Poonam Patil

@bvcoenm.edu.in

Assistant Professor Instrumentation Department
Bharati Vidyapeeth College of Engineering Navi Mumbai

Poonam Patil


               Download Compact PDF  
https://researchid.co/punam.patil

RESEARCH, TEACHING, or OTHER INTERESTS

Control and Optimization, Artificial Intelligence, Electrical and Electronic Engineering
9

Scopus Publications

Scopus Publications

  • A hybrid photonic neural network equalizer for coherent WDM fiber links using a physical Clements mesh and microring nonlinearities
    Jotiram K. Deshmukh, Prashant Vishnu Bhosale, Manisha K. Bhole, Reshma N. Pawar, Poonam J. Patil, Giridhar Urkude, Vivek S. Kadam
    Journal of Optical Communications, 2026
    A novel hybrid photonic neural network (PNN) equalizer for coherent hybrid photonic neural network (PNN) equalizer (WDM) optical communication systems is introduced and tested via differentiable simulations. The architecture features a physical 4 × 4 and extendable 16 × 16 Clements interferometer mesh responsible for broadband linear MIMO equalization and per-channel nonlinear microring resonator neurons possessing a Kerr-like corrective phase response. In addition to effectively train the PNN, an end-to-end differentiable WDM fiber channel simulator is provided that encompassing all major physical impairments like chromatic dispersion, polarization-mode dispersion, self-phase modulation, cross-phase modulation, four-wave mixing, and ASE noise. Through physics-informed optimization, the hybrid PNN learns linear and nonlinear compensations for the channel response to achieve full optical domain equalization minimizing expected DSP footprint. Reliability is confirmed through simulation results demonstrating large differences in constellation recovery and error vector magnitude (EVM) such that physical PNN-channel equalization should be possible down the line for truly high-baud rate WDM systems.
  • QR-Based Micro-Payments and Small Business Resilience: A Digital Path to Achieving SDG 8 and SDG 9
    Aman Gupta, Suma Sidramappa Hoaamani, Jie Feng, Mona Sharma, Mukesh Patil, Poonam Jagdish Patil
    Enterprise Development and Microfinance, 2025
    QR-based micro-payments have become a revolutionary digital infrastructure to small business in the developing and emerging economies, allowing small-cost interoperable and immediate transactions. This paper analyses the role of the adoption of QR-based payment in increasing the resilience of small businesses and supporting Sustainable Development Goal 8 on decent work and economic growth and SDG 9 on industry, innovation, and infrastructure. The study assesses the effects on revenue stability, operational continuity, and access to the market by a mixed-method approach involving the use of survey data collected on micro and small enterprises, transaction-level analysis as well as regression-based resilience modeling. There is empirical evidence that firms that implemented QR payments grew their transaction volume by 18-26% and their monthly income steadiness by 12-19% and their cash-managed risks by 21% relative to cash-related companies. The formal financial integration was also enhanced by the use of digital payment, with 34 per cent of companies accessing microcredit or digital savings products one year after adoption. Infrastructurally speaking, QR systems reduced the barriers to going into digital business, raised the standards of interoperability between payment platforms, and facilitated local service innovation. The article also results in the strongest gains in resilience between women-owned enterprises and informal enterprises that conduct their operations in high volatility market conditions. Altogether, the results indicate that QR-based micro-payments can offer a digital channel of scale to enhance resistance of small businesses, fasten financial inclusion, and promote sustainable economic performance in accordance with the priorities of global development.
  • Simulation and Optimization of Hvac Systems Using Machine Learning
    Manisha Kishor Bhole, Pankaj Hiraman Zope, Punam J. Patil
    2025 2nd International Conference on Integration of Computational Intelligent System Icicis 2025, 2025
    HVAC systems are critical to providing thermal comfort and maintaining indoor air quality, yet they are also among the largest energy consumers in buildings. Traditional control methods often struggle with the complex, non-linear dynamics of building environments, leading to energy waste and suboptimal performance. This study proposes modelling of HVAC system using EnergyPlus software to generate data and model-based Reinforcement Learning (RL) method to optimize the performance of HVAC system. We leverage LSTMs' ability to learn temporal dependencies in data to create a highly accurate predictive model of a building's thermal behavior. This LSTMbased model is then integrated into an optimization framework to determine the most energy-efficient control strategies for the HVAC system while ensuring occupant comfort. By using a data-driven approach, our method adapts to specific building characteristics without extensive physical modeling. We demonstrate the effectiveness of this methodology through a case study on a two-zone building case study. The results highlight the potential of using machine learning algorithms like LSTMs to create more intelligent, and sustainable building energy management systems.
  • Predictive Maintenance of Industrial Equipment Failure Using Random Forest Algorithm
    Poonam J. Pati, Pankaj H. Zope, Manisha K. Bhole
    2025 2nd International Conference on Integration of Computational Intelligent System Icicis 2025, 2025
    Predictive maintenance (PdM) has become an essential component in modern industrial systems to ensure reliability, reduce downtime, and minimize maintenance costs. This paper presents predicting failures in DC electric motors using machine learning –based approach, which are critical components in industrial setups. Real-time sensor data such as voltage, current, temperature, and vibration are used and Random Forest classifier trained with this dataset. From result it has been observed that the Random Forest model effectively classifies the motor's health status, demonstrating its potential for industrial predictive maintenance systems. We trained model on a publicly available dataset from Kaggle that simulates industrial motor conditions. The model effectively predicts motor health based on parameters like temperature, vibration, and load. We also developed a monitoring dashboard to visualize predictions and system health. The system achieves over 96% accuracy and is ready for deployment in real-time industrial applications.[1][12]
  • Optimization of HVAC (Heating Ventilation and Air Conditioning Systems) using Machine Learning
    Manisha Kishor Bhole, Pankaj Hiraman Zope, Poonam J. Patil
    Proceedings of 3rd International Conference on Intelligent Cyber Physical Systems and Internet of Things Icoici 2025, 2025
    HVAC system is one of the largest contributors to building total energy consumption with both high cost and sustainability concerns. HVAC systems are responsible for indoor thermal comfort provision and indoor air quality provision but through their traditional control techniques—rule-based, and model-based control strategies that are not capable of addressing the complexity and dynamics of real building environments. The shortcoming of these approaches causes inefficiencies, excessive energy consumption, and suboptimal control performance. Introduce a model-based Reinforcement Learning (RL) method that combines a neural network-based system approximation with Model Predictive Control (MPC). Our approach learns the internal system dynamics of the HVAC system with a neural network and then applies MPC with a random-sampling shooting method to learn optimal control policies. With a learned model instead of direct interaction, our method dramatically decreases the number of training samples needed for convergence. In addition, impose safety constraints by limiting control actions to within pre-defined safe ranges and bounding rapid changes using prior knowledge to ensure stable and reliable system behavior. Validate our suggested method using Energy Plus, a widely used building energy simulation software, on a two-zone data center case study. Our findings confirm the feasibility of model-based reinforcement learning for intelligent HVAC control, with energy savings, enhanced efficiency, and enhanced safety on actual building applications.
  • A Hybrid Hidden Markov Model and Neural Network Framework for Fraud Detection in Mobile Payment Systems
    M. Karthikeyan, T. Kohila Kanagalakshmi, Trupti Mandar Joshi, S. M. Karpagavalli, Poonam J. Patil, Suganya S
    2025 2nd IEEE International Conference for Women in Computing Incowoco 2025, 2025
    Despite the advent of mobile wallets and other mobile payment systems, a plethora of previously inaccessible transactions have become possible through mobile channels. The most common kind of mobile payment fraud is the unauthorized use of credit card or certification numbers, although there are other methods as well. This surge in fraud has occurred alongside the growth of mobile payment convenience. The lack of a tangible means of payment makes these systems more susceptible to fraud. In response, they present a framework for fraud detection that improves the accuracy of machine learning models by using thorough data preparation techniques, such as variable selection, aggregation, and feature construction. An efficient and real-time method of distinguishing between fraudulent and valid transactions is built into the framework. A high accuracy rate of 99.37% in detecting fraudulent activity with few false positives is demonstrated by the evaluation findings. These results show that the framework may make mobile payment systems much more secure and offer a realistic solution to the problems caused by mobile commerce fraud.
  • Smart Helmet for Coal Mine Workers
    Punam J Patil, Manisha Bhole, Dilip N. Pawar, Swati Nadgaundi, Reshma Pawar, Atharva Mhatre
    2023 International Conference on Integration of Computational Intelligent System Icicis 2023, 2023
    A more conventional version of the smart helmet has been developed to help miners while working in the mining sector. The mining industry frequently has dangerous occurrences, many of which end in fatalities or seriously injured parties. Using different sensors, the smart helmet able to recognize catastrophic situations such as presence of harmful gases like Carbon-Monoxide (CO), Methane (CH <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">4</inf> ), Ammonia (NH3) as well as temperature and humidity within the mine areas. Also the pulse of the coal miner monitored continuously so that it can be detected whether the miner is facing some difficulty or if any accident has occurred in the mine. Helmet wear by miner or not wear, is detected by an infrared sensor, hence negligence of the miner for not wearing the safety helmet can be avoided. Each sensor used has a threshold value that, if that value is exceeded, it causes the buzzer to activate, signaling the miners and supervisors. Wi-Fi and ThingSpeak is used for the remote transmission of information from coal mine to a central location. This technology may improve the safety and scale back accidents within the coal mines.
  • Predictive Analysis of Employee Turnover in IT Using a Hybrid CRF-BiLSTM and CNN Model
    Subha B, Irfan Abdul Karim Shaikh, Poonam Jagdish Patil, R. Sethumadhavan, M. Preetha, Harshal Patil
    International Conference on Sustainable Communication Networks and Application Icscna 2023 Proceedings, 2023
    Many people in the field of information systems are still devoting a lot of time and effort to researching and analyzing the correlation between IT spending and financial returns. More study is needed to establish the particular mechanisms or intermediate processes through which IT investments deliver value to firms, but there is growing consensus that such investments are beneficial. The proposed methodology rests on three primary pillars: data preparation, feature selection, and model training. As part of the data preparation process, discordant dimensions, redundant data, Qualitative features are useless, useless, and underutilized because they lack utility, value, and data. Extensive Optimized PCA was utilized for feature selection. CRF-BiLSTM-CNN is then used to train the models. When compared to the two most common alternatives, CNN and BiLSTM, the proposed technique performs significantly better. The success rate for the proposed method was 98.87%.
  • Model order reduction of high order LTI system using balanced truncation approximation
    Poonam J. Patil, Mukesh D. Patil
    Proceedings of 2011 International Conference on Process Automation Control and Computing Pacc 2011, 2011
    Most of the mathematical model of industrial applications have very high order, implementation and computation of such system is complicated and time consuming so there is an increasing need for obtaining suitable low order approximations of such systems. Low order models result in several advantages such as reduction in computational complexity and easy analysis of original system structure. This paper proposes a numerically efficient model order reduction method using balanced truncation for high order Linear Time Invariant(LTI) systems. Most important feature of this method is that it provides a bound of the infinity norm of approximation error, using Henkel singular values. This bound is very useful in performance evaluation. The simulation result shows the effectiveness of the proposed scheme to obtain the stable 8 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">th</sup> order reduced model from a stable 20 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">th</sup> order original system with minimum error bound.