Development of a Multi-Physics FEM model with Voxel based porosity algorithm for defect prediction in LPDC components
2022
The project involves development of a Multi-Physics FEM model of Low Pressure Die cast Aluminium alloy wheel for shrinkage porosity reducion using voxel based post processing algorithms.
Topology Optimization using Finite Element Methods based on Optimality Criteria (Summer of Science)
2021
The project involved Topology Optimization of a Structural and Thermal system using SIMP method in Sigmund-Bendsoe algo. over FEM formulations, achieving an avg. accuracy 92% by validating against standard TopOpt, Ansys app.
Parallelizing Machine Learning and Statistical algorithms for Stock Price Prediction
2021
Algorithms used Moving average, Auto Regression, KNN, Linear Regression, ANN and used OpenMP NUMBA with lowest RMSE 2.1178 and time boost by 92.35% for 8 threads on NVIDIA GPU
Development of Convolution Neural Network Speed-up with Sparsity (Industry Sponsored Project, in association with Sentry AI, USA)
2021
The project achieved a boost in the inference speed of the existing YOLOv5 model by pruning and sparsity over the neural nets. The optimal sparsity obtained was 30% for a YOLOv5 with a 3.57% Precision drop and 3.125% Recall drop
Design Optimization on the Full Car SUV vehicle suspension system to maximize riders comfort against road vibrations using Genetic Algorithm, Simulated Annealing and Particle Swarm Optimization.
2021
The project involved designing of stable suspension for road vibrations by optimizing a 12 variable Spring-Mass-Damper Laplace transformed design space achieving a maximum accuracy of 86% & 90 MiB usage for SA algorithm.
Design of Damping Energy Harvesting System for Automobile Suspension (Jaguar Land Rover, India)
2021
The project involved development of 5 different vibration energy capturing mechanisms recognized by JLR, India by modelling the car suspension system in SIMULINK with Max 37.85% conversion efficiency and 450 CC space
Dynamic Degree of Freedom Reduction and Analysis of Finite Element Assembly using Characteristic Constraint Modes with Component Mode Synthesis Modelling
2020
The proposed technique achieved the DDoF reduction obtained by Craig Bampton for a Multi-Body FE Assembly from 3142 DoF to 50 Dof with an average error of 0.0132 by introducing the interface node truncation of substructures.
Development of Artificial Neural Network Approach for Solving Ordinary and Partial Differential equations
2020
This technique which is an application of Physics Informed Neural Network, involved solving Ordinary DE (1st and 2nd Order) with MSE 0.02% and Partial DE (Linear and Non-Linear) with MSE 0.18% using a single NN hidden layer.
Warehouse Loading-Unloading of Fast Moving Consumer Goods (FMCGs) Productivity Tracking
2020
The project involved the development of KPI and productivity estimation tool for Warehouse Loading-Unloading of FMCGs. The variables comprised of 20 FMCG types, 5 Bays, 3 shifts, with Manpower, quantity as variables for optimization
Productivity Tracking for fleet management based on Object recognition using Computer Vision Techniques over low power edge computing devices like Raspberry Pi and Android
2020
The project involved the deployment of CNN models using Tensorflow on Raspberry Pi and Tensorlite on Android with optimized models of 92.36% on faster RCNN and 90.28% on SSD mobileNet accuracy.
Applications of Edge Computing for Industrial Internet of Things and Industrial Automation in Industry 4.0
2020
Master’s Seminar on “Applications of Edge Computing for Industrial Internet of Things and Industrial Automation in Industry 4.0” based on NVIDIA Jetson Nano for Object Detection, Recognition, Digital Twin, etc
CFD Simulation of Thermal Management System (Immersion Cooling) of Lithium-Ion Batteries in EVs
2020
This technique which is an application of Physics Informed Neural Network, involved solving Ordinary DE (1st and 2nd Order) with MSE 0.02% and Partial DE (Linear and Non-Linear) with MSE 0.18% using a single NN hidden layer.
Digital Twin Construction for Vibration Analysis of Spindle Head in various Machine Tools for Predictive Analytics
2020
The construction involved the development of a Cyber Twin for Predictive Analytics over Vibration signatures in the form of 3D gyroscopic acceleration of Machine-tool Spindle head and simulating it real-time on CAD software.
Predictive Maintenance of Aircraft Engines- Engine Health Monitoring and Predicting Time to Failure using various Machine Learning Models
2020
Reliability modelling and analysis of the Turbofan engine was done using the simulated run to failure data of 100 engines for 20000 cycles. ML models like regression were used to predict TTF with a min. error of ±25 cycles (MAE).