Predicting Time-Dependent Fresh Concrete Properties Using Image-Based Deep Learning

Core Concepts

A convolutional neural network can predict the time-dependent behavior of fresh concrete properties, such as slump flow diameter, yield stress, and plastic viscosity, based on stereoscopic image sequences and mix design information.

Abstract

The paper presents a method for predicting the time-dependent behavior of fresh concrete properties using a convolutional neural network (CNN). The key highlights are: The CNN receives stereoscopic image sequences of the concrete's flow behavior during the mixing process, along with information from the concrete mix design, as input. The network also receives temporal information in the form of the time difference between the image acquisition and the reference measurements of the concrete properties. This allows the network to implicitly learn the time-dependent behavior of the properties. The network predicts the slump flow diameter, yield stress, and plastic viscosity of the fresh concrete. Experiments show that using depth images, optical flow images, and mix design information as input leads to the best performance, outperforming the use of orthophotos alone. Averaging multiple predictions for the same reference values can further improve the accuracy of the predictions. The trained network is able to predict the time-dependent behavior of the fresh concrete properties, such as the gradual decrease in slump flow diameter over time. The proposed approach has the potential to enable the prediction of fresh concrete properties during the mixing process, allowing for timely adjustments to the concrete mix to achieve the desired properties at the time of placement.

Stats

The data set covers a wide range of fresh concrete properties: Slump flow diameter (δ): 30.00 cm to 63.50 cm Yield stress (τ0): 65.84 Pa to 585.40 Pa Plastic viscosity (μ): 19.76 Pa·s to 121.91 Pa·s Time difference (Δt): -49.88 min to 87.16 min

Quotes

"Increasing the degree of digitisation and automation in the concrete production process can play a crucial role in reducing the CO2 emissions that are associated with the production of concrete." "To ensure the quality of the fresh concrete, traditional quality assurance measurements like the slump test (EN 12350-5, 2019) and rheometer measurements are typically employed. However, these methods are labor intensive and are associated with relatively high uncertainties."

Key Insights Distilled From

Image-based Deep Learning for the time-dependent prediction of fresh concrete properties

by Max Meyer,Am... at arxiv.org 04-16-2024

https://arxiv.org/pdf/2402.06611.pdf

Deeper Inquiries

How could the proposed method be extended to handle a wider range of time-dependent behaviors, such as steady or increasing flow over time, rather than just the decreasing flow observed in this study?

To extend the method to handle a wider range of time-dependent behaviors, such as steady or increasing flow over time, the training data set could be diversified to include concretes with different behaviors. By including concretes that exhibit steady or increasing flow over time, the network can learn to predict a broader spectrum of fresh concrete properties. Additionally, the network architecture could be modified to incorporate more complex patterns and trends in the data. For example, the network could be designed to capture non-linear relationships between input features and time-dependent behaviors. By training the network on a more diverse set of data and adjusting the architecture to capture different types of behaviors, the method can be extended to handle a wider range of time-dependent properties.

What are the potential limitations of using only mix design information as additional input, and how could the method be improved by incorporating other data sources, such as measurements of the raw material properties?

Using only mix design information as additional input may have limitations in capturing the full complexity of fresh concrete properties. Mix design information provides details about the composition of the concrete, but it may not account for variations in raw material properties that can impact the final properties of the concrete. To improve the method, incorporating measurements of raw material properties, such as aggregate characteristics, cement properties, and water quality, can provide a more comprehensive understanding of how these factors influence fresh concrete properties. By integrating raw material data, the network can learn the specific effects of individual components on the properties of the concrete, leading to more accurate predictions and a better optimization of the mix design process.

Could the time-dependent prediction capabilities of the network be leveraged to optimize the concrete mix design and production process to achieve the desired fresh concrete properties at the time of placement?

Yes, the time-dependent prediction capabilities of the network can be leveraged to optimize the concrete mix design and production process. By accurately predicting the fresh concrete properties at the time of placement, adjustments can be made during the mixing process to ensure that the concrete meets the desired specifications. For example, if the network predicts that the slump flow diameter or rheological parameters will deviate from the target values at the time of placement, appropriate additives or adjustments can be made to the mix to achieve the desired properties. This proactive approach based on real-time predictions can help in reducing waste, improving efficiency, and ensuring the quality of the final concrete product.