| Περίληψη: |
The fresh state properties of concrete, such as slump, yield stress, and plastic viscosity, are pivotal parameters influencing construction quality and efficiency. Conventional methods for assessing workability, including the slump test and rheometer measurements, exhibit limitations such as latency and high subjectivity, failing to meet the demands of smart construction for real-time and multiparameter synchronous monitoring. Furthermore, sensor-based monitoring techniques (e.g., torque-rheological parameter mapping) still suffer from limited prediction accuracy for critical parameters like plastic viscosity and segregation resistance due to simplified physical model assumptions. In contrast, machine vision technology captures the spatiotemporal evolution characteristics of concrete flow patterns through dynamic image sequences. By integrating deep learning models to autonomously extract key features like texture and edges, it establishes a mapping between visual semantics and rheological parameters, providing an innovative solution for the real-time and dynamic assessment of concrete rheological behavior. This paper presents a systematic review and critical analysis of machine vision technology for concrete rheology assessment. It delves into the technical intricacies, comparative performance, and implementation challenges of key methodologies, including image processing techniques, deep learning architectures (e.g., CNN, vision transformer), and multimodal data fusion strategies. By comparing the performance differences between traditional testing methods and intelligent vision-based techniques, the significant advantages of machine vision in terms of real-time capability, multiparameter prediction, and dynamic adaptability are revealed. Moreover, current technical bottlenecks, such as environmental interference and the insufficient generalization ability of deep learning models on small data sets, are further discussed. Addressing these challenges, the paper prospects adaptive optimization directions in smart construction scenarios, providing theoretical references and technical roadmap guidance for related research. Practical Applications: This review synthesizes advances in machine vision for real-time quality control of fresh concrete, presenting a transformative solution for the industry. The technology uses industrial cameras installed on mixer trucks or production lines to continuously monitor concrete flow during mixing, transportation, and discharge. By applying deep learning models such as CNNs and vision transformers to analyze the captured images, it enables continuous, real-time prediction of critical performance indicators like slump and plastic viscosity. Compared to conventional approaches, the system offers three key benefits: (1) It overcomes the inherent lag and subjectivity of conventional quality control by converting experience-based judgments into objective data. (2) It allows dynamic mix adjustment by automatically regulating water and admixture levels when deviations are detected, reducing waste and avoiding rework. (3) It supports remote, centralized quality monitoring via a cloud-based platform, significantly reducing reliance on specialized on-site personnel and making standardized quality control across multiple and cross-regional sites feasible. In summary, this machine vision framework establishes the basis for a digital, traceable, and sustainable concrete production process. It significantly cuts material waste and supports the industry's transition toward greener, low-carbon operations. [ABSTRACT FROM AUTHOR] |