Manufacturing cell formation using genetic algorithm vs. neural networks

Abstract

Cellular manufacturing system (CMS) is an important application in group technology. CMS design involves identifying part families and machine groups. Genetic algoritllm (GA) is a robust adaptive optimization method based on principles of natural evolution, having the advantages of parallel processing and multi-point searching. On the other hand, neural networks can be used to identify similar patterns at a high computational rate. Therefore, GA search method and neural network method are well suited for the machine-component grouping (MCG) problem. In this work, a mathematical programming formulation is presented which considers the intercell part flow and the manufacturing cell density for the MeG problem. Two algorithms based on GA and neural networks for solving the problem are also proposed. Implementation results demonstrate that both proposed algorithms can provide feasible cell grouping solutions under machine capacity constraints. These results also demonstrate that GA can perform adequate tasks when the fitness function is carefully selected. On the other hand, the proposed modified ART-1 model can also achieve a good performance when the appropriate vigilance parameter is given.

摘要

單元製造系統爲群組技術的重要應用。單元製造系統的設計包括如何決定工件族和機器群。基因法則演算法是基於自然演化的自適最佳化方法，並具有平行處理和多點捜尋的優點。另一方面，類神經網路擁有很好的計算效率來辩識相似的物體。因此，基因法则搜尋法與類神經網路非常適合應用在機器工件分群問題上。本文提出一個考慮製造單元間工件流量及製造單元密度的機器工件分群問题數學摸式，並利用基因法則演算法與頮神經網路發展兩個演算法以解決這個問题。實際驗證結果顯示兩種演算法都能在機器產能限制的條件下提供理想的機器工件分群結果。結果並顯示當適合度函數小心選取時，基因法則演算法可以表現良好。另外，當警戒係数能適當給定的情況下，本文所提的改良式ART-1模式亦有良好的表現。

Key Words:

• group technology
• machine-cell formation
• machine-component grouping
• genetic algorithm
• neural network
• ART-I

關鍵詞:

• 群組技術
• 工件族和機器群形成
• 機器工件分群問瓸
• 基因法则演算法
• 類神經網路
• ART-1