范正成Fan, Jen-Chen臺灣大學:生物環境系統工程學研究所黃聖修Huang, Sheng-HsiuSheng-HsiuHuang2010-05-052018-06-292010-05-052018-06-292009U0001-2907200918545300http://ntur.lib.ntu.edu.tw//handle/246246/181176本研究藉由所收集之240筆現地資料，以傳統相對粒徑比可灌性推估公式，與倒傳遞類神經網路，進行可灌性之預測分析。傳統推估公式之預測準確率為45%至68%，明顯可看出其對於超微粒水泥滲透灌漿於砂性粉土層無法有效預測。而類神經網路以七個影響可灌性之因子，即土壤粒徑(d10)、土壤粒徑(d15)、孔隙比(e)、細粒料含量(FC)、土壤均勻係數(Cu)、土壤級配係數(Cz)與水灰比(W/C)，作為類神經網路輸入層之神經元，建構一適合台灣地區高細粒料含量之砂性粉土層超微粒水泥滲透灌漿可灌性的網路預測模式。依據本研究之分析結果顯示，以土壤粒徑(d15)、孔隙比(e)、細粒料含量(FC)、土壤均勻係數(Cu)、土壤級配係數(Cz)與水灰比(W/C)作為輸入層之神經元，可得到較佳的預測能力，其準確率為96%。 此外，本研究亦進行室內滲透灌漿試驗，採用與現地資料相同之水灰比(3.34、4.0及4.65)、水泥之爐石含量50%及不同細粒料含量(0%、10%、20%、30%及40%)之砂柱試體，用以針對網路模式進行驗證。以滲透灌漿試驗之結果，進行倒傳遞類神經網路預測模式的驗證，其可灌性預測準確率可達87%。 由本研究之可灌性預測模式及結果分析，前人所提出之相對粒徑比推估公式，對於超微粒水泥滲透灌漿於砂性粉土層之可灌性，明顯無法有效的推估。而應用倒傳遞類神經網路來建構可灌性預測模式，為相當可行之方法，顯示類神經網路在解決此類問題上有相當良好的功效。In this study, 240 sets of field data were collected and analyzed to evaluate the groutability by using two methods, namely the conventional formula with relative particle size ratio and the backpropagation neural network(BPN). The accuracy of the conventional formula method ranged from 45% to 68%, i.e., this method can not be successfully used to predict the groutability. Seven factors affecting the groutability were used in the BPN methods；they are： the effective soil particle size (d10), the soil particle size(d15), void ratio(e), fines content(FC), uniformity coefficient(Cu), coefficient of gradation(Cz) and the water-to-cement ratio(W/C). These factors used as the neuron of the neural network input layer to establish a suitable network model which may be used to predict the groutability of permeation grouting with microfine cement grout to the sandy silt soils with high content of fines in Taiwan. From the obtained results, it can be found that while the soil particle size(d15), void ratio(e), fines content(FC), uniformity coefficient(Cu), coefficient of gradation(Cz) and the water-to-cement ratio(W/C) were used as the neuron of the input layer, the BPN method showed a better forecast ability with an accuracy as high as 96%. Aside from these, in this study, the permeation grouting experiments were also conducted in the laboratory. The water-to-cement ratio were controlled to be 3.34, 4.0 and 4.65, which were the same as the value used in the field. The slag content of the microfine cement is 50% and five different contents of fines, namely, 0%, 10%, 20%, 30% and 40%, were used. Using the data obtained from the permeation grouting experiments, the BPN forecasting model were verified and its accuracy reached 87%. According to the results of this study, the conventional formula method could not be successfully used to predict the groutability of the permeation grouting with microfine cement grout to sandy silt soils. However, while dealing with these problems, the BPN model showed its superiority and practicality.摘要 ibstract ii錄 iii目錄 vi目錄 ix一章 研究動機與目的 1 1.1 研究動機 1 1.2 研究目的 2二章 文獻回顧 4 2.1 超微粒水泥研磨攪拌技術發展 4 2.2 水泥配比成分影響 6 2.3 滲透灌漿工法 13 2.3.1 滲透灌漿之特性 14 2.3.2 室內滲透灌漿試驗 18 2.4 滲透灌漿可灌性評估 23 2.5 類神經網路 25三章 研究方法 29 3.1 現地灌漿資料 29 3.2 可灌性預測分析 31 3.2.1 傳統可灌性經驗公式 33 3.2.2 倒傳遞類神經網路 34 3.3 室內滲透灌漿試驗 39 3.3.1 灌漿材料及性質 39 3.3.2 試驗砂土 43 3.3.2.1 試驗砂土基本性質試驗 44 3.3.3 超微粒水泥滲透灌漿模擬試驗 50 3.3.3.1滲透灌漿模擬試驗規劃 50 3.3.3.2 滲透灌漿模擬試驗設備 53 3.3.4 滲透灌漿模擬試驗步驟 58四章 結果與討論 66 4.1 傳統相對粒徑比之可灌性預測公式 66 4.2 倒傳遞類神經網路可灌性預測模式 70 4.3 現地資料分析結果 78 4.4 室內滲透灌漿模擬試驗結果與分析 79 4.5 綜合比較 83五章 結論與建議 84 5.1 結論 84 5.2 後續建議 86考文獻 87錄 94application/pdf3973245 bytesapplication/pdfen-US倒傳遞類神經網路可灌性超微粒水泥滲透灌漿backpropagation neural network (BPN)groutabilitymicrofine cementpermeation groutingthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/181176/1/ntu-98-R96622026-1.pdf