應用深度學習輔助資料庫建立及 RC 建築之大梁斷面設計

Abstract

應用深度學習模型預測建築物中二樓至頂樓的每根大梁的斷面設計，包含斷面尺寸、主鋼筋量與箍筋配置共三種類型的預測目標，每種預測目標會建立對應的人工神經網路 (ANN) 模型，並以真實設計建案作為資料庫工模型進行訓練、驗證、測試，並探討參數重要性以及模型的誤差表現。實際使用上僅需將建築物及大梁的相關參數輸入至模型即可得到模型的預測值，結構工程師可以此值作為基準去評估設計結果是否合理。

本研究的資料庫是使用 88 棟真實建案 (由築遠工程顧問公司提供) 自行建立，共有 27720 筆資料，每筆資料代表一根大梁。資料庫的特徵取自結構計算書、結構平面圖、柱配筋圖等，預測目標 (斷面尺寸、主筋量、箍筋配置) 則取自梁配筋詳圖。由於預測目標需要輸入的資訊量較為龐大，因此本研究建立一套自動化輸入流程，應用電腦視覺的技術自動讀取梁配筋圖上的資訊並輸入至資料庫，並且設立檢核機制以確保輸入資料庫的資訊是正確的；最後以真實建案做為輸入測試，從準確度及花費時間來探討此流程的效益。

Applying deep learning models to predict the cross-sectional design of each beam from the second floor to the top floor of a building, including three types of prediction targets: cross-sectional dimensions, the amount of main reinforcement, and the configuration of stirrups. A corresponding artificial neural network (ANN) model is established for each prediction target and is trained, validated, and tested using real construction projects as the database model. The importance of parameters and the model’s error performance are also explored. In practical use, the relevant parameters of the building and beams only need to be input into the model to obtain the model’s predicted values. Structural engineers can use these values as a benchmark to evaluate the reasonableness of their design results.

The database used in this study was self-built using 88 real construction projects (provided by Envision Engineering Consulting Co., Ltd.), comprising a total of 27,720 data entries, each representing a beam. The features of the database are extracted from structural calculation books, structural plans, and column reinforcement drawings, while the prediction targets (cross-sectional dimensions, main reinforcement amount, stirrup configuration) are taken from the beam reinforcement detail drawings. Due to the extensive amount of information required for the prediction targets, this study establishes an automated input process, applying computer vision technology to automatically read the information from the beam reinforcement drawings and input it into the database. A verification mechanism is set up to ensure the correctness of the information entered into the database. Finally, real construction projects are used as input tests to explore the efficiency of this process in terms of accuracy and time consumption.