Concrete convention showcases latest research on AI, 3D printing
The architecture, engineering and construction industry is well-versed in the concrete design principles needed to optimize productivity. However, the advanced technologies that could take that productivity to the next level are less mature. Structural engineering researchers have explored machine learning applications in the field for decades. But only within the past five years has the community of researchers and practitioners begun to seriously explore ways in which machine learning can improve the efficiency and accuracy of specific tasks or solve previously intractable problems, says Henry Burton, an associate professor at the University of California at Los Angeles. The same can be said for concrete 3D printing, which has the potential to transform the industry through customized structural shapes, reduced formwork and lower construction time and cost. The American Concrete Institute’s 2021 Virtual Concrete Convention offered a peek inside the latest research into the potential and challenges of these technologies.
ML, data mining key to AI in concrete
Much of the convention’s featured session on AI focused on one of its subsets, machine learning, which is the use and development of computer systems that can learn and adapt without following specific instructions by using algorithms and statistical models to analyze and draw inferences from data patterns. During the session, Burton offered four criteria to help structural engineers decide when to use machine learning:
- If you were going to use an empirical model
- If it saves time
- When a physics-based model is deemed “incomplete”
- If it creates new opportunities
One company leading the way in the use of machine learning in concrete is Giatec Scientific, an Ottawa, Ontario-based company that developed Roxi -- the first artificial intelligence program created for concrete testing. During the session, Andrew Fahim, the company’s manager of research and development, explained how the company uses data from tens of thousands of internet-of-things-based sensors to train machine learning algorithms to detect anomalies, suggest changes to concrete mixes and predict future performance.
Fahim explained that one way the company does this is through its recently announced SmartMix web dashboard, which can help concrete producers predict the compressive strength, workability and air content of concrete based on a mix’s proportions, raw material characteristics and ambient conditions. It can also help producers meet performance specifications and prescriptive requirements, including strength at different ages and slump.
But while solutions like Roxi and SmartMix get the lion's share of attention when it comes to AI in concrete, data mining is also a critical component of the AI/ML equation. Emilio Garcia Taengua, an associate professor at the University of Leeds, compared society’s view of AI development to Plato’s Allegory of the Cave, suggesting that data science often gets overlooked because most of it happens behind the scenes. One project Garcia Taengua is working on is the ACI Foundation-supported “OptiFRC” project, which involves the creation of an exhaustive database with information on hundreds of fiber-reinforced concrete mixes compiled from papers published over two decades. It also involves modeling the relationships between the relative amounts of the mix constituents, fiber geometry and dosage. The team is using the correlations between residual flexural strength parameters, the limit of proportionality and compressive strength to develop predictive tools. The final outcome is a software package where users can access the database and utilize models to optimize mix proportions.
Although efforts from Giatec and the University of Leeds incorporate vast information, data sparsity remains one of the greatest challenges facing structural engineers seeking to apply ML, Burton said. He also noted that as model complexity increases, researchers run the risk of overfitting models.
Rheology a big focus of concrete 3D printing
Rheology, or the flow of matter, was a common theme during ACI’s featured session on 3D printing. Concrete 3D printing has seen success in relatively small applications, but for the technology to scale up, researchers must find ways to reduce cost, make the material more durable and print faster with fewer defects, said Karthik Pattaje Sooryanarayana, a graduate student at the University of Illinois at Urbana-Champaign. Common issues include a lack of vertical reinforcement, jamming in the extruder head, gaps in extrusion and drying shrinkage. A big part of the equation is concrete's rheological properties, such as loss modulus, yield stress yield, yield strain yield and complex viscosity.
Wilson Ricardo Leal de Silva, product manager at The Concrete Centre at the Danish Technological Institute, said that engineers can control the material’s yield stress at the time of extrusion and its evolution over time by tailoring rheology and hydration with admixture additions in the extruder. Another method to improve 3D-printed concrete is the inclusion of coarse aggregates, said Pattaje Sooryanarayana. Coarse aggregates can create flow issues, but Sooryanarayana’s research indicates the rheology of granular suspensions like concrete can be controlled through vibration.
Rheology is also closely tied to buildability -- the ability of a deposited material bulk to retain its dimensions under increasing load. Mohamadreza Moini, an assistant professor at Princeton University, explained that a layer-wise, additively manufactured, cement-based material’s ability to extrude and to achieve shape stability depends upon the early-age rheological properties of the deposited materials.
Once the materials are extruded, buildability challenges can originate from two common failure mechanisms: yielding of the material in lower layers and buckling of the element. However, it has been unclear which rheological properties control the early-age materials’ deformation during printing processes and contribute to the resulting buildability of the elements. The key takeaway in Moini’s research is that to control buckling in additive manufacturing, engineers must manipulate shear stiffness, not yield strength. This conclusion, Moini said, could be critical to helping engineers find new ways to compose materials for 3D printing.