Method and arrangement for the detection of superficial faults in a moving sheet.

The present invention relates to the detection of surface defects on a strip during running, and in particular to a method of the type according to which analog signals representing images of successive parts of the surface of the strip are formed and these are processed. signals to detect abnormal level variations.

The field of application of the invention is more particularly that of the detection of surface defects on a laminated sheet traveling at high speed.

In the context of this application, a method of the type defined above is described in French patent No 77 27 501. The surface of the sheet is lit and images are formed by means of a television camera. reflected from successive parts of the surface of the sheet which constitute a continuous representation of the surface examined. Each image is analyzed by comparison with a maximum luminance level and a minimum luminance level. A recording is made of the number of defects detected on each image and of the percentage of the total surface of the defects relative to the surface of the strip portion corresponding to an image.

It is known that the examination of rolled sheets, and in particular hot rolled sheets at the outlet of a pickling line is difficult. Indeed, on the one hand, there are a large number of possible defects, more or less extensive, visible or serious and, on the other hand, the surface of a sheet is rough and presents stains resulting from a pickling not regular. This is why, with the known method described above, it is practically impossible to carry out an effective detection of all the real faults without introducing a large proportion of untimely detections.

Thus, in practice, a visual examination of the sheets is used. In order for the operator to have a clear image of the moving sheet, it is necessary that the speed thereof is limited. However, it is desirable to be able to carry out an examination at high speed and on both sides of the sheet to detect faults in real time. This possibility would prove particularly advantageous to allow an examination of a sheet after hot rolling and pickling and before cold rolling when the cold rolling train is connected directly to the pickling line. Indeed, in such a case, the combination of a reliable detection of dangerous faults for cold rolling and a high running speed (for example up to 6 m / s) at the control station is highly desirable.

Also, the object of the present invention is to provide a method enabling effective detection of surface defects on a strip during running, even when the strip is moved at a high speed.

This object is achieved by a method of the type defined at the head of the description and according to which, according to the invention,<ul id="ul0001" list-style="none"><li>signals representative of each image of a part of the examined surface of the strip are recorded in an image memory,</li><li>- the processing of the analog signals relating to each image is carried out during the period during which this image is formed and includes the operations of: converting analog signals into digital form by dividing the image into picture elements with each of which is associated a digital value, filtering the digital signals representative of the picture to detect relative variations in the digital values ​​of the picture elements, and analysis of the filtered signal to provide or not a fault detection signal according to the detected variations, the recording representative of the image being kept or not in response to the emission or the absence of emission of the fault detection signal, and</li><li>- each record saved is read for viewing at a control station for a sufficient time to visually check the image it represents.</li></ul>

Thus, the method according to the invention combines the reliability of the automatic detection of surface irregularities on a moving strip with the reliability of the qualification of the irregularities detected by a visual control of the still images of these irregularities.

The filtering of the digital signals representative of an image can be carried out by means of a bidirectional contour detection filter, that is to say by detecting, between picture elements located close to each other, variations of digital values ​​of amplitude greater than a first predetermined threshold. The analysis of the filtered signal can then consist, for each image, of developing, from the detected variations of amplitude greater than the first threshold, quantitative information (for example area or perimeter) of the surface irregularities of the image considered and to whether or not to supply the fault detection signal depending on whether or not this quantitative information is greater than a second predetermined threshold.

According to a particular feature of the method according to the invention, the filtering of digital signals is carried out in two different ways depending on whether these signals are associated with picture elements located in the vicinity of the edges of the strip or with picture elements located in full face of the band. To this end, the digital signals are separated into first digital signals representative of parts of the image situated along the edges of the strip and second digital signals representative of the other parts of the image, the first digital signals are processed to detect the amplitude variations corresponding to the passage from one side to the other of the edges of the strip and to provide information representing the difference between the detected position of the edge of the strip and a position of reference, and the second digital signals are processed to associate with each picture element information representing the amplitude variations of the digital signals associated with the neighboring picture elements. Thus, edge defects can be detected by examining the difference between the measured and reference positions of the edges, while full-face defects can be detected by looking for the picture elements located on the outline of a "task " of the image.

Preferably, the signals representative of each image are recorded in a mass memory addressable image by image, and the passage from one address to another in write mode is controlled in response to a fault detection signal.

The stored records are preferably read sequentially to allow viewing of each of the images which they represent for a predetermined period of time at the end of which the address at which the recording of the image which has just been viewed is displayed again available in write mode. Thus, the operator has a predetermined period of time to qualify the detected surface irregularity, period at the end of which the recording of this image can be erased by a new recording. The absence of qualification during this period can be assimilated to a serious defect.

Advantageously, a simultaneous examination of the two opposite faces of the strip is carried out by alternating the processing of image signals from successive parts of one face with the processing of image signals from successive parts of the other face. It is then possible to view the preserved recordings of images of parts of one side on a display device different from that used to view the stored records of images of parts of the other side.

The invention also aims to provide an installation allowing the implementation of the process defined above.

This object is achieved by an installation of the type comprising at least one photosensitive device which periodically generates analog signals representing the images of successive parts of the surface of the strip and a device for processing analog signals to detect abnormal level variations,

installation in which, in accordance with the invention:<ul id="ul0002" list-style="none"><li>an addressable mass memory is provided for recording signals representing each image converted by the photosensitive device,</li><li>the processing device comprises: an analog-digital converter for converting the analog signals representing each image in digital form by dividing the image into image elements with each of which is associated a digital value; a digital filter receiving the digital signals representing each image and detecting relative variations in the digital values ​​of the picture elements; and an analysis circuit connected at the output of the digital filter to generate or not, for each image, a fault detection signal as a function of the variations in detected digital values,</li><li>- Means are provided for controlling the conservation or non-conservation of the recording of an image, at the end of the period during which this image has been converted by the photosensitive device, in response to the presence or the 'absence of a fault detection signal corresponding to this image, and</li><li>- a display device is connected to the mass memory to view each stored record in the form of a fixed image in order to allow visual control of the image represented by this record.</li></ul>

1 The digital filter comprises for example a bidirectional contour detection filter comprising means for associating with each picture element digital information calculated as a function of the digital values ​​associated with neighboring picture elements; the analysis circuit then comprises means for developing quantitative information from the detections of picture elements for which the associated digital information exceeds a first predetermined threshold, and comparison means for producing a fault detection signal when said quantitative information exceeds a second predetermined threshold.

The digital filter may further include an edge detection filter for detecting the positions of the edges of the strip and providing information representing the deviations between the detected positions of the edges and the reference positions. This information is taken into account by the analysis circuit when the detected deviations exceed a predetermined threshold.

Preferably, the mass memory is an analog memory intended for recording the analog signals representing each image and addressable image by image. As analog memory, at least one magnetic video disc is advantageously used.

The writing and reading in the memory are controlled in response to the reception of fault detection signals. Advantageously, the mass memory comprises two memory units, one of which operates in write mode while the other operates in read mode, or vice versa; thus, an image recorded in one unit is viewed while each new image formed is recorded in the other unit.

More preferably, in order to carry out a simultaneous inspection of the two faces of the strip, the installation comprises first and second photosensitive devices arranged on each side of the strip to generate each of the analog signals representing the images of successive parts of a respective side of the strip, a switching circuit connected between, on the one hand, the photosensitive devices and, on the other hand, the mass memory and the processing device, and switching circuit control means for transmitting alternately to the mass memory and to the processing device an analog image signal from one side and an analog image signal from the other side. Two separate viewing devices can then be used to view the stored records relating respectively to one side of the tape and the other side.

For reasons of space, the photosensitive devices may have to be offset along the inspection line. In this case, the image signals produced by the upstream device are delayed to compensate for the offset so as to have at all times image signals relating to the opposite faces of the same section of strip.

Other particularities and advantages of the method and of the installation in accordance with the invention will emerge on reading the description given below, by way of indication but not limitation, with reference to the appended drawings in which:<ul id="ul0003" list-style="none"><li>FIG. 1 is a general diagram of an embodiment of an installation in accordance with the invention,</li><li>FIG. 2 illustrates the operation of the digital filter of the installation shown in FIG. 1, and,</li><li>FIG. 3 is a more detailed diagram of this digital filter, and</li><li>FIG. 4 is a functional flow diagram of the tasks carried out by the microprocessor of the installation illustrated in FIG. 1.</li></ul>

The installation shown in FIG. 1 is intended for the detection of surface defects on a sheet 10 during running, for example a sheet leaving a pickling line after hot rolling.

Each face of the sheet 10 is inspected by means of a respective television camera 11, 12. The cameras are located on either side of the transport plane of the sheet 10 and are arranged, in the example illustrated, in parallel to the sheet metal. The images of the faces of the sheet 10 are returned to the cameras by mirrors 13, 14 inclined at 45⁰ compared to the sheet metal. The assembly formed by a camera and the associated mirror is housed in a dust-tight housing and provided with an observation window on the side of the sheet. This bootmaker is itself housed in an enclosure provided with means, such as neon tubes and diffuser, to illuminate the sheet as homogeneously as possible.

Each camera 11, 12 is equipped with a tube of very high sensitivity with the possibility of shuttering, for example the tube known under the name ISOCON and produced by the British company called English Electric Valves. This tube can be closed electronically, which avoids having to use less reliable electromechanical devices and reducing sensitivity. With such a tube, the reading time is 20 ms and the exposure time is adjustable. For the intended application, this exposure time is preset to allow sufficient illumination without disturbing the perception of surface irregularities in a too sensitive manner. As an indication, for a sheet moving at 6 m / s, an exposure time of 300 us may be suitable, the sheet traveling only about 2 mm during installation. Each camera is equipped with a remote-controlled lens to allow adjustment of focus, iris and magnification. The frequency of shots is determined, depending on the field of the cameras, so as to form successive images which represent adjacent parts of each face of the sheet, with no space between two consecutive parts; a slight overlap between adjacent parts is preferable to be sure to avoid that an area escapes inspection.

If it is assumed that the field of view of each camera covers a length of sheet metal of 1 m, and that the speed of movement is 6 m / s, the frequency of shooting for each camera is therefore at least equal to 6 shots of view per second. As indicated below, the signals representing the images of the successive parts of the two faces of the sheet are interspersed in order to be processed sequentially. Also, the cameras 11 and 12 operate alternately and the total duration of the exposure and the reading for each image received by a camera must be less than the half-period of shots; this condition is largely satisfied if we consider the figures cited above as an example.

The cameras 11 and 12 operate continuously, but the signals produced by these cameras are taken alternately so as to apply to the input of a processing device 20 alternately a frame coming from the camera 11 and a frame coming from the camera 12. To this end, the output signals from the cameras are transmitted to the input of the device 20 by means of respective switches 15, 16, the closing of which is controlled alternately, and each time for the time just necessary for the transmission of the signals to be processed, by respective control signals S0, S1. With the figures indicated above, each of the signals SO, S1 is in the form of rectangular slots of duration equal to 20 ms and of period equal to 1/6 s.

The processing device 20 comprises a circuit 21 for adjusting the gain and analog filtering receiving the serialized image signals, a circuit 22 forming a sampler-blocker and analog-digital converter connected to the output of the circuit 21, a digital filtering circuit 23 connected to the output of the circuit 22, an analysis circuit 24 receiving the digital signals produced at the output of the filtering circuit 23, a memory circuit 25, and a microprocessor circuit 26.

The circuit 21 includes a gain-controlled amplifier in order to compensate for the variations in gray level of the bottom of the sheet and a bandpass filter to retain only the signals of the useful frequency band.

The function of the circuit 22 is to convert the analog image signals into digital form, it receives a control signal S2 which determines the sampling frequency, for example 5 MHz.

The digital words produced by the circuit 22 are received by the filtering circuit 23 whose function is to detect relative variations in the values ​​of the digital words.

The filtering circuit comprises a filter 23a for edge detection and a filter 23b for edge detection. If we refer to Figure 2, we see a part of the surface of the sheet 10 as seen from one of the cameras. The light reflected by the surface of the sheet creates a very clear contrast between the two parts of the image separated by the edge 10a of the sheet.

The purpose of the edge detection filter 23a is to carry out special processing of the signals representative of the image parts located along the edges.

During an initialization phase of the process relating for example to the first 16 lines of the first image, the addresses of the edges of the sheet are determined, addresses expressed in point numbers of a line of the corresponding frame at the shore. If we denote by I_{i, j} the digital word representing the image element located at the intersection of line i and column j, the detection of a bank is carried out by realizing the difference I_{i, j + 3} - I_{i, j}, that is to say that two picture elements of the same line are compared not successively, but separated by two picture elements. As a variant, we could make the differences I_{i, j + 2}- I_{i, j} Yes_{i, j + 4} - I_{i, j}. The bottom on each side of the sheet being uniformly black, the first transition detected corresponds to the left bank 10a of the sheet. The address j + 3 is recorded, as well as the amplitude of this transition. The detection of the other edge of the sheet is carried out when a transition of the same amplitude is measured; the address of this second transition is recorded. The procedure is similar for the first 16 lines and the transition addresses recorded for each bank are averaged. These average values ​​constitute the reference addresses of the banks A1 and A2 expressed in point numbers on a line.

As shown in FIG. 3, the successive digital words received by the circuit 23 are counted by means of a counter 230 which supplies the address j of the picture element corresponding to the digital word received. At the end of each image line (that is to say here when 256 digital words have been counted), the counter 230 returns to the initial state and a second analog counter 231 is incremented, this second counter providing l address i of the picture element corresponding to the digital word received. The filter 23a operates only for the image areas situated along the edges of the sheet, that is to say on image bands extending from A1 - N to A1 + N and A2 - N to A2 + N. We could for example choose N = 16. The operating windows of the filter 23a are defined by means of a logic circuit 232 receiving the values ​​A1, A2 and the content A. of the counter 230 and supplying a signal F1 when A1 - N ≤ A ≤ A1 + N and a signal F2 when A2 - N ≤ A ≤ A1 + N. Signals F1 and F2 are joined by an OR gate to control the opening of an ET1 gate through which the digital words received are transmitted to a register 233 with serial input and parallel outputs to permanently store the words representing four elements d successive image of the same line. A subtractor 234 is connected to outputs of register 233 to make the difference I_{i, j + z}- I_{i, j} and providing an SR signal in response to the detection of a transition corresponding to an edge. A subtractor 235 makes the difference between A_j and either the value A1 or the value A2, either of these values ​​being transmitted to the subtractor 235 under the control of the signals F1 and F2. This difference A_j- A1 or A_j- A2 is compared in absolute value to a predetermined threshold A by means of a comparator which associates with each digital word I_{i, j} a binary value Q (i, j) equal to 1 or 0 depending on whether Q (i, j)&gt; A or Q (i, j) ≤ A. At each edge detection (signal SR), the value Q (i, j) is transmitted to the analysis circuit 24. This value will be equal to 1 only if the difference between the detected bank address and that of reference is greater than B, signifying the existence of a bank fault. We can choose for B a value corresponding for example to four picture elements.

Outside the operating windows of the filter 23a, the digital words I_{i, j} are processed by means of the contour detection filter 23b which has the function of detecting variations reflecting the existence of a surface irregularity whose contour (or perimeter) is thus determined. Different types of contour detection filters are known; we could for example use a bidirectional Prewitt filter. Each image element converted into a digital word is associated with a digital quantity calculated as a function of the digital words representing the neighboring image elements. For each picture element, the quantity associated according to Prewitt's formula is: P (i, j) = | Px | + | Py | with

<maths id="math0001" num=""><img file="EP0153218A2_D0001.tif" /></maths><maths id="math0002" num=""><img file="EP0153218A2_D0002.tif" /></maths>

This amounts, for each picture element, to calculate, on the one hand, the difference between the cumulative digital words of the three adjacent picture elements of the following line and the cumulative digital words of the three adjacent picture elements of the previous line and, on the other hand, the difference between the cumulative digital words of the three adjacent picture elements of the next column and the cumulative digital words of the three adjacent picture elements of the previous column, then add up these differences. When the quantity P (i, j) exceeds a predetermined threshold B this means that the image element considered is at the limit of two relatively different gray level zones, therefore on the contour of a task or irregularity of area. It will be noted that the bidirectional nature of the filter makes it possible to detect contours oriented in any direction. P (i, j) is given a binary value, for example 1 or 0 depending on whether P (i, j)&gt; B or P (i, j) ≤ B.

As shown in Figure 3, the digital words I_{i, j} to be processed by the circuit 23b are transmitted to an addressable register 237 through an ET2 gate which is open under the control of the inverse of the ET1 gate control signal. A calculation circuit 238 is connected to the register 237 to receive from the latter the information necessary for the calculation of P (i, j). This information is read in the register 237 under the control of address signals supplied by an addressing logic circuit 239 which receives the contents A and A from the counters 230 and 231, that is to say the address (i , j) of the image element corresponding to the word received. The quantities Px and Py and P (i, j) are worked out during the presentation time of the digital word I_{i, j}. A comparator is provided in the calculation circuit 238 to compare | Px | + | Py | at threshold B in order to supply the quantity P (i, j) in binary form.

The analysis circuit 24 receives the binary signals Q (i, j) and P (i, j) produced by the filters 23a and 23b, and sums them S for the complete image. This sum is compared with a threshold C and the analysis circuit produces a fault detection signal SD when S&gt; C, and does not produce a signal when S_&lt;'C. At the same time, the sum S is stored in memory 25.

The edge detection can be refined by eliminating the isolated points, that is to say by not taking into account the quantities P (i, j) which exceed the threshold B when the quantities associated with the neighboring picture elements do not not exceed this threshold. It can be the same for the detection of the banks.

The complete processing of the image signals by means of the circuits 21 to 25 is carried out in real time, for each image, that is to say before presentation of the signals representative of the following image. The data rate is very high; it can reach 1.5 million digital words per second at the output of the converter. Also, the digital processing is carried out in wired logic, in order to have a very fast processing speed, and the microprocessor circuit 26 performs control and connection tasks with the outside, as will be described later. in more detail. It will already be noted that the circuit 26 supplies the signals SO and S1 for controlling the position of the switches 15 and 16.

As regards the edge detection, it has simply been envisaged above to determine the sum of the quantities P (i, j), which gives an indication of the total area of ​​the defects. It will be noted that algorithms known in themselves could be used to also calculate the perimeter of the defects as well as the Euler number (contours inscribed in another contour). If the quantities P (i, j) are used to calculate the total perimeter of the defects, it is this perimeter (and not the area as before) which is compared with a predetermined threshold to control or not the emission of a fault signal.

In parallel with the processing of the analog signals representative of each image, a recording of these signals is carried out. For this purpose, a recording device 30 is used which receives on its input 30a the analog signals at the output of the switches 15 and 16 and which comprises two memory units 31, 32 with magnetic video-discs addressable image by image. The received signals are recorded in one or other of the units 31, 32 according to the position of a switch 33 having a movable contact connected to the input 30a and two fixed contacts connected respectively to the recording inputs of the units 31 , 32. The recorded signals can be read in either of the units 31, 32 according to the position of a switch 34 having two fixed contacts connected respectively to the reading outputs of the units 31, 32 and a movable contact connected to the output 30b of the recording device.

The memory units 31, 32 operate under the control of signals produced by the microprocessor circuit 26. Operation in recording or playback mode is controlled by complementary signals L / E1 and L / E2, so that, in all cases, one of the units operates in recording mode when the other operates in playback mode. In addition, the circuit 26 transmits to the units 31, 32 address signals AD1, AD2 which define the addresses for recording or reading the image signals, an address corresponding to an image. Finally, the circuit 26 also supplies the signals S3, S4 which control the position of the switches 33 and 34.

The installation shown in FIG. 1 finally comprises a display device 40, an input 40a of which is connected to the output 30b of the recording device. The device 40 is intended to allow viewing of fixed images recorded in the units 31, 32. In the example illustrated, it comprises two television monitors 41, 42 associated respectively with the two faces of the sheet 10. Depending on whether the image signals read in the recording device 30 represent the image of part of the lower surface or the image of part of the upper surface of the sheet, these signals are routed to the monitor. 31 or the monitor 32 by means of a switch 43. These monitors also make it possible, if necessary, to directly view the two faces of the sheet 10 while scrolling. To this end, the output signals from the cameras 11, 12 are applied to fixed contacts of switches 44, 45, the other fixed contacts of which are connected respectively to the fixed contacts of switch 43 and the movable contacts of which are connected to the video inputs of the monitors. 31, 32. Thus, depending on the position of the switches 44, 45, are displayed either the still images read in the recording device 30, or directly the images of the faces of the sheet in movement. The positions of the switches 43, 44, 45 are controlled by signals S5, S6, S7 produced by the microprocessor circuit 26.

The installation described above works as follows.

The signals representative of each image formed by the cameras 11, 12 are analyzed by the processing device 20 while being simultaneously recorded in one of the analog memory units at an address specified by the circuit 26.

If no fault detection signal SD is produced by the circuit 24, it is not necessary to keep the recording of the image and to take into account the information stored in the memory 25. The signals representative of the next image is then recorded at the same address of the same analog memory unit and the surface irregularity information relating to this next image is stored in the memory 25 in place of the previous information.

If, on the other hand, a fault detection signal SD is produced, the information stored in the memory 25 is transferred to the random access memory 26a of the microprocessor circuit and the image signals recorded in the memory unit are preserved, either by reversing the operations of the two memory units (so that the signals relating to the next image will be recorded in the other memory unit), either by incrementing the write address (so that the signals relating to the next image will be recorded at another address in the same memory unit).

The reading in the memory units and the operation of the display device 40 are controlled by the microprocessor device to allow an operator to view in the form of still images all the images whose records are kept in the memory units (c (i.e. the images which gave rise to the emission of a fault detection signal), and to qualify the surface irregularities displayed.

The qualification of the irregularities displayed is sent to an external computer which has received the information read in the memory 25 and recorded in the random access memory 26a. This qualification is transmitted in the form of a code indicating either that it is a non-serious fault, or that it is a serious fault and, in the latter case, the nature of this fault ( hole, fold, cylinder imprint, ...).

The approximate location of the fault on the sheet can be identified in several ways. The fault signal can be sent to an external computer where it is placed in a queue until receipt of the qualification, the external computer being programmed as is known per se to follow the sheet during its entire processing, and can therefore locate the sheet metal area which was in front of the cameras 11, 12 at the time of the emission of the fault signal. It is also possible to mark the sheet by spraying paint downstream of the detection station, in response to the fault signal. One could also, knowing the speed of travel of the sheet, develop and associate with the qualification information coordinate information calculated by measuring the time elapsed from the start of the process until the emission of the fault signal.

After sending to the external computer, the characteristics of the fault recorded in RAM 26a are erased.

The duration of presentation of each still image to the operator is limited (for example to 10 s). The absence of qualification at the end of this period can be assimilated to a qualification of serious defect.

After each qualification or each end of the presentation period, a new image is automatically presented among those remaining stored in the memory units 31, 32. The order of presentation of the images is the same as that in which they were recorded. The queue is managed by the microprocessor 26b of circuit 26, as is the routing of the images read to one or the other of the monitors 41, 42 depending on whether these images relate to the underside or the upper face of the sheet. When there is no longer any waiting image in the memory units, the switches 44, 45 are controlled so as to connect the monitors 41, 42 directly to the output of the cameras 11, 12.

FIG. 3 is a functional flow diagram of the tasks performed by the microprocessor of circuit 26.

The main program 50 includes:<ul id="ul0004" list-style="none"><li>a phase 51 of initialization and loading of the parameters (in particular tape speed, queue, etc.).</li><li>a speed sampling and conversion phase 52 consisting in generating the signals S2 at a predetermined frequency, and in generating the signals S1 and control signals for the shots for the cameras 11, 12 at a frequency depending on the scrolling speed, and</li><li>a test 53 for waiting for an interrupt request: in the event of a positive response, one returns to the start of phase 52 and, in the event of a negative response, the program is terminated.</li></ul>

Each interruption triggers the execution of a subroutine. In order of priority, the interruptions provided here are those produced by: the emission of the fault detection signal (60), the intervention of the operator to qualify a fault (70), the end of the presentation period a still image (80) and a request for dialogue with an external computer to which the microprocessor circuit (90) is connected.

The reception of a fault detection signal causes the blocking of any other interruption (phase 61), the memorization of the fault characteristics by reading the content of the memory 24 and transfer to random access memory 26a (phase 62). If the queue is not full (test 63), a swapping of the memory units is carried out, or the address is saved during recording (phase 64). The permutation consists in putting in memory mode the memory unit which functioned in recording mode and in putting the other memory unit in recording mode with fast return to the first address of reading or writing), the permutation n 'is possible only if an image is not being read from the other memory unit; otherwise, an incrementation is carried out which consists in authorizing recording at the following address without permuting the operations of the memory units (that is to say, here, advancing the video-disc in recording by one track ). The next step (phase 65) consists in updating the queue of recorded images, launching the read and delay subroutines and then unmasking the interruptions is authorized. The read subroutine triggers the display of the image at the address specified in the memory unit placed in read mode and the delay subroutine triggers the start of the image presentation delay for the operator. If the result of test 63 is positive (queue saturated), we go directly to phase 65.

In response to the interruption caused by the operator (phase 71), and while allowing the interruption due to reception of a fault signal, judgment (or qualification of fault) is acquired by the microprocessor (phase 72). Then, an interrupt request is sent to the external computer (phase 73), then either a permutation of the memory units or an increment of the read address is carried out in order to read the next stored image (phase 74 ). The read and delay subroutines are launched (phase 75), then the queue is updated and the unmasking of interrupts is authorized (phase 76).

In response to the interruption caused by the end of the image presentation period, and while allowing interruptions due to the reception of a fault signal or due to the operator's action (phase 81), the first fault in the queue is validated and an interruption in dialogue with the external computer is requested (phase 82); failure of the operator to qualify the defect within the time allowed is deemed to be a serious defect. Then, we perform either a permutation of the memory units, or an incrementation of the read address (phase 83), the read and delay subroutines are launched (phase 84) if the queue n is not empty, then the queue is updated and the unmasking of interrupts is authorized (phase 85).

Finally, in response to the interruption caused by a request for dialogue with the external computer, and while allowing the other interruptions indicated above, (phase 91), the qualification of the fault and the digital information characterizing it read in the memory 25, are sent to the external computer (phase 92) and the unmasking of interruptions_' is authorized (phase 93).

We considered above the sending to the external computer of the characteristics of all the surface irregularities detected with the qualification given to these irregularities. As a variant, the characteristics of the surface irregularities qualified as serious faults can only be sent to the external computer.

The choice between swapping memory units and incrementing addresses in recording or reading is made as follows. As soon as a first fault is detected after the initialization phase, the memory units are swapped by inverting switches 33 and 34 (signals S3 and S4) and inverting L / E1 and L / E2 so as to be able to display all of following the image concerned and during this time to save in the other memory unit any images to be kept. After displaying this first image, and the fact that the first memory unit is empty, the memory units are swapped so as to be able to display the possible image or images recorded in the other unit. When it no longer remains in this other image unit to be displayed, a new permutation will be carried out, and so on. As long as there is no image preserved, or when there is no longer any to display, switches 44 and 45 are controlled (signals S6, S7) to directly display the faces of the sheet in movement.

The queue of images to be viewed is managed by the microprocessor 26b, with, for each of these images, the recording of its address in memory and of binary information indicating that this image represents a part of the underside or of the upper face of the sheet. This binary information coincides for example with the state of one of the signals SO and S1. At the time of viewing, the binary information thus recorded is used to control the switch 43 (signal S5) in order to route the video signals to the appropriate monitor.

Of course, various modifications or additions may be made to the embodiment described above of the method and the device according to the invention without thereby departing from the protective framework defined by the appended claims.

Thus, in particular, the signals representative of each image can be recorded in a digital memory, these signals being either transmitted in digital form by the photosensitive devices, or transmitted in analog form and digitized. The digital memory can be constituted by two magnetic disks similar to those described above.