應用類神經網路配合ACI規範輔助卜作嵐混凝土配比設計

Abstract

混凝土是土木營建工程中最被廣為運用之材料。混凝土本身為一變異性很高之材料，其是由各種不同之材料所組成，每種材料之特性都不一樣，目標需求所設計出的配比往往和實際有落差。而配比、製作及養護環境之不同，皆會對混凝土之性質產生影響。其中又以配比對混凝土之影響最大，再加上材料科技的進步，現行所用之混凝土已非傳統一般水泥混凝土，因混凝土配比中皆添加了卜作嵐掺料及化學掺料，這使得混凝土性質更加難以掌控。本研究之目的係建構一套電腦輔助混凝土配比系統。首先，先利用由各文獻或實驗室中所蒐集到的資料來建構一個真實混凝土配比資料庫；接著我們利用監都式學習的類神經網路來建立混凝土強度及坍度之預測模型，並同時利用電腦程式配合ACI規範及設計流程來產生合理之混凝土配比。預測模型經訓練及測試後，其所得之結果、準確性和效能皆屬不錯。再以所建立好之預測模型來預測混凝土配比之強度和坍度性質，並將此配比資料整合成一可用配比資料庫。隨後利用一試驗來驗証模型預測配比之準確度，試驗結果顯示，實際值和預測值之誤差大多在合理範圍之內，再次證明預測模型之準確性是可靠的。最後，利用歐式距離觀念、卜作嵐取代率及材料成本三種方法來將配比加以分類，並建立一配比資料庫，使其有如一混凝土配比之型錄，可提供使用者方便依其工區環境、材料使用及成本考量來作選擇及使用。

Concrete is the most widely used materials in civil engineering. Concrete is a high variability of material owing to composed with a variety of different materials. Since the mix proportioning of concrete are added pozzolanic and chemical admixtures, it became more difficult to control the properties of concrete. The purpose of this study is to establish a computer-aided system for concrete mix design. First, actual data in reference or laboratory are collected to establish a real mix proportioning of concrete database. A supervised neural network (ANN) model is then employed to establish the prediction models of strength and slump of concretes. Following, based on ACI code, a practical database of mix proportioning of concrete is produced. For verifying the correction of generated data, the strength and slump of concrete are yielded via ANN prediction models. 12 different mix designs are also verified via cylinder test in laboratory. Finally, a classification system is utilized to categorize data into 360 clusters based on strength of concrete, pozzolanic admixtures replacement rate, and materials costs. Simulation results reveal that based on these grouped 360 clusters, the proposed concrete mix computer-aided design system can provide engineering acceptable solutions (mix design) as users acquiesced required strength of concrete, pozzolanic admixtures replacement rate, and materials costs.