WO2000007145A1

WO2000007145A1 - A production line quality control method and apparatus

Info

Publication number: WO2000007145A1
Application number: PCT/IB1999/001335
Authority: WO
Inventors: Antonio Messina
Original assignee: Carretti, Ettore; Rosanelli, Stefano; Sancese, Silvio; Torrisi, Manuele
Priority date: 1998-07-30
Filing date: 1999-07-28
Publication date: 2000-02-10
Also published as: EP1105841A1; ITBO980471A1; IT1306273B1

Abstract

In a method and apparatus for controlling quality on a production line, the parameter (IM) that constitutes the distance between the image of the product and a reference image is computed and the product is considered acceptable if the distance computed falls within a defined interval (i).

Description

A production line quality control method and apparatus

Technical Field

The present invention relates to a method and apparatus for controlling quality on a production line.

Background Art

In many fields of industrial production, quality control is currently done by computerized systems which use one or more video cameras to acquire images of the products made and compare these images with stored reference images. The fields of application for systems such as these are, for example, the following:

- the manufacture of tiles in the ceramic industry;

- the manufacture of smart cards with magnetic strips or microchips; - the sorting of objects by robots;

- special printing applications (banknotes, art prints, fabrics) .

Typically, during the manufacturing process, defects of different kinds may be introduced in the structure of the product (irregular edges, uneven surfaces, blistering, pitting, cracking, rolling defects, incorrect position of components, etc.) or in the design printed on it (irregularities, non-uniformity of colour shades, stains, print defects, etc.). Moreover, when the product is inspected using a video camera, other apparent defects are usually introduced by the image capturing system as a result of random fluctuations in pixel intensity.

Disclosure of the Invention

The aim of the present invention is to provide a method and an apparatus to automatically and instantaneously identify the production defects of an object with reference to preset control parameters, irrespective of the noise introduced by the image capturing system.

In particular, the invention has for an aim to provide a computerized quality control system for a production line whose sensitivity in identifying defects in a product is similar to and not less than that of the human eye.

Another aim of the invention is to provide a system that is capable of measuring the differences between the image of a product and a reference image of the product and, hence, to identify products whose images differ from the reference image to an extent greater than a preset threshold value, thus being capable of sorting products according to quality classes defined at the very beginning of production with the procedure described below. Yet another aim of the invention is to provide a system for comparing images acquired by a video camera with a reference image and applying a single procedure to identify differences in geometry, flatness, pattern, or colour, between the images acquired and the reference images, even in the presence of translations, rotations or scale variations.

The above mentioned aims are achieved by a method and apparatus as described in the claims herein.

Further advantages and technical characteristics of the present invention are described in more detail below, with reference to the accompanying drawings, which illustrate a preferred embodiment of the invention without restricting the scope of the inventive concept, and in which:

- Figure 1 schematically illustrates an apparatus made according to the present invention; - Figure 2 is a flow chart which schematically illustrates the procedure used by the apparatus made according to the present invention;

- Figure 3 is a schematic axonometric view of an apparatus constituting another embodiment of the present invention; - Figure 4 is a schematic cross section of the apparatus illustrated in Figure 3; - Figure 5 is a schematic cross section of another embodiment of some of the parts of the apparatus illustrated in Figure 1;

- Figure 6 is a schematic axonometric view of some of the parts of the apparatus illustrated in Figure 5; - Figure 7 schematically illustrates yet another embodiment of the invention, comprising some of the parts shown in Figures 3 and 4 and others shown in Figures 5 and 6.

With reference to Figure 1, an apparatus made according to the present invention essentially comprises a video camera 1, a lighting system 2, a frame grabber 3 and a processing and control unit 4. The video camera 1 may be of the linear, colour type, with medium luminosity lenses and a colour temperature conversion filter. The video camera 1 is positioned above a conveyor belt 5 on which the products 21 (for example, ceramic tiles) lie, and is preferably pivotally mounted so that it can be directed in the ideal direction to capture the best images according to the product conveying system.

Further, the video camera 1 is preferably mounted in such a way that its height can be varied according to the size of the objects it has to focus on and according to the optical characteristics of the system.

The lighting system 2, which may consist of halogen lamps 20, is made in such a way as to appropriately illuminate the object, depending on the inspection to be made. For example, to identify differences in colour shades, the light beam must be perpendicular to the surface of the object. The system may also advantageously comprise a dichroic filter to improve the lighting of the field by the lamps 20, a lenticular array and means 30 for tilting the lamps 20 relative to the vertical. In particular, in the embodiment illustrated in Figures 5 and 6, the lighting system 2 comprises a plurality of halogen lamps 22 housed in a centring mask 23, a dichroic filter 24, a lenticular array 25, and means 26 for cooling the lamps 22. The lamps 22 are connected to the power supply (not illustrated) through connectors 28 and are fixed to the centring mask 23 by springs 27. For example, the lamps 22 may be positioned in the centring mask at a distance L of approximately 110 mm from each other and at a distance HI of approximately 10-15 mm from the dichroic filter 24. The dichroic filter 24 is in turn approximately 10-15 mm from the lenticular array 25. In this embodiment, too, there are means 30 for tilting the lighting system 2 relative to the vertical.

In the embodiment illustrated in Figures 3 and 4, the lighting system 102 comprises a diffuser 121 which makes it possible to suitably illuminate the field to be imaged with fluorescent lamps 120. Moreover, as shown in Figure 7, the two lighting systems 2 and 102 can be combined in a single production line: the halogen lamps 20 of the lighting system 2 are more effective in detecting the geometric defects (flatness, etc) of the product 21, while the fluorescent lamps 120 of the lighting system 102 more effectively detect defects in the tone of the products 21 (colour and tone) .

The two systems 2, 102 can be positioned in sequence in the direction of feed of the conveyor belt 5, as shown in Figure 7, or they may be combined into one (not illustrated) .

In the embodiment illustrated in Figures 3 and 4, there is also a video camera of the linear type, which captures the images through an aperture of width W made in the diffuser 121. The aperture may, for example, measure 3-4 mm in width W.

If the image is acquired by a video camera of the linear type, a reflective photocell positioned at right angles to the conveyor belt may be envisaged. This permits generation of Start and Stop signals, necessary to restrict image acquisition only to the parts required and to synchronize the conveyor with image acquisition.

Both the frame grabber 3 and the processing and control unit 4 may be selected according to the specific requirements of the production line concerned. In particular, the unit 4 may consist of a personal computer.

The apparatus works as described below and as illustrated in

Figure 2. At the start of production, two or more sample images of products without defects are acquired and stored under the same conditions. A reference image is then made by "averaging" the sample images. The "average" may, for example, be obtained by calculating the average of the pixel values of the acquired images. For example, if a virtual image created with computer graphics tools is available, the sample images can be generated computationally by summing the virtual image with the noise produced by the imaging system.

Next, a parameter IM (described in more detail below) is computed, for example for each sample image acquired or generated, as a measure of the "distance" from the reference image. The apparatus then computes the average value <IM> and the corresponding error σ of the IM values computed compared to a statistical distribution chosen in accordance with the quality control criteria that are to be applied to the product.

For example, if the defects are known (or assumed) to be distributed according to normal distribution and we want to identify all the products whose images differ from the reference image beyond a certain limit, then the average <IM> is the arithmetic mean of the IM values and σ the mean square deviation. The error computed is used advantageously to define an interval i(σ) (for example, i = 2σ) of acceptable values for the subsequent parameters IM for the products made .

It is stressed that this technique can also be used, for example, on a tile production line to sort tiles according to tone. To simulate the sorting that can be performed by the human eye (which, as is known, follows the Weber-Fechner law in response to external luminous input) it is sufficient to consider both the average magnitude in each acquisition channel of a colour video camera, and the colours, to obtain sorting intervals and averages similar to those obtained by the human eye. The magnitude (M) is defined as the logarithm of the luminous flux (F) received on the different channels of a colour video camera (in the red M_R=log F_R, in the blue M_B=log F_B, in the green M_s=log F_G) . The colour is the difference in magnitude in two channels (for instance,

. To obtain averages and sorting intervals that conform with the response of the human eye, two or more sample images of products with the required tone are required and stored. In this context, the term "tone" means the combination of brightness, intensity and colour. Next, the magnitudes in the different acquisition channels of the video camera are computed and averaged. Lastly, the colours as C_I=M_G-M_B, C₂=M_G-M_R and c₃=c₂-C_ι are computed and averaged. Here, the green magnitude has been used as reference since the human eye performs better in this visual band. If a more sensitive response to another colour channel is desired, the magnitude in the corresponding acquisition channel should be used as reference to compute the colour. This procedure applied to the sample images will produce values of <IM> and σ and will permit the definition of intervals for the sorting of other products .

The computational algorithm used to detect geometrical defects is based substantially on the technique of comparison between two-dimensional images s(x,y) and r(x,y) known as Symmetric Phase-Only Matched Filtering of Fourier-Mellin invariant image descriptors (FMI-SPOMF) (see Q.S. Chen et al . , Symmetric Phase-Only Matched Filtering of Fourier-Mellin Transforms for Image Registration and Recognition, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 16, No.12, December 1994). This document is incorporated by reference into the specification of this application. According to this technique, the phases of the two images are extracted and matched by applying the inverse Fourier transform to the correlation between the transforms S(u,v) and R(u,v) of the two images - or their spectra map in polar, logarithmic spaces - normalized by their amplitudes . The inverse transform provides a measure of the "distance" between the images and, in particular, in the continuous case, gives a Dirac δ function which permits identification of translations, rotations or scale factors.

Advantageously, according to the present invention, the correlation of the Fourier transform is filtered by means of a Gaussian function. Thus, performing the inverse transform, in the discrete case, provides a function consisting of the convolution of a Gaussian function with a Dirac 5 function and having a peak centred on the values of the translation, in the plane, or on the values of the rotation and scale variation. The parameter IM used is proportional to the value of the peak of the inverse transform function.

Using the Gaussian function as a filter has the advantage of eliminating high values of (u,v) and hence the noise created by the imaging system on small scales, thus permitting a more accurate identification of the peak of the inverse transform function.

Thus, once production has started, the images of the objects being made are in turn acquired and digitally processed in such a way as to compute and correct any translations, rotations or scale changes relative to the sample image. As illustrated in Figure 2, in this case, the unit 4 must permit parallel processing of the algorithm. If the type of product being made (for example, tiles or smart cards) and or the mechanics of the system can guarantee that the product is not translated, rotated or changed in scale, the computational steps described above are not necessary and the quality inspection can therefore be performed more quickly.

For each image captured and, if necessary, processed to produce the reference image, the parameter IM is computed. If its value falls within the preset interval i, then the product is considered to be free of defects, otherwise it is identified as being of unacceptable quality. Considering the statistical nature of the criterion used, the fact that the value of the parameter IM computed falls within the preset interval i guarantees the absence of defects within the percentage limits corresponding to the amplitude of the interval.

Obviously, the choice of the distribution function for IM determines the control criteria applied. In particular, the quality inspections permitted by the system disclosed by the present invention are substantially the same as those that can be made visually by a human operator, identifying only defects that are visible to a human eye. Moreover, If two or more intervals of values are defined for IM, it is possible to sort products by quality. Thus, on the basis of the values computed for IM for every image acquired, the processing and control unit 4 can send commands downstream of the production line to automatically route the products, for example, tiles) of different quality to different collection points.

Tests have shown that the system disclosed is capable of detecting even the smallest defects in products, independently of the noise introduced by the imaging system. The advantage of this is that even low-cost video cameras can be used without adversely affecting the efficiency of the system.

Since the algorithms used can be easily and very effectively ported on parallel hardware architecture, the time taken to perform the quality inspections can be advantageously reduced using a computer system comprising two or more processors. For example, by using four processors instead of one, approximately one quarter of the time is required.

Claims

1. A method for controlling the quality of production line by comparing the acquired image of a product with a reference image, characterized in that it comprises the following steps:

- finding an acceptability threshold, depending on the type of quality control required, by defining an interval (i) of values for parameter (IM) designed to measure the distance between the image acquired and the reference image;

- sequentially acquiring the images of the products on the line;

- for each image acquired, computing the distance (IM) from the reference image;

- considering the product acceptable if the distance (IM) computed falls within the defined interval (i) or considering the product unacceptable if the distance (IM) computed does not fall within the defined interval (i) .

2. The method according to claim 1, characterized in that the acceptability threshold is defined by determining the average value (<IM>) and the corresponding error (╧â) of the parameter (IM) compared to a preset statistical distribution function, the interval (i) being chosen in accordance with the error (╧â) .

3. The method according to claim 1 or 2 , characterized in that the reference image consists of the average of two or more sample images of products whose quality is acceptable.

4. The method according to claim 1 or 2 , characterized in that the reference image consists of the average of two or more sample images obtained by summing a virtual image with the noise produced by the imaging system.

5. The method according to claim 1, characterized in that two or more classes of products corresponding to the same number of intervals of the parameter (IM) are defined and each product is automatically assigned to the class corresponding to the interval in which the value of the parameter (IM) computed for the image of the product falls .

6. The method according to any of the foregoing claims, characterized in that the parameter (IM) that measures the distance between the acquired image s(x,y) of a product and the reference image r(x,y) is proportional to the height of the peak of the inverse transform of the correlation of the Fourier transforms S(u,v), R(u,v) of the two images - or their spectra map in polar, logarithmic spaces - normalized by their amplitudes, a filter consisting of a Gaussian function being applied to said correlation.

7. The method according to any of the foregoing claims, characterized in that the parameter (IM) measures the distance between the magnitudes (M_R, M_B, MQ) and the colours (c_x, c₂, c₃) of a product (21) to be inspected and the corresponding average quantities of the reference image.

8. An apparatus for controlling the quality of a production line comprising at least one video camera (1) , a lighting system (2, 102; 2, 102) for illuminating the product (21), a frame grabber (3) and a processing and control unit (4), and characterized in that it works on the basis of the method described in claims 1 to 6.

9. The apparatus according to claim 8, characterized in that the lighting system (2) comprises a plurality of halogen lamps (20) housed in a centring mask (23), a dichroic filter (24), a lenticular array (25) and cooling means (26) .

10. The apparatus according to claim 8, characterized in that the lighting system (102) comprises a plurality of fluorescent lamps equipped with a diffuser (121) .

11. The apparatus according to claim 8, characterized in that the video camera (101) is of the linear type.

12. The apparatus according to claim 8, characterized in that the lighting system (2, 102) comprises a first system (2) equipped with halogen lamps (20) housed in a centring mask (23), a dichroic filter (24), a lenticular array (25) and cooling means (26), and a second system (102) equipped with fluorescent lamps (120) with a diffuser (121); the two systems (2, 102) being positioned in sequence in the direction of feed of the conveyor belt (5) , or being combined into one.

13. The apparatus according to any of the foregoing claims from 8 to 12, characterized in that it comprises means for sorting the products (21) on the basis of the computed value of the parameter (IM) .