運用AI進行不同料源粗粒料辨識

Abstract

粗粒料是水泥預拌混凝土中的重要組成部分，其級配曲線和粗粒料之扁平、細長和扁長率對混凝土的工作性能和力學性能具有重要影響。因此，粒徑識別和分析成為了確保混凝土品質和性能的關鍵步驟。 傳統上，兩者的評估主要依賴於機械篩析法以及比例測徑器，需要耗費大量時間且操作繁瑣。然而，隨著影像處理和機器學習技術的發展，影像的粒徑分析方法逐漸受到關注。通過分析粒料影像來獲取粒徑分佈信息，具有快速、準確的優勢。 本研究針對北部預拌混凝土廠常用的花蓮和閩侯料源之粗粒料進行尺寸辨識，旨在有效即時監測三分石和六分石的資訊。為了獲取粒徑分佈的準確信息，本研究採用了影像辨識軟體Image Pro Plus來獲取粒料的影像資料，使用Python程式語言編碼機器學習模型進行數據處理以及分類，對粒料的料源和尺寸進行了詳盡的分析。利用分類結果，以及粗粒料篩析法，以獲取篩號停留重量，並建立了與影像數據的關係，最終得到不同料源之間的占比以及生成級配曲線，其成果較傳統文獻上的誤差更小，甚至在D90和D50有接近零誤差的表現。 粗粒料的扁平、細長和扁長率在傳統的影像辨識方法通常需要多個攝影機來準確捕捉。本研究通過大量平面資料的採樣以及實際重量的估算，建立了影像平均厚度與真實量測粒料厚度之間的關係，推估粒料在平面影像上的厚度，並最終獲得粒料的扁平、細長和扁長率結果，可以在短時間內得到與實際試驗結果相近，特別是扁平率以及細長率誤差較小。 綜上所述，本研究結合了影像辨識和機器學習技術，通過分析粗粒料的影像資料來評估其粒徑分佈和形狀特徵。該方法不僅具有高準確性和效率，而且能夠提供即時的資訊，支援混凝土生產和品質管理的過程，對於提高生產效率、節省成本並確保混凝土品質具有重要意義。

Coarse aggregates are essential components of concrete.Their gradation curves and flat and elongated particles, significantly affect the workability and mechanical properties of concrete. Therefore, accurate particle size distribution and analysis are crucial for ensuring concrete quality and performance. Traditionally, particle size and shape evaluation relied on labor-intensive methods such as sieving method and the use of proportioning calipers. However, with advancements in image processing and machine learning, image-based particle analysis methods have gained attention. These methods utilize digital image processing techniques to analyze particle images and provide instant and accurate information on particle size distribution. This study focuses on the commonly used coarse aggregates from Hualien and Minhou in ready concrete plants. The aim is to effectively monitor the information of three-quarter-inch and one-half-inch aggregates, as these particle sizes are representative of three-quarter-inch and one-half-inch aggregates. To obtain accurate particle size distribution information, the image processing software Image Pro Plus is used to acquire particle image data. Through machine learning classification methods, comprehensive analysis of the particle sources and sizes is performed. The classification results reveal the proportions among different sources and facilitate the generation of gradation curves. Additionally, traditional sieve analysis is employed to determine the weights of particles in each sieve layer, and the relationship between sieve weights and image data is established. This allows a thorough understanding of the proportions among different sources and the generation of accurate gradation curves, with minimal errors as compared to traditional methods, including approach zero errors in D90 and D50. Assessing flat and elongated particles traditionally requires multiple cameras for precise capture. In this study, a large amount of planar data sampling and estimation of actual particle weights are conducted to establish the relationship between average image thickness and measured particle thickness. This enables the estimation of particle flatness, elongation, and angularity from planar images, yielding results that closely match actual test results, particularly with small errors in flatness and elongation ratios. In conclusion, this study combines image recognition and machine learning techniques to evaluate the particle size distribution and shape characteristics of coarse aggregates using image data analysis. The proposed method demonstrates high accuracy and efficiency, providing real-time information to support concrete production and quality management. It is of significant importance in improving production efficiency, reducing costs, and ensuring concrete quality.