Advanced Visual Inspection Technology with 2-Dimensional Motion Images for Film-shaped Products Sumitomo Chemical Co., Ltd. Industrial Technology & Research Laboratory Osamu HIROSE Maya OZAKI This paper presents an advanced technology for visual inspection of film products. Usually, line sensor is used to inspect defects in web-shape products such as long films. However, one-dimensional image data captured via line sensors always includes restrictive optical information about defects. Therefore, inspection performance is limited. With the advantage of using area sensors, two-dimensional images containing more optical information can be obtained. The authors have established a novel imaging procedure which is based on two-dimensional motion images. The experimental results show that the new image processing framework enhances the appearance of defects. The authors also achieved an in-line web inspection system for film manufacturing lines. (a) Machine direction(md) 2 1) 2 1) 1 Fig. 1 1 (b) Line sensor Fig. 1 Light source Image processing unit Transverse direction(td) Traditional web inspection system Camera group (a) and (b) indicate bright-field mode and dark-field mode respectively. 2) 1 3) 2013
10µm 2 2 Line Composition and Integration ; LCI 4) 1 2 2 FPGA Field Programmable Gate Array ; CG Fig. 1 Transverse direction ; TD 1 Machine direction ; MD TD MD TDMD 1 1 Fig. 2 3 A B C Fig. 1 2 Defect A Defect B Defect C Line sensor Light shield (knife-edge) Fig. 2 Dark-field On-edge Bright-field ~50μm Light source ~50μm Various observation methods for diferent types of defects 2013
Fig. 2 3 50µm 1 35 Fig. 3 (a) (b) θ a d Fig. 4 a = 50µm d = 200mm θ=tan 1 (a/2d) 0.007degree Light source Fig. 4 θ a d Flat surface Distorted surface Deviation of line sensor view due to film distortion Line sensor Light source Line sensor view Reflection image of light source Defect Film Reflection image of light source Defect Line sensor view Defect Intensity Intensity TD position (a) Flat surface TD position (b) Distorted surface Fig. 3 Film distortion and its adverse affect on observation of a defect Each lower plot shows intensity profile on a dashed line (line sensor view) in the upper image. 2013
Fig. 5 Area sensor Fig. 6 (a) 1 40MHz 1 10,0004,000 112 Fig. 6 (b) FPSFrames Per Second Fig. 5 Light source Machine direction(md) Image processing unit Transverse direction(td) 2-dimentional web inspection system 10,000 1 1001,000 1,000 ; 1 10100FPS 1/10 1/100 2 Line sensor Area sensor Fig. 6 MD TD (b) 2D-images by using area sensor Comparison of image constructions 2 Fig. 7 (a) 2 1 Fig. 7 (b)p 1 p 4 Fig. 7 (a)p 1, p 2 TD position TD position scan 0 scan 1 scan 2 scan 3 : : (a) 1D-images by using line sensor MD TD Time-series frames frame 0 frame 1 frame 2 frame 3 : : t t 2013
Area sensor Angle of view MD Film Light shield Light source Position p 1 p 2 p 3 p 4 MD Dark-field On-edge Bright-field (a) Configuration of the optical system Position p 1 (Dark-field) p 2 (Dark-field) p 3 (On-edge) p 4 (Bright-field) Defect A Defect B MD TD Defect C (b) Observation of defects in 2D motion images Time sequence Fig. 7 Examples of defect images by using an area sensor p 3 p 4 Ap 2 p 3 p 4 B p 4 C C 2 2 LCI ; Line Composition and Integration 21 2, 3 Bright-field On-edge Dark-field Original image(2d motion images) FPGA LCI image (1 line/frame) Image processing unit (FA computer) Camera Link DMA Area sensor Capture board Fig. 8 Conceptual diagram about LCI 2013
Fig. 8 2 1 LCI 1 LCI LCI LCIFig. 9 3 3 LCI FIFO Fig. 9 (b) } 1 1LCI 1 Fig. 9 (a) (b) Fig. 7 (a) 2 LCI 2 FPGA Fig. 8 FPGALCI LCI LCI LCI FIFO t Frame # i 1 Frame # i Frame # i + 1 Line# #1 #2 #3 #n FIFO memory * * * * Op. Op. Op. Op. (a) Original image (b) Line composite image (c) Differential operator (d) Integration (e) LCI image Fig. 9 LCI procedure 2013
FPGA 2 Table 1 Fig. 10 Fig. 11 (a) (b) (c) LCI LCI LCI Table 1 Defect samples Defect type Slight concave Particle Height : 1 μm Less than 100 μm in Typical size Width : 1mm or more diameter Distortion of On-edge transmitted image Observation method Frame # i Frame # i+5 Frame # i+10 (b-1)bright-field (420μm from edge) (b-2)bright-field (210μm from edge) Traditional method: Distortion of transmittedimage Defect Frame # i+15 (b-3) Just on-edge Frame # i+20 (b-4)dark-field (280μm from edge) (a) Original images (b) Line composite images (c) LCI image Fig. 10 LCI experimental result (Defect type: Slight concave) Frame # i Frame # i+5 (b-1)bright-field (420μm from edge) Traditional method: On edge Frame # i+10 Frame # i+15 (b-2)bright-field (210μm from edge) Defect (b-3) Just on-edge Defect Frame # i+20 (a) Original images (b-4)dark-field (280μm from edge) (b) Line c composite images (c) LCI image Fig. 11 LCI experimental result (Defect type: Particle) 2013
LCI Fig. 12 LCI Main defect (a) Traditional method (Bright-field image via line sensor) Fig. 13 (b) LCI image An example of LCI image observed film surface with complex shape Defect and observation method Traditional method LCI image Convex defect (Dark field) Particle 1 (On edge) Slight concave (Distortion of transmitted image) Particle 2 (Bright field) Fig. 12 Comparison of defect images between LCI and traditional methods Fig. 13 10µm 0.1µm (a) (b) LCI 3 3 LCI LCI CG Fig. 14 z z =0 xtdy MDΔH T f z = z r z = z f z = z s 2013
Camera x(td) Imaging device Q Q P Lens y(md) Imaging device d x(td) P Q Q s 2 y(md) ΔH R R Film Defect model MD Film R R z r T f s 1 Focal plane z S Illumination plane F S z f z s Focal plane Illumination plane z (a) 3D view (b) Side view Fig. 14 Coordinate system for LCI simulation LCI y < 0 r M = = 1 f = r 0 1 s 1 + s 1 s 2 1 s 2 (1) (2) CG P P PQ z = z f 1 F SS P 12 MD2 2 5), 6) 1 f ΔH r 0 r [m/pixel] 1 PF(x f, y f, z f ) F x f = M x p y f = M y p z f = s 1 + s 2 + ΔH s 1 s 2 z=0 2 R QF Fig. 15 Ray1 Ray2 k(0 < k < 1) z = g(x, y) = Ae 2 x 2 + y 2 σ 2 (3) (4) A 2013
σ z=g(x, y) (x, y) x y Fig. 16 Ray 1 Ray 2 r i r i = ( x i, y i, z i ) r o = ( x o, y o, z o ) n = ( x n, y n, z n ) θ i n i, n o r o (6) Film Fig. 15 Particle defect z attenuation Variation of rays affected by defects y r o s R(x r, y r, z r ) R (x r, y r, z r ) Defect model : z=g(x, y) x o = n i n o = μx i + Γ x n y o = μy i + Γ y n z o = μz i + Γ z n where μ = n i /n o x i + S i = r i n, S o = r o n Γ = S o μs i S o n i n o S i = 1 μ 2 + μ 2 S i 2 μ S i x n (7) n b a (7) Fig. 15 R s x Fig. 16 Defect model and its normal vector g g n =,, 1 (5) x y r i n r o Fig. 17 r i r θ i in t o n i t i n o 3 R s S P1 1 S 1 d s = 2 1.22fλ/d (8) λ fd PS P P θ o r on Fig. 17 n r o The change of ray direction at the medium boundary surface LCI Fig. 18 2013
1µm 2σ 1mm (1) Original CG MD 2 1LCI 100 10 (2) Line composite images (3) LCI LCI (b) CG Base unit PC-1 PC-2 PC-n Rotary encoder Ethernet Capture board equipped with LCI (a) Configuration of in-line inspection system CMOS area sensor LED lamp with Light shield MD position (3) Generated LCI image Fig. 19 TD position (b) An example of defect detection map In-line web inspection system by using area sensors equipped with LCI technology (1) Original CG (10frames) (2) Line composite images (a) Simulation result Fig. 18 (b) Experimental result (Observed image of actual defect sample) Comparison between simulation and experimental result Defect model: Convex defect (defect height 1μm, defect width 1mm) Fig. 19 PC 1 PC LCI 1 FPGALCI LCI DMA 1 1 2 LCI LED MD CMOS CMOS LCI TD MD 2013
1 2 Fig. 18 (b) 2 LCI LCI 2 LCI CG CG 1),, (2012). 2),, (2007). 3),, (2008). 4),, 2009, 223. 5) F. Jenkins and H. White, Fundamentals of Optics, McGRAW-HILL (1957). 6),, (1988). PROFILE Osamu HIROSE Maya OZAKI 2013