Machine learning approach for estimating workability of concrete from mixing image sequences using 3D stereo camera.

Thumbnail Image

URL

Journal Title

Journal ISSN

Volume Title

Insinööritieteiden korkeakoulu | Master's thesis

Date

2021-10-18

Department

Major/Subject

Mcode

Degree programme

Master's Programme in Building Technology (CIV)

Language

en

Pages

75+2

Series

Abstract

Concrete is one of the most utilized man-made material on earth and therefore its properties are important to study. Workability of concrete is a property of fresh concrete that refers to how easily the concrete can be placed, consolidated, and finished. It is commonly measured using a slump test where more fluid concrete yields a higher slump. Slump can be categorized into multiple slump classes depending on its fluidity. Performing the slump test is a laborious task and prone to manual errors, hence there is a need to find measurement solutions based on modern technological advancements to estimate the workability. The aim of this thesis was to investigate the potential of depth sensor data to predict the slump classes of concrete using machine learning. The depth sensor was mounted on a mixer that recorded the mixing process of concrete. Visualizing the depth data of the mixing process revealed a sinusoidal pattern for slump classes with each having a different peak. This visualization was used to remove noisy data as well. Haralick features are 13 statistical properties such as entropy and contrast that can be extracted from an image as useful information. These properties were used to train various machine learning classifiers to learn from and then predict the slump class of a concrete mix. High classification accuracy of up to 94% was achieved demonstrating that using depth data with machine learning is effective in estimating the workability of concrete while mixing.

Description

Supervisor

Punkki, Jouni

Thesis advisor

Masood Mustafa, Khalid
Ojala, Teemu

Keywords

concrete, automation, workability, slump, machine learning, depth imaging

Other note

Citation