Method and apparatus for automatically determining sheet metal strain

Abstract

Disclosed is a method and apparatus for automatic measurement of strain in a formed sheet metal sample. Prior to the forming operation, a grid of circles is imprinted on the sample and during the forming operation the circles are stretched to ellipses. The formed sample is exposed to an instrumentation camera and a digitized image of the pattern on the sample is stored in a digital computer. The computer is programmed to segment the image to distinguish the elliptical patterns from the background; to extract boundary points on an elliptical pattern and fit these points to an ellipse. Strain is determined as a function of the diameters of the fitted ellipse as well as the chord lengths of the actual pattern along the axes of the fitted ellipse.

Description

FIELD OF THE INVENTION

This invention relates to sheet metal strain measurement and, more particularly, to a method and apparatus for automating the measurement of strain in formed sheet metal.

BACKGROUND OF THE INVENTION

Industry in general, and the automobile industry in particular, produces a large number of stamped sheet metal parts each year. With the greatly increased usage of the new lighter weight and higher strength alloys, the problems of producing such parts have become aggravated. The design of stamping dies is largely an art and the considerable body of experience relative to dies for stamping steel is not readily applied in connection with the newer materials. The high cost incurred in acquiring the needed information by the trial and error methods of die design necessitates a different approach. One promising approach is to develop a better understanding of the stamping process itself. A measurement frequently required in stamping studies is the amount of strain present in the metal at various locations. In addition to supporting research efforts to understand the metal forming process and die design, strain measurements are also useful as an inspection aid. The strength of metals can be correlated with the amount of strain. The maximum strain in critical parts can also be checked to determine how close they are to the failure strain.

The standard technique employed to record strain measurements is carried out as follows. Sheet stock is printed with grids of small circles prior to a forming operation. As the metal is formed, the circles stretch into ellipses in proportion to the strain. Because the circles were printed in a known size, it is possible to calculate the strain by measuring the dimensions of the ellipse patterns. The circles can be printed in a variety of sizes and configurations using a variety of printing methods. They may be printed as open rings or filled-in dots. Two commonly employed printing methods are the photo-resist and electro-etch techniques.

The measurement of many small ellipses is a very labor intensive operation. For quick, approximate measurements, a clear plastic mask is held over the sample and slid along the sample to match up the appropriate dimensions. If greater precision is required, the samples to be measured are cut into pieces which can be laid on a calibrated toolmaker's microscope stage. The microscope stage is fitted with a digital position read-out indicating the amount of stage translation. The grid circle to be measured is aligned along one of the table axes with one side centered beneath a cross hair. The position counter is then zeroed and the stage moved until the opposite side is aligned beneath the cross hair. The display then indicates the ellipse measurement in some arbitrary units. This method is currently being used and produces good results. Readings have been found to be accurate to plus or minus 0.3 percent of the true strain value. The measurement does, however, involve some subjectivity in deciding where the pattern edge lies and in precisely orienting the pattern along one of the principle axes. This is difficult to achieve in severe cases where the pattern is degraded due to the stamping process and also when the strain levels are low. A skilled technician requires approximately one to two minutes for each point measured. This technique normally requires two similar measurements--once for the major axis strain and again for the minor axis strain. When many points must be measured, it becomes a very tedious time consuming operation.

SUMMARY OF THE INVENTION

With the foregoing in mind, it is an object of the present invention to overcome both the speed and accuracy limitations of the manual methods of measuring surface strains in sheet metal resulting from a forming operation.

In accordance with the present invention, a solid-stage instrumentation camera is used to record a picture of the sheet metal surface containing an elliptical pattern to be measured. The picture is digitized as a square array of light intensity values and stored in the memory of a digital computer. The computer is programmed to determine a gradient-weighted average light intensity of the picture for use as a threshold level below which everything in the picture can be considered black. Once the threshold level has been selected for the picture, the image is segmented into solid black (pattern) and solid white (background) regions and those regions which touch the border of the image are dismissed from further consideration. For a typical field of view all but one pattern will be eliminated. Points on the boundary of the remaining pattern are extracted and fitted to an ellipse. Once an acceptable fit is attained, the major and minor strain is computed from the fitted ellipse diameters and also from a chord measurement operation which measures the dimensions of the actual elliptical pattern rather than the dimensions of the fitted ellipse.

A more complete understanding of the present invention may be had from the following detailed description which should be read in conjunction with the drawings, in which:

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 shows the apparatus of the present invention;

FIGS. 2 and 3 are a pictorial representation useful in describing the threshold selection process; and

FIG. 4 is an overall flow chart of the program executed by the computer.

DESCRIPTION OF THE PREFERRED EMBODIMENT

Referring now to the drawings and initially to FIG. 1, the sample generally designated 10 for which the strain is to be measured may be an actual sheet metal stamping or a test specimen stretched on a special press or pulled on a tensile test machine. Prior to the forming operation the sample is imprinted with a grid of circles of predetermined diameter. A commonly used circle size is 2.54 mm diameter. The technique selected to print the grid of circles onto the sheet stock depends on the type of material and the severity of the stamping operation. Preferably, the technique utilized produces a high contrast pattern which adheres well to the deforming surface while not altering the material properties. The pattern masks are normally generated on a film by a computer. One printing technique in use on steel samples is the photo-resist method, which provides excellent contrast between the patterned and non-patterned surface areas.

The sample is mounted in a fixture (not shown) similar to a microscope stage set-up. A camera indicated at 12 may be attached to an upright pole of the fixture on an adjustable slide or the camera may be a portable hand held unit. Preferably, the camera 12 is a solid state instrumentation camera whose sensing elements cover a square rather than rectangular area in the image. This permits the width of a single picture element (pixel) to be used as a convenient distance measure. Similarly, the area of a single picture element is a convenient unit of area. A suitable camera is the General Electric TN-2200 which contains a 128.times.128 matrix of C.I.D. (charge injection device) light sensing elements. The lens employed depends on the viewing area desired. For the standard grid spacing of 2.54 mm center-to-center distance, a 50 mm lens with 60 mm of extension tube has been found to work well. To ensure that the pattern will be accurately and unambiguously located in the recorded image, a high constrast image is required, i.e., one in which the pattern on the metal surface appears much darker (or much lighter) than the background. This in turn is dependent on the type and intensity of light utilized. The lighting source should be diffused and directed so as to minimize glare. One lighting arrangement which has been used successfully employs a fluorescent ring 14 placed around the lens itself. By keeping the camera integration period around 1/30th of a second, the flicker from the fluorescent ring does not interfere with the recording.

The lens of the camera 12 focuses the image of a portion of the sample including at least one of the elliptically shaped patterns onto the light sensor array. Each sensor element develops a variable charge, depending on the relative light intensity. The analog information from the camera 12 is fed to interface circuitry, generally designated 16, which digitizes the camera output to sixteen distinct levels of light intensity, and packs it into a form suitable for input through a standard DMA port of a computer, generally designated 18. The light intensity data is stored in the computer memory as 128.times.128 element array of integers whose value corresponds to the light intensity of respective picture elements.

The interface circuitry 14 also provides the necessary power and control signals to the camera and generates the necessary sweep and level signals to drive a monitor, generally designated 20. The monitor 20 is a cathode ray tube which enables the operator to locate the precise grid circle of interest and adjust the camera for contrast and focus. A suitable interface is the General Electric PN-2100A, which performs all the above functions, except for packing the data and communicating with the DMA port of the computer 18. The data packing and communication functions are well-known to those skilled in the art and may be implemented with logic which stores four digitized light intensity values per computer word and sequentially feeds this information to the computer 18.

The computer 18 is preferably the Digital Equipment Corporation, LSI-11 Microcomputer. A PDP-11VO3 package, available from the computer manufacturer, may be used in practicing the invention and includes the RT-11 operating system, a dual floppy disc drive generally designated 22, and a terminal generally designated at 24. Alternatively, the machine executable program may be stored in a read only memory. As an output medium the system includes a graphic display, generally designated 26. The display 26 is particularly useful where ambiguities arise and it is desirable to have a person verify that a picture has been correctly interpreted. A suitable display controller is the MLSI-512 manufactured by Matrox Electronic Systems. The display 26 contains an array of about 65,000 memory elements with each uniquely associated with a point on a square grid. Thirty times per second, it uses this memory to generate a television signal where the brightness of each point in the image is controlled by one of the memory elements. By writing out the values to a block of memory elements, a picture may be displayed on an attached TV monitor. The MLSI-512 display controller connects directly into the LSI-11 computer bus. In addition, it generates the composite video signal which can be fed directly into a standard TV monitor.

The program for processing the picture data is depicted in a generalized flow chart format in FIG. 4 and a detailed listing in Fortran IV and Macro-11 assembly language is provided hereinafter. Once the picture data is located into the computer memory, the first step in processing the picture is to select a threshold. The picture is assumed to contain one or more dark elliptical patterns against a lighter background. However, as the surface characteristics and the lighting vary, so does the apperance of the pattern in the picture. The process of thresholding is to establish some brightness level below which everything can be considered black (i.e., the pattern area of the picture). Because of the aforementioned variations, it is not possible to select a fixed threshold value which will work for all picture. Preferably the threshold level selected is a gradient-weighted average light intensity. To establish the threshold level a squence of rows, for example rows 32, 64 and 96, in the picture are scanned as shown in FIG. 2. The spacing between the rows is selected to ensure that at least a few true edges are crossed if an elliptical pattern is indeed contained in the image. The computer calculates the gradient at each picture element in the selected rows. For example, with reference to FIG. 3, the gradient around the picture element e would be computed as the absolute value of the sum of the differences between the light intensity values at the picture elements a, i, c and g. More specifically, the gradient g=.vertline.a-i.vertline.+.vertline.c-g.vertline.. If the gradient exceeds a minimum predetermined value, then the gradient calculated for the element e is added to the sum (SUMG) of the gradient values previously calculated and the sum of the light intensity values at a, i, c and g are multipled by the gradient and added to the sum (SUMGI) of weighted values previously calculated. When all elements in the selected rows have been evaluated, the threshold is equal to SUMGI/(4.times.SUMG) rounded to the nearest integer.

Once a threshold level has been selected, the image is segmented into solid black and solid white regions. Typically, there will be less than ten of these regions in a picture. This represents a major reduction in the volume of data from approximately 16,000 numbers to a brief description of only a few regions. In order to correctly label all pairs of connected elements as being part of the same region, it is necessary that the segmentation algorithm be capable of handling complex maze-like patterns which do in fact arise in actual pictures. Because of the large number of potential connections and the many picture elements to be considered it is important to perform these computations efficiently. A particularly efficient segmentation approach is one which employs a run length coding algorithm which sequentially examines each element in a row and records the column members where "color" transitions occur. Each row is thus grouped in segments of length defined by column members. A 6-connected algorithm is employed to determine whether segments in succeeding rows overlap previously defined segments or whether a non-overlapping segment has been encountered. Overlap occurs if a picture element is adjacent an element of the same "color". The criteria for adjacency includes each of the four sides of the element as well as the "northeast-southwest" diagonal but not the "northwest-southeast" diagonal. For example, in FIG. 3 the picture element e is connected to the elements b, c, f, h, g and d but not a and i.

When the segmentation process is completed there may, for example, be six regions as shown in FIG. 2. Five of the regions are black and numbered 1-5 while the sixth region is the white background region. The coordinates of a point on the boundary of each region is obtained during segmentation. Regions 1-4 are eliminated from further processing since these regions touch the border of the image and are therefore incomplete patterns. The pattern boundary of region 5 is extracted starting with the point on the boundary identified during the segmentation process. The coordinates (row, column) of this point are stored at the beginning of a list. The program steps through the digitized picture array one cell at a time in a counterclockwise direction around the border until the initial point is reached with coordinates of new points periodically added to the list.

The boundary data points thus identified are fitted to an ellipse using an algorithm for fitting at least-squares conic. The algorithm utilizes the coordinate pairs (x.sub.k, y.sub.k), where k corresponds to the boundary data points 1, 2 . . . , n and constructs a 5.times.5 matrix A and a 5-vector B as follows:

______________________________________A B______________________________________.SIGMA.x.sub.k.sup.4 .SIGMA.x.sub.k.sup.3 y.sub.k .SIGMA.x.sub.k.sup.2 y.sub.k.sup.2 .SIGMA.x.sub.k.sup.3 .SIGMA.x.sub.k.sup.2 y.sub.k .SIGMA.x.sub.k.sup.2.about. .SIGMA.x.sub.k.sup.2 y.sub.k.sup.2 .SIGMA.x.sub.k y.sub.k.sup.3 .SIGMA.x.sub.k.sup.2 y.sub.k .SIGMA.x.sub.k y.sub.k.sup.2 .SIGMA.x.sub.k y.sub.k.about. .about. .SIGMA.y.sub.k.sup.4 .SIGMA.x.sub.k y.sub.k.sup.2 .SIGMA.x.sub.k y.sub.k.sup.3 .SIGMA.y.sub.k.sup.2.about. .about. .about. .SIGMA.x.sub.k.sup.2 .SIGMA.x.sub.k y.sub.k .SIGMA.x.sub.k.about. .about. .about. .about. .SIGMA.y.sub.k.sup.2 .SIGMA.y.sub.k______________________________________

The sums of matrix A and vector B are over all the data points. It will be noted that matrix A is symmetric, i.e., A(i,j)=A(j,i) and that A(2,2)=A(1,3); B(1)=A(4,4); A(2,4)=A(1,5); B(2)=A(4,5); A(3,4)=A(2,5); B(3)=A(5,5). After the matrix A and the vector B are constructed the linear system AP=B is solved for the 5-vector P. The implicit equation of the fitted conic is:

P(1).multidot.x.sup.2 +P(2).multidot.xy+P(3).multidot.y.sup.2 +P(4).multidot.x+P(5).multidot.y=1 If P(2), P(4) and P(5) are replaced by P(2)/2, P(4)/2 and P(5)/2 respectively the center of the fitted conic, having coordinates x.sub.0 and y.sub.0 is: x.sub.0 =(P(2).multidot.P(5)-P(3).multidot.P(4))/.DELTA. y.sub.0 =(P(2).multidot.P(4)-P(1).multidot.P(5))/.DELTA. where .DELTA.=P(1).multidot.P(3)-P(2).multidot.P(2) Assuming that the conic is an ellipse the axes are: ##EQU1## where F=1-P(4).multidot.x.sub.0 -P(5).multidot.y.sub.0, S=(P(1)+P(3))/2, R=.sqroot.D.sup.2 +P(2).sup.2, D=(P(1)-P(3))/2. The conic is an ellipse if .vertline.S.vertline. is greater than .vertline.R.vertline.. The major axis will be .alpha. if F and S have the same sign. The angle of inclination of the ellipse is that of the .alpha. axis to the X or horizontal axis. The inclination angle .theta. can be determined from the formulae: ##EQU2## .theta. can be construed as an angle in the range -90.degree. to +90.degree. if the radical for sin (.theta.) is taken in sign opposition to P(2).

To generate the fitted conic the following parametric equations are employed:

x=x.sub.0 +.epsilon.cos(.theta.)-.eta.sin(.theta.), y=y.sub.0 +.epsilon.sin(.theta.)+.eta.cos(.theta.), where .epsilon.=.alpha.cos(.tau.) and .eta.=.beta.sin(.tau.), (.tau. ranging from 0 to 2.pi.).

Following a fit to an ellipse, the quality of the fit is checked by measuring the root-mean-square distance from the data points on the pattern boundary to the nearest point on the fitted ellipse. If the error level exceeds an established amount, an attempt to improve the fit is made by eliminating all points further than some tolerance distance from the ellipse. The remaining data points are then used to fit a new ellipse. This refitting operation is repeated a predetermined number of times or until the error level is reduced to an acceptable amount. On each successive fit the allowable error level used for discarding data points is reduced.

Once an acceptable fit is attained, the major and minor strain is computed from the fitted ellipse diameters and also from a chord measurement operation which measures the dimensions of the actual elliptical pattern rather than the dimensions of the fitted ellipse. In both cases strain is calculated by comparing the major and minor diameters of the elliptical pattern with the diameters of the original undistorted circular pattern as determined during calibration. That is to say:

STRAIN=(d-d.sub.0)/d.sub.0 where d=diameter of elliptical pattern d.sub.0 =diameter of circular pattern The chord measurement values are computed by extending the major and minor axes of the pattern as determined by the ellipse fit from the center point of the pattern until they intersect the image borders. The program then steps back through the picture data from the four points on the image border along the axes until the target pattern is reached. The major and minor diameters are then computed as the distance between the two pairs of points found on pattern boundaries along the axes.

Calibration of the system is accomplished by processing undistorted circles in the same manner as described for the elliptical patterns. The circular pattern is fitted to an ellipse so two radius values are computed (which presumably are very close to each other). The average of the two radius values is also computed and the average as well as the radius values are printed for verification by the operator as valid. The average radius for the new circle is then averaged in at equal weight with all previous averages obtained since the system was last initialized.

A complete listing in Fortran IV and Macro-11 assembly language with appropriate comments follows: ##SPC1## ##SPC2## ##SPC3## ##SPC4## ##SPC5##

Claims

1. A method of automatically determining the strain induced in a sheet metal sample by a forming operation, said sample having a grid of circles of predeterming diameter printed thereon prior to said operation, which circles are stretched into patterns of generally elliptical shape as a result of said operation, said method comprising the steps of:

obtaining a digital representation of the image of a portion of the sample including at least one of said patterns, which digital representation is a sequence of integers, each integer corresponding to the average light intensity of the image at discrete image element locations,

storing said integers in addressable locations in the memory of a digital computer, said computer being programmed to perform the steps of:

(a) segmenting the image into binary regions, one region type comprising contiguous image elements whose light intensity levels are less than a threshold level and the other region type comprising contiguous image elements whose light intensity levels are equal to or greater than said threshold level;

(b) identifying a plurality of points residing on the boundary of one region type which is totally within the field or view defined by the image and which is greater than a predetermined size;

(c) fitting an ellipse to said points;

(d) calculating a diameter of the fitted ellipse and;

(e) calculating the strain induced in the sample as a function of said calculated diameter and the predetermined diameter of said circles.

2. A method of automatically determining the strain induced in a sheet metal sample by a forming operation, said sample having a grid of circles of predetermining diameter printed thereon prior to said operation, which circles are stretched into patterns of generally elliptical shape as a result of said operation, said method comprising the steps of:

obtaining a digital representation of the image of a portion of the sample including at least one of said patterns, which digital representation is a sequence of integers, each integer corresponding to the average light intensity of the image at discrete image element locations,

storing said integers in addressable locations in the memory of a digital computer, said computer being programmed to perform the steps of:

(a) segmenting the image into binary regions, one region type comprising contiguous image elements whose light intensity levels are less than a threshold level and the other region type comprising contiguous image elements whose light intensity levels are equal to or greater than said threshold level;

(b) identifying a plurality of points residing on the boundary of one region type which is totally within the field of view defined by the image and which is greater than a predetermined size;

(c) fitting an ellipse to said points;

(d) calculating the distance between two of said points lying on one of the axes of the fitted ellipse and;

(e) calculating the strain induced in the sample as a function of said distance and the predetermined diameter of said circles.

3. A method of automatically determining the strain induced in a sheet metal sample by a forming operation, said sample having a grid of circles of predetermining diameter printed thereon prior to said operation, which circles are stretched into patterns of generally elliptical shape as a result of said operation, said method comprising the steps of:

obtaining a digital representation of the image of a portion of the sample including at least one of said patterns, which digital representation is a sequence of integers, each integer corresponding to the average light intensity of the image at discrete identified image element locations,

storing said integers in addressable locations in the memory of a digital computer, said computer being programmed to perform the steps of:

(a) determining from selected groups of said integer values a threshold level of light intensity which is a gradient-weighted average light intensity in transition areas of the image where the gradient of light intensity exceeds a predetermined value;

(b) segmenting the image into binary regions, one region type comprising contiguous image elements whose light intensity levels are less than said weighted average and the other region type comprising contiguous image elements whose light intensity levels are equal to or greater than said weighted average;

(c) identifying the location of a plurality of points residing on the boundary of one region which is totally within the field of view defined by the image and is greater than a predetermined size;

(d) fitting an ellipse to said points;

(e) identifying those points within a predetermined tolerance distance from the fitted ellipse and refitting an ellipse to the points so identified, reducing said tolerance distance and repeating this step (e) a predetermined number of times or until a predetermined root-mean-square percent error is attained;

(f) calculating at least one of the major and minor diameters of the ellipse fitted in step (e) and;

(g) calculating the strain induced in the sample as a function of the calculated diameter and the predetermined diameter of said circles.

4. Apparatus for automatically determining the strain induced in a sheet metal sample by a forming operation, said sample having a grid of circles of predetermining diameter printed thereon prior to said operation, which circles are stretched into patterns of generally elliptical shape as a result of said operation, said apparatus comprising:

means for obtaining a digital representation of the average light intensity of discrete elements of the image of a portion of the sample including at least one of said patterns;

digital computer means for storing said digital representation in addressable memory locations, said computer being programmed to:

(a) determine a threshold level of light intensity which is a gradient-weighted average light intensity in transition areas of the image where the gradient of light intensity exceeds a predetermined value;

(b) segment the image into binary regions, one region type comprising contiguous image elements whose light intensity levels are less than said threshold and the other region type comprising contiguous image elements whose light intensity levels are equal to or greater than said threshold;

(c) identify a plurality of points residing on the boundary of one region type which is totally within the field of view defined by the image and is greater than a predetermined size;

(d) fit an ellipse to said points;

(e) identify those points within a predetermined tolerance distance from the fitted ellipse and refit an ellipse to the points so identified, reduce said tolerance distance and repeat this step (e) a predetermined number of times or until a predetermined root-mean-square percent error is attained;

(f) calculate at least one of the diameters of the fitted ellipse and;

(g) calculate the strain induced in the sample as a function of the calculated diameter and the predetermined diameter of said circles.

5. Apparatus for automatically determining the strain induced in a sheet metal sample by a forming operation, said sample having a grid of circles of predetermining diameter printed thereon prior to said operation, which circles are stretched into patterns of generally elliptical shape as a result of said operation, said apparatus comprising:

means for obtaining a digital representation of the average light intensity of discrete elements of the image of a portion of the sample including at least one of said patterns;

digital computer means for storing said digital representation in addressable memory locations, said computer being programmed to:

(a) segment the image into binary regions, one region type comprising contiguous image elements whose light intensity levels are less than a threshold and the other region type comprising contiguous image elements whose light intensity levels are equal to or greater than a threshold;

(b) identify a plurality of points residing on the boundary of one region type which is totally within the field of view defined by the image and is greater than a predetermined size;

(c) fit an ellipse to said points;

(d) calculate at least one of the diameters of the fitted ellipse and;

(e) calculate the strain induced in the sample as a function of the calculated diameter and the predetermined diameter of said circles.