Such a control assumes particular relevance in industrial making processes of elongated metal products, where guaranteeing the desired surface quality requires continuous measuring.

Exemplary suchlike industrial processes are those concerning the continuous making or working of pipes, bars, billets.

Hereinafter, for simplicity's sake reference will likewise be made to the specific instance of the working of billets, i. e. of unfinished steel products of elongated shape from which, by subsequent rolling, bars, rods and sections are obtained.

Manual inspection techniques are commonly employed, based on the observation performed by an inspecting technician who visually and constantly observes the uncooled product in search of thermal inhomogeneities revealing the presence of defects and/or cracks.

The inspection can take place, e. g. , among the stands of a rolling mill, while the product flows even at rather high speeds (-500mm/s) and the defects under inspection are rather reduced in size (<lmm width).

Moreover, such an inspection generally occurs only on one side of the product; therefore, some defects may be kept out of sight.

Aim of the automated inspection is to integrate the operator's judgment by attaining a system capable of providing answers altogether more reliable, objective and reproducible over time.

The technical problem underlying the present invention is to provide a method and an apparatus for control of surface quality of elongated products with the overcoming of the drawbacks mentioned with reference to the known art.

Hence, for this purpose it is provided a method for identification and classification of surface defects of an elongated product during its working, comprising the following steps: - capturing a plurality of images of said product ; and - processing said images, wherein said step of processing said images is apt to provide indications useful for recognition of said defects on the basis of variations of shape and/or of luminous

intensity.

The present invention further provides an apparatus for identification and classification of surface defects of an elongated product during its working, comprising: - one or more assemblies for taking images of said product; and - means for acquiring and processing said taken images, wherein said processing means is apt to provide indications on the presence of any defects on the surface of said product during a step of working the same.

The main advantage of the present invention lies in that the inspection relies on measuring the radiation generated directly by the hot product, using high-resolution CCD (Charged Coupled Device) cameras, and therefore for defect perception it utilizes a technique fully reproducing that successfully used by humans.

Other advantages, as well as the features and the modes of employ of the present invention will be made apparent in the following detailed description of a preferred embodiment thereof, given by way of a non-limiting example, making reference to the figures of the annexed drawings, wherein: Figure 1 is a schematic top view of a rolling mill plant comprising an apparatus according to the present invention; - Figures 2A and 2B are side views of the two image taking assemblies of Figure 1; - Figure 3 is a schematic view of a case of one of the cameras according to the present invention; and - Figure 4 is a flow chart illustrating the main steps of a method according to the present invention.

To describe the present invention, hereinafter reference will be made to the aboveindicated figures.

An apparatus according to the present invention comprises one or more image taking assemblies based on linear cameras capturing a plurality of images of the product and a system for acquiring and processing captured images for detection of defects.

Each image taking assembly utilizes linear CCD cameras, i. e. capable of capturing images along a line transversal with respect to the feed direction of the elongated product, like, e. g. , a billet.

The two-dimensional image of the product is obtained by exploiting the motion of the product, which, by crossing the scanning line, exposes its subsequent cross- sections to sight, thereby enabling to reconstruct a complete image. This technique is utilized as it enables to standardize taking conditions as much as possible along the

entire length of the product.

The image processing software recognizes and classifies defects according to <BR> <BR> predetermined typologies of interest (e. g. , cracks), recognizing them on the basis of the analysis of their contrast with respect to the remainder of the image, as well as of their characteristic shape.

The image acquiring and processing system is implemented on a computer, e. g. a Personal Computer (PC), equipped with acquiring and processing means (in particular, a dedicated software) suitable for processing images, locating defects, computing a graphic map of defect location on each billet, signaling alarms to an operator, printing statistical reports summarizing the working.

With reference to Figure 1, it shows a rolling mill plant 1 for working elongated metal products 2 designated as billets.

According to the present invention, such a plant is equipped with an inspection apparatus comprising two image taking assemblies 3,4, each equipped, as it will be detailed hereinafter, with two cameras.

A first image taking assembly 3, equipped with two cameras 31,32, is arranged downstream of one of the rolling stands, in particular the rolling stand L5 (V). The image taking assembly 3 is preset for observation of the just-rolled lateral sides of the billet 2.

The arrangement of the cameras 31,32 of the image taking assembly 3 is evident in the next Figure 2A.

Downstream of a second rolling stand L4 (0), it is advantageously provided a second image taking assembly 4.

In this instance, the two cameras 41,42 of the image taking assembly 4 are arranged in a manner such as to observe the top and bottom sides of the billet 2 under working. The arrangement of the cameras 41,42 is evident in Figure 2B.

Advantageously, each of the cameras 31,32, 41,42, along with the relevant lenses, is housed in a corresponding case C1, C2, C3, C4, constituting a protection for the camera.

Each case ends in a flat spout, leaving a thin front opening through which the linear camera may sight the billet. An exemplary and schematic view of such a case is reported in Figure 3.

The spout portion nears the surface of the billet, in a position transversal with respect to the direction of motion thereof.

Advantageously, the spout end portion is hinged to the case body in a manner such as to be horizontally pivoted, so as to automatically get out of the way when accidentally hit by the piece under inspection.

Air under pressure is inlet, via ducts 10, 11, inside the case, exiting therefrom via the front opening.

The air, by exiting the spout at high speed, attains the triple purpose of : 1) cooling the camera; 2) preventing water splashes from entering from the beak toward the camera optics; and 3) removing steam present in front of the billet in the taking zone.

Moreover, additional jets of compressed air may be utilized to create a circulation/elimination of the steam about the billet, and in order to prevent water drops from propagating onto the top surface thereof, sliding until reaching the inspection zone.

The compressed air for camera cooling and cleaning is generated by two service centrifugal fans 20,21, one for each pair of cameras and provided with suitable filters.

Advantageously, some or all of the image taking assemblies, like, e. g. , the image taking assembly 4 of the bottom surface, may be differentiated from the others by creating a deflection of the image taking path by means of a mirror inside the case C4 of the camera 42, in order to reduce the dimensions of the system and enabling a simplified assembling thereof.

Advantageously, as it is evident in Figures 2A and 2B, each taking assembly (horizontal for taking lateral and vertical sides and vertical for bottom and top surfaces) is anchored to a support connected to a wheel-mounted extractable frame, in order to allow connecting and disconnecting the instruments over short times.

The apparatus according to the present invention further comprises a system for acquiring and processing images provided by the cameras of the image taking assemblies.

In the specific embodiment, such an acquiring and processing system is a PC, equipped with means for acquiring images in analog and/or digital format and for their storing on filing media.

The images acquired are processed by a software obtaining information useful for identification of any defects on the end product.

Next, figure 4 is a flow chart illustrating the main steps of a method according to the present invention. According to the present invention, the method is implemented by means of software carrying out the individual steps of the method according to the present invention, as described hereinafter.

The image of each of the four cameras is processed separately from the other ones according to a procedure like that indicated in Figure 4.

The image yielded by a camera appears as a continuous sequence of lines, each corresponding to a subsequent scanning of the product.

For each line, several operations are executed that may advantageously be grouped into some functional sections.

With reference to Figure 4, it will be given a description of each section and of the corresponding operative modules of which such sections are made.

It is understood that each of the following sections and each of the following modules may comprise one or more of the steps of the inspection method according to the present invention, as it is claimed.

SECTION A-Image preprocessing Object of this section is to preprocess the camera image with the intention of making it homogeneous and utilizable by a subsequent section (SECTION C) that will perform a search for defects.

Section A processes the camera-acquired image. The processing sequence may be subdivided into the following operative modules: Module A. 1 For each of the lines of the image taken by the camera, there are singled out the product edges on the line, in order to separate, in the framed area, the region actually occupied by the product.

Module A. 2 By means of an interpolation of the product-occupied area singled out in the preceding passage, image sizes are modified (enlarged or reduced) to be brought to constant and predetermined values.

The image thus generated is inputted to the subsequent SECTION B.

SECTION B-Singling out of an average calibration profile Aim of SECTION B is to yield an average calibration profile to be used for a subsequent image normalization step.

In fact, the billet exhibits different intensities corresponding to different physical situations, e. g.: high intensity for usually uncovered surface, low intensity for regions covered with ferrous oxide, very high intensity for defective regions.

For each position of the average calibration profile, the software implementing this section attempts to single out the intensity of the uncovered surface, which therefore is taken as reference of normal condition. This will enable, downstream of a subsequent normalization, to highlight the defective regions, as exhibiting intensities greater than the normal condition.

The singling out of the intensity of the uncovered surface is performed hypothesizing that the uncovered portions be the most intense ones, with the

exception of the defective regions that anyhow occupy merely a limited percentage of the surface.

By performing a search for peaks, with the exception of a limited percentage of higher values, the desired value can be singled out.

SECTION B operates on an input represented by the image generated as described with regard to Module A. 2.

Hereinafter, the operative modules of Section B are described: Module B. 1 Image lines are grouped into packets, e. g. of 256 lines.

Module B. 2 For each line packet, distribution histograms of the intensities of each column are computed.

Module B. 3 For each column it is singled out the intensity splitting the histogram into two regions defined by the percentage of values belonging to each region, e. g. , 95% in the low region and 5% in the high region. This operation corresponds to the search for peaks with the exception of a limited percentage of higher values.

Module B. 4 The set of values obtained at the preceding passage for each column forms a vector, onto which it is performed a moving average of definite width, in order to standardize the profile.

The resulting output is an average calibration profile obtained following the processing of the line packet.

This profile is utilized in the subsequent image normalization step.

Module A. 3-Image normalization.

Such a normalization is necessary, as the various billet regions (image center and lateral regions) exhibit different intensity conditions, due to the bending and to the different working conditions of the product itself. Normalization is aimed at bringing the average intensity level to a constant value over the whole area of the product.

This normalization is carried out by dividing, pixelwise, the product image provided by Module A. 2 by the average normalization profile singled out in SECTION B.

The normalized image obtained at the end of the hereto-described routine is inputted to a subsequent SECTION C.

SECTION C-Singling out defects This section performs the search for potentially defective areas. These areas can occur according to two main typologies.

The first typology, called"open crack", stands out by occurring as a region exhibiting a higher intensity with respect to the average value of the image, due to the higher temperature of the material inside the crack.

The second typology, called"folded crack", stands out by occurring as formed by a slightly more intense edge immediately adjacent to a dark area; the dark area extends in the opposite direction, slowly degrading to normal intensity (edge folded and cooled).

SECTION C implements distinct algorithms for singling out both defect typologies, hence it actually splits into two parallel branches.

Always having as input the normalized image, a first processing routine is represented by modules C. 1, C. 2 and C. 3 described hereinafter: Module C. 1 Starting from the inputted image, there are singled out pixels exhibiting intensity greater than a predefined threshold; Module C. 2 The pixels singled out in the preceding passage and distant thereamong of less than a predefined distance are considered as belonging to a single defect called"open crack" ; Module C. 3 The image portion corresponding to the region comprising the"open crack"thus singled out is sent to a subsequent SECTION D.

A second processing routine, operating in parallel to the first one, is represented by modules C. 4, C. 5 and C. 6 described below: Module C. 4 Starting from the inputted image (normalized image) there are singled out the regions exhibiting rapid variations of intensity (low/high) and in which, progressing toward the dark region, the intensity slowly degrades to the normal value; Module C. 5 The pixels singled out in the preceding passage and distant thereamong of less than a predefined distance are considered as belonging to a single defect called "folded crack" ; Module C. 6 The image portion corresponding to the region comprising the"folded crack"thus singled out is sent to a subsequent SECTION D.

SECTION D-Defect classification This section performs a detailed analysis of the potentially defective image portions generated and provided by SECTION C, in order to confirm the existence of

the defect and to reduce the number of possible erroneous signaling.

The processing sequence is subdivided into the modules described hereinafter: Module D. 1 For each received image portion there are computed parameters, called"features" identifying geometric and luminous intensity distribution features.

By way of a non-limiting example, said parameters are: luminous intensity average, luminous intensity moments, perimeter and area of the light/dark regions of the image, barycenter positions of said regions.

Module D. 2 For each of the potential defects signaled, it is checked whether the values of the corresponding"features"fall within predefined acceptance intervals. In the affirmative case, the potential defect is sent to a subsequent SECTION E, otherwise it is considered as erroneous signaling and therefore ignored.

SECTION E-Operator confirmation This section enables a skilled operator to review and manually confirm the defect signaling produced at the end of SECTION D.

Confirmation by a skilled operator ensures the total certification of the defects observed by the system.

Manual certification is viable and advantageous since the system produces a very limited number of signaling, relieving the operator from the task of continually observing the product.

SECTION E operates on an input constituted by the images and the information relevant to the defects singled out in the preceding section (hereinafter, we will call "defect"together the image and the corresponding information) and the processing sequence is as follows: Module E. 1 For each defect obtained at the end of SECTION D the corresponding image portion is displayed on video.

Module E. 2 The operator may confirm the defect, or signal that it is an erroneous signaling.

Lastly, all images of confirmed defects and relevant numerical and positional information are saved on a central database of the system.

The technique described hereto enables to appreciate thermal inhomogeneities sensibly revealing the presence of cracks.

The takes yield images quite similar to those observable by a human operator, with light or dark areas depending on the surface temperature. Hence, for the possibility of detecting the defect to subsist, it is necessary that:

the defect, internal or surfacing, generate thermal alterations optically observable at the outside of the product; and that the defect sizes be appreciable by the image taking system, which discerns regions of a size proximal to or greater than the taking resolution.

Therefore, the applicability of the system is verified in all those cases in which the defect searched for induces perceptible thermal variations on the surface, i. e. , when the seriousness of the defect is directly recognizable from the outward appearance of the product. In these instances it is possible to generate a reliable defectiveness response.

The present invention has hereto been described according to a preferred embodiment thereof, given by way of example and not for limitative purposes.

It is understood that other embodiments may be envisaged, all to be construed as falling within the protective scope thereof, as defined by the appended claims.