College of Engineering, Pune

Electric two wheelers are becoming very common in Indian cities, but their charging units often work in harsh and uncontrolled places such as home balconies, shared parking areas, and roadside stalls. In these places, changes in temperature, voltage quality, and usage patterns put extra load on the internal parts of the charger. This paper studies the reliability of the Power Factor Correction (PFC) stage and the DC-DC converter stage of a two wheeler EV charger using Failure Mode and Effects Analysis (FMEA) along with an Accelerated Life Testing (ALT) protocol. Field reports show that faults like MOSFET over temperature trips, battery over under charge protection, and short circuit protection are common in these chargers. The FMEA method is used to find the Risk Priority Number (RPN) of each part, and results show that MOSFETs and electrolytic capacitors carry the highest risk because of thermal stress and aging. To check these results, a charger sample was tested under thermal cycling, voltage transients, and power cycling, and the same fault patterns seen in the field were reproduced in the lab. The study also finds that fast voltage changes affect the PID control loop of the DC-DC converter, causing overshoot and unwanted protection trips. Based on these findings, hardware, thermal, protection, and control loop improvements are suggested, ranked from low cost to high cost. The paper also discusses using field data logging to keep improving FMEA scores over time. This work gives a simple and repeatable method to improve the safety and life of two-wheeler EV chargers.

The soil properties have a strong impact on agricultural productivity, crop selection, and sustainable land management. Soil pH is one of the most significant factors influencing nutrient availability, microbial activity, and soil fertility. Conventional soil testing approaches have proven to be time-consuming, taking time, problems with symmetry, and expensive, making them difficult to scale for fieldwork in agriculture. This work explores the use of DL methodologies that will utilize basic soil attributes such as electrical conductivity (EC), phosphorus, potassium, organic carbon, and lime to predict soil pH. A real time soil data collected from different villages in Sangli, Miraj district, Maharashtra, India and developed neural network model. The model was fitted with regression metrics, resulting in a MAE 0.732, showing depth of its prediction capabilities on soil pH. Soil analysis based on Feedforward Deep Neural Network provides a data-driven technique for precision agriculture with applications in the real-time monitoring of soil health and the precise application of fertilizers. The results emphasize how AI can revolutionize conventional agriculture by enabling farmers to make data-driven decisions based on precision soil data, leading to sustainable agricultural practices and improved crop yields. Ongoing research will involve a larger data corpus, further environmental parameters, and network architecture optimization to improve prediction accuracy.

The key breakthrough to production of deepfake speech poses daunting difficulty to the most widely adopted detection processes. A new hybrid system taking advantage of speech's self-supervised learning advantages of its representations, and by close-in spectro-temporal analysis to enhance the effectiveness of deepfake detection is proposed. The system uses WavLM embeddings and phase-wise dynamic feature extraction during the case of the scenario of contrastive learning, thus allowing to uncover of both high-level semantic variations and low-level signal anomalous variations. The network uses phase-aware attention to emphasize fine variations of fraudulent speech and implements the usage of a temperature-scaled contrastive adapter to attain vigorous then the characteristic variations between genuine and forged specimens. The addition of cross-modal attributes is useful to the interrelation between the interrelation between the characteristic of time and characteristic of frequency. The system is tested with the ASVspoof 2019 dataset with our system registering a validation accuracy of 97.98% with keeping the test accuracy across to 91.99%, with unparalleled usage to widely adopted classes of spectrum spoofing attack techniques. The system shows exemplary proficiency to suppress the effect of the interrelation of interfered noise and thus overcomes a key challenge with regards to real-world application. Systematic ablation experiments confirm interdependence advantages of the attributes of phases and of learning by contrast within the detection system. The research undertaken here serves a seminal milestone to the literature, with a culmination of another of merging of learning by itself and of forensic signal processing analysis to achieve of a considerable verification to the effect of detecting of deepfake speech.

Copper Indium Gallium Diselenide (CIGS) solar cells are regarded as promising due to their good electrical and optical features. In this work we have simulated substrate structure of CIGS thin film solar cell using SCAPS-1D software. The current study investigates the effect of different window layers on the electrical parameters of a solar cell. Different window layers have been used, including ZnO, In2S3, ITO, AZO, and ZnMgO. The dependence of efficiency on window layer and defect density is clearly visible in this work. Efficiency shows variation from 25.81% to 28.14%. Apart from efficiency, the current and voltage generated are dependent on the window layer and the defect concentrations connected with it. AZO yields the efficiency of about 28 % whereas, ITO contribute least efficiency near about 25%. Thickness optimization of window layers simulated in this work was carried out. This study demonstrates that AZO and ZnMgO are favorable materials for creating the window layer in CIGS solar cells. Small deviation was found in efficiency upon change in defect density. The thickness of window layers employed in this work was optimized. Thickness of window layers shows an adverse effect on photovoltaic features. Band alignment shows variation upon change in window layer.