Muh-Cherng Wu

A fuzzy CBR technique for generating product ideas

This paper presents a fuzzy CBR (case-based reasoning) technique for generating new product ideas from a product database for enhancing the functions of a given product (called the baseline product). In the database, a product is modeled by a 100-attribute vector, 87 of which are used to model the use-scenario and 13 are used to describe the manufacturing/recycling features. Based on the use-scenario attributes and their relative weights - determined by a fuzzy AHP technique, a fuzzy CBR retrieving mechanism is developed to retrieve product-ideas that tend to enhance the functions of the baseline product. Based on the manufacturing/recycling features, a fuzzy CBR mechanism is developed to screen the retrieved product ideas in order to obtain a higher ratio of valuable product ideas. Experiments indicate that the retrieving-and-filtering mechanism outperforms the prior retrieving-only mechanism in terms of generating a higher ratio of valuable product ideas.

An effective application of decision tree to stock trading

Abstract This paper presents a stock trading method by combining the filter rule and the decision tree technique. The filter rule, having been widely used by investors, is used to generate candidate trading points. These points are subsequently clustered and screened by the application of a decision tree algorithm C4.5. Compared to previous literature that applied such a combination technique, this research is distinct in incorporating the future information into the criteria for clustering the trading points. Taiwan and NASDAQ stock markets are used to justify the proposed method. Experiment results show that the proposed trading method outperforms both the filter rule and the previous method.

Analysis on machined feature recognition techniques based on B-rep

Abstract Solving the machine feature recognition problem has been widely recognized as a cornerstone for developing an automated process planning system directly linked to a cad system. Various recognition techniques have been developed; however, they are in general deficient in robustness. That is, valid machined features may not be recognized and features which are recognized may not be valid in practice. This paper is intended to analyse the existing machined feature recognition techniques, which are based on the B-rep solid modelling scheme, in order to give the reasons why the robustness problem would occur. The pros and cons for recognizing machined features are also analysed. Finally, a cutter selection methodology, known as process requirement modelling, is introduced; this methodology seems to provide a promising way to solve the machined feature recognition problem.

An enhanced Taguchi method for optimizing SMT processes

Taguchi quality control method is a cost-effective experimental design technique for characterizing the optimal operating parameters for a manufacturing process. Most previous applications of the Taguchi method have been concerned with the optimization of a single-response process and a large amount of satisfactory implementation results have been published. However, for multi-response cases, an effective and user-friendly technique is still lacking. The multi-response technique presented in this paper modifies the traditional Taguchi method by adopting an empirical loss function to sum up the impact of each process variable. Compared with previous methods, this technique is distinguished in two aspects: (1) it theoretically optimizes the single-response case in a more accurate way; (2) it can optimize multi-response processes involving discrete and continuous data types. Considering a surface mount technology (SMT) process for a VGA card (a typical multi-response process involving six variables and nine responses), experiments applying the proposed technique were conducted and showed satisfactory results. The yield rate for the SMT process improved from 88% to 98%.

A neural network approach to the classification of 3D prismatic parts

This paper presents a neural network approach to the classification of 3D prismatic parts based on their global shape information modelling. In this approach, a 3D part is modelled by the contours of its three projected views, which are approximately represented by three rectilinear polygons. The global shape information of each polygon is modelled by its simplified skeleton, which originally is of a tree structure and can be represented by several vectors by a conversion method. These vectors are the input to a polygon classifier which is constructed on the basis of the back-propagation neural network model. The classification results of polygons can be used to group the 3D prismatic parts into families in a hierarchical manner, by setting different levels of similarity criteria. The proposed method for classifying 3D workpieces can be used to enhance the productivity of design and manufacturing processes. By retrieving and reviewing similar parts from the part families, the designers or process planners could be greatly assisted in performing a new task. That is, they can avoid the reinvention of an existing design and can create a new design by modifying existing ones.

A short-term capacity trading method for semiconductor fabs with partnership

This paper presents a capacity trading method for two semiconductor fabs that have established a capacity-sharing partnership. A fab that is predicted to have insufficient capacity at some workstations in a short-term period (e.g. one week) could purchase tool capacity from its partner fab. The population of such a capacity-trading portfolio may be quite huge. The proposed method involves three modules. We first use discrete-event simulation to identify the trading population. Secondly, some randomly sampled trading portfolios with their performance measured by simulation are used to develop a neural network, which can efficiently evaluate the performance of a trading portfolio. Thirdly, a genetic algorithm (GA) embedded with the developed neural network is used to find a near-optimal trading portfolio from the huge trading population. Experiment results indicate that the proposed trading method outperforms two other benchmarked methods in terms of number of completed operations, number of wafer outs, and mean cycle time.

Design of BOM configuration for reducing spare parts logistic costs

This paper proposes an approach to reduce the total operational cost of a spare part logistic system by appropriately designing the BOM (bill of material) configuration. A spare part may have several vendors. Parts supplied by different vendors may vary in failure rates and prices - the higher the failure rate, the lower the price. Selecting vendors for spare parts is therefore a trade-off decision. Consider a machine where the BOM is composed of s critical parts and each part has k vendors. The number of possible BOM configurations for the machine is then k^s. For each BOM configuration, we can use OPUS10 (proprietary software) to calculate an optimum inventory policy and its associated total logistic cost. Exhaustively searching the solution space by OPUS10 can yield an optimal BOM configuration; however, it may be formidably time-consuming. To remedy the time-consuming problem, this research proposes a GA-neural network approach to solve the BOM configuration design problem. A neural network is developed to efficiently emulate the function of OPUS10 and a GA (genetic algorithm) is developed to quickly find a near-optimal BOM configuration. Experiment results indicate that the approach can obtain an effective BOM configuration efficiently.

A GA methodology for the scheduling of yarn-dyed textile production

This paper presents a scheduling approach for yarn-dyed textile manufacturing. The scheduling problem is distinct in having four characteristics: multi-stage production, sequence-dependent setup times, hierarchical product structure, and group-delivery (a group of jobs pertaining to a particular customer order must be delivered together), which are seldom addressed as a whole in literature. The scheduling objective is to minimize the total tardiness of customer orders. The problem is formulated as a mixed integer programming (MIP) model, which is computationally extensive. To reduce the problem complexity, we decomposed the scheduling problem into a sequence of sub-problems. Each sub-problem is solved by a genetic algorithm (GA), and an iteration of solving the whole sequence of sub-problems is repeated until a satisfactory solution has been obtained. Numerical experiment results indicated that the proposed approach significantly outperforms the EDD (earliest due date) scheduling method-currently used in the yarn-dyed textile industry.

Dispatching for make-to-order wafer fabs with machine-dedication and mask set-up characteristics

This paper develops a dispatching algorithm to improve on-time delivery for a make-to-order semiconductor wafer fab with two special characteristics: mask set-up and machine dedication. A new algorithm is proposed for dispatching series workstations. Simulation experiments show that the algorithm outperforms the previous methods both in on-time delivery rate, cycle time, and only slightly less than the best benchmark in throughput. The experiments are carried out in 10 test scenarios, which are created by the combination of two product-mix ratios and five mask set-up times.

A cost model for justifying the acceptance of rush orders

Rush orders are immediate customer demands which exceed the expectation of the currently effective MPS (master production schedule). Even though such orders are quite common to companies in a dynamic market, most existing studies published in the relevant literature seldom discuss the economical justification of accepting such an order. This paper proposes a mixed integer programming model for computing the cost of accepting the production of a rush order. The computed cost value could serve as a valuable reference for justifying the economics of accepting a rush order, and help determine its pricing strategy.