應用電子式反彈錘及適應性類神經模糊推論系統預測混凝土抗壓強度

Application of the Adaptive Neuro-Fuzzy Inference System Model in Predicting the Concrete Compressive Strength from the Silverschmidt Hammer

本研究旨在探討 SilverSchmidt 電子式混凝土試驗反彈錘之準確性，並利用適應性類神經模糊推論系統 （adaptive neuro-fuzzy inference system,ANFIS），來建立混凝土強度推估模型，以提昇混凝土抗壓強度之預測準確性。ANFIS是一將模糊邏輯的概念融入類神經網路之軟體計算技術。本研究運用變異量解釋能力 （variance account for, VAF） 與均方根誤差 （root mean squareerror, RMSE） 來評估類神經模糊推估系統的預測效能。研究結果顯示，電子式混凝土試驗反彈錘所得到之抗壓強度與混凝土圓柱試體抗壓強度實際值比較，其平均絕對誤差率 （mean absolute percentage error,MAPE） 為27.07%。而本研究建立之適應性類神經模糊推論系統之預測能力，平均絕對誤差率已降至5.55%。如此說明利用電子式混凝土試驗反彈錘配合ANFIS的方法來提昇混凝土抗壓強度預測準確性與信賴度是可行的。

The main purpose of this study is to investigate the accuracy of thedigital concrete rebound hammer (the SilverSchmidt) using the AdaptiveNeuro-Fuzzy Inference System (ANFIS) to establish a model of predictingconcrete compressive strength in order to enhance predicting accuracy.ANFIS is a soft computing technique which incorporates the concept offuzzy logic into neural networks. The data collected in the study weredivided into training sets, checking sets and testing sets, and were used to train the construction of the prediction model. In this study, variousstatiscal performance indices such as VAF (variance accounts for), RMSE(root mean square error) were considered to make a comparison of theprediction performance of the constructed neuro-fuzzy model.The final results show that the compressive strength predicted by thedigital concrete rebound hammer when compared with the actual concretecompressive strength, the MAPE (mean absolute percentage error) wasabout 27.07 %. In this study, the MAPE of the ANFIS-based model was to5.55 % even. This demonstrates the the proposed ANFIS-based predictionmodel can be used to produce a more accurate and reliable predictionfrom the SilverSchmidt concrete rebound hammer.