Chien-Chih Wang

Machine vision-based gray relational theory applied to IC marking inspection

In the semiconductor industry, IC marking error remains a problem. The objective of this study is to identify IC marking using gray relational analysis. The gray theorem determines the gray relational grades of all of the selected factors by choosing the highest gray relational grade, even under incomplete information circumstances. In an IC marking identification procedure, an image is rotated and segmented first. Second, thresholding and thinning operations are applied to reduce the calculation complexity and extract features from the segmented image. Finally, the gray relational analysis method is applied to inspect the IC markings. The identification rate reaches 97.5%. As compared to traditional methods, there are three advantages in gray relational analysis: 1) No large amount of data is needed; 2) No specific statistical data distribution is required; and 3) There is no requirement for the independency of the factors to be considered. It is an easy and practical method in the field of IC marking inspection.

Liquid crystal display surface uniformity defect inspection using analysis of variance and exponentially weighted moving average techniques

Display quality is part of the final liquid crystal display (LCD) inspection process before shipping. A ‘limited sample’ is provided based on the agreement between the manufacturer and the customer. This inspection usually includes an operator who compares the LCD product with the limit sample using the naked eye. The procedure often causes controversy in the manufacturing plant and between the manufacturer and customers. This study attempts to establish a more objective, automatic method to determine the MURA-type defects in LCD panels. A luminance meter is used as the measurement device. An LCD panel is divided into 144 areas. Five points are measured to obtain the luminances. Analysis of variance and the exponentially weighted moving average techniques are applied to determine the existence of MURA defects. Fifty normal LCD panels and 50 MURA defects panels were used to test the inspection method. All 50 LCD panels with MURA defects were correctly identified using the proposed inspection method. The proposed inspection method can help LCD manufacturers reduce the variation in LCD panel inspection results and establish a better relationship with customers through a common inspection mechanism.

Using data mining techniques for multi-diseases prediction modeling of hypertension and hyperlipidemia by common risk factors

Research highlights? This paper proposes a two-phase analysis procedure to simultaneously predict hypertension and hyperlipidemia. ? Common risk factors of these diseases picked up by data mining and majority vote. ? This study uses common risk factors to build MARS predictive models for hypertension and hyperlipidemia. Many previous studies have employed predictive models for a specific disease, but fail to note that humans often suffer from not only one disease, but associated diseases as well. Because these associated multiple diseases might have reciprocal effects, and abnormalities in physiological indicators can indicate multiple associated diseases, common risk factors can be used to predict the multiple associated diseases. This approach provides a more effective and comprehensive forecasting mechanism for preventive medicine. This paper proposes a two-phase analysis procedure to simultaneously predict hypertension and hyperlipidemia. Firstly, we used six data mining approaches to select the individual risk factors of these two diseases, and then determined the common risk factors using the voting principle. Next, we used the Multivariate Adaptive Regression Splines (MARS) method to construct a multiple predictive model for hypertension and hyperlipidemia. This study uses data from a physical examination center database in Taiwan that includes 2048 subjects. The proposed analysis procedure shows that the common risk factors of hypertension and hyperlipidemia are Systolic Blood Pressure (SBP), Triglycerides, Uric Acid (UA), Glutamate Pyruvate Transaminase (GPT), and gender. The proposed multi-diseases predictor method has a classification accuracy rate of 93.07%. The results of this paper provide an effective and appropriate methodology for simultaneously predicting hypertension and hyperlipidemia.

Application of classification techniques on development an early-warning system for chronic illnesses

Highlights? The paper used the health examinations to uncover risk factors. ? The risk factors used to construct the early warning criteria by data mining. ? The early-warning criteria will lead to detect chronic illnesses earlier. ? The patients can reduce the risk of suffering from the relate complications. Chronic disease has gradually become a major fatality cause in Taiwan. Being afflicted with such illnesses elevate vulnerability to other complications as well. Therefore, this paper adopts a preventative perspective and ascertains the impacts of important physiological indicators and clinical test values for various chronic illnesses. The paper investigates five chronic diseases: hypertension, diabetes, cardiovascular disease, disease of the liver, and renal disease. Utilizing chronic diseases risk factors to establish early-warning criteria may reduce the complication occurrence rate. K-nearest neighbor, linear discriminant analysis, and sequential forward selection are utilized, which is divided into two parts. The first part classifies and screens both healthy persons and those affiliated with the abovementioned chronic illnesses for characteristic value determination. The second part determines the critical value of the important risk factors of each chronic illness and builds early-warning criteria to recognition the chronic illnesses. This paper uses data from a medical center in Taiwan to verify the proposed methodology. The results reveal that classifying materials and screening important factors are both positively efficient with a corrected rate of 80%. Additionally, through the important factors of early-warning criteria, not only can help patients understand the risks of suffering diseases, but also effectively offer diagnosis reference criteria for medical personnel.

Machine vision and background remover-based approach for PCB solder joints inspection

A machine vision and background remover (BR)-based inspection method is proposed in this paper for solder defects inspection on a printed circuit board (PCB). The solder location is identified first using an unloaded PCB as the BR, thus reducing the amount of information needed for the following process. A set of candidate features is then calculated based on both binary and gray-level images. The defects are classified based on box plots of the feature value. The classification correctness reaches 97.3%. This methodology combines solder joints location identification and defects classification, making the inspection of PCB solder joints easier, without the requirement for special lighting or special instruments such as ultra-sonic and thermal sensors.

Machine Vision-Based Defect Detection in IC Images Using the Partial Information Correlation Coefficient

The normalized cross correlation coefficient is a prevalent pattern-matching algorithm in machine vision for industrial inspections. Despite its common use, there are problems with practical applications. For instance, false alarms occur since it is highly sensitive to environmental changes or inspection equipment, not to mention it requires complex calculations. This paper proposes the partial information correlation coefficient (PICC) method to improve the traditional normalized cross correlation coefficient (TNCCC). The PICC uses the technique of significant points to calculate the correlation coefficient. An experiment is also conducted to demonstrate the application through many image samples from the IC industry, such as PCBs, BGAs, and ICs. The results show that the PICC can effectively reduce false alarms in defect detection.

Using simulation techniques to determine optimal operational region for multi-responses problems

Most industrial problems involve multiple-quality characteristics. The methods used for optimising such a multiple-response problem are those normally used for dealing with a single response problem, such as the experimental design, response surface methods (RSM) and regression techniques. There is a major difficulty with this traditional approach on the point estimation of the optimal solution, thus the obtained solution may not be practically feasible. In this research, a simulated regression approach via Monte Carlo simulations and the nonlinear optimisation via the trust-region methods were used to determine an optimal operational region (OOR). In practice, an engineer can refer to this region for adjusting manufacturing parameters in the production environment to minimise the response variations. An example with three controllable variables is used to demonstrate the application of the proposed methodology.

Multivariate analysis-based image enhancement model for machine vision inspection

Image enhancement is an essential procedure in machine vision-based inspection. In practical applications, image enhancement is usually a part of image pre-processing, intended to make the following inspection more effective. The image enhancement method is usually selected by trial-and-error or on the basis of experience. This paper presents an automatic procedure for fast and effective image enhancement. The procedure uses multivariate analysis to automatically construct an optimal image enhancement model. First, an optimally enhanced image was selected from the literature as a basis for the model. Then, the image features were identified and Wilks’ statistic was used for feature selection. Next, discriminate functions were built to select the optimal image enhancement method. To verify the model, 53 training images from the literature and 12 test images from a local company were used in an experimental analysis. The model achieved 98.11% accuracy in selecting the most suitable image enhancement method, and the average increase in contrast was 98% for the 53 training images. The enhancement method selection results for the 12 test images were also in agreement with the 53 training images from the literature. The results show that the proposed method is effective and appropriate for quickly improving image contrast.

Automatic bubble defect inspection for microwave communication substrates using multi-threshold technique based co-occurrence matrix

The microwave communication substrate is currently using manual inspection from ultrasonic image. The bubble defect ratio is an important index for evaluating the quality of the substrate. It is difficult to formalise standards for the quality inspection procedure by an operator since different numbers of bubbles may be realised. This paper proposes a novel inspection procedure for the calculation of the bubble ratio of microwave communication substrate. First, the optimal threshold is determined for binary image segmentation on the basis of a co-occurrence matrix. Further, the image of the microwave communication substrate is divided into three parts, namely, the image skeleton area, soldering area, and IC working area. The image subtraction technique is used to obtain the global and critical regions for bubbles. After data analysis, if the critical region bubble ratio is larger than 10% of the critical region, then it is judged to be a defect. When a substrate is judged to be non-defective, we should further proceed with the global area detection. If the global region bubble ratio is larger than 12.3% of the global region, then it is judged as a defect. An experiment was conducted to demonstrate the application of this technique. The results showed that the inspection accuracy reached 97.49%. The results of this study provide an effective solution for the inspection of the interface bond quality of a microwave communication substrate.

Singular Value Decomposition Based Feature Extraction Technique for Physiological Signal Analysis

Multiscale entropy (MSE) is one of the popular techniques to calculate and describe the complexity of the physiological signal. Many studies use this approach to detect changes in the physiological conditions in the human body. However, MSE results are easily affected by noise and trends, leading to incorrect estimation of MSE values. In this paper, singular value decomposition (SVD) is adopted to replace MSE to extract the features of physiological signals, and adopt the support vector machine (SVM) to classify the different physiological states. A test data set based on the PhysioNet website was used, and the classification results showed that using SVD to extract features of the physiological signal could attain a classification accuracy rate of 89.157%, which is higher than that using the MSE value (71.084%). The results show the proposed analysis procedure is effective and appropriate for distinguishing different physiological states. This promising result could be used as a reference for doctors in diagnosis of congestive heart failure (CHF) disease.