FR3074295A1 - METHOD FOR DETECTING ROLL DEFECTS IN PRINTED GLASS - Google Patents

Abstract

The present invention relates to a method for detecting rolling defects in a printed glass. It also relates to a device implementing such a method. The method for detecting laminating defects in a printed glass moving in a first direction comprises (a) a step of calculating (E200) a spectrogram by the Fourier transform of a thermogram composed of surface temperature profiles printed glass, acquired at the output of the rolling rolls, in a second direction perpendicular to the first direction; (b) a filter step (E201), by a bandpass filter; (c) a step of searching (E202) frequency values for which one or more filtered Fourier transforms satisfy a criterion of severity severity Ψ0.

Description

Method for detecting rolling defects in a printed glass

The present invention relates to a method of detecting a rolling defect in a printed glass. It also relates to a device implementing such a method.

Printed glasses, sometimes also called textured glasses, are manufactured using a process including a hot-rolling operation of a glass sheet between two generally metallic rollers. The surface of the rollers may have a relief making it possible to mechanically print a pattern or a texture on the surface of the glass sheet. The terms "printed glass" and "textured glass" are often used to refer to glass obtained by such a process, whether or not it has a pattern or texture on its surface. They therefore also denote a glass, obtained by rolling, the surface of which is smooth or devoid of any particular texture. The temperature at which the rolling operation is carried out depends on the composition of the glass. For example, for a standard soda-lime glass, the temperature of the glass is generally between 700 and 800 ° C. Examples of the manufacturing process for printed glass are described in US1519314 A and DE102004034694 B3.

When manufacturing printed glasses, defects of different types and origins may appear. Some of these defects are characteristic of the hot rolling operation and appear at the end of this operation.

In the technical field of printed glass, these defects have several names. They are called “Pas de Chat” (PC) in French, “Cat FootPrints” (CFP) in English, “BlasenStreifen” (BS) in German, because of their form and their frequency. Their origin is not well known. According to the current state of knowledge, their origin is considered to be related to the creation of bubbles, within the glass, which, under the effect of convection movements generated by the rolling rollers, periodically form marks or imprints at the surface of the glass. These marks or imprints are areas of the glass surface whose optical properties are altered. They are generally visible as periodic optical defects on the surface of the glass under inclined or grazing lighting.

The formation of PC defects can generally be stopped by inducing the bubbles to move towards the edges of the sheet either by raising the upper rolling roller or by means of a hook. However, these PC faults can generally only be detected visually when the glass has been cooled at the end of the manufacturing process. During the lag time between the time when the defect is generated and the time when it is visually detected, the production of glass is not stopped, and the glass produced during this period can neither be sold to the customer nor used due the presence of the fault. It is therefore a double material and financial loss for the manufacturer. It is therefore desirable to detect the presence of PC faults as soon as possible in order to be able to act quickly on the manufacturing process so as to eliminate the source and thereby reduce glass losses.

However, the state of the art does not disclose any means of such detection. The present invention overcomes this problem. A subject of the invention is therefore a method of detecting rolling defects in a printed glass moving in a first direction at the outlet of the rolling rolls, said method comprising the following steps:

at. the computation, implemented by computer, of a spectrogram by the Fourier transformation of each of N lines of a thermogram of dimension Μ XN, said thermogram being composed of M temperature profiles of the surface of the printed glass, acquired in exit of the rolling rolls, in a second direction perpendicular to the first direction, with a digital spatial resolution of N pixels in said second direction, an acquisition frequency f _e of each profile greater than the frequency ωθ of rotation of rolling rolls , and a pixel size equal to or less than the dimension <5θ of the defects;

b. the filtering, implemented by computer, by a bandpass filter of natural frequency ü) _c = κω ₀ , of each of the N Fourier transforms, where K is a definite positive real number greater than 1, and the value of the factor of quality Q is equal to or greater than a defined value χθ;

vs. the search, implemented by computer, of the set Ω of frequency values for which one or more of the N filtered Fourier transforms satisfy a severity criterion ψ ₀ previously defined.

In the present text, the expression “printed glass” is used to denote a glass obtained by a hot rolling operation as described above, whether or not the glass has a pattern or a texture on its surface.

During the laminating operation, the surface temperature of the printed glass at the outlet of the laminating rollers oscillates around an average value. The frequency of these oscillations is correlated with the frequency ωθ of rotation of the rollers. In the context of research on the temperature signal r (t) of the surface of the glass printed at the end of rolling, it has been observed that, when PC faults appear, said temperature signal r (t) comprises a phenomenon high frequency different from that which can be associated with the oscillations caused by the rolling rollers. It has been surprisingly empirically found that this high frequency phenomenon is linked to the frequency of occurrence of PC faults.

The invention is based on the exploitation of the high frequency signal contained in the temperature signal to detect and locate possible PC faults directly at the rolling output even when the glass temperature is still too high to allow a visual inspection. of its surface. The invention allows rapid and early detection of faults and thus acts quickly on the parameters of the manufacturing process to make them disappear.

The implementation of the present invention is dependent on the characteristics of the hot rolling process with which the printed glass is produced. These characteristics influence in particular the values of the parameters f _e , δ ₀ , K and χθ. The values of these parameters cannot be known a priori. They may depend in particular on the frequency of rotation ωθ of the rolling rolls, variations in the temperature of the printed glass and the composition of the glass. Consequently, it is advantageous, for each process, to carry out a preliminary characterization of the defects on the glasses printed at the end of manufacture using appropriate optical means. This characterization makes it possible to determine the frequency Ψρ £ of appearance of the PC faults as well as their shape and their dimension <5θ. On the value of these parameters depend those of the other parameters f _e , K, χθ, the definitions of which are detailed below.

In step (a) of the method of the invention, the acquisition frequency f _e with which the M temperature profiles were acquired is greater than the frequency Ψρ £ of appearance of the PC faults. Otherwise, the high frequency phenomenon contained in the temperature signal r (t) may not be detected. In addition, there may be a risk of aliasing during the Fourier transformation of each of the N lines. It is recommended that the acquisition frequency f _e be greater than twice the frequency Ψρ £ of occurrence of PC faults. In one embodiment of the invention, the acquisition frequency f _e is between 20 and 150 Hz, preferably between 20 and 100 Hz. This frequency range is advantageous because it can cover most of the processes for manufacturing printed glass by rolling. The successive acquisition of the M temperature profiles is a temporal sampling according to a time interval between the acquisition of two consecutive profiles corresponding to -. A fortiori the acquisition frequency f _e is greater than the frequency ωθ of rotation of the rollers. There is no upper limit on the value of the acquisition frequency compared to that of rotation of the rollers.

The digital resolution of N pixels and the size of the pixels are dependent on the acquisition means used for the acquisition of the temperature profiles over the entire width of the glass sheet in the second direction. In order to be able to distinguish faults, it is preferable that the resolving power of the means for acquiring temperature profiles is equal to, or less than, the size or dimension < ₅₀ of the faults. The resolving power is generally defined using the Rayleigh criterion:

Δτ = 1.220 ^ where the parameter λ is the wavelength of the monochromatic radiation used for the observation, the parameter D is the aperture diameter of the diaphragm and the parameter f is the distance to the focal plane.

A fault is visible by the acquisition means if 2Δτ <<5 ₀ . Preferably, the size of the pixels is equal to, or less than, Δτ. The maximum width of the glass sheet on which the temperature profiles in the second direction can be acquired is a function of the size, ε, of the pixels and their number, N. This width f can be estimated according to the formula f = N - ε

Defects generally do not have a regular shape in the first and second directions. The dimension < ₅₀ can then correspond to an average physical dimension of the defects, such as the average value of the Feret diameters, measured in the surface plane of the glass, of each defect, or even to the average size in said second direction.

The acquisition of the M temperature profiles can be carried out on part or all of the glass sheet at the outlet of the rolling rolls. Preferably, the temperature profiles are acquired over the entire width of the glass sheet. The digital resolution of N pixels is adapted to the width of the area or of the entire glass sheet. The higher N, the higher the spatial resolution of the temperature profiles in the second direction. As a nonlimiting example, the digital resolution can be 1000 pixels, which corresponds to most of the acquisition means such as infrared beam scanning scanners available on the market.

In one embodiment of the invention, the detection method further comprises, before step (a), a step of acquisition, at the outlet of the rolling rolls, of M temperature profiles of the surface of the printed glass. , in the second direction perpendicular to the first direction, with a digital spatial resolution of N pixels in said second direction so as to create a thermogram of dimension Μ XN.

In a particular embodiment, the acquisition step is carried out using an infrared beam scanning digitizer. Infrared radiation is particularly suitable for measuring temperature profiles on printed glasses whose temperatures can vary between 500 ° C and 1000 ° C at the outlet of the rolling rolls.

In order to simplify the expression of mathematical formulas, the Dirac notation for the Fourier transforms is used in the rest of the text. The Fourier transform Τ (ω) of the temperature signal r (t) and its inverse transform are respectively noted:

-J + co

Τ (ω) = (T (t) | e ^_iùJt ) = - = IT (t) e ^_iùJt dt \ 2π J_co

-J + co r (t) = (e ^ÎÙJt | T (a))) = - = IT (a)) e ^ÎÙJt da) \ 2π J_co

The set of M temperature profiles with a digital resolution of N pixels can be directly acquired using a digital acquisition means or by sampling an analog signal coming from an analog acquisition means. In the latter case, the thermogram of dimension Μ X N corresponds to M and N samples of the signal in the first and second directions respectively.

The thermogram can be represented in a matrix form: a = (Tm, n) _MN G n, m en, with the space of matrices of dimension M xN with values in IR. The coefficient T _{m> n} represents the value of the temperature at the m, n coordinates of the thermogram.

The temperature signal T _n (t) for a line n of the set of N lines of the matrix a is a series T _n (t) = T _mn , t G {0, ..., M - 1], of the M terms of the line îl, let T _n (0) = T _On , T _n (l) = τ _1η , ..., τ _η (Μ - 1) = The Fourier transform of the series is a transform of Discreet Fourier. According to the notation adopted, it is the series of M terms defined by

M-l

Σ ωί

Tn (t) e ““ ω = 0

Generally with:

r „. m „m„ 1. ..

ω G l (J, 1, ..., - - 1, - -, ..., —Il if M is even, or r- Γη Λ (Μ-1) Λ (Μ-1) J η4. · ω G ΙΟ, 1, ..., —--- 1, ---—, ..., - 11 if Μ is odd.

The set of N discrete Fourier transforms obtained for each of the N lines can be represented in matrix form:

A = (T _m , n \ _<m < _MN £ M _{Q <miM: N} (T), Ν, ΜΕΚ

The matrix A is called spectrogram. The spectrogram can be represented graphically in the form of an image using the real part of the coefficients T _mn , their imaginary part, or their module.

Discrete Fourier transformations can be calculated using one or more calculation units and a fast Fourier transformation algorithm such as the Cooley-Tukey algorithm, the Good-Thomas algorithm, the Rader or the Bluestein algorithm. These algorithms are generally suitable for calculating the Fourier transform of a series whose number of terms is a power of 2.

In one embodiment of the invention, the number M of profiles is a power of two, 2 ^P , with p an integer between 9 and 15, preferably between 10 and 14. This number

M provides sufficient temporal sampling of temperature profiles for most processes for manufacturing printed glass using laminating rollers.

The signal of the high frequency phenomenon corresponding to the PC faults can be advantageously extracted from the spectrogram using a bandpass filter of natural frequency ü) _c = κωθ, where K is a definite positive real number greater than 1. The value of the parameter K depends on the frequency Ψρ £ of occurrence of PC faults previously determined. It corresponds to the ratio of the frequency ωθ to the frequency Ψρ £: K =

It has been found empirically that the value of the parameter K is between 3 and 6, or even between 4 and 5. This range of values seems to be suitable for most processes for manufacturing printed glass by means of laminating rollers.

A bandpass filter can be expressed mathematically in the form of a transfer function H (ω) in the frequency domain. The filtered Fourier transform, Τ * (ω) or Τ _η * (ω), can be written:

Τ * (ω) = Τ (ω) Η (ω)

Τ _η * (ω) = Τ _η (ω) // (ω), Vn G {0, ..., 1V}

The filtered temperature signal T * (t) or T _n * (t) is obtained by the inverse Fourier transformation of Τ * (ω) or Τ _η * (ω) respectively, that is τ * (ί) = (β ^ίωί | Τ * (ω)}

M — l ^,, _t V ^{1 ωπι} τ _η \ ΐ) = (ε ^ίωί | Τ _η * (ω)) = Τ _η * (ω) β ^ι2π ^, η Ε {Ο ..... Ν] ω = 0

Bandpass filters are generally qualified using a parameter called quality factor, noted Q. The quality factor is defined as the ratio of the natural frequency, (jù _c , over the width of the passband, Δω, ie Q = The width of the passband Δω is the difference between the two cut-off frequencies for which the maximum response intensity at the central frequency is generally reduced by 3 decibels (dB). It is also possible to define a another reduction value, different from 3 dB, for the calculation of the cut-off frequencies.

By way of nonlimiting example, for a Gaussian bandpass filter defined by the ~ (ω + ω _£ ) ² - (ω-ω _£ ) ² transfer function, Η (ω) = - L_ + -, the frequencies are the

V27T 2 V27T ^{1 1} frequencies for which the intensity of the frequency response, (jù _c , is reduced by a factor

VT

Depending on the nature of the bandpass filter for implementing the present invention, the value of the quality factor Q varies, in particular through the value of the width of the passband, Δω. The value of the quality factor is then defined equal to or greater than a certain value χθ determined during the preliminary characterization of the PC faults. During this characterization, an evaluation spectrogram can be calculated so as to determine the width of the signal peak at the frequency Ψρο The value χθ is generally defined so that the width of the passband of the filter is equal or greater at the width of the signal peak at the frequency Ψρο In certain cases, in particular when the noise intensity in the temperature signal is high, the value χθ can advantageously be defined so that the width of the passband of the filter is less than the width of the peak depending on the level of noise reduction sought.

In a particular embodiment of the invention, the bandpass filter is a super Gaussian bandpass filter of power 10. An example of a transfer function representing a super-Gaussian filter is the following function:

/ / (ω + ω _£ ) λ ¹⁰ / (ω-ω _£ )

H (ω) = C iev ^σ '+ e' ^σ

The parameters C and ff are fixed parameters defining the intensity of the filter and the width of the frequency window in which the filter is applied. The advantage is that the filters

band passers of this type generally have a profile with a flat top, the width of which is a function of the value of <J, which is suitable for filtering PC faults.

A PC fault is considered detected when at least one peak in the spectra corresponding to the N filtered Fourier transforms satisfies a severity criterion ψ ₀ previously defined by the operator as a function of the results of the preliminary characterization of the PC faults. The use of a severity criterion is advantageous in that it makes it possible to set a threshold beyond which a peak significantly corresponds to a PC fault and thus to avoid detection artifacts. The severity criterion can be considered as a sensitivity factor. The set Ω contains the frequencies for which their peak satisfies the defined severity criterion.

In one embodiment of the invention, the severity criterion ψ ₀ is chosen for one or more peaks from a threshold value of the intensity, a threshold value of making under peak or a threshold value of eccentricity of a line level.

A filtered thermogram can be obtained by graphic representation of the filtered matrix a * = GM _M ' _N (W), N, MGH, with the space of matrices of dimension Μ XN with values in IR. The coefficient T * _mn represents the value of the filtered temperature at the coordinates m, n of the image.

In one embodiment of the invention, the detection method comprises, after step (b), a step of displaying the N filtered Fourier transforms in the form of an image of dimension Μ X N. L ' image is a graphic representation of the spectrogram

the imaginary part.

T * _mn of the matrix A * = (r _m , _n ) _M , / W _MJV (Z), iV, MEN, with the space of matrices of dimension Μ XN with values in Z.

This representation is advantageous because it makes it possible to graphically highlight the location of the defects over the width of the glass sheet in the second direction. It can be subject to prior image processing such as binarization. This treatment often makes it possible to remove any noise that may be annoying for highlighting the areas of interest of the spectrogram, that is to say areas in which the defects are present.

One of the aspects of the technical effect which can be sought by the invention is the automation of the detection of defects during manufacture. It can, in fact, be sought that the time of a human operator is not monopolized by the surveillance of a screen on which the filtered spectrograms and thermal images are displayed to warn of the possible appearance of faults. On the contrary, it may be advantageous for an operator, busy with other tasks, to be notified only when a fault is detected. To this end, in another embodiment of the invention, the method further comprises, after step (c), a step of visual and / or sound signaling when the assembly Ω is not empty.

All the embodiments described can advantageously be combined.

The detection method of the invention is implemented by computer. The subject of the invention is also a computer program comprising instructions for the execution of the steps of the method of the invention in all the possible embodiments. Any type of programming language compiled into a binary form or directly interpreted can be used to implement the process steps by a series of arithmetic or logical instructions executable by a computer or any programmable information processing system. The computer program may be part of software, that is, a set of executable instructions and / or one or more datasets or databases.

In this sense, the invention also relates to a computer-decipherable storage medium on which is recorded a computer program comprising instructions for the execution of the steps of the detection method of the invention. Preferably, this storage medium is a non-volatile or non-volatile computer memory, for example a magnetic or semiconductor mass memory (solid state drive, flash memory). It can be removable or integrated into the computer which deciphers its contents and executes the instructions. It can also be integrated into a computer, called "server", different from the one that executes the instructions, called "client". To execute the instructions contained in the storage medium, the "client" computer can access the memory of the "server" computer by means of physical and / or aerial telecommunications. The "server" computer can also decrypt the storage medium on which the computer program is stored and communicate the instructions in binary form to the "client" computer by any means of telecommunication.

It may be advantageous for the storage medium to be a removable medium or accessible remotely by a telecommunication means so as to facilitate its distribution in all places where a detection method according to the invention is likely to be used.

The invention also relates to a device for detecting a rolling defect in a printed glass moving in a first direction at the outlet of the rolling rolls, said device comprising the following modules:

a module for calculating a spectrogram by the Fourier transformation of each of the N lines of a thermogram of dimension Μ XN, said thermogram being composed of M temperature profiles of the surface of the printed glass, acquired at the output of the rolling rolls , in a second direction perpendicular to the first direction, with a digital spatial resolution of N pixels in said second direction, an acquisition frequency f _e of each profile greater than the frequency ωθ of rotation of rolling rollers, and a size of pixel equal to or less than the dimension < ₅₀ of the defects;

a filter module with a natural frequency bandpass filter ù) _c = κωθ of each of the N Fourier transforms, where K is a definite positive real number greater than 1, and the value of the quality factor Q is equal or greater at a defined value χ ₀ ;

a search module for the set Ω of frequency values for which one or more of the N filtered Fourier transforms satisfy a severity criterion ψ ₀ previously defined.

The modules of the detection device of the invention can advantageously include one or more calculation units. Computing units are included in the Central Processing Units. Central processing units are generally integrated into computers which also include a set of other electronic components, such as input-output interfaces, volatile and / or non-volatile storage systems and BUSes, necessary for the transfer of data. between central processing units and communication with external systems, here the different modules.

The number and speed of calculation of the calculation units, and a fortiori of central processing units, necessary for the execution of the calculation steps of the detection method of the invention can be adjusted according to the number of temperature profiles. acquired.

In one embodiment of the invention, the detection device further comprises a module for acquiring, at the output of the rolling rolls, M temperature profiles of the surface of the printed glass, in the second direction perpendicular to the first direction, with a digital spatial resolution of N pixels in said second direction so as to create a thermogram of dimension Μ X N.

In particular, the acquisition module may comprise an infrared beam scanning digitizer, suitable for acquiring temperature profiles on the surface of a printed glass at the outlet of the laminating rolls.

It may be advantageous for an operator, busy with other tasks, to be notified as soon as a fault is detected. To this end, in another embodiment of the invention, the detection device can further comprise a visual and / or audible signaling device when the assembly Ω is not empty. The signaling device can be, for example, a buzzer or a mobile device, such as a smartphone or a telephone, on which a warning message is sent by an aerial telecommunication means.

To allow the location, by a human being, of the defects detected by the detection device over the width of the printed glass sheet, the detection device may further comprise a display device, decipherable by a human being, of the N Fourier transforms filtered in the form of an image of dimension Μ X N.

In a particular embodiment of the device of the invention, the calculation, filtering and search modules are virtual modules. For example, they can be modules instantiated in the form of objects by a computer program or computer software from classes in the random access memory, possibly assisted by a virtual memory, of a computer. The computer can include several central processing units, storage media and I / O interfaces. Where appropriate, it advantageously comprises telecommunication means with the acquisition module, the display device and the signaling device.

The computation of the discrete Fourier transforms is generally carried out using a fast Fourier transformation algorithm such as one of those mentioned above. However, the computation of the Fourier transforms and of the successive inverse Fourier transforms can require significant resources in calculation units and present some disadvantages when these resources are limited. It is also quite difficult to implement it by parallelism of the computing units. It also does not make it possible to filter the temperature signal on the "fly", without a step of calculating the Fourier transforms, which generally reduces the computational load for limited resources. It may then be advantageous to use an alternative calculation method called the kernel method, which is less precise but faster.

In the context of the invention, this method must be considered equivalent to that presented above, involving Fourier transforms. The kernel method requires the calculation of a single inverse Fourier transform corresponding to the kernel, of the transfer function, Ζ / (ω), of a bandpass filter. The filtered temperature signal is obtained by the convolution of said nucleus and the temperature signal in time space.

Z- + co

Τ * (ω) = Τ (ω) // (ω) pp τ * (ί) = (k * τ) (ί) = I fc (0) r (t - 0) άθ

J — CO with fc (t) = (β ^ίωί \ Η (ω))

By way of nonlimiting example, for a Gaussian bandpass filter with a transfer function

Η (ω) = ~ (ω + ω _£ ) ^{1 2} - (ω-ω _£ ) ² le 2σ ² le 2σ ²

Æ 2 V2ÏÏ the nucleus Zc (ü) is given by:

fc (t) = (β ^ίωί \ Η (ω))

Using the delay theorem, (e ^{iu> t} | // (ω - a> _c )) = e ^{iu> ct} (e ^iuit \ H (ω)}, it simply comes:

2V2tt (e

~ (ω + ω _£ ) ² - (ω-ω _£ ) ² ) the 2σ ² the 2σ ²

Æ 2 V2ÏÏ + e

- ^ = cos (M _c t) le ^{l (ùt}

2V2ÏÏ \

-g ² t ² ^ == eos (ù) _c t) e 2

The filtered temperature signal can then be obtained on the fly, that is to say without calculating the Fourier transforms of each of the N lines of the thermogram, by convolution:

fc (0) r (t - 0) άθ

When the kernel method is used, the severity criterion ψ ₀ must be adapted to the temperature signal in the time domain. In the time domain, after filtering the temperature signal, if PC faults are present, the high frequency or short period phenomenon associated with these faults should appear more intense. The severity criterion ψ ₀ can be, for example, a threshold for the value of the amplitude of the filtered signal.

Thus, in an alternative mode of the detection method of the invention, steps (a) to (c) are replaced by the following steps:

a ¹ , the computation, implemented by computer, of the nucleus of a band pass filter of natural frequency ü) _c = κω ₀ , where K is a definite positive real number greater than 1, and the value of the quality factor Q is equal to or greater than a defined value χ ₀ ;

b ¹ , the filtering, implemented by computer, of a thermogram of dimension Μ XN by calculating the convolution of the core of said filter with each of the N lines of the thermogram of dimension Μ XN, said thermogram being composed of M profiles of temperature of the surface of the printed glass, acquired at the outlet from the rolling rolls, in a second direction perpendicular to the first direction of movement of the printed glass at the outlet from the rolling rolls, with a digital spatial resolution of N pixels in said second direction, an acquisition frequency f _e of each upper profile at the frequency of rotation ωθ rolling rolls, and a pixel size equal to or smaller than the size <5 ₀ defects;

c ¹ . the search, implemented by computer, of the set Ω of the lines among the N lines of the filtered thermogram which satisfy a severity criterion ψ ₀ previously defined.

The features and advantages of the invention are illustrated by the figures and examples described below.

Figure 1 is a photograph of "no cats" defects on a sheet of printed glass.

FIG. 2 is a diagrammatic representation of the detection method of the invention.

FIG. 3 is a diagrammatic representation of an embodiment of the detection method of the invention.

FIG. 4 is an example of a thermogram of a sheet of glass printed at the outlet of the rolling rolls and having no "no cats" defects.

FIG. 5 is an example of a thermogram of a sheet of glass printed at the outlet of the rolling rolls and having "no cat" defects.

FIG. 6 is the spectrogram corresponding to the thermogram of FIG. 5 after filtering in frequency by a super-Gaussian filter of power 10 and of quality factor Q> 0.8.

FIG. 7 is the thermogram of FIG. 5 after filtering by a supergaussian filter of power 10.

FIG. 8 is an example of a spectrum extracted from the spectrogram of FIG. 6.

FIG. 9 is the spectrogram of FIG. 6 representing the set Ω of frequency values for which one or more filtered Fourier transforms satisfy a severity criterion ψ ₀ of threshold value of the intensity of their module.

FIG. 10 is a schematic representation of a first embodiment of the detection device of the invention.

FIG. 11 is a schematic representation of a second embodiment of the detection device of the invention.

Examples of PC no-cat faults appearing on a printed glass sheet after its manufacture are illustrated by the photograph in FIG. 1. In the figure, the printed glass sheet has been suspended by gripping means in order to facilitate observation of faults using an optical observation device. When observed with the naked eye or with the aid of an optical observation device under inclined or grazing lighting, the defects appear as periodic optical defects in the form of dark spots, sometimes elongated. In the figure, PC faults are indicated by arrows.

The detection method of the invention is illustrated in Figure 2 in the form of a diagram. The method of detecting a rolling defect in a glass printed in rectilinear movement in a first direction DI at the outlet of the rolling rolls comprises the following steps:

at. the calculation E200, implemented by computer, by the Fourier transformation of each of the N lines of a thermogram D200, a = (r _m , n) _MN GM _mw (IR), Ν, ΜΕΝ, of dimension MxN, said thermogram being composed of M temperature profiles of the surface of the printed glass, acquired at the output of the rolling rolls, in a second direction D2 perpendicular to the first direction Dl, with a digital spatial resolution of N pixels in said second direction D2, a acquisition frequency f _e of each profile greater than the frequency ωθ of rotation of rolling rollers, and a pixel size equal to or less than the dimension <5 ₀ of the defects:

M-l

Σ ωί

T _n (t) e "ω = 0

Γ „. M „M. 1. ..

ω G 10,1, ..., - - 1, - -, ..., —Il if M is even, or r- Γη Λ (M-l) λ (M-l) -il -.

ω G ΙΟ, 1, ..., —--- 1, ---—, ..., - 11 if M is odd;

b. the filtering E201, implemented by computer, by a bandpass filter of natural frequency (at _c = κωθ, of each of the N Fourier transforms, where K is a definite positive real number greater than 1, and the value of the factor of quality Q is equal to or greater than a defined value χ ₀ :

Τ _η * (ω) = Τ _η (ω) // (ω), Vn G {0, ..., 1V} m _c = κω ₀

Q> Xo;

vs. the search E202, implemented by computer, of the set Ω of frequency values for which one or more of the N filtered Fourier transforms satisfy a severity criterion ψ ₀ previously defined:

Ω = [ω | T _n (ω) satisfies ψ ₀ }.

FIG. 3 illustrates, in diagram form, a particular embodiment of the method of the invention. The process includes the following steps:

at. the calculation E200, implemented by computer, by the Fourier transformation of each of the N lines of a thermogram D200 of dimension Μ XN, said thermogram being composed of M temperature profiles of the surface of the printed glass, acquired at the output of rolling rollers, in a second direction D2 perpendicular to the first direction Dl, with a digital spatial resolution of N pixels in said second direction D2, an acquisition frequency f _e of each profile greater than the frequency ωθ of rotation of rollers rolling, and a pixel size equal to or less than the dimension < ₅₀ of the defects:

Μ-l

Σ ωί

T _n (t) e "ω = Ο

Γ „. Μ 4 Μ „1” 4 ω G ΙΟ, 1, ..., - - 1, - -, ..., —It if Μ is even, or

,. r- Γη Λ (Μ-1) λ (Μ-1) -11 1.4. · Ω G ΙΟ, 1, ..., —--- 1, ---—, ..., - 11 if Μ is odd;

b. the definition of a super-Gaussian power filter having the transfer function:

/ f (ω + ω _£ ) 0 ° f (ù>-ù> _c ) ^ ¹⁰ \

Η (ω) = C iev σ J + ev ^σ J

vs. the filtering E201, implemented by computer, by a bandpass filter of natural frequency ü) _c = κωθ, of each of the N Fourier transforms, where K is a definite positive real number greater than 1, and the value of the factor of quality Q is equal to or greater than a defined value χ ₀ :

Τ _η * (ω) = Τ _η (ω) // (ω), Vn G {0, ..., Α] m _c = κω ₀

Q> Xo;

d. the graphic representation E302 of the filtered spectrogram A * = (T * _{m> n} ) GM _m . _w (Z), Ν, ΜΕΝ-,

e. the search E202, implemented by computer, of the set Ω of frequency values for which one or more of the N filtered Fourier transforms satisfy the criterion of severity Ψο · The criterion of severity Ψο is a threshold value of the intensity of the module fixed at ε in base 10 logarithm. The set Ω is the set of frequencies for which log ₁₀ (| T * _mn |)> ε:

Ω = {ω | log ₁₀ (| T * _mn |)>ε];

f. E303 visual or audible signaling when the Ω assembly is not empty.

FIG. 4 represents an example of a thermogram obtained by the acquisition at 20 Hz, at the output of the rolling rolls, of 1024 temperature profiles of the surface of the printed glass devoid of PC defect, with a digital spatial resolution of 1000 pixels in the second direction D2 perpendicular to the first direction DI of the movement of the glass at the outlet of the rolling rolls. The temperature profiles are represented on the abscissa in the time domain. The position, expressed in pixels, is represented on the ordinate for

890 of the 1000 pixels. The grayscale scale represents the temperature scale. The rotation of the laminating rollers causes periodic variations in temperature, which are manifested in particular by the periodic appearance of regions of the glass surface where the temperature is higher.

FIG. 5 represents an example of a thermogram obtained by the acquisition at 20 Hz, at the output of the rolling rolls, of 1024 temperature profiles of the surface of the printed glass having PC defects, with a digital spatial resolution of 1000 pixels in the second direction D2 perpendicular to the first direction Dl of the movement of the glass at the outlet of the rolling rolls. The temperature profiles are represented on the abscissa in the time domain. The position, expressed in pixels, is represented on the ordinate for 890 of the 1000 pixels. The grayscale scale represents the temperature scale. The rotation of the laminating rollers causes periodic variations in temperature, which are manifested in particular by the periodic appearance of regions of the glass surface where the temperature is higher. Between pixels 150 and 200 on the ordinate, a line of PC faults in the form of a periodic phenomenon of high frequency is visible. The analysis of the thermogram as such does not make it possible to detect and characterize this phenomenon with sufficient sensitivity for this detection to be precise.

FIG. 6 represents the spectrogram corresponding to the spectrogram of FIG. 5 after filtering in frequency by a super-Gaussian filter of power 10 and of quality factor Q> 0.8 at 3dB. The position, expressed in pixels, is represented on the ordinate for 816 of the 1000 pixels. The frequency is shown on the abscissa. The grayscale scale represents the scale of the values of the module of the Fourier transforms in arbitrary unit. On the spectrogram, the two zones located, on the ordinate, between 150 and 200 pixels and between 0 and 50 pixels, which have the highest module values for the same frequency correspond to two lines of PC faults. The position of these areas on the ordinate makes it possible to locate them precisely on the sheet of printed glass. It can be noticed that the invention makes it possible to detect a line of PC faults which was not a priori visible on the thermogram of FIG. 5.

FIG. 7 is the thermogram of FIG. 5 after filtering by a supergaussian filter of power 10. It was obtained by calculating the inverse Fourier transforms of the 816 lines of the filtered spectrogram of FIG. 6. With respect to FIG. 5, the PC fault lines located, on the ordinate, between 150 and 200 pixels and between 0 and 50 pixels, are clearly revealed.

FIG. 8 is an example of a spectrum in the positive frequency domain of a line of the spectrogram of FIG. 6 crossing the zone of high modulus between 150 and 200 pixels. The super-Gaussian power filter 10 advantageously isolates the characteristic frequency of the PC fault line. The width of the peak is related to the dimension of the PC faults.

FIG. 9 is the spectrogram of FIG. 6 representing the set Ω according to a severity criterion ψ ₀ of threshold value of the intensity of the module fixed at 2.7 in base 10 logarithm. The figure represents the spectrogram of FIG. 6 after binarization at the threshold value 2.7. The points for which the module value is less than 2.7 appear in black and the points for which it is greater than 2.7 appear in clear.

A first embodiment of the detection device of the invention is shown diagrammatically in FIG. 10. The device for detecting a rolling defect in a glass printed in rectilinear movement in a first direction DI at the outlet of the rolling rolls, comprises the following modules:

a computation module 1007a of a spectrogram by the Fourier transformation of each of the N lines of a thermogram of dimension Μ XN, said thermogram being composed of M temperature profiles of the surface of the printed glass, acquired at the output of the rolls of rolling, in a second direction D2 perpendicular to the first direction Dl, with a digital spatial resolution of N pixels in said second direction D2, an acquisition frequency f _e of each profile greater than the frequency ωθ of rotation of rolling rollers, and a pixel size equal to or less than the dimension < ₅₀ of the defects;

a filter module 1007b by a natural frequency bandpass filter O) _c = κωθ of each of the N Fourier transforms, where K is a defined positive real number greater than 1, and the value of the quality factor Q is equal to or greater than a defined value χ ₀ ;

a search module 1007c of the set Ω of frequency values for which one or more of the N filtered Fourier transforms satisfy a severity criterion ψ ₀ previously defined.

In the present embodiment, the calculation modules 1007a, filtering 1007b and research 1007c are virtual modules. They are modules instantiated in the form of objects by a computer program or computer software from classes in the random access memory 1007, possibly assisted by a virtual memory, of a computer 1006.

The computer can include several central processing units, storage media and I / O interfaces. It also includes telecommunication means 1005 and 1008 with the acquisition module 1004, the display device 1009 and the signaling device 1010.

For purely illustrative and nonlimiting purposes, the acquisition module 1004 can be an infrared beam scanning digitizer. The acquisition module 1004, digitizes, using an infrared laser line 1004a, the surface 1001b of a printed glass sheet 1001a having defects 1003 at the output of the rolling rolls 1002a and 1002b. The signaling device 1010 can be an audible warning, for example a siren. It can be replaced or supplemented by a visual signaling device such as an indicator light. The display device 1009 can be a screen provided with a graphical interface on which information decipherable by a human being is displayed. This information can be filtered and unfiltered thermograms and / or spectrograms, or information relating to the detection of PC faults. The computer 1006 can also communicate with a physical or virtual input interface (not shown) so as to interact with the detection device by communicating data to it.

In the present embodiment of the detection device, the modules and elements other than the calculation modules 1007a, filtering 1007b and search 1007c are optional. In Figure 10, they are shown for illustrative and indicative purposes. It should be noted that they can be adapted according to the needs of those skilled in the art who would like to implement the device on a printed glass manufacturing installation.

A second embodiment of the detection device of the invention is shown in FIG. 11. In this mode, the signaling device 1101 is one or more mobile devices, for example, one or more smartphones, in aerial telecommunications with the computer 1006. When a PC fault is detected, the computer can send an audible and / or visual message to the mobile device 1101 to warn an operator. This message can also contain information such as an extract from a filtered and binarized spectrogram so as to allow an operator present on the production line to locate the PC defect on the sheet of glass and to intervene directly. Aerial telecommunications can be a WiFi or Bluetooth type telecommunications, or even by a cellular network.

In the second embodiment of the detection device, the modules and elements other than the calculation modules 1007a, filtering 1007b and research 1007c are shown for illustrative and indicative purposes.

Example

The advantages of the invention are illustrated by the example described below. A thermogram of the surface of a printed glass presenting PC defects was acquired at the output of the rolling rolls using an infrared beam scanning digitizer such as the infrared laser scanner LSP-HD 50 from the company LAND Metek. The rotation frequency ωθ of the rollers is 0.07 Hz.

The thermogram includes 1024 temperature profiles acquired with a frequency f _e of 20 Hz, with a digital spatial resolution of 1000 pixels in the second direction D2 perpendicular to the first direction DI of the movement of the printed glass at the outlet of the rolling rolls. The thermogram a = ( ^T m, n) ₁₀₂ 4 ₈₉₀ ^ 1024.890 ^with coefficient the T _mn which represents the value of the temperature at the coordinates m, n of the thermogram, is represented in figure 5. The temperature profiles are represented on the x-axis in the time domain. The position, expressed in pixels, is represented on the ordinate for 890 of the 1000 pixels. The grayscale scale represents the temperature scale.

Each of the 890 lines includes 1024 measurement points. The calculation of the discrete Fourier transform of each of the 890 lines was carried out using a fast Fourier transform calculation algorithm, according to the following formula, with T _n (t) = T _mn , t G { 0, ..., 1024- 1} the 1024 terms of line n:

M-l

Σ Ojt τ _η (ύ) β ~ ^ιλπ ^ ω = 0 ω e 0,1, ..

1024

1024

A super-Gaussian bandpass filter of power 10 with a quality factor Q = 0.84 to 3 dB was applied to each of the 890 discrete Fourier transforms according to the formula:

Τ _η * (ω) = Τ _η (ω) // (ω), n G {0, ..., N} with // (ω) defined by

a> _c = κω ₀ = 0.308 κ = 4.5 σ = 10

Figure 6 shows the filtered spectrogram. In this figure is represented, in the positive frequency domain, the module (| T * _mn |) _{0 <m <1024 890} ^c ' ^{c rnalTlcc} & * = ( ^T * m, n) _{1024 890} ^{e W} lO24.89o ( ^) ·

The position, expressed in pixels, is represented on the ordinate for 890 of the 1000 pixels. The frequency is shown on the abscissa. The grayscale scale represents the scale of the values of the modulus of the discrete Fourier transforms in base 10 logarithm of an arbitrary unit. Two clear areas appear on the spectrogram. In these zones, the module values are maximum. They are located, on the ordinate, between 150 and 200 pixels and between 0 and 50 pixels, in the frequency domain filtered from 0.13 Fiz to 0.49 Fiz by the bandpass filter.

An example of a spectrum is provided by FIG. 8 in the positive frequency domain of a line of the spectrogram of FIG. 6 crossing the zone of high modulus between 150 and 200 pixels. The super-Gaussian power filter 10 advantageously isolates the characteristic frequency of the PC fault line. The width of the peak is related to the dimension of the PC faults.

FIG. 7 is the thermogram obtained by calculating the inverse discrete Fourier transforms of the 890 lines of the filtered spectrogram of FIG. 6.

The severity criterion ψ ₀ used for the search for the set Ω is a threshold value of the intensity of the module fixed at 2.7 in logarithm base 10. In other words, the set Ω is the set of frequencies for which l ° gio (| T * m, n |) - 2.7, that is Ω = {ω | log ₁₀ (| T * m, n I) £ 2.7} _ Figure 9 is the spectrogram of Figure 6 representing the set Ω. In the present case, it was obtained by binarization image processing at the threshold value 2.7. The points for which the module value is less than 2.7 appear in black and the points for which it is greater than 2.7 appear in clear.

The two zones located, on the ordinate, between 150 and 200 pixels and between 0 and 50 pixels, in the frequency domain filtered from 0.13 Hz to 0.49 Hz satisfy this criterion.

- 22 This example clearly shows that the detection method of the invention makes it possible to detect PC faults appearing on a printed glass at the outlet of the rolling rolls according to a defined severity criterion.

Claims (17)

claims 1. Method for detecting rolling defects in a printed glass moving in a first direction (Dl) at the outlet of the rolling rolls, said method comprising the following steps: 5 a. the computation, implemented by computer, of a spectrogram by the Fourier transformation of each of the N Egnes of a thermogram of dimension Μ XN, said thermogram being composed of M temperature profiles of the surface of the printed glass, acquired in exit of the rolling rolls, in a second direction (D2) perpendicular to the first direction (Dl), with a digital spatial resolution of N pixels in said second direction (D2), an acquisition frequency f _e of each upper profile at the frequency ü) ₀ of rotation of rolling rollers, and a pixel size equal to or less than the dimension δ ₀ of the defects; 15. b. the filtering, implemented by computer, by a bandpass filter of natural frequency W _c = κωθ, of each of the N Fourier transforms, where K is a definite positive real number greater than 1, and the value of the quality factor Q is equal to or greater than a defined value χθ; vs. the search, implemented by computer, of the set Ω of frequency values for which one or more of the N transforms of Fourier filtered meet a severity criterion ψ ₀ previously defined. 2. Detection method according to claim 1, such that it further comprises, before step (a), a step of acquisition, at the outlet of the rolling rolls, of M temperature profiles of the surface of the printed glass. , in the second direction 25 (D2) perpendicular to the first direction (Dl), with a digital spatial resolution of N pixels in said second direction so as to create a thermogram of dimension Μ X N. 3. Detection method according to claim 2, such that the acquisition step is carried out using an infrared beam scanning digitizer. 30 4. Detection method according to any one of the preceding claims, such that the bandpass filter is a super Gaussian filter of power 10. 5. Detection method according to any one of claims 1 to 4, such that the acquisition frequency f _e is between 20 and 150Hz, preferably between 20 and 100Hz. 6. Detection method according to any one of claims 1 to 5, such that the number M of profiles is a power of two, 2 ^P , with p an integer between 9 and 15, preferably between 10 and 14. 7. Detection method according to any one of claims 1 to 6, such that the value of the parameter K is between 3 and 6, or even between 4 and 5. 8. Detection method according to any one of claims 1 to 7, such that the severity criterion ip _Q is chosen for one or more peaks from a threshold value of the intensity, a threshold value of the area under peak or a threshold value of eccentricity of a level line. 9. Detection method according to any one of claims 1 to 8, as it comprises, after step (b), a step of displaying the N filtered Fourier transforms in the form of a dimension image Μ x N. 10. Detection method according to any one of claims 1 to 9, as it further comprises, after step (c), a step of visual and / or audible signaling when the assembly Ω is not empty. 11. Computer program executable by a computer comprising instructions allowing the execution of the steps of a detection method according to one of claims 1 to 10. 12. A storage medium decipherable by a computer comprising a computer program according to claim 11. 13. Device for detecting a rolling defect in a printed glass moving in a first direction (Dl) at the outlet of the rolling rolls, said device comprising the following modules: a module for calculating a spectrogram by the Fourier transformation of each of the N lines of a thermogram of dimension M x N, said thermogram being composed of M temperature profiles of the surface of the printed glass, acquired at the output of the rolls of rolling, in a second direction (D2) perpendicular to the first direction (Dl), with a digital spatial resolution of N pixels in said second direction (D2), an acquisition frequency f _e of each profile greater than the frequency ω ₀ rolling rollers, and a pixel size equal to or less than the dimension δ ₀ of the defects; a filter module with a natural frequency band pass filter C0 _c = kîüq of each of the N Fourier transforms, where K is a definite positive real number greater than 1, and the value of the quality factor Q is equal to or greater than a defined value; a search module for the set Ω of frequency values for which one or more of the N filtered Fourier transforms satisfy a severity criterion ψ ₀ previously defined. 14. Detection device according to claim 13, such that it further comprises a module for acquiring, at the output of the rolling rolls, M temperature profiles of the surface of the printed glass, in the second direction (D2) perpendicular to the first direction (Dl), with a digital spatial resolution of N pixels in said second direction (D2) so as to create a thermogram of dimension MxN. 15. Detection device according to claim 14, such that the acquisition module comprises an infrared beam scanning digitizer. 16. Detection device according to any one of claims 13 to 15, as it further comprises a display device, decipherable by a human being, of the N Fourier transforms filtered in the form of a dimension image Μ X N. 17. Detection device according to any one of claims 13 to 16, such that it further comprises a visual and / or audible signaling device when the assembly Ω is not empty.