WO1998044338A1 - Method for examining an unwinding strip surface and analysing the strip defectiveness - Google Patents

Abstract

The invention concerns a method for examining an unwinding strip (10) surface comprising the following steps: forming at least a digital image of at least one strip surface, consisting of a set of elements each with an assigned digital value; detecting the strip surface defects; processing the surface defects to identify the type of fault corresponding to each defect and analysing said faults for determining the strip (10) defectiveness. The invention is characterised in that during the analysing step, said identified faults of each type are classified according to a set of surface fault sub-classes each corresponding to the intrinsic damaging nature of each type of faults and the use of said strip is validated when the population of each fault sub-class is less than a maximum acceptable value for said sub-class and for said use.

Description

 Method for inspecting the surface of a moving strip and analyzing the defect of the strip ".

The present invention relates to a surface inspection method of a moving strip, in particular of a laminated sheet moving at high speed, for detecting a surface defect. In conventional surface inspection systems, in which a laminated sheet is inspected automatically, at least one digital image of at least one face of the strip is formed, made up of a set of affected image elements. each of a digital value, said at least one digital image is filtered for the detection of surface irregularities of the strip by detection of relative variations of said digital values, said at least one filtered digital image is processed for identification of the type of defect corresponding to each irregularity and said defects are analyzed for the determination of the defectiveness of the strip.

In systems of this type, the analysis of defects is carried out by associating with each type of defect a criterion of gravity which corresponds in fact to a criterion of harmfulness of each type of defect for a particular product, which depends on a part of the nature and dimensions of the defect, but also of the subsequent use of this product.

This last criterion being a parameter which varies over time, it is not possible, using such systems, to have a fixed scale of appreciation over time. Consequently, using such systems, it is not possible to observe the development over time of the operating drifts of the manufacturing installations placed upstream. The object of the invention is to overcome this drawback.

It therefore relates to a method of inspecting the surface of a moving strip, for detecting a surface defect, comprising the steps consisting in:

- forming, using image-taking means, at least one digital image of at least one of the faces of the strip, consisting of a set of lines of image elements each assigned a digital value , filtering said at least one digital image for the detection of surface irregularities, by detecting relative variations of said digital values,

processing said at least one filtered digital image for identifying the type of surface defect corresponding to each detected irregularity, and analyzing said identified defects for determining the defectiveness of the strip, characterized in that during said step analysis of faults, said identified faults of each type are classified according to a set of sub-classes of surface defects each corresponding to a predefined intrinsic harmfulness for each type of defect and a use of said inspected strip is validated when the population of each defect subclass is less than a predetermined maximum permissible value for said subclass and for said use.

The method according to the invention may also include one or more of the following characteristics: - prior to the image processing step, a general characterization of the irregularities is carried out by determining for each of them the value of predetermined parameters characteristic of defects surface, and a prior classification of said irregularities is carried out, on the basis of determined values of said parameters according to a set of predefined classes, said processing step being carried out on each class;

a second mode of characterization of the irregularities is determined for each predefined class the number of characteristic parameters of which is less than the number of characteristic parameters of general characterization and, after the prior classification step, the following is determined for each irregularity detected value of the characteristic parameters of the second specific characterization mode of said class to which the irregularity belongs, from the values of the characteristic parameters of general characterization;

- during said image processing step to identify the type of defect corresponding to each detected irregularity, in response to a detection of an image element of an irregularity, a memory area is delimited in a storage area for lines of image elements successively delivered by the photographing means and comprising at least an image element corresponding to at least one irregularity, each storage area is segmented into suspect areas each having at least one surface irregularity, suspect areas of successive storage areas are matched and corresponding to the same irregularity and 1 ' compares the total number of lines of picture elements of the matched suspect areas with a very long fault detection threshold, and, if said threshold is exceeded, said step of processing said at least one filtered digital image is carried out only on one of the suspect zones paired, the result of the treatment and analysis of the identified fault being assigned to the other zones above paired pects;

The processing step further comprises the steps of counting the number of identified defects of the same type per unit of length, and of comparing said number of defects of each type with a predetermined threshold value representative of the minimum number of defects from of which said faults are likely to be periodic, with a view to detecting periodic faults; - during the image processing stage for

The identification of the type of defect corresponding to each detected irregularity, a specific classification of the irregularities is carried out according to a set of elementary classes, and the population of said elementary classes is analyzed with a view to detecting periodic faults;

said image processing step for identifying the type of defect corresponding to each irregularity comprises a step of determining, for each detected irregularity, the value of parameters characteristic of surface defects, a step of grouping identified defects using a set of predefined criteria, in particular geometric and / or topographic criteria, and a step of combining the parameters characteristics of the grouped surface defects, said step of analyzing said defects being carried out on the basis of the value of said combined parameters for the grouped defects. Other characteristics and advantages will emerge from the following description, given solely by way of example and made with reference to the appended drawings in which:

- Figure 1 is a general diagram of an embodiment of a surface inspection installation according to the invention; - Figure 2 shows a part of an image delivered by the shooting means of the installation of Figure 1 and stored in the memory;

- Figures 3a to 3e show different images of the surface of a strip during a step of dividing images;

- Figure 4 is a flowchart illustrating the general operation of one installation of Figure 1; FIG. 5 is a flowchart showing the different stages of processing the filtered digital images;

- Figures 6a and 6b are diagrams showing, as a function of the length and the width of the surface defects, the different classes of defects, respectively for a semi-finished product (DKP) and for a galvanized product; and

- Figure 7 is a flowchart showing the steps of a detected surface defect analysis program.

The installation shown in FIG. 1 is intended for the detection of a surface defect in a strip 10 during high-speed scrolling, for example a rolled sheet leaving a rolling line.

The surfaces of the sheet 10 are inspected by means of a camera 12 delivering to a filtering stage 14 digital images of the surface of the strip. In the embodiment shown, the installation comprises a single camera 12 pointing at one of the surfaces of the strip, but of course, the installation can be equipped with two cameras adapted to form images of each surface of the strip 10.

The camera 12 can consist of any type of camera suitable for the intended use, the field width of which is substantially equal to the width of the inspection zone of the strip 10, which zone of inspection which may consist of the entire bandwidth. It can thus be constituted either by one or more matrix cameras delivering images of finite length, by considering the direction of travel of the strip, or by a camera or several linear cameras delivering images of infinite length.

In the case where a matrix or linear camera is not sufficient to cover the entire width of the inspection area of the strip, several cameras are used distributed over the width of the strip.

With reference to FIG. 2, the camera 12 forms lines i of M picture elements I _i>) , or pixels, addressable, for a marker of the pixels according to the length of the strip 10, per line n ° i and, according to the width, by column n ° j of picture elements, each picture element being associated with a numerical value representative of a gray level.

The lines of picture elements are stored in a memory 18 of the filtering stage under the control of a management circuit 20.

According to a first example, the camera consists of a linear camera delivering to memory 18 10,000 lines of 2048 pixels per second, these lines being stored in the memory at successive addresses.

According to another example, the camera consists of two matrix cameras distributed over the width of the strip to cover the width of the strip and adapted to take 10 images / s. Each image delivered by a single camera is made up of 1024 lines of 1024 pixels, delivered to memory 18.

Thus, the shooting system continuously delivers lines of picture elements, each element image being associated with a numerical value representing a gray level. It is understood that it is clocked by line if it is a linear camera and clocked by group of lines if it is a matrix camera. Referring again to FIG. 1, it can be seen that the filtering stage 14 further comprises a filtering circuit 21 constituted by an image processing operator ensuring the detection of relative variations in the digital values of the image elements or pixels for the detection of surface irregularities.

Preferably, the filtering circuit is constituted by a contour detection circuit, for example a detector of the "Prewitt" type, detecting variations in gray levels between picture elements located close to each other, which makes it possible to detect areas of the sheet 10 having surface irregularities.

As shown in FIG. 1, the output of the filtering circuit 14 is connected to a signal processing unit 22 comprising a first stage 24 for segmenting the digital images into zones of image elements each delimiting a surface irregularity detected by the filtering stage 14, and a second stage 26 of signal processing, constituted by a calculation circuit 28 associated with a corresponding memory 30, in which are stored processing algorithms for recognizing and identifying fault of surface, for each area with a surface irregularity.

The installation shown in FIG. 1 is further provided with a display device 32 connected at the output of the processing unit 22, an input of which is connected to an output of the calculation circuit 28 and allowing the visualization of surface defects detected, associated with information relating to the type of fault and to parameters representative of the severity of these faults, as will be described in detail below.

The description of the operation of the installation which has just been described will now be made with reference to FIGS. 2 to 7. In FIG. 3a, a part of the sheet 10 is shown which has a set of surface irregularities such as 34.

The field of the camera 12 preferably covers the entire width of the strip 10. With reference to FIG. 4, during a first step 36, the camera takes successive lines d image elements of the surface of the strip 10, these image elements being stored in memory 18, associated with a gray level value.

During this first step 36 of taking pictures, the management circuit 20 performs, if necessary, a fusion of the images delivered by the taking camera 12, by grouping together the successive pixels, on the one hand in the direction of the width of the strip 10 in the case where several cameras are used to cover the entire width of the inspection area, to obtain in the memory 18 an image whose width corresponds to that of the inspected area, and, on the other hand, in the direction of the length of the strip 10, in the case where the camera 12 uses one or more matrix cameras, by merging the groups of lines of pixels successively delivered.

The image, stored in memory 18, hereinafter called "raw image", consists of a set of image elements I _lrj , i designating the address of the line in memory, varying from 1 to N, and j designating the number of an image element of each line and varying from 1 to M, M being equal to for example 2048, each image element being associated with a numerical value of gray level. It should be noted that the value N depends on the capacity of the memory. This capacity must be adapted for storing a sufficient number of lines with regard to the subsequent processing to be carried out. For example, for storing an image corresponding to a length of 15 m of sheet metal with a number of lines of picture elements equal to 1024 / m, N is preferably equal to 15360 lines.

When the memory capacity is full, successive incoming lines are memorized in place of oldest pixel lines previously stored and normally processed.

When the memory 18 is full and said oldest pixel lines have not been processed, a saturation alarm is emitted to indicate that an area of the strip will not be inspected.

In this case, the non-inspected area is identified on the strip, by identification and storage in a file of the non-stored successive lines, for example with a view to a statistical analysis of portions of the strip not inspected.

However, taking into account the average speed of tape travel and the average density of surface irregularities to be identified for a given type of tape, it is possible to determine an average computing power necessary corresponding to an average processing speed beyond which in practice no longer risks erasing untreated lines.

Preferably, the processing modules are therefore sized so that the instantaneous processing speed is greater than this average speed.

Thus, in addition to its role of fusion of images, memory

18 acts as a buffer enabling variations, and in particular increases, in the processing load due to an increase in the density of surface irregularities to be absorbed.

During the following step 38, a binary image representing contour lines of surface irregularities is associated with each image stored in the memory 18. To do this, during this step, the successive lines of raw picture elements are filtered by the filtering circuit 21, constituted as mentioned previously for example by a bidirectional Prewitt filter of conventional type, having the function of detecting variations in the gray levels of the raw image elements reflecting the existence of surface irregularities with a view to determining their contour inscribed in the associated binary image.

In the embodiment described, it is considered that the filter used is a Prewitt filter, but of course, any other type of filter suitable for the intended use can also be used.

The Prewitt filter ensures the location of the surface irregularity contour position by detecting, on each line of a raw image, image elements likely to belong to an irregularity contour line, these elements d 'images being designated thereafter "suspicious pixels".

The filter used here assigns a numerical value "1" to each binary image element associated with each suspect pixel of the raw image delivered by the camera 12, the other pixels of the binary image being kept at 0.

This filtering step 38 thus makes it possible to form in the memory 18 a binary image consisting of a set of binary image elements B _lrJ to each of which is assigned a binary value equal to 1 for a pixel belonging to a contour of an irregularity equal to zero for a pixel not belonging to a contour of a surface irregularity.

During the next step 40, the binary image stored in the memory 18 is processed using a conventional connectivity operator which applies a mask to this image to force the pixels "1" the binary image having a zero value and located between two suspect image elements relatively close, in order to obtain and define continuous lines for each contour detected.

After having undergone this processing, the raw and binary images are cleaned to eliminate the spots delimited by an outline whose surface is less than a determined threshold, for example of 3 × 3 pixels. A binary image is then obtained, superimposed on the raw image delivered by the camera 12, and showing the contours delimiting the surface irregularities detected in the raw image. The binary image and the raw image are then ready for processing.

During the following step 42, the management circuit 20 successively analyzes each line of the stored binary image, for the detection of the binary elements of value "1", that is, suspects. As soon as a suspicious pixel is detected, the management circuit 20 locates the number of the corresponding line, opens a storage area of predetermined capacity in the form of a window in the memory 18 (step 44) from this number line and keep this window open as long as the management circuit detects suspicious pixels in the following lines.

This window, subsequently designated by "suspicious window", thus contains suspicious pixels, that is to say pixels likely to belong to a surface irregularity.

The management circuit 20 closes the suspect window as soon as no suspect pixel is no longer detected in a predetermined number of successive lines of the binary image, by recording the number of the last line in which a suspect pixel has been identified. .

The suspect window thus defined in the memory 18 represents a raw image segment, associated with a corresponding binary image segment, and contains at least one surface irregularity which it is sought to identify and recognize.

In particular, the window, opened during step 44, is kept open as long as the number of the last successive lines of image elements stored in said window not containing a suspect pixel does not exceed a predetermined threshold number of successive bit lines, this threshold being at least equal to 1.

Thus, during the following step 45, the number of successive lines of image elements not containing a suspect pixel is compared with this threshold number and, in the event of a tie, the suspicious window is closed (step 46).

Furthermore, during step 47, the number of lines recorded in the open window is compared with a predetermined threshold known as "detection of very long window" or "detection of a long fault". This predetermined threshold corresponds to the predetermined maximum capacity of the storage areas in the memory 18.

If the number of rows recorded exceeds this threshold, the window is closed (step 48) and it is then decided from the following step 50, that the window is a window known as "suspect of great length", which contains a surface irregularity the number of lines of picture elements of which is greater than the long fault detection threshold. It will also be noted that, in the embodiment described, the suspicious windows are opened successively.

The surface inspection process continues with the phases of dividing the suspicious windows stored in the memory 18 into so-called "suspicious areas", each presenting a surface irregularity, using either the component corresponding to the raw image, or the component corresponding to the binary image.

To do this, the stage 24 performs, during the following steps 58 to 64, a calculation, using appropriate means, for example software means, of the accumulation profile of the digital values or of the binary values, respectively for each raw image or each binary image, on the one hand in the longitudinal direction and, on the other hand, in the width direction, by projection of the numerical values or the binary values according to two perpendicular axes and by thresholding of the profiles of so as to delimit suspect areas each incorporating a surface irregularity.

Although the calculation of this profile can be performed from the digital values associated with the picture elements of the raw image or from the binary values of the image stored after processing, in the following description, it will be considered that image processing is carried out from the binary image. This calculation operation begins with a segmentation phase of each suspicious window into a suspect strip including irregularities, each strip then being segmented into one or more suspect areas.

First, during step 58, the stage 24 calculates, using a calculation circuit 24-a (FIG. 1), the sum of the binary values of each line of the suspect window for obtain, on M columns, a first transverse profile, in the direction of the width of the strip. This gives the curve shown in Figure 3b. During the following step 60, this profile is presented at the input of a framing circuit 24-b, so as to be framed so as not to separate image elements of an irregularity situated close to each other . The framing circuit 24-b can be constituted by any suitable type of filter, such as a RIF filter with finite impulse response, or RII with infinite impulse response, but is preferably constituted by a filter of the sliding window type making it possible to deliver a framed profile r (x) whose values are determined according to the following relation:

K r (x) = Σ F (x-i) x Q (i) (1) i = -K

in which K denotes the width of the sliding window,

F (x-i) designates the value of the column (x-i) of the profile to be framed,

Q denotes the coefficient of the sliding window filter, chosen for example equal to 1, and x denotes the column number of the framed profile. The profile thus framed is then thresholded using a threshold circuit 24-c, during the following step 62, by comparison with a threshold value for detecting irregularities.

The cropped and thresholded profile represented in FIG. 3c is thus obtained delimiting suspect bands, represented with the help of dotted lines in FIG. 3a, each including one or more surface irregularities. As mentioned previously, the suspect bands thus defined are then segmented into suspect zones, each having a surface irregularity.

To do this, during the following step 64, steps 58, 60 and 62 are carried out again and applied independently to each line of image elements of each suspect band, so as to obtain an accumulation profile binary values in the longitudinal direction, as shown in Figure 3d. This longitudinal profile is then framed and thresholded, as previously, to obtain the image represented in FIG. 3e in which suspect zones are defined, such as 66, which each delimit a detected surface irregularity, each irregularity possibly, of course, have more than one object or segment of irregularity.

Each suspect zone thus defined therefore contains a raw image segment and the corresponding binary image segment. Preferably, the suspect areas 66 thus delimited are further presented at the input of a second calculation circuit 24-d, connected at the output of the thresholding circuit 24-c, by means of which the irregularities of small dimensions are eliminated. To do this, during the following step 68, each suspect area of the binary image is treated independently using a conventional labeling algorithm in order to delimit objects constituting a surface irregularity, each object being defined by a set of suspicious picture elements in contact with each other.

The area of each object is then calculated, as well as the average area of objects belonging to the same suspect area.

Small objects are eliminated from the processing. To do this, we decide to eliminate objects whose individual area is less than a predetermined percentage of the average area calculated.

One thus obtains, at the output of the calculation circuit 24-d, suspect zones each containing an irregularity, from which the small objects have been eliminated.

These suspect areas thus cleaned are then stored in the memory 30 of the calculation circuit 28 for processing, as will be described in detail below with reference to FIG. 5. It should be noted that the calculation circuits 24 -a, framing 24-b, and thresholding 24-c and calculation 24-d are conventional type circuits. They will therefore not be described in detail below. In the case where a suspicious window has been qualified as a suspicious window of great length during the preceding step 50, the image processing step is preceded by a phase of elimination of the processing of certain suspicious areas, which reduces the load on the calculation circuit 28.

To this end, as soon as a suspicious window of great length is detected (steps 47, 48 and 50) and that it is cut into suspect zones as previously described, at least one zone is identified during the next step 70 suspect of this window whose bottom line of picture elements belongs to that of said window. This suspect area thus identified is then qualified as "suspect area cut at the bottom".

The suspicious window following a very long suspicious window is called a "suspicious extension window".

It is understood that a suspicious extension window can also be very long.

After cutting, as previously described, of a window suspicious of extension into suspect zones, at least one suspect zone of this window is identified, the upper line of image elements of which belongs to that of said window, this suspect zone being then qualified as "suspect area cut at the top" or "suspect area extended" (step 71).

We match the suspect areas "cut at the bottom" of the very long window and those, "cut at the top", of the suspicious window of extension (step 72).

During the following step 73, it is determined whether the window suspecting for extension is itself of great length. If this is the case, at least one suspicious area of this window is identified, the lower line of picture elements of which belongs to that of said window, this suspect area then being described as previously "suspect area cut at the bottom" and the same processing of recomposition of this suspect zone is carried out with the suspicious zones "cut at the top" of the following window, called the extension window (step 74). As the matching or association of the suspect areas cut from one window to the next progresses, the length of each defect is updated. During the following step 75, the processing unit 22 compares the length of each fault with the length of a window of great length, that is to say with the long fault detection threshold previously mentioned.

As soon as this length exceeds that of a very long window, the fault is qualified as being a long fault (step 76) and a "long fault group" is opened, defined by an area of the memory of the processing stage in which places all the successive cut and associated suspect zones which in fact constitute a single defect called "long defect". Then remove from the image processing all the suspect areas of extension which belong to "long fault" groups; thus, in each "long fault" group, image processing is only carried out on the first suspect area ("cut down") and, to simplify processing, the result of this processing is assigned to all areas suspects of extension of the same group "long fault".

As the suspect zones cut from one window are paired or associated, updating the length of each defect associated with suspect zones which correspond from one window to the next. , it can be seen during step 75 that this fault is not a long fault.

The segmentation of such a fault cannot take place over more than two successive windows, otherwise it would be qualified as a long fault.

In this case, a storage area is opened in the memory 30 in the form of a suspect area called "recomposition" in which the two suspect areas cut from the same defect, suitably butted and centered, are placed, the size of said window being adapted to frame said defect as in the case of uncut suspect areas (step 77). The suspect redial zones are then treated like all other suspect zones.

The segmentation phase of the raw and binary images into suspect areas to be treated now being completed, we then proceed to the treatment of each suspect area delimited in steps 58 to 68, with the exception of the suspect areas of group extension "long fault" .

The processing of each suspect zone will now be described with reference to FIGS. 5 and 7. This processing begins with a step 78 of calculating fault identification parameters, generally qualified as a parameter extraction step.

In a manner known per se, the nature of the parameters likely to characterize the defects or irregularities of the surface of the strip to be inspected, and necessary to recognize and identify them in a precise and reliable manner, is determined.

The mode of calculation of these parameters is also determined, in particular as a function of values of picture elements of the raw or binary image of a suspect area containing said defect or said surface irregularity.

Conventionally, among these parameters, we generally find the length, width and surface of a surface irregularity in a suspect area, the average intensity of the gray levels of the elements of the raw image inside. of the defect, the standard deviation of these gray levels ...

The number of parameters necessary for precise and reliable recognition, denoted hereafter P, can be very high and reach for example 65.

The nature and mode of calculation of the defect parameters now being defined for a type of strip to be inspected, the P parameters are calculated for each suspect area. Each suspect area or irregularity can thus be represented by a point in a P-dimensional space.

This high number P of parameters is a handicap, having regard to the time and means of processing for recognizing suspect areas. In order to avoid, at least to limit, this handicap, a roughing step 80 is carried out which makes it possible to considerably simplify the treatment of each suspect area by classifying the irregularities according to a set of roughing classes. This roughing step, which constitutes a prior classification of irregularities, according to a set of predefined classes, makes it possible to divide the general problem of the analysis of irregularities into a set of problems which are simpler to deal with. In particular, within each roughing class, a set of elementary classes or families of defects is defined, the number of which is limited.

In order to be able to carry out the roughing step, it is necessary to have provided for a preliminary phase of definition of the roughing classes and, possibly, of their associated simplified benchmark, generally before the implementation of the method according to the invention.

This preliminary phase is specific to a type of strip to be inspected. As an example of a preliminary phase leading to the definition of roughing classes, we proceed by learning as follows.

A surface inspection is carried out, as previously described up to this stage of the process, of a sufficient number of samples of the same type of strip to arrive at a sufficiently large and representative population of suspect areas, of which each irregularity is represented. by a point in the P-dimensional space previously mentioned.

According to the known method, moreover, of factorial correspondence analysis, we can see how these points are grouped in clouds in this space.

It is then considered that each region of the space delimiting a cloud makes it possible to define a typology of defects, and the defects of the same cloud therefore have elements in common and will be able to be possibly represented in a simplified reference proper to this cloud or to this typology.

To define the axes of a simplified coordinate system specific to a typology or to a given cloud, we can use the main axes of inertia of this cloud, whose positions and directions can be calculated in a manner known per se.

Thus, all the faults of the same class can be represented in the same simplified reference in a space whose dimension is less than P, that is to say that all the faults of the same class can be characterized by a reduced number of parameters, less than P.

Using conventional mathematical methods, we establish reference change matrices which allow us to move from a representation of defects in a P-dimensional space to a representation of the same defect in a simplified reference frame of reduced dimensions.

Thus, in this preliminary phase intended to prepare the roughing, we defined typologies or "roughing classes" of defects and a simplified benchmark of defect representation, specific to each roughing class.

According to a specific example, these roughing classes can be defined from the length (L) or the width (1) of the irregularities; with reference to FIGS. 6a and 6b, for example 5 and 6 roughing classes are defined, respectively for a "DKP" sheet and for a galvanized sheet, namely a class of small defects (pt), a class of fine and short defects (fc), a class of fine and long faults (f1), a class of medium and short faults (me), a class of medium and long faults (ml) and a class of wide faults (la); each class being associated with a simplified representation benchmark.

After the step of extracting the parameters, it is now possible to carry out the step 80 of prior classification or roughing-out, properly speaking.

To do this, each defect or irregularity in the surface of the suspect area is distributed among the different roughing classes previously defined, as a function of the value of the P parameters of a defect and of the characteristics which define these classes.

This prior distribution of defects in roughing classes makes it possible to considerably simplify the recognition of faults, by carrying out this recognition on each roughing class.

As a variant, all the defects of the same class are represented in the simplified coordinate system associated with this class, using the matrix for changing the coordinate system of this class, applied to the P parameters. This then leads to a simplified characterization of all the defects, by a reduced number of parameters, which limits the quantity of calculations to be performed during recognition. The subsequent step 82 of the processing consists in recognizing and identifying the defects of each roughing class.

The identification and recognition processing is specific to each roughing class and is generally defined beforehand according to the types of defect that are likely to be encountered in each class.

This identification and recognition processing can consist of a classification based for example on the so-called "Coulomb spheres" method. Other known methods can also be used, such as the discriminant analysis method, the decision tree method or the method which involves determining the nearest neighbor "K".

According to the Coulomb spheres method, the fault typologies, specific to a given roughing class, are represented by spheres, identifiable, in position and size, in the simplified space associated with this class.

Each sphere corresponds to a type of fault and / or a fault identification name.

Thus, in order to recognize and identify a defect in a given roughing class, in step 83, one identifies which sphere the defect belongs to and assigns the identification name associated with this sphere (step 84). Advantageously, this recognition and identification operation can be carried out very quickly because, the number of spheres and the number of parameters being reduced due to the previous roughing step, the classification calculations can be performed on a reduced number of criteria.

In the particular case where, within a given roughing class, we would encounter a defect not belonging to any sphere, we assign it the name of identification of the nearest sphere.

Thus, at the end of step 84 of assigning a fault identification name to each irregularity, all the irregularities are identified as corresponding to a particular type of defect.

The next step 86 consists in carrying out a second classification using a second classification stage of the calculation circuit 28, from a reduced number of classes, in order, for example, to confirm the result provided by the first classification stage and to remove certain uncertainties which could have appeared in the identification of certain faults, or in order, for example, to differentiate in more narrow typology faults of the same type that one would have decided not to differentiate at the level of the first classification stage, lack of sufficient classification performance at this level.

In order to be able to implement this second classification step 86, it is necessary to have provided for a prior qualification phase for each elementary class. In this preliminary phase, statistical processing of validation or non-validation of the classification carried out for the identification of faults is carried out, using the method which has just been described, so as to identify the elementary classes which contain the most d 'fault classification errors.

These elementary classes, in reduced number, which contain the greatest number of classification errors, are qualified as "elementary classes of uncertain identification", the others, which contain the least classification errors, are qualified as "classes elementary of certain identification ".

The second classification, implemented in step 86, is only carried out on faults or irregularities classified in elementary classes of uncertain identification.

The second classification stage uses for example one of the classification methods mentioned above.

It is suitable, for example, to validate or not the membership of faults in these classes of uncertain identification. In the event of non-validation, the defect is then considered as not being a defect and is eliminated from the processing.

It can also be adapted to distribute the defects of certain elementary classes of uncertain identification into precise identification classes, predefined according to a more narrow typology. It should be noted that this additional classification relates to a very small number of classes of defects and can therefore be carried out very quickly.

At the end of these steps 80 to 86, each defect is identified and recognized, that is to say assigned to an elementary class.

The image processing phase ends with a step 88 of merging the data during which certain defects are grouped, using previously defined criteria, relating in particular to the geometry and topology of the defects (for example: distance of the faults between them, identical position above and below the strip, proximity to the edge of the strip, ...).

This merger phase makes it possible to remedy certain imperfections likely to appear during the recognition of faults and to resolve some specific problems of confusion, without calling into question the results already confirmed.

The decision to group faults is made after confrontation of information coming from the close vicinity of an object to be recognized, of the order of a meter for example, of other cameras (for example pointing towards the other face of the strip), or data relating to the processing of the strip (nature of the strip, breakpoint, ...). In particular, it is decided to group together faults for which an ambiguity on the name remains, as well as faults of the same nature.

Furthermore, faults with particular proximity relationships are grouped together, for example, faults located nearby, on the same face of the strip or on an opposite face, as well as faults located in the same longitudinal or transverse alignment.

Thus, for example, in the case of a galvanized sheet, a defect of the "granular drag" type results in a multitude of surface irregularities located in the vicinity of the edge of the sheet. In this case, the identification of the fault is not completely reliable. Indeed, each of these irregularities can be recognized as belonging to a "grainy streak", or be recognized individually as a defect of another type, in particular an "exfoliation", or a "blistering".

In this particular case, the irregularities located in the vicinity of the edge of the sheet metal and aligned with respect to each other are merged and identified as belonging to a "grainy trail" type defect.

Similarly, according to another example, during this melting step, the defects located in the same position are grouped together on the upper and lower faces of the sheet and they are given an identical name.

During this merging step and as described above, with reference to step 76 in FIG. 4, we also group together long faults, cut during the opening of suspicious windows, by assigning, as mentioned previously, the name from the defect of the suspect area of great length to the faults of the suspect areas of extension of the same group.

During this merging step, the population of each elementary defect class over a given length of strip is also analyzed, that is to say the number of defects per unit of length having the same identification.

This population is then compared to a predetermined threshold, called the presumption of default threshold. periodic. This threshold is determined for the same given strip length.

When the population of an elementary class exceeds said threshold, it is considered that defects of this class are probably of a periodic nature.

To validate this characteristic, a conventional method of detecting periodic faults can be used.

For example, we plot the histogram of the distance between each defect in this class and, if this histogram highlights a periodicity (fundamental or harmonic), we open in memory a specific group "periodic defect" and we group in this same group the periodic faults of this class.

According to a variant, this step of detecting and grouping periodic faults can be carried out after the extraction of the parameters but before identification and recognition, or even before roughing or prior classification.

This variant then supposes a specific, relatively summary classification processing since it must be based on the characterization of the defects according to a high number P of parameters and, for the detection of periodic defects, the population of the elementary classes defined in this is then analyzed. specific classification. This variant has the advantage of displaying a result which does not depend on the performance of the recognition modules (roughing and downstream classification).

After having detected, recognized and possibly grouped together the faults corresponding to detected irregularities, the subsequent phase of the inspection process consists in analyzing the faults with a view to determining their severity, to allow the determination of the defectiveness of the strip. This phase will now be described with reference to FIG. 7. Beforehand, before the implementation of the method, for each class or each type of defect, as a function of different possible intrinsic harmfulnesses of the type of defect, a set of sub -classes, each subclass being associated with a possible intrinsic harmfulness of the type of defect, for example a harmfulness evaluated from photometric, geometric and topographical parameters predefined for each type of defect. Each subclass may possibly be assigned an intrinsic gravity coefficient.

It is understood that each surface irregularity is at this stage, identified and therefore characterized by characteristic parameters, in particular by a reduced number of parameters. During the first step 90 of this phase of analysis of the defects, the defects, grouped in a fusion group in the preceding step, are assimilated to a single defect called "fusion defect". To this end, for these grouped defects, the parameters characterizing the fusion defect are calculated by linear combination of the values of the parameters characterizing each defect or irregularity of the fusion group.

From the values of the parameters characterizing the non-grouped faults and the fusion faults, an additional classification of these faults is carried out in the following step 92 according to the set of subclasses specific to each type of fault.

This additional classification can be carried out according to the same type of methods as those used during the recognition of faults.

This additional classification results in an independent result from subsequent uses of the sheet.

At the end of this additional classification, a band "intrinsic defect profile" can be defined by a list giving the population of each "severity" subclass of each type or "elementary class" of defect, this population being reported to a unit of tape length; this profile can for example be represented in the form of histograms of the population of each subclass, arranged side by side in a predetermined order (subclasses after subclasses, classes after classes).

In parallel, for a given use of the band, one can define according to the same formalism (for example: histograms in the same order) a "profile of allowable defect ", that is, for each" severity "subclass of each possible defect type, a maximum allowable population for that given use (always referred to the same unit of tape length). This" allowable defect profile " is not defined "once and for all" for a given use; it may even vary depending, for example, on the evolution of the specifications for this use.

Next, in step 94, the intrinsic defect profile of the strip inspected is compared with the admissible defect profile of the desired use of said strip.

Thus, during step 94, if it is found that the intrinsic defect profile of the inspected strip fits (or is contained) into the admissible defect profile of the intended use of this strip, this strip is considered to be acceptable or validated for this use (step 96).

If this is not the case, this inspected strip is considered to be unacceptable or "defective" with regard to this use (step 98).

In order to avoid discarding this inspected strip, the use is then sought in the admissible defect profile of which the intrinsic defect profile of this inspected strip fits (or is contained), and this strip is assigned to this other use.

It is known in fact that a sheet having a predetermined number of defects of a given severity and of a particular type may not be defective for one use, but may be defective for another use.

For example, a sheet having a scratch is defective if it is not laminated during a subsequent processing step but is considered not to be defective if it is re-rolled, the scratches being therefore crushed.

The decisive advantage of this method of evaluating the defectiveness of a strip by measuring an intrinsic defect profile is that this measurement is independent of the subsequent and downstream use of the strip, and of the evolution concerning the criteria to be satisfied for this use.

Another advantage of this method for evaluating the defectiveness of a strip is that it makes it possible to select, from a strip, strip segments having a predetermined profile of permissible defect.

To this end, in a strip whose intrinsic defect profile has been established as previously described, segments are cut which correspond to the admissible defect profile; this selection forms a "batch" which can be used for the use attached to this admissible defect profile.

These strip segments can be cut lengthwise or crosswise ("slitting"). This use of the method can be compared to a tape "repair" method.

Advantageously, the intrinsic defect profiles of the bands inspected can be used, on the contrary, to follow the development and any drifts of the methods of manufacturing these bands, according for example to the manufacturing campaigns; one can thus for example identify possible drifts in the behavior of the upstream rolling chain.

The intrinsic fault profiles of the inspected tapes can also be used to identify deviations on the inspection system itself.

According to a simplified variant of the defect analysis method, a coefficient can be assigned to each "severity" subclass of the types of defects, the value of which is a function of the estimated severity for a given use, and define the profile failure of a band by the sum of the populations of each subclass multiplied by the corresponding coefficient. To validate this use, it is then simply checked that the result obtained, namely said sum, does not exceed a predetermined value defined for this use.

Other simplified variants, based on the use of coefficients, can be envisaged.

Claims

1. Method for inspecting the surface of a moving strip (10), for detecting a surface defect, comprising the steps consisting in: - forming, with the aid of image-taking means (12), at least one digital image of at least one of the faces of the strip (10), consisting of a set of lines of image elements (I _irj , B _{1 # J} ) each assigned a digital value,

- filtering said at least one digital image for the detection of surface irregularities (34), by detection of relative variations of said digital values,

- processing said at least one filtered digital image for identifying the type of surface defect corresponding to each irregularity (34) detected, and - analyzing said identified defects for determining the defectiveness of the strip (10), characterized in that that during said step of defect analysis, said identified defects of each type are classified according to a set of sub-classes of surface defects each corresponding to a predefined intrinsic harmfulness for each type of defect and one validates a use of said inspected strip when the population of each defect subclass is less than a predetermined maximum admissible value for said subclass and for said use.

2. Method according to claim 1, characterized in that, prior to the step of processing said at least one digital image, a general characterization of the irregularities is carried out by determining for each of them the value of predetermined parameters characteristic of defects surface and a prior classification of said irregularities is carried out, on the basis of determined values of said parameters, according to a set of predefined classes, said processing step being carried out on each class.

3. Method according to claim 2, characterized in that a second mode of characterization of the irregularities is determined for each predefined class whose number of characteristic parameters is less than the number of characteristic parameters of general characterization and, after the prior classification step, the value of the characteristic parameters of the second specific characterization mode of said class to which the irregularity belongs is determined for each irregularity detected, from the values of the characteristic parameters of general characterization.

4. Method according to any one of claims 1 to 3, characterized in that during said image processing step for the identification of the type of defect corresponding to each detected irregularity, in response to a detection of an image element of an irregularity, a memory area (18) delimits a storage area for lines of image elements (I _{i # j} , B _lrj ) successively delivered by the image-taking means (12 ) and comprising at least one image element corresponding to at least one irregularity (34), each storage area is segmented into suspect areas (66) each having at least one surface irregularity (34), suspect areas are matched ( 66) of successive storage areas and corresponding to the same irregularity (34) and the total number of lines of image elements of the suspect areas (66) matched with a threshold for detecting a fault of great length is compared, and if the Said threshold, said step of processing said at least one digital image filtered only on one of the matched suspect areas is carried out, the result of the processing and analysis of the identified defect being assigned to the other matched suspect areas (66).

5. Method according to any one of claims 1 to 4, characterized in that the processing step further comprises the steps of counting the number of identified defects of the same type per unit of length, and of comparing said number of defects of each type with a predetermined threshold value representative of the minimum number of faults from which said faults are likely to be periodic, with a view to detecting periodic faults.

6. Method according to one of claims 1 to 4, characterized in that during said image processing step for the identification of the corresponding type of defect at each detected irregularity, a specific classification of the irregularities is carried out according to a set of elementary classes, and the population of said elementary classes is analyzed with a view to detecting periodic faults.

7. Method according to any one of claims 1 to 6, characterized in that said image processing step for identifying the type of defect corresponding to each irregularity comprises a step of determining, for each detected irregularity, the value of characteristic parameters of surface defects, a step of grouping identified defects using a set of predefined criteria, in particular geometric and / or topographic criteria, and a step of combining the characteristic parameters of the combined surface defects, said step of analysis of said faults being carried out from the value of said combined parameters for the grouped faults.

8. Installation of surface inspection of a strip

(10) scrolling for the implementation of a method according to any one of claims there 7, characterized in that it comprises means for taking pictures (12) of at least one of the faces of the strip , a filtering circuit (21) of at least one image delivered by the image-taking means (12) for the detection of surface irregularities and a processing unit (22) of the signals coming from the filtering circuit (21 ) suitable for identifying the type of surface defect corresponding to each irregularity and analyzing the identified defects for determining the defectiveness of the strip by classifying said defects according to a set of subclasses each corresponding to a predefined intrinsic harmfulness for each type of fault.