Aniruddh Herle

A. Herle, C. M. O'Riordan, S. Vegetti

Convolution Neural Networks trained for the task of lens finding with similar architecture and training data as is commonly found in the literature are biased classifiers. An understanding of the selection function of lens finding neural networks will be key to fully realising the potential of the large samples of strong gravitational lens systems that will be found in upcoming wide-field surveys. We use three training datasets, representative of those used to train galaxy-galaxy and galaxy-quasar lens finding neural networks. The networks preferentially select systems with larger Einstein radii and larger sources with more concentrated source-light distributions. Increasing the detection significance threshold to 12$σ$ from 8$σ$ results in 50 per cent of the selected strong lens systems having Einstein radii $θ_\mathrm{E}$ $\ge$ 1.04 arcsec from $θ_\mathrm{E}$ $\ge$ 0.879 arcsec, source radii $R_S$ $\ge$ 0.194 arcsec from $R_S$ $\ge$ 0.178 arcsec and source Sérsic indices $n_{\mathrm{Sc}}^{\mathrm{S}}$ $\ge$ 2.62 from $n_{\mathrm{Sc}}^{\mathrm{S}}$ $\ge$ 2.55. The model trained to find lensed quasars shows a stronger preference for higher lens ellipticities than those trained to find lensed galaxies. The selection function is independent of the slope of the power-law of the mass profiles, hence measurements of this quantity will be unaffected. The lens finder selection function reinforces that of the lensing cross-section, and thus we expect our findings to be a general result for all galaxy-galaxy and galaxy-quasar lens finding neural networks.

Aniruddh Herle, Janamejaya Channegowda, Dinakar Prabhu

The Electric Vehicle (EV) Industry has seen extraordinary growth in the last few years. This is primarily due to an ever increasing awareness of the detrimental environmental effects of fossil fuel powered vehicles and availability of inexpensive Lithium-ion batteries (LIBs). In order to safely deploy these LIBs in Electric Vehicles, certain battery states need to be constantly monitored to ensure safe and healthy operation. The use of Machine Learning to estimate battery states such as State-of-Charge and State-of-Health have become an extremely active area of research. However, limited availability of open-source diverse datasets has stifled the growth of this field, and is a problem largely ignored in literature. In this work, we propose a novel method of time-series battery data augmentation using deep neural networks. We introduce and analyze the method of using two neural networks working together to alternatively produce synthetic charging and discharging battery profiles. One model produces battery charging profiles, and another produces battery discharging profiles. The proposed approach is evaluated using few public battery datasets to illustrate its effectiveness, and our results show the efficacy of this approach to solve the challenges of limited battery data. We also test this approach on dynamic Electric Vehicle drive cycles as well.

Aniruddh Herle, Janamejaya Channegowda, Kali Naraharisetti

Electric Vehicles (EVs) are rapidly increasing in popularity as they are environment friendly. Lithium Ion batteries are at the heart of EV technology and contribute to most of the weight and cost of an EV. State of Charge (SOC) is a very important metric which helps to predict the range of an EV. There is a need to accurately estimate available battery capacity in a battery pack such that the available range in a vehicle can be determined. There are various techniques available to estimate SOC. In this paper, a data driven approach is selected and a Nonlinear Autoregressive Network with Exogenous Inputs Neural Network (NARXNN) is explored to accurately estimate SOC. NARXNN has been shown to be superior to conventional Machine Learning techniques available in the literature. The NARXNN model is developed and tested on various EV Drive Cycles like LA92, US06, UDDS and HWFET to test its performance on real world scenarios. The model is shown to outperform conventional statistical machine learning methods and achieve a Mean Squared Error (MSE) in the 1e-5 range.

Aniruddh Herle, Janamejaya Channegowda, Dinakar Prabhu

Electric Vehicle (EV) fleets have dramatically expanded over the past several years. There has been significant increase in interest to electrify all modes of transportation. EVs are primarily powered by Energy Storage Systems such as Lithium-ion Battery packs. Total battery pack capacity translates to the available range in an EV. State of Charge (SOC) is the ratio of available battery capacity to total capacity and is expressed in percentages. It is crucial to accurately estimate SOC to determine the available range in an EV while it is in use. In this paper, a Temporal Convolution Network (TCN) approach is taken to estimate SOC. This is the first implementation of TCNs for the SOC estimation task. Estimation is carried out on various drive cycles such as HWFET, LA92, UDDS and US06 drive cycles at 1 C and 25 °Celsius. It was found that TCN architecture achieved an accuracy of 99.1%.