Toly Chen

Feasibility Evaluation and Optimization of a Smart Manufacturing System Based on 3D Printing: A Review

As a major enabling factor for Industry 4.0, three‐dimensional (3D) printing faces technical and managerial concerns that may hinder its sustainable development. In this study, four technical challenges are reviewed as follows: time‐consuming 3D object design, limited types of usable materials, low precision, and low productivity. Seven managerial concerns are also discussed as follows: 3D object database management, intellectual property rights of 3D printing, business innovation, ubiquitous manufacturing, lean manufacturing, globalization and deglobalization, and feasibility evaluation and optimization. Then, this study asserts that technical challenges should be addressed to ensure the feasibility of a 3D printing application in a manufacturing context, whereas managerial concerns should be addressed to advance and optimize a 3D printing application. Based on the discussion, to maximize profit, a smart manufacturing system based on 3D printing should continually provide 3D objects of interest to customers, or join as many ubiquitous manufacturing networks as possible.

A heterogeneous fuzzy collaborative intelligence approach for forecasting the product yield

Abstract For manufacturers, forecasting the future yield of a product is a critical task. However, a yield learning process involves considerable uncertainty, rendering the task difficult. Although a few fuzzy collaborative intelligence (FCI) methods have been proposed in recent years, they are not problem-free. Hence, to overcome the challenges associated with these methods and to improve the accuracy of future yield forecasts, a heterogeneous FCI approach is proposed in this study. In this method, an expert applies mathematical-programming-based or artificial-neural-network-based methods (i.e., heterogeneous methods) to model an uncertain yield learning process. Subsequently, fuzzy intersection narrows the possible range of the future yield, and finally, an artificial neural network derives a crisp (representative) value. The effectiveness of the proposed heterogeneous FCI approach was successfully demonstrated by considering data obtained from a factory manufacturing dynamic random access memory devices. The approach achieved an average increase of 21% in the forecasting accuracy compared with existing approaches.

An innovative fuzzy and artificial neural network approach for forecasting yield under an uncertain learning environment

Most methods for fitting an uncertain yield learning process involve using fuzzy logic and solving mathematical programming (MP) problems, and thus have several drawbacks. The present study proposed a novel fuzzy and artificial neural network (ANN) approach for overcoming these drawbacks. In the proposed methodology, an ANN is used instead of an MP model to facilitate generating feasible solutions. A two-stage procedure is established to train the ANN. The proposed methodology and several existing methods were applied to a real case in a semiconductor manufacturing factory, and the experimental results showed that the methodology outperformed the existing methods in the overall forecasting performance.

An ANN approach for modeling the multisource yield learning process with semiconductor manufacturing as an example

The future yield of a product in a wafer fabrication factory is estimated.An innovative approach is proposed for modeling the yield learning process with artificial neural networks.The effects of various sources of yield learning were separated. Estimating the future yield of a product is a crucial task for semiconductor manufacturers. However, existing methods cannot differentiate the effects of various sources of yield improvement. To address this concern, this study proposes an innovative approach for modeling the yield learning process of a semiconductor product with artificial neural networks, which enable separating the effects of various sources of yield learning. Two real cases were used to demonstrate the effectiveness of the proposed methodology.

A fuzzy collaborative intelligence approach for estimating future yield with DRAM as an example

To increase the ecological sustainability of manufacturing, enhancing the yield of each product is a critical task that eliminates waste and increases profitability. An equally crucial task is to estimate the future yield of each product so that the majority of factory capacity can be allocated to products that are expected to have higher yields. To this end, a fuzzy collaborative intelligence (FCI) approach is proposed in this study. In this FCI approach, a group of domain experts is formed. Each expert constructs an artificial neural work (ANN) to fit an uncertain yield learning process for estimating the future yield with a fuzzy value; in past studies, however, uncertain yield learning processes were modeled only by solving mathematical programming problems. In this research, fuzzy yield estimates from different experts were aggregated using fuzzy intersection. Then, the aggregated result was defuzzified with another ANN. A real dynamic random access memory case was utilized to validate the effectiveness of the proposed methodology. According to the experimental results, the proposed methodology outperformed five existing methods in improving the estimation accuracy, which was measured in terms of the mean absolute error and the mean absolute percentage error.

Integer nonlinear programming and optimized weighted-average approach for mobile hotel recommendation by considering travelers’ unknown preferences

The existing mobile hotel recommendation systems are usually subject to a difficult problem—travelers choose dominated hotels. This problem is difficult to resolve because there is no reason to recommend a hotel that is inferior to another in all aspects. To address this problem, an artificial dimension is added to each hotel to model unknown personal preferences. The possible values along the artificial dimension and the weight associated with it are derived by solving an integer nonlinear programming problem. Thus, the proposed methodology hybridizes objective and subjective weights. An illustrative example is provided to show the applicability of the proposed methodology. In addition, a field study was conducted in a small region of Seatwen District, Taichung City, Taiwan to evaluate the possible advantages of the proposed methodology over existing methods. The experimental results showed that the proposed methodology outperformed five existing methods in improving the successful recommendation rate, with the most significant advantage being up to 33xa0%. Furthermore, the recommendation results generated using the proposed methodology were found to be less risky.

Optimal multiple-period scheduling and sequencing of operating room and intensive care unit

Operating rooms (ORs), one of the most crucial hospital resources that generate among the highest costs, are prone to bottlenecking. This paper, therefore, proposes optimization models for multiple-period scheduling of patients in ORs and intensive care units (ICUs) as well as for sequencing of patients in the operating theatre. The first model considers scheduling at a minimal total cost, comprising hospitalization, undertime, overtime, and cancelation costs. The second model involves ICU scheduling and patient sequencing in the operating theatre at a minimal total overtime cost. The scheduling and sequencing models were implemented at a hospital that offers comprehensive surgical procedures for all surgical specialties. The research results are expected to improve patient satisfaction and resource efficiency as well as reduce hospital expenses by minimizing overtime, undertime, hospitalization costs, and cancelation costs. These optimization models may provide substantial assistance to decision makers and planners in facilitating coordination among available hospital resources, including ORs, surgeons, and patients, thereby increasing the productivity and efficiency of ORs and ICUs in providing high-quality care. Future research should consider OR scheduling for unexpected events.

A multi-granularity approach for estimating the sustainability of a factory simulation model: semiconductor packaging as an example

Dynamic factory simulation has been considered as an effective means to control a factory. However, the large amount of money, time, efforts, and know-how required for conducting a factory simulation study force a factory to pursue the persistent application of the factory simulation model, i.e. the sustainability of the factory simulation model. Therefore, strategies are required to facilitate the rapid establishment of the factory simulation model, to lower the technical requirements of the model, and to reduce the effort and time spent on simulation tasks, thus increasing users’ willingness to continue the application of the model. However, such issues have rarely been discussed. In addition, no method is available for estimating the sustainability of a factory simulation model. To address this problem, short-time evidence was analyzed rather than observing data over a long period. Then, a multi-granularity approach is proposed to estimate the sustainability of a factory simulation model based on these evidences. The proposed methodology has been applied to the simulation of a real semiconductor packaging facility. According to the experimental results, the multi-granularity approach reduced the input space by 89% and maintained a very high estimation accuracy. In addition, it also saved considerable time in building the models for estimating sustainability. Furthermore, without the multi-granularity approach, the sustainability of the factory simulation model could be observed only after a long period.