標題: 應用人工智慧技術輔助設計混凝土配比
Application of Applied AI models in concrete mixture proportion
作者: 呂夙修
Lu, Su-Hsiu
洪士林
Hung, Shih-Lin
土木工程學系
關鍵字: 混凝土配比設計;K-Means演算法;類神經網路;Concrete proportioning;K-Means;artificial neural network
公開日期: 2010
摘要: 混凝土在土木建築結構工程中是最為廣泛運用的營建材料。混凝土是敏感性很強的材料,在每一個製作過程,如配比、拌合、澆置及養護等等,皆對混凝土有重要的影響,尤其是配比部份。但是依照傳統的配比設計方法,並不一定能保證得到需求目標。整個過程不僅費時也浪費資源,若混凝土試體試驗失敗,不僅是資源成本的浪費,在時間成本的損失更是難以計價。如能藉由電腦輔助設計混凝土配比,不僅能降低資源成本,更能提升工程效率。 至今已有電腦輔助設計混凝土配比的方法,常以成本最佳化為設計目的。然而在不同的環境下,使用者會有不同的需求。本論文應用K-Means演算法分析資料庫,並建立混凝土配比設計系統;依使用者所需的目標設計混凝土配比,並提供多樣性的配比設計,讓不同環境下的使用者,依自身需求使用不同的混凝土配比;並利用類神經網路建立資料庫中缺乏的混凝土配比資料。模擬驗證結過顯示本系統可藉由K-Means演算法快速找尋可能解,若資料庫無近似解存在,則可由ANN提供答案。
Concrete is one of most utilized construction materials in civil and infrastructural engineering. Concrete is a highly sensitive material to the issues in production process, such as proportioning, mixing, pouring, curing, etc. Among those factors, propositioning is the most important aspect. However, concrete mix, designed based on conventional methods, are not guaranteed to satisfy the required aim. Meanwhile, if the concrete specimen test does not pass, it results in not only wasting cost, but also loss of time. Recently, computer-aided design of concrete mix proportioning is a feasible approach with the aspect of reducing the resource costs and increasing construction efficiency. Based on cost optimization approach, there are, currently, many schemes of computer-aided concrete mix proportioning design. However, cost is not the only aim for concrete mix design. This work attempts to employ K-Means algorithm to analyze the pre-collected database to design concrete mix proposition based on the predefined requirements and provide a diversity of design to satisfy requirements of engineering. In addition, ANN model can generate solutions, if K-Means algorithm cannot find solutions in database. Simulation results reveal that the system is feasible and practicable in concrete mix design.
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT079716526
http://hdl.handle.net/11536/44841
Appears in Collections:Thesis


Files in This Item:

  1. 652601.pdf