Project 1 : Developing CA models using machine learnt rules from computationally expensive PDE solvers Duration : Dec 2019 Client : Internal Research Project Personal contribution : Data generation using expensive PDE solvers. Identification of key variables and relevant data for the analysis. Rule engine extraction using deep neural networks, to detect the rules for given neighbourhood. Build the CA Model using developed rule engine. Objectives achieved : The above methodology is applied for a simple transient heat transfer analysis in a plate, a time reduction of 4 times is observed when compared with the commercial available FEM packages. With minimum domain expertise, we can build the CA for complex phenomena which otherwise are developed in using computationally expensive PDE solvers. By building the CA models the computational cost, as well as time are drastically decreased. Project 2 : Developed a recommendation system using hierarchical deep learning architecture for abrasive wheel selection process. Duration : Jan 2019 – August 2019 Client : Saint Gobain Personal contribution : Built a hierarchical machine learning model using deep neural network architecture from scratch. Augmentation of original dataset. Training and testing of the model. Created a probabilistic model for the calculation of confidence score for each possible recommendation. Objectives achieved : Reduced the domain expert’s dependency in the abrasive wheel selection process. Reduce the timeline of wheel selection process from an average of 6 months to 1 month which subsequently results in overall cost reduction and efficiency improvement. ©