Standardization and optimization of index for 28 day strength for cement made from standard clinker

Abstract: This project regards the prediction of 28 day compressive strengths of cement. Using traditional multivariate analysis in combination with Artificial Neural Networks indexes have been developed which makes these predictions possible. Compressive strength is highly dependent on the cement hydration and clinker reactivity and literature on these topics have been studied followed by statistical analyses in Unscrambler X (Camo AS) and the Neural Networks model developed at Heidelberg Technology Center (HTC). After studying the theory behind compressive strength, some key parameters are identified including the Alite content, particle sizes as well as a variety of other parameters. Following this, a data set has been collected and formatted for the use in the project. Data on cement properties including compressive strength has been compiled by the quality engineer in Slite making the data collection simple. Having the data, the procedure includes a start with traditional multivariate analysis in Unscrambler to identify significant parameters in an effort of reducing the number of variables in the final model. In Unscrambler, Partial Least Squares regression has been used with uncertainty analysis as a selected option for parameter selection. Following the analysis in Unscrambler, the data set for each cement type is inserted into the neural networks models and the significant parameters are selected to act as input data, predicting either 1d or 28d strength. Before insertion into the Neural Networks model, the parameters are manually vetted with support of the literature and the accepted theories on cement hydration as correlations not necessarily mean that there is a causal link. Results are presented using the verification set of these indexes, indicating the prediction capacity of the indexes. Scenarios have also been used to study the underlying correlations between various properties and the compressive strengths. The results have shown good performance of the indexes created, and the procedure has proven to be fast and effective in creating these indexes. This opens up possibilities of using similar approaches to other areas of the plant in the future efforts to improve environmental and financial sustainability. Included in the final section of the report are also a few recommendations that would simplify the future work on this topic.