Method for determining a bent-wires indicator for a roller brush

Abstract

A method for determining a bent-wires indicator, relative to a roller brush (1) made up of metallic wires, the method comprising: - s1: acquiring a base image of a portion of a lateral surface (2) of the roller brush, - s2: determining a first processed image by intensity-filtering of the base image using a first input-output intensity filtering function, and determining a second processed image by intensity-filtering of the base image using a second input- output intensity-filtering function, the second input-output intensity-filtering function being higher than the first input-output intensity-filtering function for intermediate intensity levels, - s3: determining the bent-wires indicator by comparing the first processed image and the second processed image.

Description

Method for determining a bent-wires indicator for a roller brush

[0001] The technical field is that of characterizing a roller brush, more specifically characterizing a roller brush made up of metallic wires.Technical background

[0002] Due to the strong development of electrically-driven vehicles, so-called “electric” steels are produced in ever-increasing quantities. The demand for electric steel coils and sheets in particular is very high and increasing. Electric steels, which are steels with specific magnetic properties well suited for parts of electric motors, generators, relays or transformers. They usually have a substantial silicon content (typically above 1% in mass, for instance from 1% to 7% in mass). For electric steel grades, the manufacturing is usually very challenging, and may differ substantially from the manufacturing of more conventional grades. For instance, in annealing furnaces designed for electric stees, roller brushes are employed instead of conventional plain rollers. Such roller brushes are made up of metallic wires, to withstand the very high temperatures in the annealing furnace (above 600°C, or even above 700°C). An annealing furnace for manufacturing electric steel typically comprises tens of such roller brushes, each being expensive and easily damaged.

[0003] Employing roller brushes in good condition allows for high quality processing of the steel strip. In particular, it is preferable that each roller brush has wires ends homogeneously distributed on the lateral surface of the roller brush (with no wire agglomeration, or splay), and that the wires ends have a limited blur. It is thus useful to be able to test if a roller brush is in good condition, either to detect wear or damages for a roller brush that has been in operation for a while, or to assess the quality of a newly received roller brush (a brand-new roller brush).

[0004] In this contest, there is thus a need for a method for testing the condition of a metallic roller brush.Summary

[0005] In this context, a method according to claim 1 is provided.

[0006] In this method, a bent-wire indicator is determined, by image analysis. Indeed, the inventors have observed that for a roller brush made of metallic wires, bent wires can be frequent and can even be the predominant defect for the roller brush. When some of the wires are bent, the density of wires ends on the surface is lowered or increased locally and the distribution of wires ends on the roller brush surface becomes uneven, inhomogeneous, which is not desirable in terms contact quality. In the instant method, to assess the quality of the roller brush, the presence of bent wires is thus tested, and possibly quantified. In practice, the bent wires indicator is determined by comparing the first processed image and the second processed image.

[0007] In the absence of bent wires, the base image comprises of:multiple high intensity (i.e.: bright) zones, corresponding to the multiple wires ends present in the image,- and a low-intensity (i.e.: dark) background (corresponding to the inner volume of the roller brush).

[0008] In the presence of bent wires, the base image comprises also zones of intermediate intensity, corresponding to portions of the lateral surfaces of the bent wires, seen from the side. These zones are somehow located between the foreground, which comprises the wires ends, and the background. In these zones, the intensity is intermediate in that it has intermediates values, comprised between the intensity of the wires ends and the intensity of the dark background. As the intensity in these zones is intermediate, they are filtered differently by the first intensity filter and by the second intensity filter. More precisely, the intensity in these zones is enhanced in the second processed image (see figure 3), compared to the first processed image (see figure 2).

[0009] In the absence of bent wires, the first and second processed images are almost identical (see figures 4 and 5, for instance). While in the presence of bent wires, the second processed image is markedly different from the first processed image which enables detecting the presence of bent wires.

[0010] This method is robust, as it is a somehow a differential detection method. And it is independent of the wires density, diameter or arrangement (as wire bundles, or not). It can thus be applied to different types of roller brushes, without needing a preliminary calibration, specific to the type of roller brush considered (such a calibration may still be achieved, for other quality check purposes, but it is not mandatory).

[0011] The method according to the invention may comprise one or several additional, optional features, defined in claims 2 to 13, considered alone or in combination.

[0012] The instant technology also concerns an electronic device according to claim 14, a system according to claim 15 and a computer program according to claim 16, and a non- transitory computer-readable media, such as a hard drive or a flash memory, comprising such a program.Detailed description

[0013] The instant technology will now be described in more detail and illustrated by examples without introducing limitations, with reference to the appended figures.

[0014] Figure 1 represents partially a base image of a portion of a lateral surface of a roller brush, in a case for which bent wires are present.

[0015] Figure 2 represents partially a first processed image, obtained by intensity-filtering of the base image using a first input-output intensity filtering function.

[0016] Figure 3 represents partially a second processed image, obtained by intensity-filtering of the base image using a second input-output intensity filtering function.

[0017] Figure 4 and figure 5 are other examples of portions of the first and second processed images, in a case for which no or just a few bent wires are present.

[0018] Figure 6 schematically represents a lighting device employed to illuminate the roller brush in order to capture the base image, the lighting device being in an open configuration.

[0019] Figure 7 schematically represents the lighting device.

[0020] Figure 8 schematically represents the portion of the lateral surface of the roller brush captured in the base image.

[0021] Figure 9 schematically represents the first input-output intensity filtering function and the second input-output intensity filtering function.

[0022] Figure 10 schematically represents another example of first and second input-output intensity filtering functions that could be used to implement the instant method.

[0023] Figure 1 1 is a bloc diagram of steps of a quality check for the roller brush.

[0024] As above mentioned, the instant method concerns the determination, by image analysis, of a bent-wire indicator for a roller brush made up of metallic wires 4, and more generally a quality check method for such a roller brush. The roller brush typically has a length above 1 meter and a diameter above 10 cm or even above 20 cm.

[0025] General aspects of the method will be presented first, and an exemplary embodiment will be presented then in more detail.General aspects of the method

[0026] The instant method comprises:- s1 : acquiring a base image Img (figure 1 ) of a portion 3 of a lateral surface 2 of the roller brush 1 (figures 7 and 8),- s2: determining a first processed image Img1 (figure 2) by intensity-filtering of the base image Img using a first input-output intensity filtering function fi (see figure 9 or 10)), and determining a second processed image Img2 (figure 3) by intensity-filtering of the base image Img using a second input-output intensity-filtering function f2, the second input-output intensity-filtering function f2being higher than the first input-output intensity-filtering function fi for intermediate intensity levels t,- s3: determining the bent-wires indicator by comparing the first processed image Img1 and the second processed image Img2.

[0027] The base Image Img may be colour image or a grey-levels image. When a colour image, it may be converted into a grey-level image before being intensity filtered to determine Img1 and Img2. Step s1 may comprise capturing the base image, and then transmitting it to processing device (eg.: computer) or module which thus acquires the base image. Alternatively, step s1 may comprise acquiring (in other words charging, that is receiving and storing) the base image, by a processing device, without comprising the image capture itself.

[0028] By intensity filtering using the first input-output intensity-filtering function fi, it is meant that, to determine the first processed image Img1 , for each pixel of the base image Img, an intensity level x (or, in other words, a brightness level) of the pixel is replaced by a modified intensity level y=fi(x), where fi is the first input-output intensity-filtering function. The same applies for the intensity filtering using the second input-output intensity-filtering function f2.

[0029] In the base image Img, the intensity level(s) for the wire ends e is typically very high (due, inter alia, to fact that the wires are metallic wires), often equal to highest possible intensity level ax in the base image (for instance equal to 255 if Img is an 8-bits grey levels image). An average intensity level, for the wire ends e, in Img, is denoted le.

[0030] On the contrary, the intensity in the background of the base image, which corresponds to the inner volume of the roller brush, is very low (i.e.: the background is very dark). An average intensity level for the background, in Img, is noted lbg-

[0031] The intermediate intensity levels lint, for which the second input-output intensity f2is higher than the first input-output intensity filtering function fi , are comprised between lbgand le.

[0032] The intermediate intensity levels Lt may, like here, form a range of intensity levels, from a lower bound I. to an upper bound In practice, I. may be set as being equal to c.lbgwhere c is a coefficient from 2 to 4 (to have the intermediate intensity levels sufficiently different from the background average intensity), while l+is set as being equal to I. + c’.(le- lbg) with c’ from 0,1 to 0,40. (so that the intermediate intensity levels span over a substantial portion of the intensity dynamic le- lbg). For example, for 8-bits grey levels images, in a case with lbg=45 and le=250, I. and l+could be respectively set to 130 and 160, for example (which corresponds to c=2.88 and c -0.14). And in case with lbg=20 and le=255, I. and l+could be respectively set to 50 and 115 (which corresponds to c=2.5 and c -0.27), or to 50 and 78 (c=2.5 and c’=0.12), for example.

[0033] For intensity levels above the intermediate intensity levels Lt, and possibly also for intensity levels below the intermediate intensity levels Lt, the first input-output intensity filtering function fi and the second input-output intensity filtering function f2may be equal to each other or substantially equal to each other (i.e.: equal within 5%).

[0034] Conversely, for the intermediate intensity levels Lt, the second input-output intensity filtering function f2may be at least 1 .5 higher, or even at least two times higher than fi.

[0035] In practice, the first and second intensity filterings may correspond to binarizations of the base image Img (in which case the first and second input-output intensity filtering functions fi and f2are step functions, as illustrated in figure 9). In this case, the first intensity filtering is a binarization of Img using a first intensity threshold th 1 while the second intensity filtering is a binarization of Img with a second intensity threshold th2 lower than the first intensity threshold th1. In this case, the upper and lower bound of the intermediate intensity levels range, I. and L, are th2 and th1.

[0036] Still, the first and second intensity filterings may be achieved using other kinds of shapes for the filtering functions, for instance smooth, continuous ones (see figure 10, for instance) instead of step functions.

[0037] Regarding the bent-wires indicator, it can be a binary indicator specifying:- either that bent wires are present (more precisely that a significant proportion of the wires are bent, or in other words slanted compared to a normal radial orientation or the wires);- or, on the contrary, that no bent wires are detected (no significant proportion of wires bent).

[0038] The bent-wires indicator could also be a continuous, real-valued indicator whose value is all the higher as the bent-wires are abundant.

[0039] The step s3, during which img1 and img2 are compared to each other, may comprise:- s30: determining a first wire-filed area S1 , which is a total area occupied, in the first processed image Img1 , by wires ends e and by possibly present wires side-surfaces portions s, and determining a second wire-filed area S2, which is a total area occupied, in the second processed image Img2, by wires ends and by possibly present wires side-surfaces portions, s31 : determining the bent-wires indicator based on a comparison of the first wire-filed area S1 with the second wire-filed area S2, or based on a comparison of a first and a second wire-filed area fractions S1 % and S2% corresponding respectively to the first wire-filed area S1 divided by a total image area S, and to the second wire-filed area S2% divided by the total image area S.

[0040] In particular, step s30 may comprises:Identifying zones z corresponding to wires ends e or corresponding possibly also to wires side-surfaces portions s, in the first processed image Img1 , using a segmentation algorithm or a particle detection algorithm,Determining the first wire-filed area S1 by summing the individual areas of said zones z identified in the first processed image Img1 ,Identifying zones z corresponding to wires ends e or possibly to wires side-surfaces portions s, in the second processed image Img2, using the segmentation algorithm or the particle detection algorithm,Determining the second wire-filed area by summing the individual areas of said zones z identified in the second processed image Img2.

[0041] Using such a segmentation of particle detection algorithm to determine S1 and S2, enables to reduce the influence of inevitable noise present in the images (compared for instance to a direct count of bright pixels - that is pixel with an intensity level above I., in the first, or second processed image). In particular, one thus reduces the influence of noisy brightpixels isolated in the dark background, not belonging to the wider zones corresponding to wires ends or sides.

[0042] The segmentation or particle detection may be achieved by outer edge detection, for instance using a cany filter, or based directly on the thresholded image (when the intensity filterings are binarizations). The segmentation or particle detection preferably comprises eliminating particle whose area is below a given area threshold (this threshold may below 0.9 or 0.8 times the area corresponding to the nominal section of the kind of wire employed, for instance).

[0043] In step s31 , the comparison of S1 with S2, or the comparison of S1 % with S2% may be achieved by computing the difference thereof, or the ratio thereof, and comparing the difference, or the ratio, with a predetermined threshold. Still, other comparison techniques (like applying a logistic function to the difference, for instance) could be employed.

[0044] In the exemplary implementation described below, in step s31 , S1 % is compared with S2% by computing the difference S2% - S1% and testing whether it is above a detection threshold dt. The detection threshold dt may be determined in advance by calculating the quantity S2%-S1 % for an image of a reference roller brush considered in good condition, or for multiple images of such reference roller brushes, and multiplying the result by a coefficient, for instance equal to or higher than 2. In practice, the detection dt is typically from 2% to 10%, for instance equal to 8%.

[0045] It is noted that, from an algorithmic point of view, step s2 and step s30 could be achieved during a same processing step, for instance during a same segmenting step (parametrised differently to measure S1 , and to measure S2).Exemplary implementation of the method

[0046] 1 . image acquisition

[0047] In step s1 comprises, the roller brush 1 is illuminated in a controlled manner using one, or, like here, multiple light sources 13. A portion 3 of the lateral surface 2 of the roller brush 1 , that is a portion of the circumferential, cylindrical surface or the roller brush, is illuminated from the front, so that the light emitted by the one or more light sources reaches the wires ends (that is the wires cut extremity) first, and then penetrates in the inner volume of the roller brush. In other words, the wires are not illuminated from the side, but from the front. Besides, surrounding light other than the light produced by the light sources 13 is blocked, so that the surrounding light does not reach said portion 3 of the lateral surface 2 of the roller brush 1 .

[0048] To this end, the lighting device 10 of figures 6 and 7 is employed. This lighting device 10 comprises:- a casing 16 made of an opaque material (e.g.: back-painted wood) and which, on one side of the casing 1 1 (its lower side 11 ), fits the lateral surface 2 of the roller brush,here in a substantially light-tight manner; to this end, the side 11 of the casing has a round shape,- the light sources 13 housed in the casing 16,

[0049] The casing 16 has at least a first opening 12 or window (see figure 8), on the side of the casing 1 1 that fits the lateral surface 2 of the roller brush, for the light produced by the light sources 13 to reach the portion 3 of the lateral surface 2 of the roller brush (on which the casing is applied).

[0050] The lighting device 10 comprises also a diffuser (for instance a leaf of white paper) interposed between the light sources 13 and the first opening 12 or window.

[0051] The light sources 13 are arranged, relative to the first opening 12, so that this opening is illuminated in a multidirectional manner (i.e.: with light coming several different directions), and in a homogeneous manner (i.e. with an intensity that is homogenous over the whole opening 12).

[0052] To obtain a multidirectional lighting, the light sources 13 are arranged peripherally, around the opening 12. The light sources are arranged all around the opening 12, on each of the four sides of this opening. Thanks to this multidirectional lighting, the wire ends reflect light towards an image acquisition device used to capture the base image Img (such as a digital camera or a smartphone), and thus appear very luminous in the image, whatever the orientation of the wire ends. Here, the light sources 13 are LEDs of a LEDs ribbon that is applied along a rectangular inner edge of the casing 16.

[0053] To obtain a homogeneous lighting, the light sources 13 are arranged at a distance from the opening 12. To this end, the casing 16 has a width which is at least twice, or even four times the width of the first opening.

[0054] The homogeneity and mu Itidirectionallity of the lighting is also improved by the diffuser 14.

[0055] As represented in the figures, the casing 16 has a second opening 15, on a side opposite side 1 1 , that enables to capture the base image Img using the image acquisition device, which is placed outside of the casing 16. Still, in other implementations, the casing could be deprived of such a second opening, the image acquisition device being then placed directly within the casing.

[0056] Step s1 comprises here:- turning on the one or more light sources 13 and check that the light inside the box (at the first opening 12) is equal to 86 ±2 lux before,- placing the casing (the box) on the roller brush at the location chosen for the measurement; this location may be chosen by visual study, be the worst-condition location on the roller brush;- placing image acquisition device at the second opening 15;- capturing the base image Img using the image acquisition device, making sure that the entire study window (i.e. the opening 12) is seen in; checking that the image is not blurred, recapturing it if necessary; and verifying that the portion 3 is illuminated homogeneously, recapturing if necessary.

[0057] 2. image analysis

[0058] 2.1 Bent-wires indicator

[0059] The base image Img is converted to grey levels and cropped to the study window, then scaled so that actual areas (in square cm, for instance) can be measured directly from the base image.

[0060] Then Img1 is determined by binarizing Img using the first intensity threshold th1 , here equal to 160 (step s2).

[0061] Then, automatic detection and analyses of particles in Img1 is achieved, to determine the total area fraction S1% occupied by wire ends or side surfaces, in Img1 (step s30). To this end, a minimum particle size is set in the particle detection algorithm (to eliminate noisy pixels not corresponding to wire ends / sides). The minimum particle size set as equal to 0.8 times the nominal section of the wires.

[0062] Img2 is determined by binarizing Img using a second intensity threshold th2, here equal to 130 (step s2).

[0063] Then, automatic detection and analyses of particles in Img2 is achieved, similarly as for Img1 , to determine the total area fraction S2% occupied by wire ends or side surfaces, in Img2 (step s30).

[0064] Then (step s31 ) it is tested if S2%-S1% is above the detection threshold dt, in which case the bent-wires indicator, which is here a binary indicator, specifies that bent wires are present. A corresponding bent-wires non-conformity message is displayed by a humanmachine interface.

[0065] 2.2 Burr rate

[0066] In this exemplary implementation, the burr rate testing is carried out based either on Img1 , or on another thresholded image obtained by binarizing Img, with a intensity threshold possibly different from th 1 .

[0067] Anyhow, the binarized image considered is filtered using watershed segmentation, to separate wires ends that touch each other, in the image. The watershed segmentation comprises calculating an Euclidian distance map (distance between the back-ground and the pixel considered, within the particle identified) and finding the ultimate eroded points (which are the maxima for this distance). It then dilates each of the ultimate eroded points until the edge of the particle is reached, or until the edge touches a region of another growing ultimate eroded point.

[0068] Then particle detection is applied to count the number N of wire ends e in the (processed) image, and to compute the total area Se occupied by the wire ends. N is corrected using a linear correction, to determine a corrected number of wires N’. The linear correction to be applied is determined preliminary, based on manual counting and calibration.

[0069] A burr rate br is then computed as follow: br=Se / (N’.Sn) where Sn is the nominal section of the wires.

[0070] A normalized burr rate nbr is computed as follow: nbr=br / bro, where bro is a reference burr rate.

[0071] bro is obtained by applying the above described operations to an image of a reference roller brush, considered with no burr.

[0072] In practice, for the tests achieved, bro was close to 1 .85. Even if the wires are cut perfectly, with no burr, bro is higher than 1 , in practice, due to the limited resolution of the image acquisition device, and to the fact that it may be not perfectly focused on the lateral surface 2 of the roller brush.

[0073] Normalizing the burr rate in this way makes the burr detection independent of possible imperfections of the imaging system.

[0074] Then, if the normalized burr rate is substantially higher than 1 , for instance higher than 1.2, burr is considered too high, corresponding to a non-quality case for the roller brush. A corresponding excessive-burr non-conformity message is then displayed by the human- machine interface.

[0075] 2.3 Density of wires per unit surface

[0076] The density of wires per unit surface, ds, if computed by dividing N; by a total area S of the image.

[0077] If ds is not equal to a nominal wire density for that roller, dso, within a given acceptable precision (here within 10%), a non-quality indicator is output. For instance, a density nonconformity message may be displayed by the human-machine interface.

[0078] By nominal density, it is meant the density the roller brush is supposed to have, when brand new and in the absence of bent-wires, on the lateral surface (the contact surface) of the roller brush.

[0079] Quality check

[0080] The quality check for the roller brush is passed if the three tests, regarding the presence of bent wires, the burr rate, and the density of wires, are all passed (figure 11 ).

[0081] A message is displayed on the human-machine interface to specify that the quality check is either passed or failed.

[0082] Exemplary measurements

[0083] Values of wire-filled area fractions, and other quantities computed when executing this method are presented in table 1 for seven tested roller brushes and the reference roller brush. Roller brush 2 did not pass the quality check.

[0084] Table 1Device and system for determining a bent-wires indicator for a roller-brush

[0085] Expect for the steps of illuminating the roller brush and capturing the base image, the method described above can entirely be executed by an electronic device having the structure of a computer (in particular comprising at least a processor and a memory) such as a laptop computer, a smartphone or a remote (and possibly distributed) server. In this regard, it is noted that the instant technology also concerns such an electronic device, configured (in practice programmed) for executing the above method.

[0086] The instant technology also concerns a system comprising the electronic device in question, the lighting device 10, and, optionally, the image acquisition device. The image acquisition device and the electronic device may take the form of a single device having both image acquisition and image processing capabilities such as a smartphone.

[0087] Different alternatives are possible for the method and device presented above, in addition to the ones already mentioned. For instance, the first and second processed images could be compared differently than by computing S1 % and S2%. For instance, a dissimilarity level between Img1 and Img2, such as root-mean-square of the difference between the two images, could be computed directly, the bent-wire indicator being then derived from this dissimilarity level.

[0088] Besides, more than two processed images could be employed. For instance, different values of a wire-filed area Se% (in fact, of the detected-particles-filed area) could be computed for different values of a binarizing threshold th (or, more generally, for different intensity filterings), the presence of bent wires being then detected from the curve Se%, as a function of the threshold th. For the bent wires, a depth indicator, relative to a depth (depth relative to the surface 2 of the roller brush) at which the bent-wires are bent (position at which they arefolded), may be determined from such multiple processed images. For instance an indication specifying that the position of the wires foldings is either close to the surface of the roller brush, or far from it (for instance a binary indication) could be derived from the curve Se% as a function of the threshold th.

[0089] More generally, using several processed images, instead of just two, may allow for determining a tilting profile for the wires, that is a tilt as a function of the depth.

Claims

CLAIMS1 . A method for determining a bent-wires indicator, relative to a roller brush (1 ) made up of metallic wires (4), the method comprising:- s1 : acquiring a base image (Img) of a portion (3) of a lateral surface (2) of the roller brush (1 ),- s2: determining a first processed image (Img1 ) by intensity-filtering of the base image (Img) using a first input-output intensity filtering function (fi), and determining a second processed image (Img2) by intensity-filtering of the base image (Img) using a second input-output intensity-filtering function (f2), the second input-output intensity-filtering function (f2) being higher than the first input-output intensityfiltering function (fi) for intermediate intensity levels ( t),- s3: determining the bent-wires indicator by comparing the first processed image and the second processed image.

2. A method according to claim 1 , wherein step s3 comprises:- s30: determining a first wire-filed area, which is a total area occupied, in the first processed image, by wires ends and by possibly present wires side-surfaces portions, and determining a second wire-filed area, which is a total area occupied, in the second processed image, by wires ends and by possibly present wires sidesurfaces portions, s31 : determining the bent-wires indicator based on a comparison of the first wire- filed area with the second wire-filed area, or based on a comparison of a first and a second wire-filed area fractions corresponding respectively to the first wire-filed area divided by a total image area, and to the second wire-filed area divided by the total image area.

3. A method according to claim 2, wherein step s30 comprises:Identifying zones (z) corresponding to wires ends (e) and possibly also zones (z) corresponding to wires side-surfaces portions (s), in the first processed image (Img1 ), using a segmentation algorithm or a particle detection algorithm, Determining the first wire-filed area by summing the individual areas of said zones (z) identified in the first processed image,Identifying zones (z) corresponding to wires ends (e) and possibly also zones (z) corresponding to wires side-surfaces portions (s), in the second processed image (Img2), using the segmentation algorithm or the particle detection algorithm,Determining the second wire-filed area by summing the individual areas of said zones identified in the second processed image.

4. A method according to claim 2 or 3 wherein the bent-wires indicator is a binary indicator specifying that bent wires are either present or absent, and wherein, in step s31 , if a difference between the second wire-filed area fraction and the first wire-filed area fraction is above a detection threshold, then, the bent-wires indicator specifies that bent wires are present.

5. A method according to anyone of the previous claims, wherein the intensity filtering employed for determining the first processed image (Img1 ) is a binarization of the base image (Img) using a first intensity threshold (th1 ), and wherein the intensity filtering employed for determining the second processed image (Img2) is another binarization of the base image (Img) using a second intensity threshold (th2) lower than the first intensity threshold (th1 ).

6. A method according to anyone of the previous claims, wherein the first input-output intensity filtering function and the second input-output intensity filtering function are equal to each other for high intensity levels that are above said intermediate intensity levels.

7. A method according to anyone of the previous claims, further comprising:- determining a number of wires ends in the base image and a total area occupied by the wires ends in the base image, the base image being possibly processed before determining said number and total area,- computing a burr rate as equal to said total area divided by the number of wires ends, and divided by a nominal wire section,- detecting a burr abnormality by comparing the burr rate with a reference burr rate obtained for a reference roller brush.

8. A method according to the previous claim wherein determining the number of wires ends in the base image comprises binarizing or segmenting the base image and then applying a watershed filtering.

9. A method according to claim 7 or 8, further comprising outputting a quality check indicator, which specifies that the roller brush passed a quality check provided at least that:- the bent-wires indicator specifies an absence of bent wires, or is below a given threshold;- and that no burr abnormality is detected.

10. A method according to anyone of the previous claims, comprising:Determining one or more additional processed images, by intensity-filtering of the base image (Img) using input-output intensity filtering functions that are different one from another, and that are different from the first (fi) and the second (f2) inputoutput intensity filtering functions,Determining a tilt profile of the wires, or a depth indicator relative to a depth at which the bent-wires are bent, form the first processed image, the second processed image and the one or more additional processed images.1 1 . A method according to anyone of the previous claims wherein step s1 comprises:- illuminating the roller brush (1 ) using one or more light sources (13) arranged to illuminate said portion (3) of the lateral surface (2) of the roller brush (1 ) from the front,- capturing the base image (Img) using an image acquisition device.

12. A Method according to the preceding claim, wherein step s1 comprises, before capturing the base image, blocking surrounding light other than the light produced by the one or more light sources, so that the surrounding light does not reach said portion of the lateral surface of the roller brush.

13. A Method according to the claim 11 or 12, wherein the light sources (13) are distributed peripherally, around said portion (3) of the lateral surface (2) of the roller brush (1 ).

14. An electronic device comprising at least a processor and a memory, configured for executing the method according to anyone of claims 1 to 10.

15. A System comprising:- the electronic device according to claim 14,- an image acquisition device,- a lighting device (10) which comprises:o a casing (16) made of an opaque material and which, on one side of the casing (11), fits the lateral surface (2) of the roller brush, o one or more light sources (13) housed in the casing (16), o the casing having at least one opening (12) or window, on the side of the casing (11) that fits the lateral surface (2) of the roller brush, for the light produced by the one or more light sources (13) to reach a portion (3) of the lateral surface of the roller.

16. Computer program comprising instructions whose execution by a computer makes the computer to execute the method according to anyone of claims 1 to 10.

Images