Patent Application
20190047895
The present disclosure relates to methods and apparatus for inspecting edge surface quality of a moving web of glass material.
Continuous processing of ultra-thin glass web, for example glass web measuring less than or equal to about 0.3 mm in thickness, is a relatively new field and presents a number of manufacturing challenges. A conventional process for producing such webs includes employing a roll-to-roll technique in which a glass web is conveyed in a continuous transport between a supply roll and a take-up roll. To produce final products, for example glass for flat panel displays or other products, the glass web must be cut or sliced during the roll-to-roll conveyance of the glass web. A laser cutting technique (or other suitable cutting technique) may be employed to slit the glass web to remove bead portions (i.e., thickened portions that are located at the peripheral edges of the glass web that occur when forming the web) during transport. The glass web may also be cut during roll-to-roll conveyance to achieve desired width dimensions for later processing.
The final piece parts delivered to customers often must exhibit very smooth, particle free edges, with minimal edge defects and/or edge corner defects. After removal of the beads and/or cutting the web to width, however, the quality of the edge surface(s) might not be within tolerances. Conventional approaches for cutting and inspecting the glass web, however, have not provided the ability to inspect and evaluate edge surface quality during the roll-to-roll conveyance of the glass web in a continuous transport system.
Accordingly, there are needs in the art for new methods and apparatus for inspecting edge surface quality of a moving web of glass material.
The present disclosure relates to methods and apparatus for inspecting edge surface quality of a moving web of glass material, for example during the removal of beads and/or during the cutting of the glass web to desired widths.
Whether a laser cutting technique is employed or some other cutting technique, edge surface defects generally occur randomly as they are the result of imperfect process parameters and/or varying conditions during the cutting and transport process. It is generally understood that edge surface defect types may be classified into the following categories: chips (see,
During a roll-to-roll, or continuous transport cutting process, it would be highly advantageous to be able to perform real-time edge surface inspection, quantification of edge surface defects (or quantification of edge surface quality). The prior state of the art, however, does not permit real-time edge surface inspection and quantification capabilities. Thus, edge surface inspection and quality assessments are randomly checked off-line using suitable techniques, for example by way of automated high resolution microscope systems, which can generate edge surface images for specially prepared samples. Such systems, however, have proven to exhibit very limited speeds and, thus, are used for a very small number of samples. Moreover, the images produced by the commercially available high resolution microscope systems must be interpreted by a trained scientist, which is very tedious, expensive, and exacerbates an already slow process. Due to the tedious and excessively slow inspection techniques available to production personnel, many operators prefer to simply slide their fingers along the edge surface of a cut glass web to obtain what limited edge surface quality information may be available from a tactile inspection.
Whether a sophisticated high resolution microscope system is employed or whether a tactile inspection is performed, the results are either for far fewer than a 100% real-time inspection or are so crude as to be of questionable value. Consequently, present production techniques fail to include real-time, systematic and reliable defect quantification in connection with determining edge surface quality in continuous conveyance glass web cutting processes.
In accordance with one or more embodiments herein, new methods and apparatus have been developed in which an inline glass edge inspection system is employed to measure, identify, classify, and quantify edge surface defects in the glass web in real time. The inspection system may include back lit illumination of the edge surface of the glass web, high resolution optical imaging of such edge surface, mechanically driven in situ auto-focusing, and a defect classification and quantification algorithm. The defect classification and quantification algorithm analyzes the brightness contrast of the edge surface images to identify, classify, and quantify various defects on or in the edge surfaces.
Advantages and benefits of one or more embodiments herein include any of the following:
Can provide in situ process feedback capability (to change process parameters) for a glass web roll-to-roll cutting process involving edge slitting, bead removal, edge grinding, and/or chamfering.
Can provide 100% inspection of edge surfaces, which allows for the capture of transient events, trends, etc.
Can provide instantaneous feedback to the glass edge shaping and separation processes, which permits modifications to processing parameters to improve edge surface quality.
Can provide a quality control tool to capture statistical drift of a continuous conveyance process.
Can provide a non-destructive, non-intrusive, automated edge surface inspection process, which is tolerant of three dimensional (3D) glass web motion.
In accordance with one or more embodiments, methods and apparatus disclosed herein may provide for: sourcing a glass web, the glass web having a length and a width transverse to the length; continuously moving the glass web from the source to a destination in a transport direction (also known as a down-web direction) along the length of the glass web; cutting the glass web, at a cutting zone, along the length into at least first and second glass ribbons as the glass web is moved in the transport direction from the source to the destination, such that respective first and second edge surfaces are produced on the first and second glass ribbons; and optically inspecting at least one of the first and second edge surfaces in real-time as the first and second glass ribbons of the glass web are moved in the transport direction to the destination.
In accordance with one or more embodiments the inspecting operation(s) may include one or more of: (i) taking at least one image of the at least one of the first and second edge surfaces as the first and second glass ribbons of the glass web are moved in the transport direction, (ii) extracting one or more features of the at least one of the first and second edge surfaces from the at least one image, and (iii) detecting one or more defects, and identifying one or more types of the one or more defects, based on the one or more extracted features.
The types of the one or more defects may include chips, hackles, Wallner lines, arrest lines, frictive damage, and scratches.
The methods and apparatus may further provide for one or more of: directing incident light onto and through an opposing edge of at least one of the first and second glass ribbons that is laterally opposite to (also known as a cross-web direction from) the at least one of the first and second edge surfaces; propagating the light through the at least one of the first and second glass ribbons, transversely with respect to the transport direction, such that the light exits through the at least one of the first and second edge surfaces; and directing an optical axis of an imaging sensor substantially perpendicularly toward the at least one of the first and second edge surfaces, to receive the light exiting the at least one of the first and second edge surfaces, such that the imaging sensor produces the at least one image.
Additionally or alternatively, the methods and apparatus may further provide for one or more of: monitoring a distance from the imaging sensor (and/or a reference position) to the at least one of the first and second edge surfaces as the first and second glass ribbons of the glass web are moved in the transport direction to the destination; and automatically adjusting a position of focus of the imaging sensor as a function of the distance such that the at least one image remains in focus.
In accordance with one or more embodiments, the methods and apparatus may further provide for detecting the one or more defects, and identifying the one or more types of the one or more defects, to include one or more of: enhancing one or more defect features as compared to background features in the at least one image; applying a segmentation process to the enhanced defect features to separate high contrast features from lower contrast features, thereby producing a plurality of segments; and extracting features from each of the plurality of segments by analyzing each segment as to one or more of the following features: (i) a total area of the segment, (ii) an eccentricity and/or elongation of the segment, (iii) a width of the segment, (iv) a height of the segment, and (v) a fill ratio of the segment.
Additionally or alternatively, the methods and apparatus may further provide for determining and identifying the one or more types of the one or more defects based on the analysis of the segments.
For example, a determination that one or more of the segments represent a chip when one or more of: (i) a total area of the one or more segments is relatively large, within a range of relatively small to relatively large, (ii) an eccentricity and/or elongation of the one or more segments is relatively low, within a range of relatively low to relatively high, and (iii) a fill ratio of the one or more segments is relatively high, within a range of relatively low to relatively high.
By way of further example, a determination that one or more of the segments represent a hackle when one or more of: (i) a width of the one or more segments is relatively small, within a range of relatively small to relatively large, and (ii) a location of the one or more segments is relatively close to a periphery of the edge surface.
By way of further example, a determination that one or more of the segments represent a Wallner line when one or more of: (i) a total area of the one or more segments is relatively large, within a range of relatively small to relatively large, (ii) an eccentricity and/or elongation of the one or more segments is relatively high, within a range of relatively low to relatively high, and (iii) a height of the one or more segments is relatively small, within a range of relatively small to relatively large.
By way of further example, a determination that one or more of the segments represent an arrest line when one or more of: (i) a total area of the one or more segments is relatively large, within a range of relatively small to relatively large, (ii) an eccentricity and/or elongation of the one or more segments is relatively high, within a range of relatively low to relatively high, (iii) a width of the one or more segments is relatively small, within a range of relatively small to relatively large, and (iv) a height of the one or more segments is relatively large, within a range of relatively small to relatively large.
Additionally or alternatively, the methods and apparatus may further provide for automatically adjusting one or more parameters of cutting the glass web at the cutting zone based on the detection and identification.
By way of example, the cutting of the glass web at the cutting zone may include heating an elongated zone of the glass web using a laser delivery apparatus followed by cooling the heated portion of the glass web to propagate a fracture in a direction opposite to the transport direction, thereby producing the first and second ribbons. In such an embodiment, the one or more parameters of the cutting of the glass web may include a power level of incident laser light from the laser delivery apparatus, and a focus of the incident laser light from the laser delivery apparatus.
Other aspects, features, and advantages will be apparent to one skilled in the art from the description herein taken in conjunction with the accompanying drawings.
For the purposes of illustration, there are forms shown in the drawings that are presently preferred, it being understood, however, that the embodiments disclosed and described herein are not limited to the precise arrangements and instrumentalities shown.
With reference to the drawings wherein like numerals indicate like elements there is shown in
With reference to
As will be discussed in greater detail later herein, it is very desirable that the resulting edge surface(s) of the glass ribbon 103A (e.g., from removing the bead(s) and/or cutting the glass web 103 to a specified width) have very high quality, and that any significant degradation of the quality of the edge surface(s) should be detected, defects classified, and corrective actions taken. Therefore, as schematically illustrated in
The source 102 of glass web 103 may include a spool, onto which the glass web 103 was first wound, e.g., following a fusion down draw process. Typically, the spool used as source 102 would be provided with a relatively large diameter to present a relatively low bending stress to accommodate the characteristics of the glass web 103. Once coiled onto the spool used as source 102, the glass web 103 may be uncoiled from that spool and introduced into the transport mechanisms of the apparatus 100.
The destination zone of the apparatus 100 may include any suitable mechanisms for accumulating the glass ribbon 103A (and waste beads, not shown). In the example illustrated in
The apparatus 100 includes a transport mechanism having a number of individual elements that cooperate to continuously move the glass web 103 from the source 102, for example a spool of wound glass, to the destination spool 104 in the transport direction. This transport function may be accomplished without degrading the desirable characteristics of the ribbon edge surfaces, as produced from the cutting operation, or either (pristine) major surface of the central portion 205 of the glass web 103 and/or ribbon 103A. In short, the transport function is accomplished without degrading desirable characteristics of the glass ribbon 103A.
By way of example, the apparatus 100 may include a plurality of noncontact support members 106, 108, rollers, etc., to guide the glass web 103 and glass ribbon 103A through the system from the source 102 to the destination spool 104. The non-contact support members 106, 108 may be flat and/or curved to achieve desirable directional conveyance of the respective work pieces. Each of the noncontact support members 106, 108 may include a fluid bar and/or a low friction surface to ensure the glass web 103 and glass ribbon 103A are suitably conveyed through the system without damage or contamination. When a given non-contact support member 106, 108 includes an fluid bar, such element includes a plurality of passages and ports configured to provide a positive fluid pressure stream (for example air), and/or a plurality of passages and ports configured to provide a negative fluid pressure stream, to the associated surface of the glass web 103 and/or glass ribbon 103A to create an air cushion for such noncontact support. A combination of positive and negative fluid pressure streams may stabilize the glass web 103 and glass ribbon 103A during transport through the system.
Optionally, a number of lateral guides (not shown) may be employed proximate to the edge portions 201, 203 of the glass web 103 and/or the edge surfaces of the glass ribbon 103A to assist in orienting the glass web 103 and/or glass ribbon 103A in a desired lateral position relative to the transport direction. For example, the lateral guides may be implemented using rollers that engage a corresponding one of the opposed edge portions 201, 203 of the glass web 103, and/or one or more edge surfaces of the glass ribbon 103A. Corresponding forces applied to the edge portions 201, 203 by the corresponding lateral guides may shift and align the glass web 103 in the proper lateral orientation as the glass web 103 is conveyed through the apparatus.
Due to its high modulus, notch sensitivity and brittleness, however, it is beneficial for the glass web 103 to have very consistent and symmetrical stress and strain fields in the cutting zone 147 in order to exhibit suitable edge characteristics (minimal fractures) after cutting. Therefore, the apparatus 100 includes a tensioning mechanism 130 (for example, as would be understood by one of ordinary skill in the art, a dancer, a web accumulator, a roller with a break) operating to provide consistent and symmetric stress fields and strain fields in the cutting zone 147. In accordance with one or more embodiments herein, tensioning is carefully and independently (from edge portions 201, 203) controlled in the glass ribbon 103A in order to achieve consistent and symmetric stress and strain fields. This approach is intended to result in a very fine, particle free edge surfaces that minimizes edge and/or edge corner defects.
In order to achieve the aforementioned tensioning functionality, the tensioning mechanism 130 monitors the tension, determines whether the tension is within prescribed limits, and varies the force based on the determination to ensure the tension is within prescribed limits. As schematically illustrated in
The apparatus 100 further includes a cutting mechanism 120 that cuts or severs the glass web 103 in the cutting zone 147 as the glass web 103 passes over, for example, the noncontact support member 108. As will be described in more detail later herein, the cutting mechanism 120 may make a single cut or may simultaneously make multiple cuts; however, a significant characteristic of the cutting process is that the resultant glass ribbon 103A (and/or further numbers of ribbons) will exhibit edge surface(s) that are subject to defects, for example chips, hackles, Wallner lines, arrest lines, frictive damage, and scratches.
With reference to
The optical delivery apparatus may include a radiation source, for example a laser, although other radiation sources may be employed. The optical delivery apparatus may further include other elements to shape, adjust direction and/or adjust the intensity of an optical beam, for example one or more polarizers, beam expanders, beam shaping apparatus, etc. Preferably, the optical delivery apparatus produces a laser beam 169 having a wavelength, power, and shape suitable for heating the glass web 103 at a location on which the laser beam 169 is incident.
It has been found that a laser beam 169 of significantly elongated shape works well. The boundary of the elliptical footprint of the laser beam 169 may be determined as the point at which the beam intensity has been reduced to 1/e2 of its peak value. The elliptical footprint may be defined by a major axis that is substantially longer than a minor axis. In some embodiments, for example, the major axis may be at least about ten times longer than the minor axis. However, the length and width of an elongated radiation heated zone 227 are dependent upon the desired severing speed, desired initial crack size, thickness of the glass ribbon, laser power, etc., and the length and width of the radiation zone may be varied to suit the particular cutting conditions.
The cooling fluid source 181 operates to cool the heated portion of glass web 103 by application of cooling fluid, preferably a jet of fluid, for example though a nozzle or the like. The geometry of the nozzle, etc., may be varied to suit the particular process conditions. The cooling fluid may include water, however, any other suitable cooling fluid or mixture may be employed that does not damage the glass web 103. The cooling fluid may be delivered to the surface of the glass web 103 to form a cooling zone 319, where the cooling zone 319 may trail behind the elongated radiation zone 227 to propagate a fracture (initiated by an induced crack). The combination of heating and cooling effectively severs the glass web 103 to create the glass ribbon 103A, while minimizing or eliminating undesired residual stress, micro-cracks or other irregularities in edge surface(s) created by the cut, which result in the aforementioned defects.
As noted above, the apparatus 100 includes the inspection mechanism 180 to address the problem of such defects in the edge surface(s) or the glass ribbon 103A. The inspection mechanism 180 optically inspects one or more edge surfaces of the glass ribbon 103A (and/or further glass ribbons) as the glass web 103 is moved in the transport direction 105 to the destination. From a functional point of view, the inspection mechanism 180 executes actions, including: (i) taking at least one image of the edge surface(s) as the glass ribbon 103A is moved in the transport direction 105, (ii) extracting one or more features of the edge surface(s) from the at least one image, and (iii) detecting one or more defects, and identifying one or more types of the one or more defects, based on the one or more extracted features.
With reference to
The light source 182 directs incident light onto and through a proximal edge surface of the glass ribbon 103A, which is an opposing edge of the glass ribbon 103A that is laterally opposite to (cross-web from) the edge surface that is being inspected (a distal edge surface). As illustrated in
In one or more embodiments, it is considered very desirable to generate a “brightfield” image of the defects on and/or in the edge surface of the glass ribbon 103A. Fine grayscale features present under brightfield illumination enable more accurate sizing and classification of edge features than high-contrast geometries. Notably, however, “darkfield” geometries can be very useful and effective when sensitive detection of features is desired, for instance when using low optical magnification optics to detect particles or glass chips. In both cases, care should be taken to maximize the dynamic range of features on the resulting images, including preventing saturation and/or bottoming out of the features, whereby the details of the defect features are lost within poor contrast. It has been found that good results may be obtained when the imaging sensor 184 has 8-bits of grayscale and sufficient lighting is provided to interact with the defect features; indeed, such a combination has been found to produce a high contrast image. Based on use of the aforementioned high brightness LED and efficient lighting geometry (due to the waveguide approach), it may be a relatively simple matter to position the LED to within a few centimeters of the proximal edge surface of the glass ribbon 103A. Nevertheless, light diverges heavily from a bare LED. Thus, one may employ one or more additional optical elements between the light source (whether an LED, halogen lamp, laser, or any other illuminator) and the proximal edge surface of the glass ribbon 103A to increase coupling efficiency or otherwise condition the lighting. For instance, one may employ a condensing lens to collect more light coming from the light source and focus the light to a point at or near the proximal edge surface of the glass ribbon 103A. This may greatly increase brightness. Additionally and/or alternatively, one may employ a diffuser between the light source and the proximal edge surface of the glass ribbon 103A to even the output spatially. Still further, a color filter and/or a polarizer may be employed to affect the light reaching the imaging sensor 184 to target some property of the defects in question.
To obtain coupling of the light from the light source into the proximal edge of the glass ribbon 103A (To achieve waveguiding) it is desirable to have a suitable edge finish (for example a straight, mirror-like finish attained via the aforementioned laser cut). In contrast, when the proximal edge of the glass ribbon 103A exhibits the characteristics of a ground edge, an attempt to couple light from the light source into the proximal edge of the glass ribbon 103A exhibits high loss because the edge scatters a lot of light. To compensate, one may increase the intensity of the light source to provide sufficient illumination for adequate defect contrast, or instead use a reflection geometry.
The imaging sensor 184 is preferably provided such that an optical axis thereof is directed substantially perpendicularly toward the edge surface being inspected. The imaging sensor 184 receives the light exiting the edge surface being inspected and produces at least one image of the edge surface. By way of example, the imaging sensor 184 is preferably a high resolution image acquisition device, with sufficient speed to be effective despite a conveyance speed of the glass ribbon 103A in the range of about 200 mm/s or higher, and a relatively large field of view (with reference to the edge surface dimensions). To achieve a high resolving power, the imaging sensor 184 may employ a high numerical aperture (NA) lens (e.g., above 0.2) and a light sensitive sensor (e.g., a charge couple device (CCD) array, for example a CCD having 7 um per pixel resolution). High NA optics are beneficial for highlighting fine defect features on the edge surface (for example hackle lines). High NA optics also are characterized by small depths-of-field (DOF), which may be less than about 50 μm. Consequently, as discussed below, one should very carefully measure, track, and adjust for any change in distance from the imaging sensor 184 to the edge surface to remain within the aforementioned DOF.
With reference to
Alternatively and/or additionally, the automatic focus mechanism 186 may interface with an adjustable lens system of the imaging sensor 184 to automatically adjust the lens system as a function of a varying distance D to adjust the aforementioned focal length. When the imaging sensor 184 includes such an adjustable lens system, the optics of the lens itself may be adjusted to vary a focal length of the lens, and therefore the motion stage (and the resultant translational movement of the imaging sensor 184 towards and/or away from the edge surface of the glass ribbon 103A) may be avoided. Indeed, the position of the imaging sensor 184 may remain fixed. Nevertheless, adjustments in the focal length of the imaging sensor 184 may be made to account for variations in the distance D from the imaging sensor 184 to the edge surface of the glass ribbon 103A during conveyance thereof.
Examples of the images of the edge surface being inspected have been presented above (
By way of example, the processing and control unit 190 may include a computer processor operating under the control of a computer program, which may be stored in a digital storage medium. When the computer program is executed by the computer processor, the computer program causes the computer processor to carry out the actions of detecting the one or more defects, and identifying the one or more types of the one or more defects. More specifically, the algorithm may include one or more of: (i) enhancing one or more defect features as compared to background features in the at least one image; (ii) applying a segmentation process to the enhanced defect features to separate high contrast features from lower contrast features, thereby producing a plurality of segments; and (iii) extracting features from each of the plurality of segments by analyzing each segment as to one or more predetermined features. Such extracted features may include: (i) a total area of the segment, (ii) an eccentricity and/or elongation of the segment, (iii) a width of the segment, (iv) a height of the segment, and (v) a fill ratio of the segment.
Further details as to how to enhance one or more defect features, apply a segmentation process, and extract features from each of the plurality of segments will be provided later herein. At present, however, it has been discovered that the above-noted features of the segments may be used to detect and identify types of defects.
For example, the processing and control unit 190 of the inspection mechanism 180 employs the algorithm to make a determination that one or more of the segments represent a chip when one or more of: (i) a total area of the one or more segments is relatively large, within a range of relatively small to relatively large, (ii) an eccentricity and/or elongation of the one or more segments is relatively low, within a range of relatively low to relatively high, and (iii) a fill ratio of the one or more segments is relatively high, within a range of relatively low to relatively high.
Additionally or alternatively, the processing and control unit 190 of the inspection mechanism 180 may employ the algorithm to make a determination that one or more of the segments represent a hackle when one or more of: (i) a width of the one or more segments is relatively small, within a range of relatively small to relatively large, and (ii) a location of the one or more segments is relatively close to a periphery of the edge surface.
Additionally or alternatively, the processing and control unit 190 of the inspection mechanism 180 may employ the algorithm to make a determination that one or more of the segments represent a Wallner line when one or more of: (i) a total area of the one or more segments is relatively large, within a range of relatively small to relatively large, (ii) an eccentricity and/or elongation of the one or more segments is relatively high, within a range of relatively low to relatively high, and (iii) a height of the one or more segments is relatively small, within a range of relatively small to relatively large.
Additionally or alternatively, the processing and control unit 190 of the inspection mechanism 180 may employ the algorithm to make a determination that one or more of the segments represent an arrest line when one or more of: (i) a total area of the one or more segments is relatively large, within a range of relatively small to relatively large, (ii) an eccentricity and/or elongation of the one or more segments is relatively high, within a range of relatively low to relatively high, (iii) a width of the one or more segments is relatively small, within a range of relatively small to relatively large, and (iv) a height of the one or more segments is relatively large, within a range of relatively small to relatively large.
With reference to
Turning now to further details concerning how the algorithm within the processing and control unit 190 determines the existence of and types of defects on the edge surface being inspected, a number of features of various types of defects will be discussed.
With reference to
With reference to
With reference to
With reference to
The one or more portions of the algorithm that identify the types of defects within the one or more images of the edge surface generally include four modules; namely, a feature enhancement module, a segmentation module, a feature extraction and classification module, and a grouping module.
The feature enhancement module serves to highlight the defect regions and suppress background signals within the image of the edge surface (see image 50 in
The features of the defect(s) within the image of the edge surface are represented by intensity changes and may be enhanced, for example, using a difference filter to implements the aforementioned local threshold. One example of such a difference filter is a two scale Haar wavelet filter, which enhances features under different scales. The Haar filter may be represented by the following function:
where it may be assumed that the glass ribbon 103A is being transported along an X-axis, and the variable d corresponds to a scale of interest. A smaller d may enhance fine scale features, for example hackle lines, while a larger d may enhance other defect features (as well as suppressing noise).
An example of a Haar filtered image (see image 52 in
The segmentation module receives the enhanced difference image (filtered image) from the feature enhancement module and separates features of defects that appear to be connected in the enhanced difference image. For example, features of Wallner lines may appear to connect with features of hackle lines. As mentioned above, different defect types exhibit features having different contrast intensities. For example, Wallner lines may have very smooth intensity variation while other defects, for example arrest lines or chips, may have very strong intensity changes. On the other hand, the features of different defects may appear to be connected, for example the features of Wallner lines and the features of hackle lines (see, upper left and lower left portions of image 50 in
By way of example, a dual threshold segmentation technique may be employed to implement the segmentation module, for example using modified hysteresis thresholding. Hysteresis thresholding is a technique to first identify high response pixels, and then recursively connecting adjacent pixels that are above a lower response threshold. In one or more embodiments, hysteresis thresholding may be applied along an X-axis (left-to-right in
The feature extraction and classification module detects and extracts features (called blob features or simply blobs, which are the features of each Binarized Linear Object (BLOB) from each of the segments, which are then used to classify the subject features into types of defects. By way of example, the feature extraction and classification module may be implemented via a binary decision tree classification technique, which permits artisans to observe, test and select key features for satisfactory classification results. The classifier algorithm may be a straight rule-based classification, a neural network, an m-of-n classifier, etc. In this regard, a set of standard and non-standard blob features may be used. The standard features may include one or more of: area, bounding box, eccentricity, orientation, centroid, and/or fill rate (which is the area/convex Area). The non-standard blob features may include one or more of: averaged horizontal integration, effective sub-blob count, effective width, and/or effective degree of elongation. Notably, the non-standard blob features were found to produce better classification results.
The averaged horizontal integration may be defined as the column average of summation along each row. The purpose of the averaged horizontal integration is to reflect overall intensity changes along the X-axis. It has been found to be useful in identifying arrest lines, especially when the intensity change along the X-axis is small but the overall intensity change is strong.
The effective sub-blob may be employed to calculate the actual number of blobs when tightly packed thin lines are recognized as one blob after the threshold process is employed. Calculating the effective number of sub-blobs has been found to be very useful in identifying hackle lines without having to find a way to split the blob into sub-blobs. The technique operates to count the number of false pixels (black) blocks along each row within the blob. The average number of black blocks for all the rows strongly relates to the number of sub-blobs in a super blob. For example, in a segment having three tightly packed blobs (e.g., semi-vertical “white” lines), most rows would have three white blocks separated by two black blocks, indicating the super blob appears as an aggregate of three sub-blobs. The effective width of the sub-blobs may be calculated as the average width (average of row summation, assuming the primary orientation of the defect is vertical, extending along the Y-axis) divided by the effective sub-blob count. The effective width may be a useful feature in identifying hackle lines, since the thickness of a hackle line is very small.
The Effective Degree of Elongation may be considered to be similar to eccentricity. Eccentricity for an ellipse is defined as:
where a is the major axis length, b is the minor axis length, and a and b are obtained by fitting an ellipse to the blob. Conveniently, one may use the ratio of b to a (width/height ratio) to indicate the eccentricity of the blob. The eccentricity may be useful in identifying elongated structures; however, eccentricity might not be accurate enough to represent the degree of elongation for rib-shaped features, which might be better represented by a length/thickness ratio. It has been found that the more curved the structure is, the lower the eccentricity.
Additionally and/or alternatively, one may define an effective degree of elongation as:
where A is the area of the blob, and the minor axis length b of the eccentricity formula is replaced with A/a. The effective degree of elongation has been found to work better for rib shaped structures, although it is not an exact representation of the length/thickness ratio.
A plurality of the extracted blob features are used for blob level classification (see, image 58 in
For detecting and classifying a defect as a chip, it has been found that suitable results may be obtained by subjecting the image of the edge surface to a Haar (scale 3) filter, applying hysteresis thresholding to produce segments, and identifying blobs as those with relatively high averaged horizontal integration values. With such processing, the resultant extracted blob features would suggest a chip when one or more of: (i) a total area of a blob is relatively large, within a range of relatively small to relatively large, (ii) an eccentricity and/or elongation of a blob is relatively low, within a range of relatively low to relatively high, and (iii) a fill ratio of a blob is relatively high, within a range of relatively low to relatively high.
For detecting and classifying a defect as one or more hackle lines, it has been found that suitable results may be obtained by subjecting the image of the edge surface to a Haar (scale 1) filter, and applying a single lower-level threshold to produce segments. With such processing, the resultant extracted blob features would suggest one or more hackle lines when one or more of: (i) an effective width of the blob(s) is/are relatively small, within a range of relatively small to relatively large, and (ii) a location of the blob(s) is/are relatively close to a periphery of the edge surface of the glass ribbon 103A.
For detecting and classifying a defect as one or more Wallner lines, it has been found that suitable results may be obtained by subjecting the image of the edge surface to a Haar (scale 3) filter, and applying a single lower-level threshold to produce segments. With such processing, the resultant extracted blob features would suggest one or more Wallner lines when one or more of: (i) a total size (e.g., height, width and/or area) of the blob(s) is/are relatively large, within a range of relatively small to relatively large, and (ii) an eccentricity and/or elongation of the blob(s) is/are relatively high, within a range of relatively low to relatively high.
Additionally and/or alternatively, for detecting and classifying a defect as one or more Wallner lines, it has also been found that suitable results may be obtained by subjecting the image of the edge surface to a Haar (scale 3) filter, and applying an hysteresis threshold to produce segments. With such processing, the resultant extracted blob features would suggest one or more one or more Wallner lines when one or more of: (i) a total size (e.g., height, width and/or area) of the blob(s) is/are relatively large, within a range of relatively small to relatively large, (ii) an eccentricity and/or elongation of the blob(s) is/are relatively high, within a range of relatively low to relatively high, and (iii) a height of the blob(s) is/are relatively small, within a range of relatively small to relatively large.
For detecting and classifying a defect as one or more arrest lines, it has been found that suitable results may be obtained by subjecting the image of the edge surface to a Haar (scale 3) filter, applying hysteresis thresholding to produce segments, and identifying blobs as those with relatively high averaged horizontal integration values. With such processing, the resultant extracted blob features would suggest one or more arrest lines when one or more of: (i) a total area of the blob(s) is/are relatively large, within a range of relatively small to relatively large, (ii) an eccentricity and/or elongation of the blob(s) is/are relatively high, within a range of relatively low to relatively high, (iii) a width of the blob(s) is/are relatively small, within a range of relatively small to relatively large, and (iv) a height of the blob(s) is/are relatively large, within a range of relatively small to relatively large (for example a total height across 90% of the edge surface).
The above-noted features have been found to provide a relatively high level of separation between different defects and to perform fairly well at the blob-level. Notably, the hackle lines may produce high responses in both scale 1 and scale 3 Haar filtered images; however, the scale 3 filtered images have been found to possibly obscure the thin line-like features of hackle lines, and thus may result in a false classification as chips.
By way of example, reference is made to image 58 of
In some cases, due to web motion or certain glass surface defects, portions of the edge surface of the glass ribbon 103A may not be well captured within the image. For example, a dark area on the edge surface may correspond to an out-of-focus region, where not enough photons are captured by the imaging sensor 184, while an overly bright region may suggest an inclined mirror surface. It is highly unlikely that some form of image processing will recover features in such areas, and thus it may be advantageous to map out these areas for future analysis. The algorithm herein may include a module to map out these areas using a global thresholding technique. It is desirable, however, that the module avoid the inclusion of background regions in the image because background regions also appear as dark regions. This may be difficult, since most defects exist on the periphery of the edge surfaces, thereby creating a very uneven surface for foreground extraction. In addition, diffraction on a corner of the edge surface and/or a flat surface may obscure the true boundary of the edge surface in the image. Furthermore, the edge surface may not be exactly horizontal, due slight distortions of the glass ribbon 103A or stress therein. To compensate for these effects, the algorithm herein introduces a new way to extract the sample area from the edge surface images using a user-defined thickness. It is assumed that the sample area of the edge surface is horizontally positioned within a small window, and thus one may implement a box matched filter, the height of which is the same as the given sample area thickness. The filtered image will have a local maximum along the Y-axis at the center of the sample area. With the calculated sample area center and user-defined thickness, the sample area may be successfully extracted. A global thresholding may then be carried out on the masked sample area to avoid inclusion of the background.
The grouping module may be employed to address the possibility that the segmentation module produces over-segmented blobs, especially for chips. After blob-level classification, it may be desirable to merge spatially close blobs to form logically more accurate representations of defects. Grouping of blobs is mostly used for hackle lines and chips (see, image 60 in
For hackle lines, the algorithm herein provides for grouping and performing another level of classification, since one of the features of hackle lines is that they are spatially segregated. The grouping of hackle lines may be performed by iteratively looking at adjacent regions for each hackle line, which is based on the assumption that hackle lines appear segregated. Next, thresholds of width and height may be used to eliminate small segregations of hackle lines.
For chips, the algorithm herein also provides for grouping and performing another level of classification. The simplest form of a chip may include two blobs, each representing a “mirror” face and a “dark” face of a chip. In a real case, chips often appear in groups. It may be desirable to show a group of closely packed chips as one chip. In addition, certain adjacent defects, which are too small to be determined or classified under blob-level classification, may also be grouped with a chip to achieve a complete and logical segmentation. One issue in grouping is to determine whether a non-chip defect blob should be connected to a chip blob. Although possible, it is desirable not to use only spatial closeness as the rule of the grouping determination. Thus, the algorithm herein may operate on the assumption that a non-chip blob is connected to a chip blob when a predefined portion of the perimeter of the non-chip blob resides in the dilated chip blob. While other ways to define connectivity are also possible, it has been found that the foregoing approach is suitable for most cases. It may be desirable to iteratively connect undetermined areas, Wallner lines, and bright regions to form logically correct segmentation.
Again, the images of
Although the disclosure herein has been described with reference to particular embodiments, it is to be understood that these embodiments are merely illustrative of the principles and applications of the embodiments herein. It is therefore to be understood that numerous modifications may be made to the illustrative embodiments and that other arrangements may be devised without departing from the spirit and scope of the present application. For example, the various features may be combined as set forth in the specific exemplary embodiments below.
A method, comprising:
The method of embodiment 1, wherein the inspecting step includes:
The method of embodiment 2, wherein the step of directing the imaging sensor toward the at least one of the first and second edge surfaces includes:
The method of any one of embodiments 1-3, wherein the step of detecting the one or more defects, and identifying the one or more types of the one or more defects, includes:
The method of embodiment 4, further comprising grouping at least some of the segments together to form at least one aggregated segment.
The method of embodiment 4 or embodiment 5, further comprising determining and identifying the one or more types of the one or more defects based on the analysis of the segments.
The method of embodiment 6, further comprising making a determination that one or more of the segments represents a chip when:
The method of embodiment 6, further comprising making a determination that one or more of the segments represent a hackle line when:
The method of embodiment 6, further comprising making a determination that one or more of the segments represent a Wallner line when:
The method of embodiment 6, further comprising making a determination that one or more of the segments represent an arrest line when:
The method of any one of embodiments 1-10, further comprising automatically adjusting one or more parameters of the step of cutting the glass web at the cutting zone based on the detection and identification, wherein
An apparatus, comprising:
The apparatus of embodiment 12, wherein the inspection mechanism comprises:
The apparatus of embodiment 13, further comprising an automatic focus mechanism comprising:
The apparatus of any one of embodiments 12-14, wherein the inspection mechanism includes a computer processor configured to operate under the control of a computer program, which when executed by the computer processor causes the computer processor to carry out the actions of detecting the one or more defects, and identifying the one or more types of the one or more defects, by:
The apparatus of embodiment 15, wherein the inspection mechanism is further configured to carry out an action of grouping at least some of the segments together to form at least one aggregated segment.
The apparatus of embodiment 15 or embodiment 16, wherein the inspection mechanism is further configured to carry out an action of determining and identifying the one or more types of the one or more defects based on the analysis of the segments.
The apparatus of embodiment 17, wherein the inspection mechanism is further configured to carry out an action of making a determination that one or more of the segments represent a chip when:
The apparatus of embodiment 17, wherein the inspection mechanism is further configured to carry out an action of making a determination that one or more of the segments represent a hackle when:
The apparatus of embodiment 17, wherein the inspection mechanism is further configured to carry out an action of making a determination that one or more of the segments represent a Wallner line when:
The apparatus of embodiment 17, wherein the inspection mechanism is further configured to carry out an action of making a determination that one or more of the segments represent an arrest line when:
The apparatus of any one of embodiments 12-21, further comprising a feedback mechanism configured to automatically adjust one or more parameters of the cutting mechanism and of cutting the glass web at the cutting zone based on the detection and identification, and wherein
This application claims the benefit of priority under 35 U.S.C. § 119 of U.S. Provisional Application Ser. No. 62/299,750 filed on Feb. 25, 2016, the content of which is relied upon and incorporated herein by reference in its entirety.
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/US2017/019006 | 2/23/2017 | WO | 00 |
| Number | Date | Country | |
|---|---|---|---|
| 62299750 | Feb 2016 | US |
