Long-Sheng Chen

Using Functional Virtual Population as assistance to learn scheduling knowledge in dynamic manufacturing environments

When a scheduling environment is static and system attributes are deterministic, a manufacturing schedule can be obtained by applying analytical tools such as mathematical modelling technology, dynamic programming, the branch- and-bound method or other developed searching algorithms. Unfortunately, a scheduling environment is usually dynamic in a real manufacturing world. A production system may vary with time and require production managers to change schedule repeatedly. Therefore, the main aim here was to find a scheduling method that could reduce the need for rescheduling. An approach called Functional Virtual Population was proposed as assistance to learn robust scheduling knowledge for manufacturing systems under rationally changing environments. The used techniques include machine learning with artificial neural networks and IF-THEN scheduling rules. To illustrate the study in detail, a simulated flexible manufacturing system consisting of four machines, four parts, one automatic guided vehicle and eight buffers was built as the foundation for learning the concept. Also, Pythia software (a back-propagation-based neural networks) was employed as the learning tool in the learning procedure.

Knowledge acquisition through information granulation for imbalanced data

Abstract When learning from imbalanced/skewed data, which almost all the instances are labeled as one class while far few instances are labeled as the other class, traditional machine learning algorithms tend to produce high accuracy over the majority class but poor predictive accuracy over the minority class. This paper proposes a novel method called ‘knowledge acquisition via information granulation’ (KAIG) model which not only can remove some unnecessary details and provide a better insight into the essence of data but also effectively solve ‘class imbalance’ problems. In this model, the homogeneity index (H-index) and the undistinguishable ratio (U-ratio) are successfully introduced to determine a suitable level of granularity. We also developed the concept of sub-attributes to describe granules and tackle the overlapping among granules. Seven data sets from UCI data bank, including one imbalanced diagnosis data (pima-Indians-diabetes), are provided to evaluate the effectiveness of KAIG model. By using different performance indexes, overall accuracy, G-mean and Receiver Operation Characteristic (ROC) curve, the experimental results comparing with C4.5 and Support Vector Machine (SVM) demonstrate the superiority of our method.

Using SVM based method for equipment fault detection in a thermal power plant

Due to the growing demand on electricity, how to improve the efficiency of equipment in a thermal power plant has become one of the critical issues. Reports indicate that efficiency and availability are heavily dependant upon high reliability and maintainability. Recently, the concept of e-maintenance has been introduced to reduce the cost of maintenance. In e-maintenance systems, the intelligent fault detection system plays a crucial role for identifying failures. Data mining techniques are at the core of such intelligent systems and can greatly influence their performance. Applying these techniques to fault detection makes it possible to shorten shutdown maintenance and thus increase the capacity utilization rates of equipment. Therefore, this work proposes a support vector machines (SVM) based model which integrates a dimension reduction scheme to analyze the failures of turbines in thermal power facilities. Finally, a real case from a thermal power plant is provided to evaluate the effectiveness of the proposed SVM based model. Experimental results show that SVM outperforms linear discriminant analysis (LDA) and back-propagation neural networks (BPN) in classification performance.

A neural network based approach for sentiment classification in the blogosphere

Recognizing emotion is extremely important for a text-based communication tool such as a blog. On commercial blogs, the evaluation comments by bloggers of a product can spread at an explosive rate in cyberspace, and negative comments could be very harmful to an enterprise. Lately, researchers have been paying much attention to sentiment classification. The goal is to efficiently identify the emotions of their customers to allow companies to respond in the appropriate manner to what customers have to say. Semantic orientation indexes and machine learning methods are usually employed to achieve this goal. Semantic orientation indexes do not have good performance, but they return results quickly. Machine learning techniques provide better classification accuracy, but require a lot of training time. In order to combine the advantages of these two methods, this study proposed a neural-network based approach. It uses semantic orientation indexes as inputs for the neural networks to determine the sentiments of the bloggers quickly and effectively. Several actual blogs are used to evaluate the effectiveness of our approach. The experimental results indicate that the proposed approach outperforms traditional approaches including other neural networks and several semantic orientation indexes.

An association-based case reduction technique for case-based reasoning

Case-based reasoning (CBR) is a type of problem solving technique which uses previous cases to solve new, unseen and different problems. Although a larger number of cases in the memory can improve the coverage of the problem space, the retrieval efficiency will be downgraded if the size of the case-base grows to an unacceptable level. In CBR systems, the tradeoff between the number of cases stored in the case-base and the retrieval efficiency is a critical issue. This paper addresses the problem of case-base maintenance by developing a new technique, the association-based case reduction technique (ACRT), to reduce the size of the case-base in order to enhance the efficiency while maintaining or even improving the accuracy of the CBR. The experiments on 12 UCI datasets and an actual case from Taiwans hospital have shown superior generalization accuracy for CBR with ACRT (CBR-ACRT) as well as a greater solving efficiency.

Intelligent ICA-SVM fault detector for non-Gaussian multivariate process monitoring

Recently, the independent component analysis (ICA) has been widely used for multivariate non-Gaussian process monitoring. For principal component analysis (PCA) based monitoring method, the control limit can be determined by a specific distribution (F distribution) due to the PCA extracted components are assumed to follow multivariate Gaussian distribution. However, the control limit for ICA based monitoring statistic is determined by using kernel density estimation (KDE). It is well known that the KDE is sensitive to the smoothing parameter, and it does not perform well with autocorrelated data. In most cases, the calculated ICA based monitoring statistic is usually autocorrelated. Thus, this study aims to integrate ICA and support vector machine (SVM) in order to develop an intelligent fault detector for non-Gaussian multivariate process monitoring. Simulation study indicates that the proposed method can effectively detect faults when compare to methods of original SVM and PCA based SVM in terms of detection rate.

Integrating independent component analysis and support vector machine for multivariate process monitoring

This study aims to develop an intelligent algorithm by integrating the independent component analysis (ICA) and support vector machine (SVM) for monitoring multivariate processes. For developing a successful SVM-based fault detector, the first step is feature extraction. In real industrial processes, process variables are rarely Gaussian distributed. Thus, this study proposes the application of ICA to extract the hidden information of a non-Gaussian process before conducting SVM. The proposed fault detector will be implemented via two simulated processes and a case study of the Tennessee Eastman process. Results demonstrate that the proposed method possesses superior fault detection when compared to conventional monitoring methods, including PCA, ICA, modified ICA, ICA-PCA and PCA-SVM.

C-Kano model: a novel approach for discovering attractive quality elements

The theory of attractive quality (the Kano model) can offer a better understanding of how customers evaluate products, and helps practitioners to focus on the most important quality attributes to improve. However, in practice, users of the Kano model usually cannot find an attractive or one-dimensional quality element due to an improperly-designed questionnaire, poorly-defined quality attributes or lifecycle of quality attributes. Kano also indicated that attractive quality creation to bring a must-be product back to an attractive product is a tough task. To solve these issues, this study proposed a novel Creativity-based Kano model (C-Kano model) which integrates the creativity techniques, TRIZ and SCAMPER, into the traditional Kano model. The proposed C-Kano model not only discovers customer needs, but also creates attractive quality elements. In this study, an actual quality survey of Massively Multiplayer Online Role Playing Game (MMORPG) was used to evaluate the effectiveness of the C-Kano model. Compared with the traditional Kano model, the experimental results show that our proposed model is superior in discovering attractive quality elements.

A process monitoring scheme based on independent component analysis and adjusted outliers

Recently, the independent component analysis (ICA) has been widely used for non-Gaussian multivariate process monitoring. An elliptical type measure is traditionally used for ICA-based process monitoring. However, it will not work appropriately since the extracted ICA components exhibit skewed distribution. Thus, this study aims to develop a novel process monitoring scheme for ICA. The basic idea of the proposed method is to first screen out outliers in order to describe well majority for training dataset. Hereafter, a rectangular type measure is applied to monitor the process. The efficiency of proposed monitoring scheme will be implemented via a five variables simulation example and a case study of Tennessee Eastman process. Results indicate that the proposed method cannot only deal with the contaminated training dataset but also shows superior fault detection ability when compared with alternative methods.

A neural network based information granulation approach to shorten the cellular phone test process

In the cellular phone OEM/ODM industry, reducing test time and cost are crucial due to fierce competition, short product life cycle, and a low margin environment. Among the inspection processes, the radio frequency (RF) function test process requires more operation time than any other. Hence, manufacturers need an effective method to reduce the RF test items so that the inspection time can be reduced while maintaining the quality of the RF function test. However, traditional feature selection methods such as neural networks and genetic algorithm lead to a high level of Type II error in the situation of imbalanced data where the amount of good products is far greater than the defective products. In this study, we propose a neural network based information granulation approach to reduce the RF test items for the finished goods inspection process of a cellular phone. Implementation results show that the RF test items were significantly reduced, and that the inspection accuracy remains very close to that of the original testing process. In addition, the Type II errors decreased as well.