Table 2 Summary of publicly accessible datasets in concrete science for machine learning (ML) applications.
From: Machine learning in concrete science: applications, challenges, and best practices
Release year | Dataset | Description | Sample size | ML application |
|---|---|---|---|---|
2007 | Compressive strength data of high-performance concrete made with ordinary portland cement and cured under normal conditions, covering different curing ages from 1 to 365 days | 1030 | ||
2009 | Slump, slump flow, and 28-day compressive strength data of high-performance concrete | 103 | ||
2011 | Concrete fire dataset217 | Mechanical properties (e.g., compressive strength, elastic modulus, and tensile strength), thermal properties (e.g., heat diffusivity and thermal conductivity), and physical properties (e.g., mass loss and spalling) of unreinforced concrete under elevated temperatures | Up to 1932 | Property prediction191 |
2015 | Concrete creep and shrinkage datasets96 | So far largest worldwide collection of creep and shrinkage laboratory data, covering long measurement periods (some over 12 years) and the influence of admixtures in modern concrete mixtures (approximately 800 creep and 1,050 shrinkage curves contain admixtures) | About 1400 creep and 1800 shrinkage curves | |
2017 | Concrete tensile strength dataset79 | Splitting tensile strength data of concrete with manufactured sand at different curing ages ranging from 1 to 388 days | 714 | |
2018 | Concrete crack image dataset219 | Color images of walls and floors of concrete buildings with a resolution of 227 × 227 pixels, divided into cracked and noncracked classes (20,000 images each) | 40,000 | |
2018 | Structural defect dataset (SDNET2018)221 | Color images of concrete bridge decks, walls, and pavements with a resolution of 256 × 256 pixels, divided into cracked and noncracked classes (8484 and 47,608 images, respectively); crack width varies from 0.06 mm to 25 mm | 56,092 |
