| KEYWORD |
Area Engineering
Machine Learning for optimal concrete production
Thesis abroad
Reference persons GIUSEPPE CARLO MARANO
External reference persons melchiorre Jonathan, Anerdi Costanza
Description Concrete has been cast for centuries in buildings, roads, bridges and more. Structures made with concrete have survived wars and natural disasters, outlasting many of the civilizations that built them. In addition to its strength and resilience, concrete is also a building staple because it is cheap and easy to produce. Worldwide, 30 billion tons of concrete are used each year. On a per capita basis, this is three times more than 40 years ago, and the demand for concrete is growing faster than that for steel or wood .
Concrete buildings and structures are, in many ways, ideal for climate-resilient construction, given their versatility and durability. But the cement industry, the main component in the concrete design mix, has an unsustainable footprint to date, accounting for 8%-10% of global CO2 emissions (to get an idea, this is more than double the emissions generated by air and shipping) . It is therefore clear that the concrete and cement industry has a key role to play in the deep decarbonization process that our society is experiencing on many fronts.
Concrete is made by mixing sand, gravel, cement and water, in well-defined quantities. Mix design, or the definition of this recipe, is a complex, multiphase process in which the best composition of ingredients must be found to create a concrete that performs as well and reliably as possible in terms of strength and durability. In addition, the chemical processes underlying the hydration and hardening of concrete are irreversible, so any errors in the design of the concrete mix (i.e., actual amount of water) could result in unforeseen costs for the investor, both during construction and in the later stages of the structure's life.
Currently, there are various approaches taken to mitigate the impact of the cement industry, ranging from the use of alternative binders (e.g., waste materials such as fly ash) to optimizing individual components, including the amount of cement. To understand how these technologies are able to make a difference, it is necessary to know what are the mechanisms by which the mix-design process is regulated.
In the literature, as well as in widespread practice, there are a number of methods for evaluating the quantities of different components, the most popular of which are those derived from the so-called "Three Equations Method," which still has many uncertainties. In order to guarantee the minimum mechanical properties, and in particular the characteristic compressive strength, manufacturers tend to secure themselves by using more cement than is actually needed, with the consequent repercussions in terms of environmental and other costs.
In fact, the mechanical properties of concrete are strongly influenced by the mix that makes it up, the environmental characteristics in which it was produced, the environmental characteristics in which it was cast and hardened, and other factors such as the amount of "hidden" water that is difficult to account for except by ad hoc evaluation. This results in great uncertainty at the production stage about what the characteristics of the final product will be.
The need to reduce these uncertainties has focused attention on the use of next-generation algorithms; for example, Machine Learning has gained relevance in the evaluation and optimization of concrete mix-design due to its ability to do pattern recognition and self-learning. ML models can thus be used to predict output data, based on a defined input data set .
Required skills Machine Learning, neural network
Deadline 24/09/2023
PROPONI LA TUA CANDIDATURA