Indian Institute of Technology Madras

About: Indian Institute of Technology Madras is a facility organization based out in Chennai, Tamil Nadu, India. It is known for research contribution in the topics: Catalysis & Heat transfer. The organization has 20118 authors who have published 36499 publications receiving 590447 citations.

Dynamic Energy Management of Hybrid Energy Storage System With High-Gain PV Converter

Abstract: In this paper, fast acting dc-link voltage-based energy management schemes are proposed for a hybrid energy storage system fed by solar photovoltaic (PV) energy. Using the proposed control schemes, quick fluctuations of load are supplied by the supercapacitors and the average load demand is controlled by the batteries. Fast dc-link voltage, effective energy management, and reduced current stress on battery are the main features achieved from the proposed control schemes. The effectiveness of the proposed control schemes are compared with the unified cascaded control. Small-signal control gains are formulated to design the voltage and current loops of the proposed energy management schemes. Detailed stability analysis is also presented to find the boundary values of compensator gains. In addition, a high-gain PV converter is proposed for extraction of maximum power from the solar panels. High voltage gain, reduced reverse recovery of diodes, and less duty cycle operation are the key features obtained from the proposed high-gain converter. The validity of the proposed energy management schemes with high-gain converter is verified by the detailed simulation and experimental studies.

Blur-Invariant Deep Learning for Blind-Deblurring

Abstract: In this paper, we investigate deep neural networks for blind motion deblurring. Instead of regressing for the motion blur kernel and performing non-blind deblurring outside of the network (as most methods do), we propose a compact and elegant end-to-end deblurring network. Inspired by the data-driven sparse-coding approaches that are capable of capturing linear dependencies in data, we generalize this notion by embedding non-linearities into the learning process. We propose a new architecture for blind motion deblurring that consists of an autoencoder that learns the data prior, and an adversarial network that attempts to generate and discriminate between clean and blurred features. Once the network is trained, the generator learns a blur-invariant data representation which when fed through the decoder results in the final deblurred output.