Apurv Darji

End-to-End deep-learning-based tool wear anomaly detection using motor current data -- Acquired Current data, cleaned & processed from complex JSON. -- Compared LSTM & 1dCNNs in Autoencoder & Hybrid NN architectures. Reconstructed input data from compressed representation, classified 'wear' state based on high reconstructing error. -- Thesis shows motor-current data alone achieves 98% accuracy in determining tool wear. -- 'Transfer Learning' approach improves classification from 0.62 to F1-score of 0.84.

− Stock simulator web application development − Responsibilities covered: addressing the defects, integrating new features, and deploying changes to the docker container − Yahoo Finance and Quandl data scrapping from the API − Stock data forecasting using ARMA-GARCH and Markowitz models − Visualization of data distribution and statistical analysis − Frameworks: Django (Python) and React (JavaScript) − DevOps tools: Docker container, bitbucket repository

− Full-stack developments − SQL, C#, .NET framework

− A team of 3 students, called 'Agile Sensors'. − Developed prototype: an autonomous robot that handles the industrial inspection routines. − I have extensively worked on developing the MATLAB GUI app that can schedule new measurement tasks and can live-track the measured data by the robot.