Multi-objective optimization from PBD-CCD experiments adopting quadratic model and GA-optimized ANN, pareto optimality
Description
The Plackett-Burman screening design was followed by CCD experiments on two responses. The CCD data were further used to generate quadratic models and ANN modeling. ANN model architecture was optimized using GA. Both the model variants were used to identify optimized process responses (maximum/minimum). Subsequently, MOO strategy was adopted. Additionally, pareto optimal solutions were offered using GA-multiobjective function
Files
Steps to reproduce
a1_ANN_GA_one_resp.m file requires an input dataset from CCD design containing 3 predictor and two responses as a table. The script performs independent ANN modeling from the dataset optimizing the network size, weight, bias, and storing the network. a2_ANN_Single_min_max.m performs optimization of the response using GA and the ANN models within the design space and saves the optimal conditions and responses (max/min) in cell array. a3_ANN_multiobj.m performs the multi-objective optimization using composite desirability, based on user-defined weight for each response. a4_Desirability_multi.m performs MOO on pareto-optimality and stores the result. b1_PBD_CCD_analysis.m develops MLR model using CCD Experiments, stores the models in a structure. b2_CCD_Single_min_max.m and b4_Desirability_multi.m are equivalent version of the ANN based approach described above.