Farzaneh Abdollahi
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Farzaneh Abdollahi has a total of 56 co-authors. Here are the top co-authors by number of shared publications:
- Hamed Rezaee
Department of Mechanical Engineering, Imperial College London, London, U.K.
10 shared publications - Heidar Ali Talebi
Department of Electrical Engineering, Amirkabir University of Technology, Tehran, Iran
9 shared publications - Amin Hajizadeh
AAU ENERGY, Aalborg University, Esbjerg, Denmark
4 shared publications - Amin Jalal Aghdasian
Department of Electrical Engineering Amirkabir University of Technology (Tehran Polytechnic) Tehran Iran
4 shared publications - Elaheh Sadat Ahmadi Mousavi
Department of Electrical Engineering Amirkabir University of Technology (Tehran Polytechnic) Tehran Iran
4 shared publications Hamed Rezaee
Department of Electrical and Electronic Engineering Imperial College London London UK
4 shared publications
Latest Publications
- 1 October 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Vehicles
- Vol. 11 (10) , 1075-1084
Tackling Snow-Induced Challenges: Safe Autonomous Lane-Keeping With Robust Reinforcement Learning
- 1 October 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Vehicles
- Vol. 11 (10) , 1075-1084
Tackling Snow-Induced Challenges: Safe Autonomous Lane-Keeping With Robust Reinforcement Learning
- Open Access
- 21 August 2026
- Published by Elsevier in SSRN Electronic Journal
Detecting and Isolating Lithium-Ion Battery Unseen Overcharge Faults UsingPhysics-Informed Unsupervised Deep Learning
This paper presents an innovative framework for detecting and isolating lithium-ion batteries (LIBs) overcharge faults using thermodynamicanalysis and unsupervised deep learning. Trained exclusively on temperature data from healthy cells, the methodaddresses three challenges: (a) model-based methods cannot capture complex internal fault dynamics, (b) labeled fault data isscarce, and (c) the issues regarding using voltage sensors, a common overcharge cause. The proposed unsupervised approachdetects deviations from normal operation and isolates slight overcharge faults through a physics-informed process based on thermodynamicsand incremental capacity analysis (ICA). Experimental validation is performed on 18650 NMC cells under healthycycling, various overcharge levels, over-discharge, low-temperature, and high-temperature conditions. A 1D convolutional neuralnetwork autoencoder initially detects anomalies with a robust threshold, achieving 98.67% F1-score on unseen data. Subsequently,the physics-informed isolation process, leveraging the unique thermal signature of overcharge identified via thermodynamic analysis,isolates the overcharge faults with 97% accuracy. This single-sensor approach offers a computationally efficient and reliablesolution for practical real-world BMS
- 21 August 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessDetecting and Isolating Lithium-Ion Battery Unseen Overcharge Faults UsingPhysics-Informed Unsupervised Deep Learning
This paper presents an innovative framework for detecting and isolating lithium-ion batteries (LIBs) overcharge faults using thermodynamicanalysis and unsupervised deep learning. Trained exclusively on temperature data from healthy cells, the methodaddresses three challenges: (a) model-based methods cannot capture complex internal fault dynamics, (b) labeled fault data isscarce, and (c) the issues regarding using voltage sensors, a common overcharge cause. The proposed unsupervised approachdetects deviations from normal operation and isolates slight overcharge faults through a physics-informed process based on thermodynamicsand incremental capacity analysis (ICA). Experimental validation is performed on 18650 NMC cells under healthycycling, various overcharge levels, over-discharge, low-temperature, and high-temperature conditions. A 1D convolutional neuralnetwork autoencoder initially detects anomalies with a robust threshold, achieving 98.67% F1-score on unseen data. Subsequently,the physics-informed isolation process, leveraging the unique thermal signature of overcharge identified via thermodynamic analysis,isolates the overcharge faults with 97% accuracy. This single-sensor approach offers a computationally efficient and reliablesolution for practical real-world BMS
- 10 May 2026
- Published by Elsevier in Journal of Energy Storage
- Vol. 168
Diagnosing long-term slight overcharging in commercial NMC lithium-ion batteries via novel electrochemical features
- 10 May 2026
- Published by Elsevier in Journal of Energy Storage
- Vol. 168
Diagnosing long-term slight overcharging in commercial NMC lithium-ion batteries via novel electrochemical features
- Open Access
- 21 February 2026
- Published by Elsevier in SSRN Electronic Journal
Detecting and Isolating Lithium-Ion Battery Unseen Overcharge Faults Using Physics-Informed Unsupervised Deep Learning
This paper presents an innovative framework for detecting and isolating faults in lithium-ion batteries (LIBs) by integrating physical insights obtained through thermodynamic analysis alongside deep learning techniques. By training a data-driven model exclusively on temperature data of healthy cells, the proposed methodology addresses several principal challenges in LIB overcharge detection: (a) the inability of model-based methods to capture the complex dynamics of LIB internal faults, (b) the lack of labeled faulty data, and (c) the issues associated with utilizing voltage sensor data, which may be a primary cause of overcharging. The proposed unsupervised methodology, trained without any faulty data, identifies deviations from normal operation and isolates slight overcharge faults through a novel, physics-informed isolation process derived from thermodynamic principles and incremental capacity analysis (ICA). The presented method is validated through experimental data collected from 18650 NMC cells subjected to healthy cycling, various levels of slight overcharge, and a broad range of other fault conditions, including over-discharge, low-temperature, and high-temperature. A 1D convolutional neural network autoencoder initially detects anomalies with high accuracy; a systematically established robust threshold achieves an F1-score of 98.67% on unseen validation data, maintaining above 96% even in the presence of 5% sensor noise. Subsequently, the physics-informed isolation process, leveraging the unique thermal signature of overcharge identified via thermodynamic analysis, effectively isolates the overcharge faults with 97% accuracy. This integrated, single-sensor approach offers a computationally efficient and reliable solution suitable for practical implementation within real-world BMS.
- 21 February 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessDetecting and Isolating Lithium-Ion Battery Unseen Overcharge Faults Using Physics-Informed Unsupervised Deep Learning
This paper presents an innovative framework for detecting and isolating faults in lithium-ion batteries (LIBs) by integrating physical insights obtained through thermodynamic analysis alongside deep learning techniques. By training a data-driven model exclusively on temperature data of healthy cells, the proposed methodology addresses several principal challenges in LIB overcharge detection: (a) the inability of model-based methods to capture the complex dynamics of LIB internal faults, (b) the lack of labeled faulty data, and (c) the issues associated with utilizing voltage sensor data, which may be a primary cause of overcharging. The proposed unsupervised methodology, trained without any faulty data, identifies deviations from normal operation and isolates slight overcharge faults through a novel, physics-informed isolation process derived from thermodynamic principles and incremental capacity analysis (ICA). The presented method is validated through experimental data collected from 18650 NMC cells subjected to healthy cycling, various levels of slight overcharge, and a broad range of other fault conditions, including over-discharge, low-temperature, and high-temperature. A 1D convolutional neural network autoencoder initially detects anomalies with high accuracy; a systematically established robust threshold achieves an F1-score of 98.67% on unseen validation data, maintaining above 96% even in the presence of 5% sensor noise. Subsequently, the physics-informed isolation process, leveraging the unique thermal signature of overcharge identified via thermodynamic analysis, effectively isolates the overcharge faults with 97% accuracy. This integrated, single-sensor approach offers a computationally efficient and reliable solution suitable for practical implementation within real-world BMS.
- Open Access
IRL-DAL: Safe and Adaptive Trajectory Planning for Autonomous Driving via Energy-Guided Diffusion Models
This paper proposes a novel inverse reinforcement learning framework using a diffusion-based adaptive lookahead planner (IRL-DAL) for autonomous vehicles. Training begins with imitation from an expert finite state machine (FSM) controller to provide a stable initialization. Environment terms are combined with an IRL discriminator signal to align with expert goals. Reinforcement learning (RL) is then performed with a hybrid reward that combines diffuse environmental feedback and targeted IRL rewards. A conditional diffusion model, which acts as a safety supervisor, plans safe paths. It stays in its lane, avoids obstacles, and moves smoothly. Then, a learnable adaptive mask (LAM) improves perception. It shifts visual attention based on vehicle speed and nearby hazards. After FSM-based imitation, the policy is fine-tuned with Proximal Policy Optimization (PPO). Training is run in the Webots simulator with a two-stage curriculum. A 96\% success rate is reached, and collisions are reduced to 0.05 per 1k steps, marking a new benchmark for safe navigation. By applying the proposed approach, the agent not only drives in lane but also handles unsafe conditions at an expert level, increasing robustness.We make our code publicly available.
Open AccessIRL-DAL: Safe and Adaptive Trajectory Planning for Autonomous Driving via Energy-Guided Diffusion Models
This paper proposes a novel inverse reinforcement learning framework using a diffusion-based adaptive lookahead planner (IRL-DAL) for autonomous vehicles. Training begins with imitation from an expert finite state machine (FSM) controller to provide a stable initialization. Environment terms are combined with an IRL discriminator signal to align with expert goals. Reinforcement learning (RL) is then performed with a hybrid reward that combines diffuse environmental feedback and targeted IRL rewards. A conditional diffusion model, which acts as a safety supervisor, plans safe paths. It stays in its lane, avoids obstacles, and moves smoothly. Then, a learnable adaptive mask (LAM) improves perception. It shifts visual attention based on vehicle speed and nearby hazards. After FSM-based imitation, the policy is fine-tuned with Proximal Policy Optimization (PPO). Training is run in the Webots simulator with a two-stage curriculum. A 96\% success rate is reached, and collisions are reduced to 0.05 per 1k steps, marking a new benchmark for safe navigation. By applying the proposed approach, the agent not only drives in lane but also handles unsafe conditions at an expert level, increasing robustness.We make our code publicly available.
- 2 January 2026
- Published by Springer Nature
- p. 185-202
A Soft Actor-Critic Approach for Energy Management in Microgrids
- 2 January 2026
- Published by Springer Nature
- p. 185-202
A Soft Actor-Critic Approach for Energy Management in Microgrids
- Open Access
- 1 January 2026
- Published by Institution of Engineering and Technology (IET) in IET Control Theory & Applications
- Vol. 20 (1)
Fault‐Tolerant Temperature Control of HRSG Superheaters: Stability Analysis Under Valve Leakage Using Physics‐Informed Neural Networks
Faults and operational disturbances in heat recovery steam generators (HRSGs), such as valve leakage, present significant challenges, disrupting steam temperature regulation and potentially causing efficiency losses, safety risks, and unit shutdowns. Traditional PI controllers often struggle due to inherent system delays, nonlinear dynamics, and static gain limitations. This paper introduces a fault‐tolerant temperature control framework by integrating a PI plus feedforward control strategy with physics‐informed neural networks (PINNs). The feedforward component anticipates disturbances, preemptively adjusting control actions, while the PINN adaptively tunes control gains in real‐time, embedding thermodynamic constraints to manage varying operating conditions and valve leakage faults. A Lyapunov‐based stability analysis confirms the asymptotic convergence of temperature tracking errors under bounded leakage conditions. Simulation results using operational data from the Pareh‐Sar Combined Cycle Power Plant demonstrate significantly improved response times, reduced temperature deviations, enhanced fault resilience, and smooth gain adjustments. The proposed adaptive, data‐driven methodology shows strong potential for industrial deployment, ensuring reliable operation, autonomous fault recovery, and enhanced performance in HRSG systems.
- 1 January 2026
- Published by Institution of Engineering and Technology (IET) in IET Control Theory & Applications
- Vol. 20 (1)
Open AccessFault‐Tolerant Temperature Control of HRSG Superheaters: Stability Analysis Under Valve Leakage Using Physics‐Informed Neural Networks
Faults and operational disturbances in heat recovery steam generators (HRSGs), such as valve leakage, present significant challenges, disrupting steam temperature regulation and potentially causing efficiency losses, safety risks, and unit shutdowns. Traditional PI controllers often struggle due to inherent system delays, nonlinear dynamics, and static gain limitations. This paper introduces a fault‐tolerant temperature control framework by integrating a PI plus feedforward control strategy with physics‐informed neural networks (PINNs). The feedforward component anticipates disturbances, preemptively adjusting control actions, while the PINN adaptively tunes control gains in real‐time, embedding thermodynamic constraints to manage varying operating conditions and valve leakage faults. A Lyapunov‐based stability analysis confirms the asymptotic convergence of temperature tracking errors under bounded leakage conditions. Simulation results using operational data from the Pareh‐Sar Combined Cycle Power Plant demonstrate significantly improved response times, reduced temperature deviations, enhanced fault resilience, and smooth gain adjustments. The proposed adaptive, data‐driven methodology shows strong potential for industrial deployment, ensuring reliable operation, autonomous fault recovery, and enhanced performance in HRSG systems.
- Open Access
- 1 January 2026
- Published by Institution of Engineering and Technology (IET) in IET Renewable Power Generation
- Vol. 20 (1)
History‐Agnostic Lithium‐Ion Batteries State of Health and Remaining Useful Life Estimation Using Fusion‐Based Bayesian Neural Network
This paper presents an innovative, history‐agnostic framework for lithium‐ion battery prognostics that addresses critical limitations of existing methods by removing the necessity for historical degradation data through the application of Bayesian neural networks (BNNs) and feature‐fusion techniques. A significant contribution is the introduction of four new health indicators (HIs) based on the voltage–time derivative (), which are fused with six other well‐established HIs derived from a single charge‐discharge cycle as inputs to the BNN. This comprehensive feature set facilitates two distinct BNNs to perform direct, non‐iterative predictions of the state of health (SoH) and remaining useful life (RUL), complete with principled uncertainty quantification. The effectiveness of the framework is validated through experiments conducted on the Oxford battery degradation dataset, employing a leave‐one‐out cross‐validation strategy. The results indicate exceptional performance, with the SoH model achieving values exceeding 0.998 and MAE below 0.002 across all test cells. The direct RUL prediction model also demonstrates high accuracy, with values consistently above 0.97 and a MAPE as low as 8.5%. By offering robust, uncertainty‐aware predictions without reliance on historical data, this approach presents a practical and scalable solution for real‐world battery management systems, particularly in applications involving second‐life batteries.
- 1 January 2026
- Published by Institution of Engineering and Technology (IET) in IET Renewable Power Generation
- Vol. 20 (1)
Open AccessHistory‐Agnostic Lithium‐Ion Batteries State of Health and Remaining Useful Life Estimation Using Fusion‐Based Bayesian Neural Network
This paper presents an innovative, history‐agnostic framework for lithium‐ion battery prognostics that addresses critical limitations of existing methods by removing the necessity for historical degradation data through the application of Bayesian neural networks (BNNs) and feature‐fusion techniques. A significant contribution is the introduction of four new health indicators (HIs) based on the voltage–time derivative (), which are fused with six other well‐established HIs derived from a single charge‐discharge cycle as inputs to the BNN. This comprehensive feature set facilitates two distinct BNNs to perform direct, non‐iterative predictions of the state of health (SoH) and remaining useful life (RUL), complete with principled uncertainty quantification. The effectiveness of the framework is validated through experiments conducted on the Oxford battery degradation dataset, employing a leave‐one‐out cross‐validation strategy. The results indicate exceptional performance, with the SoH model achieving values exceeding 0.998 and MAE below 0.002 across all test cells. The direct RUL prediction model also demonstrates high accuracy, with values consistently above 0.97 and a MAPE as low as 8.5%. By offering robust, uncertainty‐aware predictions without reliance on historical data, this approach presents a practical and scalable solution for real‐world battery management systems, particularly in applications involving second‐life batteries.
- Open Access
- 29 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 917-938
Dynamic Modeling of a Bio-Inspired Snake-Like Soft Robot Using a Parallel Robotic Architecture
This study proposes a novel dynamic model for a single-segment soft robotic actuator capable of snake-inspired bending motion. Dynamic modeling of soft robots remains a critical challenge due to their inherent nonlinearities and compliance, yet it is foundational for advancing mechanical design and control strategies. To address these challenges, we present an innovative framework rooted in parallel robot mechanisms, enabling precise kinematic and dynamic characterization of the snake-like soft robotic system. A key aspect of this work is the mapping from the task space of the parallel robot to the arc space, which is essential for deriving the dynamic model of the single-segment bending Snake-like Soft Robot (SSR). Furthermore, this paper outlines the construction of a physical prototype, along with the parameter identification for the data model. A brief discussion on a semi-Proportional Derivative (semi-PD) controller is also included. To validate the modeling accuracy, the response of the system is analyzed by simulations and real-world implementations of the closed-loop system with the controller. The results from simulations and practical applications confirm the precision of the proposed model.
- 29 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 917-938
Open AccessDynamic Modeling of a Bio-Inspired Snake-Like Soft Robot Using a Parallel Robotic Architecture
This study proposes a novel dynamic model for a single-segment soft robotic actuator capable of snake-inspired bending motion. Dynamic modeling of soft robots remains a critical challenge due to their inherent nonlinearities and compliance, yet it is foundational for advancing mechanical design and control strategies. To address these challenges, we present an innovative framework rooted in parallel robot mechanisms, enabling precise kinematic and dynamic characterization of the snake-like soft robotic system. A key aspect of this work is the mapping from the task space of the parallel robot to the arc space, which is essential for deriving the dynamic model of the single-segment bending Snake-like Soft Robot (SSR). Furthermore, this paper outlines the construction of a physical prototype, along with the parameter identification for the data model. A brief discussion on a semi-Proportional Derivative (semi-PD) controller is also included. To validate the modeling accuracy, the response of the system is analyzed by simulations and real-world implementations of the closed-loop system with the controller. The results from simulations and practical applications confirm the precision of the proposed model.
- Open Access
Tackling Snow-Induced Challenges: Safe Autonomous Lane-Keeping with Robust Reinforcement Learning
This paper proposes two new algorithms for the lane keeping system (LKS) in autonomous vehicles (AVs) operating under snowy road conditions. These algorithms use deep reinforcement learning (DRL) to handle uncertainties and slippage. They include Action-Robust Recurrent Deep Deterministic Policy Gradient (AR-RDPG) and end-to-end Action-Robust convolutional neural network Attention Deterministic Policy Gradient (AR-CADPG), two action-robust approaches for decision-making. In the AR-RDPG method, within the perception layer, camera images are first denoised using multi-scale neural networks. Then, the centerline coefficients are extracted by a pre-trained deep convolutional neural network (DCNN). These coefficients, concatenated with the driving characteristics, are used as input to the control layer. The AR-CADPG method presents an end-to-end approach in which a convolutional neural network (CNN) and an attention mechanism are integrated within a DRL framework. Both methods are first trained in the CARLA simulator and validated under various snowy scenarios. Real-world experiments on a Jetson Nano-based autonomous vehicle confirm the feasibility and stability of the learned policies. Among the two models, the AR-CADPG approach demonstrates superior path-tracking accuracy and robustness, highlighting the effectiveness of combining temporal memory, adversarial resilience, and attention mechanisms in AVs.
Open AccessTackling Snow-Induced Challenges: Safe Autonomous Lane-Keeping with Robust Reinforcement Learning
This paper proposes two new algorithms for the lane keeping system (LKS) in autonomous vehicles (AVs) operating under snowy road conditions. These algorithms use deep reinforcement learning (DRL) to handle uncertainties and slippage. They include Action-Robust Recurrent Deep Deterministic Policy Gradient (AR-RDPG) and end-to-end Action-Robust convolutional neural network Attention Deterministic Policy Gradient (AR-CADPG), two action-robust approaches for decision-making. In the AR-RDPG method, within the perception layer, camera images are first denoised using multi-scale neural networks. Then, the centerline coefficients are extracted by a pre-trained deep convolutional neural network (DCNN). These coefficients, concatenated with the driving characteristics, are used as input to the control layer. The AR-CADPG method presents an end-to-end approach in which a convolutional neural network (CNN) and an attention mechanism are integrated within a DRL framework. Both methods are first trained in the CARLA simulator and validated under various snowy scenarios. Real-world experiments on a Jetson Nano-based autonomous vehicle confirm the feasibility and stability of the learned policies. Among the two models, the AR-CADPG approach demonstrates superior path-tracking accuracy and robustness, highlighting the effectiveness of combining temporal memory, adversarial resilience, and attention mechanisms in AVs.