminated from the phase image. This elimination of the relevant pixels can be achieved, for example, by replacing the phase values ​​of the pixels with phase values ​​of surrounding pixels.

[0027] Input data can only be formed from a single recording. However, it is preferred that the input data be an image consisting of multiple recordings and comprising at least a partial area of ​​the tire, in particular a sector of the tire. An image can, for example, be composed of two recordings that capture different sectors of the tire. However, it is preferred that the entire circumference of the tire is captured, in particular, by multiple recordings. The individual recordings can be processed beforehand so that they can be combined to form one image. For example, this is particularly suitable for the area of ​​the bead and the sidewall, since these usually represent an almost symmetrical ring-sector-shaped recording of the partial area of ​​the tire. The symmetrical ring-sector-shaped recording is transformed in such a way that it ultimately represents a rectangle.For example, the shorter base side is stretched and the longer base side is compressed until the base sides are equal in length, thus creating a rectangle. The areas within the image are distorted accordingly.

[0028] Analysis means examining the input data to determine whether there is an error.

[0029] Determining means identifying a fault location.

[0030] Categorization means that the identified fault location is given an assessment as to whether an area containing the fault location is to be regarded as a critical area or a non-critical area.

[0031] A critical area is defined as an area that has a significant defect. A significant defect could be, for example, an air pocket or a crack in the tire, which could lead to a shortened tire lifespan.

[0032] A non-critical area is an area that does not have a defect or a location that is recognized as a defect but which, for example, only represents an air pocket that does not lead to a reduced service life of the tire or to a significant shortening of the service life outside the specified service life of the tire.

[0033] In an advantageous embodiment of the method, a non-critical area can also represent a location identified as a defect that is merely a non-relevant anomaly. A non-relevant anomaly is, for example, an irrelevant deformation of the tire surface caused by a repair location, a repair patch, a balancing patch, an RFID sensor, or a pressure sensor.

[0034] As previously described, in the inventive embodiment of the method, the area containing the irrelevant anomaly, i.e., the non-critical area, is removed or not taken into account in the result data. This removal may involve inserting a blank space in the removed area. The cleaned result data is then particularly suitable as training data.

[0035] To create input data, a sector boundary and / or a boundary line for the image is advantageously determined, wherein the sector boundary and / or the boundary line delimits a defined section of the surface of the tire, wherein the input data comprise the delimited and defined section, and wherein the area lying outside the defined section is preferably removed. The amount of data to be processed is reduced by cropping. The contrast image is particularly suitable for identifying the sector boundary and / or the boundary line. Furthermore, this step is advantageous for creating an image from more than one image. Furthermore, overlapping areas can be avoided so that when several images are combined to form one image, no areas are displayed twice.

[0036] To create the image, localization information about the position of a camera relative to the tire at the time the image was taken is advantageously used. At least two images are expediently combined to form the image, taking their respective localization information into account. The localization information preferably includes the distance of the camera from the recorded part of the tire surface and / or tire geometry information, in particular an axial section. The localization information can make it possible to assign a precise position to the individual images, so that a seamless image of the tire can be created, consisting of a plurality of images.

[0037] In a further advantageous embodiment, a functional structure of the tire is derived from the localization information, wherein the functional structure preferably comprises at least one, in particular two or preferably three, of the following partial regions of the tire: inner first bead, outer first bead, inner first sidewall, outer first sidewall, inner upper shoulder, outer upper shoulder, tread, inner second bead, outer second bead, inner second sidewall, outer second sidewall, inner lower shoulder, outer lower shoulder. In this way, additional functional information can be given to the individual section. This is particularly advantageous for the weighting described below, since defect locations in sensitive regions, such as the tread, can be differentiated from defect locations in less sensitive regions, such as the sidewall.

[0038] Furthermore, the image can advantageously be normalized based on the localization information. The size and / or resolution of the image are expediently scaled to a reference scale. For this purpose, the scaling utilizes, in particular, information comprising the distance and orientation of the camera to the area to be recorded. Consequently, the actual size of the area captured by a pixel and also the size of the tire are known. Accordingly, the image can be scaled to a normalized size so that, preferably, all input data and / or result data have the same reference size and / or resolution and / or scale. Since the training data set then consists of uniformly scaled training data, this data can subsequently be used universally for various camera distances and image scales.The amount of training data required to perform the most accurate analysis of different tire sizes is thus significantly reduced.

[0039] When using multiple phase images, it is advantageous to standardize the phase values, particularly by scaling them pixel by pixel to a uniform reference scale. This allows the phase values ​​to be directly compared without additional processing steps, even for different tire sizes and / or measurement distances.

[0040] Advantageously, the phase images are demodulated before normalization, especially scaling. Demodulation in the field of shearography is known, for example, from DE 196 39 213 A1. This allows for a simple and reliable rescaling of a phase image.

[0041] To avoid demodulation, which is generally prone to error, a large number of intermediate images are advantageously taken in step a) to create a phase image between two different tire loading conditions to calculate the instantaneous phase value. Advantageously, the phase values ​​of consecutive intermediate images are then subtracted pairwise, and the resulting difference values ​​are summed. A sufficiently large number of intermediate images ensures that the load change between two consecutive intermediate images is reliably within one order, even if the total load change spans more than one interferometric order.To capture the phase images during the transition from a first to a second load condition, a spatial phase shift method is advantageously used, as described, for example, in DE 39 30 632 A1.

[0042] To determine and / or categorize the critical area and / or the non-critical area in step c), the input data is advantageously weighted based on the functional classification. As previously described, a defect in a sensitive area such as the tread is more relevant than in the sidewall. Accordingly, the functional classification of the map allows for a correlated weighting of the defect location, so that, for example, the threshold at which an area is considered a critical area is exceeded earlier.

[0043] When determining and / or categorizing the critical area and / or the non-critical area in step c), the size of a defect and / or the frequency of a defect and / or the distance between two defect locations is advantageously taken into account. Often, a critical area is not created by a single defect location, but by several defect locations that are close to one another. However, it is also possible that the size of an individual defect location determines whether it is a critical area that could cause a shortened service life of the tire.

[0044] Advantageously, the result data according to step d) is stored in at least two result data sets. The result data sets are expediently assigned to the tire sub-areas resulting from the functional structure. This has the advantage that the result data sets are smaller than the result data, since they each refer to only a sub-area of ​​the tire. Furthermore, the separately stored result data sets can also be evaluated independently of one another. Furthermore, the separately stored result data also stores the data in separate training data sets.

[0045] Advantageously, the input data and / or the result data are displayed on a display unit. The critical area is expediently highlighted. The operator can then check the analysis. The operator can also make corrections. The corrections are also incorporated into the training data, so that over time the operator has to make fewer corrections and the artificial system can reliably distinguish between critical and non-critical areas. It is particularly preferred that the critical areas are highlighted in the display. The operator can thus find the relevant areas more quickly. Once the system training is complete, the analysis and categorization of the input data by the artificial system no longer requires any control from the operator and can instead carry out the test independently. Then, no further training data is generated or required.

[0046] A further aspect of the invention relates to a data processing system comprising means for carrying out the steps of the method described above. The means comprise: a test chamber with a camera for generating an image of at least a partial area of ​​the tire surface; a processor for generating input data comprising an image consisting of at least one image, and for analyzing the input data to determine and categorize a critical area comprising a relevant defect location and a non-critical area comprising a non-relevant anomaly; and a storage medium for storing the analyzed image as result data adjusted for the non-critical area and the adjusted result data as training data in the training data set.

[0047] A further aspect of the invention relates to a computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out the steps of the method described above.

[0048] A further aspect of the invention relates to a computer-readable storage medium on which the computer program product is stored.

[0049] A further aspect of the invention relates to a testing device for testing a tire using a method as described above. The testing device comprises a camera for generating an image of at least a partial area of ​​the tire's surface; a processor for generating input data comprising an image consisting of at least one image, and for analyzing the input data to determine and categorize a critical area and a non-critical area, wherein a critical area comprises a relevant defect location; and a storage medium for storing the analyzed image as result data adjusted for the non-critical area and the adjusted result data as training data in the training data set.

[0050] An exemplary embodiment of the invention is explained below with reference to the drawings. The drawings show: Figure 1: Representation of a testing device suitable for carrying out the method according to the invention on a tire; Figure 2 : Arrangement of the test sections of a tyre during sector-by-sector scanning; Figure 3 : Alignment and positioning of the measuring head for testing the outer bead and sidewall of a tire; Figure 4 : Alignment and positioning of the measuring head for testing the inner tire belt / tread of a tire; Figure 5 : Alignment and positioning of the measuring head for testing the inner tire belt / tread of a tire; Figure 6 : Representation of an axial section; Figure 7 : Schematic representation of a photograph of the bead and sidewall area of ​​a tire; Figure 8 : Schematic representation of a photograph of the inside of the tire belt / tread of a tire; Figure 9 : compilation of the images bead to bead and all sectors; Figure 10: the overall presentation of all individual sectors from Figure 9 after transformation of the sector boundaries and boundary lines into a common object coordinate system; Figure 11a, 11b , 11c : the representation of the images of all individual sectors of a real measurement on a truck tire; Figure 12 : the representation of the image comprising the images of all individual sectors of a real measurement on a truck tire after transformation of the image data into a common object coordinate system; Figure 13 : Training a deep learning module (DLM) using images of defect locations from the bead area of ​​a tire; Figure 14 : Training a deep learning module (DLM) using images of defect locations from the sidewall of a tire; Figure 15 : Verification of the evaluation by the Deep Learning Module (DLM) using test data from the bead area of ​​a tire; Figure 16: Verification of the evaluation by the Deep Learning Module (DLM) using test data from the sidewall of a tire; Figure 17 : Scheme of the evaluation of tire data using the trained Deep Learning Module (DLM) and entry of the found defect locations in the overall representation of the tire; Figure 18 : the representation according to Figure 10 with the defect locations found by the deep learning module in all functional sections of the tire; Figure 19 : the representation according to Figure 18 with the defects found using image processing functions, Figure 20 : the representation of a tire with defects in different sections of the tire, Figure 21 : the representation of another tire as an interferometric result image, Figure 22 : the representation of the tire according to Fig. 21 as a video image, Figure 23 : the representation of a tire according to Fig. 22with a fault location found by the Deep Learning module, here a repair location, and Figure 24 : the representation of a tire according to Fig. 21 , eliminating the repair point.

[0051] The Figures 1 to 5 show a testing device 10 with a tire 101, 102 suitable for carrying out the method according to the invention. The testing device 10 is integrated into an artificial system 500, which has a processor 502 with a storage medium 504, which is connected to the testing device 10.

[0052] The testing device 10 has a measuring head 20 with a camera 21 for taking images 200. The measuring head 20 can be moved horizontally and vertically relative to the object to be tested, the tire 101, 102, via a positioning device 24, 23. The horizontal positioning device 24 is fixedly connected to a base frame 11, so that the measuring head 20 can be moved horizontally and vertically relative to the tire 101, 102. Furthermore, the inclination of the optical axis 22 of the camera 21 relative to the tire 101, 102 can be changed via a pivot axis 25. The testing device 10 further has a turntable 12 for rotating the tire 101, 102 relative to the measuring head 20. This enables testing of the tire 101, 102 over its entire circumference.For geometrically detecting the tire contour, in particular for creating an axial section, a distance sensor 30 with a vertically downward-directed laser beam 31 is located between the conveyor module 40 and the testing device 10. In addition, the testing device 10 has a distance sensor 32 mounted below the measuring head 20. When the measuring head 20 moves into the tire 102 through the rim hole, it detects an axial section from the inside of the tire. The laser beam 33 of the sensor 32 first scans the upper sidewall 113 of the tire 102, then the inside of the bead 114, and finally the inside of the tread of the tire 102.

[0053] The two determined axial sections of the tire 101, 102 are combined to form one axial section, as in Figure 6shown. The solid line is the actually measured contour line, whereas the dashed parts of the contour line are calculated by mirroring. The half of the outer axial section located below the tire center plane RME was generated by mathematically mirroring the upper half of the outer axial section at the tire center plane RME. Furthermore, the contour line was mirrored about the rotation axis Ar and calculated so that the axial section reflects the contour line of the tire 101, 102. The combined axial sections result in a more accurate model of the tire 101, 102. This allows the spatial coordinates of imaged points on the surface of the test object to be calculated with greater accuracy. Furthermore, the axial section represents an exact source of information that can be used as localization information.

[0054] Furthermore, the axial section after testing the tire 101, 102 is used to transfer the measurement data into an object coordinate system.

[0055] The Figure 2 shows the schematic arrangement of the test sectors when testing a tire 101, 102 with eight sectors in the bead and sidewall area 113, 114, 115, 116 and eight sectors in the area of ​​the tread 111 and the shoulders 110, 112 of the tire 101, 102. In practice, depending on the shape and size of the tire 101, 102 or the field of view 130 of the measuring head 20, a higher or lower number of sectors may be required and / or used.

[0056] The Figure 3shows the schematic structure of a testing device 10 with the typical orientation of the measuring head 20 for testing the outside of the bead 114, 116 and sidewall 113, 115 of the tire 101, 102. The camera 21 of the optical measuring head 20 looks perpendicularly or in the axial direction onto the sidewall 113 of the tire 101, 102.

[0057] The Figure 5 shows the schematic structure of a testing device 10 with the typical orientation of the measuring head 20 for testing the tread area of ​​the tire 101, 102 from the inside. The two shoulders 110, 112 and the tread center 111 of the tire 101, 102 are detected.

[0058] The Figure 7shows a schematic representation of an image 200 of the bead and sidewall area 113, 114 of the tire 101, 102 within the boundary of the field of view 130. The area relevant for the further processing steps lies within the sector boundaries 120, 121 and the inner and outer boundary lines 123 and 122 in the radial direction of the tire 101, 102. The field of view 130 is determined by the camera optics and the distance to the detected surface. After processing the image 200, the input data 210, in a preferred embodiment of the method, only includes the area enclosed by the sector boundaries 120, 121 and boundary lines 123, 122.

[0059] The Figure 8shows the schematic representation of a recording 200 of the inside of the tire belt or the tread 110, 111, 112 of the tire 101, 102. Within the boundary of the field of view 130, the area relevant for the further processing steps lies within the sector boundaries 120 and 121.

[0060] The Figure 9 shows the schematic compilation of the 200 bead-to-bead and all sectors. The test was carried out for both sidewalls 113, 115 of the tire 101, 102 and for the tread area 110, 111, 112 of the tire 101, 102 with eight sectors each according to Figure 2 This shows that a simple line-up of the images 200 cannot create a seamless image 205.

[0061] The Figure 9 shows the overall presentation of all individual sectors according to Figure 8after transforming the sector boundaries and boundary lines into a common object coordinate system. All partial areas of the tire surface shown in the images 200 within the sector boundaries are located in a correct position relative to one another. Image points located outside the sector and tire areas are eliminated, resulting in a clear representation of the entire tire surface. For this purpose, the sections of the tire are also scaled such that the sector enclosed by the sector boundaries and boundary lines has a rectangular shape. Accordingly, for example, the approximately trapezoidal shape of the enclosed area must be stretched in certain areas so that the individual areas can be combined to form an overall rectangular image of the tire.This pictorial representation of the test results is fed as an overall image or divided into several sub-areas to an artificial system, in particular a deep learning module, for evaluation.

[0062] The Figures 11a to 11c show the compilation of all images 200 of a bead-to-bead test on a truck tire. The images 200 from the outside of the first sidewall 113 are in Figure 11a The images 200 from the inside of the tread 110, 111, 112 are shown in Figure 11b The receptacles 200 from the outside of the second side wall 115 are shown in the Figure 11c The images of the tread 110, 111, 112 in Figure 11b show belt edge solutions in the upper and lower tire shoulder 110, 112, which extend over several sectors.

[0063] The Figure 12shows the representation of all individual sectors of a real measurement on a truck tire after transformation of the image data into a common object coordinate system comparable to the schematic representation in Figure 10 The belt edge solutions in the shoulder areas 110, 112 are joined across sectors and thus visible in their true circumferential extent. The extension of the belt edge solutions into the sidewall areas 11, 115 is now also clearly visible.

[0064] The Figure 13schematically shows the training of a deep learning module (DLM) using images 205 of defect locations from the bead area 114, 116 of a tire 101, 102. Defect locations with a predominantly radial orientation or significant extension in the circumferential and radial directions are used as training data for the defect locations to be located. Defect locations that, on the other hand, only have a significant extension in the circumferential direction are used as negative examples for the purpose of so-called negative mining, since these usually only represent artifacts at the visible edge of the bead 113, 116. The parameters or parameter weightings determined by the deep learning module DLM are stored in a first parameter set PS1 for further use, which can be used, for example, as training data.

[0065] The Figure 14shows the training of a deep learning module (DLM) using images of defect locations from the sidewall 113, 115 of a tire 101, 102. Detectable defect locations are used as training data. Irrelevant artifacts are used as negative examples. The parameters or parameter weightings determined by the deep learning module (DLM) are stored in a second parameter set (PS2) for further use.

[0066] The Figure 15 shows schematically the verification of a Deep Learning Module DLM using test data from the bead area 114, 116 of a tire 101, 102. For this purpose, test data from the bead areas 114, 116 of the overall representation according to Figure 10 used in combination with the parameter set PS1. The result is verified according to the usual verification strategy for a deep learning module DLM.

[0067] The Figure 16shows the verification of a Deep Learning Module DLM using test data from the sidewall 113, 115 of a tire 101, 102. Test data from the sidewall areas 113, 115 of the overall representation according to Figure 10 used in combination with the parameter set PS2.

[0068] The Figure 17 shows the scheme of the evaluation of tire data using the trained deep learning module DLM and the entry of the found defect locations into the overall representation of the tire 101, 102. The deep learning module DLM provides the position and boundary lines of the found defect locations.

[0069] The Figure 18 shows the representation according to Figure 12 with the defect locations found by the deep learning module for all areas of the tire. The position and size of the detected defect locations are shown as rectangles.

[0070] The Figure 19 shows the representation according to Figure 18with the contours of the detected defects post-processed using image processing functions. Overlapping rectangles are combined to form a single defect, and the precise contour of the defect is determined from this.

[0071] The Figure 20shows the representation of a tire with defect locations 201 to 204 in different tire sections. Depending on the tire section, the defect locations are evaluated differently by the Deep Learning Module DLM according to the section-specific parameter sets. Defect location 201 is located in tread 111 and, with this extent and shape, represents minor damage. Defect location 203 has approximately the same size and shape as defect location 201, but represents significant damage because it is located in the area of ​​bead 116. Defect location 204 in bead 116 has a predominantly horizontal extent and is therefore not critical in bead 116. Defect location 202 in sidewall 115 is comparable in shape and size to defect location 204. However, because it is located in sidewall area 114, it is significant. Figure 21shows a further exemplary representation of an interferometric result image of the tread of a tire, in particular a vehicle tire, with the areas of upper shoulder 110, tread center 111, and lower shoulder 112. In the resulting image shown, isolated small air inclusions are visible. In the lower shoulder 112, a defect caused by a repair patch can be seen in the circumferential direction between 135° and 160°. The defect location 250 represents an example of an image area that cannot be evaluated using the artificial system. Other such image areas that cannot be evaluated can also be caused by one of the elements listed below, but not exhaustively: repair areas, balancing patches, RFID sensors, or pressure sensors. If the image were viewed alone, this image area would represent a significant defect. Figure 22 shows that Figure 21Corresponding video image in a standardized, full-circumference view of the tire 101, 102, analogous to Figure 21. The video image shown was recorded using non-coherent illumination. In the area between 135° and 160° in the circumferential direction, the repair site, which has a repair patch, is more clearly visible.

[0072] Figure 23 shows the video image from Figure 22 with the position and area of ​​the repair patch determined by a deep learning module, which is in Figure 23 This image area is, according to one aspect of the invention, excluded from further evaluation of the resulting image from Figure 21 excepted because this area is a non-critical area 212, which includes a non-relevant anomaly.

[0073] The Figure 24shows the elimination of the image area not to be evaluated in the result image by replacing the relevant image area with a blank space. In summary, the testing device 10, and in particular the camera 21, takes a picture 200 of each sector of the tire 101, 102. The picture 200 is then processed and compiled into an image 205. The image 205 is then normalized and transformed into input data 210. The input data 210 is loaded into the deep learning module DLM. There, the input data 210 is analyzed. During the analysis, defect locations are identified and categorized into critical areas 211 and non-critical areas 212. The analyzed image 205 is stored as result data 220, which is a result image. The result data 220 is additionally integrated as training data 230 in a training data set 235. Reference sign list

[0074] 10Test fixture 11Base frame 12Rotary table 13Conveyor belt 20Measuring head 21Camera 22Optical axis Camera 23Vertical linear system 24Horizontal linear system 25Swivel axis 30Distance sensor 31Laser beam for distance sensor 30 32Distance sensor 33Laser beam for distance sensor 32 40Conveyor module 41Conveyor belt 42Incremental encoder 50Programmable logic controller 101, 102Tire 110Upper shoulder area 111Tread area 112Lower shoulder area 113First sidewall 114First bead 115Second sidewall 116Second bead 120left sector border 121right sector border 122outer boundary line 123inner boundary line 130field of vision 200Recording 205Figure 210Input data 211Critical region 212Non-critical region 220Result data 221First result data set 222Second result data set 230Training data 235Training data set 250 Defect location 251 Tread defect location 252 Sidewall defect location 253 Bead defect location 254 Bead defect location 262relevant error location 264non-relevant anomaly 500artificial system 502processor 504storage medium 506training data set A r axis of rotation of the tire B max maximum width B s width at the tire shoulder B w width at the tire bead B wi width at the tire bead D a outer diameter D ai outer diameter on the inside D i inner diameter P a outer limit point P i inner limit point RM tire center plane F k principal point of camera optics W k aperture angle of the camera field of view A k axial or vertical position of the camera R k radial or horizontal position of the camera DLMDeep Learning Module PS1 Parameter set 1 PS2 Parameter set 2

Claims

1. A method for interferometric testing of a tyre (101, 102) by means of an artificial system (500) which has a processor (502) and a storage medium (504) which are suitable for carrying out the following method steps, the artificial system (500), which is trained by means of at least one training data set (235), being suitable for detecting relevant defect locations (250), the method comprising the following method steps: a) Generating a picture (200) by means of a camera (21) of at least a partial area of the surface of the tyre (101, 102); b) generating input data (210) comprising an image (205) consisting of at least one picture (200); c) analysing the input data (210) to determine and categorise a critical region (211) comprising a relevant defect location (262) and a non-critical region (212) comprising a non-relevant anomaly (264); d) Saving the analysed image (205) as result data adjusted for the non-critical region (220); and e) Saving the adjusted result data (220) as training data (230) in the training data set (235).

2. The method according to claim 1, characterised in that the non-relevant anomaly (264) comprises an irrelevant deformation of the surface of the tyre (101, 102).

3. The method according to claim 1 or 2, characterised in that the picture (200) is a phase image, a contrast image corresponding to the phase image or a video image.

4. The method according to any one of claims 1 to 3, characterised in that a sector boundary (120, 121) and / or a boundary line (122, 123) to the picture (200) is determined for the creation of input data (210), wherein the sector boundary (120, 121) and / or the boundary line (122, 123) delimits a defined section of the surface of the tyre (101, 102), wherein the input data (210) comprises the bounded and defined section, and preferably removing the area outside the defined section.

5. The method according to any one of claims 1 to 4, characterised in that localisation information about the position of a camera (21) in relation to the tyre (101, 102) at the time of the picture (200) taken by means of the camera (21) is used to create the image (205); wherein preferably at least two pictures (200) are combined to form the image (205), taking into account the respective localisation information; wherein further preferably the localisation information comprises the distance and / or the orientation of the camera (21) to the recorded part of the surface of the tyre (101, 102) and / or a tyre geometry information, in particular an axial section.

6. The method according to claim 5, characterised in that a functional structure of the tyre (101, 102) is derived from the localisation information; wherein preferably the functional structure comprises at least one, in particular two or preferably three, of the following sub-regions of the tyre (101, 102): inner first bead (114I), outer first bead (114A), inner first sidewall (113I), outer first sidewall (113A), inner upper shoulder (110I), outer upper shoulder (110A), tread (111), inner second bead (116I), outer second bead (116A), inner second side wall (115I), outer second side wall (115A), inner lower shoulder (112I), outer lower shoulder (112A).

7. The method according to claim 5 or 6, characterised in that the image (205) is normalised on the basis of the localisation information; wherein preferably the size and / or the resolution of the image (205) are scaled to a reference scale.

8. The method according to any one of claims 3 to 7, characterised in that when using a plurality of phase images, the phase values are normalised, in particular scaled to a reference scale pixel by pixel; wherein the phase images are preferably demodulated before normalisation, in particular scaling; wherein preferably in step a) a plurality of intermediate images are taken to calculate the phase value in order to create a phase image between two different load states of the tyre, and wherein preferably the phase values of successive intermediate images are subtracted in pairs and the resulting difference values are totalled.

9. The method according to any one of claims 6 to 8, characterised in that for determining and / or categorising the critical region (211) and / or the non-critical region (212) in step c), the input data (201) are weighted on the basis of the functional classification.

10. The method according to any one of claims 1 to 9, characterised in that the size of a defect location (250) and / or the frequency of a defect location (250) and / or the distance between two defect locations (250) are taken into account when determining and / or categorising the critical region (211) and / or the non-critical region (212) in step c).

11. The method according to any one of claims 1 to 10, characterised in that the result data (220) according to step d) is stored in at least two result data sets (221, 222); wherein the result data sets (221, 222) are preferably assigned to the sub-areas of the tyre (101, 102) resulting from the functional classification; wherein preferably the input data (210) and / or the result data (220) are displayed on a display unit; wherein the critical region (211) is further preferably highlighted.

12. A system for data processing comprising means for carrying out the method according to any one of claims 1 to 11, wherein the means comprise: a camera (21) for producing a picture (200) of at least a partial area of the surface of the tyre (101, 102); a processor (502) for generating input data (210) comprising an image (205) consisting of at least one picture (200), and analysing the input data (210) to determine and categorise a critical region (211) comprising a relevant defect location (262) and a non-critical region (212) comprising a non-relevant anomaly; and a storage medium (504) for storing the analysed image (205) as result data (220) adjusted for the non-critical region (212), and the adjusted result data (220) as training data (230) in the training data set (235).

13. A computer program product comprising instructions which, when the program is executed by a computer, cause the apparatus of claim 12 to perform the method of any one of claims 1 to 11.

14. A computer-readable storage medium on which the computer program product according to claim 13 is stored.

15. A testing device (10) for testing a tyre (101, 102) by means of a method according to any one of claims 1 to 11, comprising: a test chamber; a camera (21) for producing a picture (200) of at least a partial area of the surface of the tyre (101, 102); a processor (502) for generating input data (210) comprising an image (205) consisting of at least one picture (200), and for analysing the input data (210) to determine and categorise a critical region (211) and a non-critical region (212), wherein a critical region (211) comprises a relevant defect location (262); and a storage medium (504) for storing the analysed image (205) as result data (220) adjusted for the non-critical region (212) and the adjusted result data (220) as training data (230) in the training data set (235).