基於動態數位影像之粗粒料即時顆粒特性分析

Abstract

混凝土是建築中廣泛使用的基本材料。在一般預拌混凝土中，粒料之組成佔有大約混凝土體積的60%~70%，扮演相當重要之角色，其形狀與粒徑分佈狀況將對混凝土的整體力學表現與工作性產生影響。於此，擁有良好的品管與監測是不容忽視的工作。 目前實務上對於粒料的品質控管主要採用「粗粒料篩析法」及「粗粒料中扁平、細長或扁長顆粒含量試驗法」，透過兩者求得級配曲線與扁平、細長、扁長率，然而這種方法既費時又費工，還存在人工量測誤差的問題。針對如此反覆且單調的實驗步驟，隨著科技的進步與發展，許多利用數位影像分析粒徑的方法逐漸受到重視。 本研究針對由台中預拌廠提供之粗粒料進行研究，旨在建立一套快速且準確的分析系統，能即時預測粒料之級配曲線和扁平、細長、扁長率。 本研究採用一種動態攝影的方式，透過定點高速連拍自由掉落的粒料，獲得多張不同面向的平面影像，經由影像辨識與特徵篩選處理後，藉由模擬真實搖篩原理分類粒料，求得數量百分率級配曲線。計算過程中對粒料形狀進行了假設，因此需對預測結果進行尺寸校正，並透過統計估算出不同尺寸顆粒的平均重量，將數量百分率級配曲線轉換成重量百分制，根據結果顯示，3/4”、1/2”、3/8”及 #4的精準度均達9成以上，最大誤差為9.65%。 本研究從多角度的粒料平面影像裡獲得長寬平面與寬厚平面影像，並透過影像辨識取得長度、寬度及厚度資訊，計算扁平、細長、扁長率，結果顯示，三者中整體的最大誤差僅有1.94%。 綜合上述，本研究結合了動態攝影與影像辨識，透過適當的特徵篩選與尺寸校正，加上重量與數量間的轉換，即時推算出級配曲線，另一方面，透過推測粒料的長寬平面與寬厚平面，求得扁平、細長、扁長率。此方法不僅快速且具有一定準確性，能有效節省人力與時間成本，對於混凝土生產和品質管理具有重要意義。

Concrete is a fundamental material widely used in construction. In general ready-mixed concrete, aggregates constitute about 60% to 70% of the concrete volume, playing a crucial role. The particle size distribution of aggregates significantly impacts the overall mechanical performance and workability of the concrete. Therefore, maintaining good quality control and monitoring is essential. Currently, the practical quality control of aggregates primarily employs the "coarse aggregate sieve analysis method" and the "test method for flat, elongated, or long particles in coarse aggregates." These methods yield the gradation curve and the ratios of flat, elongated, and long particles. However, these methods are time-consuming, labor-intensive, and prone to human measurement errors. Given these repetitive and monotonous experimental procedures, the advancement of technology has led to increased attention on digital image analysis methods for particle size. This study focuses on the coarse aggregates provided by a Taichung ready-mixed plant, aiming to establish a rapid and accurate analysis system capable of instantly predicting the gradation curve and the ratios of flat, elongated, and long particles. The study adopts a dynamic photography approach, capturing multiple planar images of freely falling aggregates through high-speed continuous shooting at fixed points. After image recognition and feature selection processing, aggregates are classified by simulating the real sieving principle to obtain the particle size distribution curve in terms of quantity percentage. During the calculation process, assumptions about aggregate shapes are made, necessitating size correction of the predicted results. By statistically estimating the average weight of particles of different sizes, the quantity percentage gradation curve is converted into weight percentage. The results show that the accuracy for 3/4", 1/2", 3/8", and #4 is over 90%, with a maximum error of 9.65%. This study obtains length-width and width-thickness planar images from multi-angle planar images of aggregates. Through image recognition, length, width, and thickness information is acquired to calculate the ratios of flat, elongated, and long particles. The results show that the overall maximum error among the three is only 1.94%. In conclusion, this study combines dynamic photography and image recognition. Through appropriate feature selection and size correction, along with the conversion between weight and quantity, the gradation curve is instantly calculated. Additionally, by estimating the length-width and width-thickness planes of aggregates, the ratios of flat, elongated, and long particles are derived. This method is not only rapid but also reasonably accurate, significantly saving labor and time costs, and holds significant importance for concrete production and quality management.