Data-driven multicollinearity-aware multi-objective optimisation of green concrete mixes

A multicollinearity-aware multi-objective optimisation (MA-MOO) framework was developed to minimise the main environmental issues and the cost of production of green concrete, while preserving the compressive strength in a desirable range with the help of machine learning modelling. A novel set of constraints were proposed to restrain the search space and eliminate the known statistical trap of multicollinearity. To test the framework, a comprehensive dataset of 2644 concrete mixes incorporating five supplementary cementitious materials (SCMs) was collected from the literature on which the extreme gradient boosting machine (XGBM) could achieve the best performance (RMSE 4.3 MPa). XGBM was deployed within the framework to design mixes with a similar multicollinearity structure to the training data. The mixes could reach up to more than two times lower cost of production and environmental issues.

Automated summary

A multicollinearity-aware multi-objective optimisation (MA-MOO) framework was developed to minimize environmental issues and production cost of green concrete, while maintaining compressive strength. This was achieved using machine learning and a novel set of constraints to eliminate multicollinearity. Testing the framework with a dataset of 2644 concrete mixes, the extreme gradient boosting machine (XGBM) achieved the best performance (RMSE 4.3 MPa). The framework allowed for the design of mixes with significantly reduced production cost and environmental impact.