| Flexible Printed Circuit board(FPC)of membrane keyboard based on polyester film as the backing material and metallic silver as wire is with the high density of wiring,light weight,thin thickness and good bending property characteristics.It is a kind of using the new base material developed in recent years with the new wire material and new process of plate,which is mainly used as a laptop keyboard circuit board.The manufacturing process of flexible circuit board of membrane keyboard involves more than 90 processes,among which the process involving apparent defect detections of more than 10 processes,such as silver line printing quality detection,silicone cap pasting quality detection,etc.At present,the main process is visual detection,and the rate of leakage detection is difficult to control.Silver wire printing is the first process of manufacturing flexible circuit board of membrane keyboard.Through the screen printer,the silver paste is printed on the polyester film.However,the silver wire defects can be caused by the grid blockage or breakage,and the thin or thick silver paste solution.The width and spacing of the silver line are usually 0.2-0.4 mm,and the wiring is dense.While the printing cycle of the silver line is 2-4 seconds,the printing area can be over 0.3 square meters,and the leakage rate is difficult to control through manual visual detection.However,due to the excellent bending property of polyester film substrate,it is easy to generate flexible deformation.Therefore,the detection algorithm developed previously for typical silver wire defects of PCB or slightly deformed FPC is no longer applicable.At the same time,new process and new materials have brought new non-typical defects.In the process of silicone cap pasting,the combination of silicone cap and key cap constitutes to the key switch.When the deviation of silicone cap pasting exceeds the standard,it will affect the assembly cooperation with the key cap,resulting in the key failure.The current detection method is to buckle the finished product on the mold.If the silicone cap is not pasted,all the silicone caps can enter the mold hole.This method is slow in detection and can only be suitable for sampling inspection.This paper mainly studies the quality detection methods of the two links of silver wire printing and silicone cap pasting,among which the two typical defects of circuit break and embossment are mainly studied in the silver wire printing defects,as well as the detection problem of the non-typical grout defect.In the process of silicone cap pasting,the problem of pasting excursion detection is studied.The primary researches are as follows.In the aspect of detecting silver wire circuit break: After analyzing the morphological feature of trace circuit break and mathematical mechanism of bare FPC flexible deformation,we proposed the detection idea based on topological structure invariance.Taking advantage of describing the trace region with the homotopic skeleton,the circuit break is detected qualitatively using the topological variation of the pure skeleton without noise branches,and the circuit break is located accurately via the pure skeleton endpoints.The model of Dynamic Template Scale Space(DTSS)is proposed for acquiring pure skeleton,in which the process of removing skeleton branches is considered as gradual multiscale image sequences according to a rule and removing the branches turns into extracting specific members from the sequence.This method can be applied in the cases requiring removing branches,and it is a novel ideal and method of removing branches.Then according to the characteristic that the drainage basin of skeleton endpoints caused by circuit break always intersect with other skeletons after the extension following the flow and combining the priori information such as the basic configuration of the trace,the model of circuit break measurement is built.The algorithm is validated on the established image database,and the experiment results show that the FDR is as low as 1.97 percent on the premise of without missing inspection,which decreases 8.50 percent at least comparing to other algorithms and proves the robustness to flexible deformation of this method.In the aspect of detecting silver wire embossment: A method of minimum distance of boundary sequence is proposed to detect the embossment,which can be used to measure the distance value and distance mutation of any disjoint curvilinear pairs.It can be extended to other similar situations in digital image processing that can be expressed by two basic parallel lines,and it is not affected by the change of the width and direction of the line due to the flexible deformation of the board.A novel 3D curved surface description model called Cylindrical Rectification Model(CRM)is proposed.This model is extended on the basis of the 2D shape Point Distribution Model(PDM).In this paper,PDM and CRM are combined to distinguish them according to the differences of the contour and gray distribution surface between the embossment and background speckle noise on the two-dimensional plane.The experimental results show that:(1)In terms of embossment detection,the average detection accuracy of the method in this paper is 99.13%,which is improved by at least 6.5% compared with other typical methods.This shows the advantages of the method in this paper and shows the excellent anti-deformation ability of the method compared with other methods.(2)In terms of distinguishing false convex noises,the error rate of the method in this paper is as low as 4.21%.In the aspect of detecting silver wire grout: The defect detection of grout is studied from the aspect of texture,and a novel texture description mode called Local Markov Pattern(LMP)is proposed.LMP describes the strength and weakness of the gradation relation between adjacent texture units,which can distinguish the noise of grout with strong gradation relation from the noise with weak gradation relation.In terms of the design of overall detection scheme,the idea of detecting from rough to fine is put forward in this paper.Firstly,grout pixels are treated as outliers and the quadrangle method is used for preliminary extraction based on the probability distribution theory.Then the accurate extraction is realized according to the definition of LMP texture description.The experimental result shows that this method is superior to other methods,which indicates that:(1)In the aspect of grout detection,the LMP feature defined in this paper is effective,which provides a new feasible texture description method for the digital image processing field.(2)The whole scheme developed in this paper is effective for the grout detection and has important practical application value.The above research methods of naked plate printing defects are tested online on the spot.The test results showed that the error detection rate is still low under the guarantee with no defect omission.Specifically,the error detection rates of circuit breaking,embossment and grout are 0.88%,1.33% and 2.6% respectively,indicating the effectiveness and stability of the algorithm in this paper.At the present,the system has been applied to the online quality inspection of thin film keyboard and mass production.In the aspect of detecting rubber cap skewing: The center of the rubber cap on the image is used as the feature point,and the registration is carried out through flexible space transformation and the theoretical center of rubber cap on CAD graph.And then the skewing is detected through the spatial distance value.Among them,three key techniques of flexible registration,namely the definition of similarity measure,the calculation of space transformation parameters and the optimization strategy,are lucubrated.In terms of similarity measure,a similarity measure index with the minimum tolerance deviation is defined,which mainly uses the relative position relation between feature points and takes the influence of outlier value into account.In terms of space transform method,based on the classical image registration method of B spline,a space transform parameter solution method with feature point transform quantity and control grid moving quantity as non-closed loop is proposed,in which the basic idea is to derive the deformation field of the target point by using the optimal spatial transformation of the feature points of the 8-neighborhood grid range of the control grid where the target point is located,and then to solve the spatial transformation parameters of the target point.This method avoids the closed-cycle effect of space transform and deformation field iteration of the classical method.In terms of optimization strategy,this paper proposes a piecewise hierarchical multiresolution optimization method based on fuzzy connectivity,mainly using fuzzy link under the control of the upper grid transformation after the fragmentation feature points according to the evaluation of attribute.And then only the feature points against the conditions are refined,so as to realize the saving of resources in space and time,which provides a new solution idea and method of parameter optimization.The method in this paper has been verified by experiment and tested on the production site.The results show that on the premise of without missing inspection,the FDR is as low as 9.4 percent,which is superior to the current artificial visual method in the detection accuracy or accuracy,and has great practical application value. |
PDF Full Text Request