Microstructure Quantification and Structure-Property Correlation of Dual-phase Steels
In this thesis, a study has been made to find the influence of chemistry and processing parameters on the microstructures and properties of dual-phase (DP) steel. A machine learning-based approach is employed to develop the composition-process-structure-property correlations using artificial neural network (ANN). The data collected from published literature as well as self-generated data are employed for the learning of the network. Suitable image processing algorithms are devised for quantification of the volume fraction of phases and morphology of the microstructure towards the quantification of the parameters. The data extracted from the microstructure are also used for mapping the correlation between composition and processing with microstructure as well as between microstructure and properties. The ANN models are used as objective functions for a genetic algorithm (GA) based searching of the suitable combinations of composition, processing and microstructural parameters to achieve an improved strength-ductility balance of the steel.
Primarily various artificial neural network models have been developed for strength, ductility, yield ratio and strain hardening exponent from the literature data. The ANN models are used to identify the role of parameters from composition-process-property, composition-process-microstructure and composition-microstructure-property correlations through sensitivity analyses and simulation studies. multi-objective genetic algorithm is applied to find the optimum values for all the variables to reach good strength and ductility. While searching for the optimum microstructural parameters, the insufficiency of the reported information regarding microstructure is revealed. Hence data was generated experimentally in a systematic manner for the compositional, heat-treatment and microstructural parameters along with the properties. For this purpose, three different steels are subjected to various heat treatments and characterized. A few algorithms are proposed for automatic classification and quantification of martensite morphology using image processing methods. For measuring the volume fraction of phases, a noise removing algorithm is applied to the thresholded images. The shape, size and distribution of the martensite phases have been quantified by three separate routes proposed here.
The generated data on the composition, processing, microstructure and properties are used for developing various ANN models Four types of correlation models of composition, processing, microstructure and property are used for analyzing the role of the parameters again. These newly generated models are used as objective functions for optimization of different combinations of properties for multi-objective optimization of strength and ductility. The Pareto optimal solutions show that IA and IQ types of processing provide higher ductility but SQ type of heat-treatment process provides better strength, as these heat-treatment parameters play a significant role to control the volume fraction, shape, size and distribution of martensite. It is also seen that lower volume fraction, particle size, higher shape factor and better distribution of martensite leads to better ductility.