Effective battery thermal management is very essential to enhance the safety, efficiency, and lifespan of battery systems in Electric Vehicles (EVs). Various factors influencing the thermal performance of battery system include fast charging, discharging, aging, manufacturing defect, environmental conditions, etc. To study the effect of thermal performance on battery, accurate heat generation prediction is very essential. The main issues in sensor-based temperature measurement and heat generation estimation are inaccuracy and reliability due to point-point measurement and physical contact with battery system. To address these issues, thermal image-based battery surface temperature monitoring and surface heat generation prediction based on Artificial Neural Network (ANN) is proposed to study the thermal performance of 24V,100Ah Li-ion prismatic battery cell under different charging and discharging cycles. The proposed methodology involves collecting real-time thermal images of the battery under different charging and discharging conditions using TIX580 infrared thermal imaging camera and analysing its performance. To automate heat generation prediction, Levenberg-Marquardt (LM) and Bayesian Regularization (BR) ANN models are developed. Finally, the performances of the prediction models are compared. The results show temperature uniformity and accuracy of heat generation prediction and addresses thermal runaway risks.

Accurate state-of-charge (SoC) estimation of lithium-ion (Li-ion) batteries is essential for the reliable operation of the battery management system (BMS) in electric vehicles (EVs). Conventional unscented transformation based unscented Kalman filters (UT-UKF) are very useful for moderately non-linear systems with Gaussian noise distribution data during state update. However, it provides satisfactory results only for highly non-linear and nonGaussian data. This paper provides a robust generalized unscented transformation based unscented Kalman filter (RGenUT-UKF) for SoC accurate estimation of real-world drive cycle. This proposed approach effectively captures highly non-Gaussian and non-linear characteristics using distribution-free non-linear transformation with optimally tuned noise covariance matrices ensuring reliability against unseen real-time data. The generalized unscented transformation technique uses higher order moments, in addition to mean and covariance for sigma points selection for enhancement of filter performance. This study deals with the development of an equivalent circuit model (ECM)-based GenUT-UKF for the estimation of SoC of Turnigy Graphene Li-ion battery for different drive cycles that include LA92, US06, and UDDS drive cycle. The efficacy of the proposed algorithm has been demonstrated by comparing it with UT-UKF and GenUT-extended Kalman filter (GenUT-EKF). Additionally, analysis of the robustness of the proposed algorithm has been made with LA92 drive cycle at 40 degrees C against data uncertainties such as 5 % sensor noise to the battery current, ambient temperature variations (0 degrees C, 10 degrees C and 25 degrees C) and model uncertainty in the noise covariance matrices (Q and R). The proposed GenUT-UKF showed outperformance over UT-UKF and GenUT-EKF against sensor noise, ambient temperature variations, and model uncertainty with RMSE of 0.2780%, 0.7842 %, and 0.7843 % respectively. Finally, a RGenUT-UKF has been designed by optimally tuned noise covariance matrices for the LA92 drive cycle using the Bayesian optimization technique. Improved performance of the proposed method has been ensured with a minimum RMSE of 0.4099 %. In addition, the efficacy of the proposed algorithm has been demonstrated under complex working conditions through simulation as well as experimental studies. The practicability of realizing the algorithm has been demonstrated with RT-LAB-based real-time simulator using software-in-the-loop (SIL) configuration.

Robust State of Charge (SoC) estimation under dynamic drive cycles and varying ambient temperatures is crucial for Li-ion battery. Traditional model-based methods offer accurate SoC estimation under specific operating condition with generalizable model, while data-driven methods outperform under varying operating conditions with interpretable model. To address these issues, a robust hybrid estimation algorithm is proposed for improved SoC prediction by combining Long Short-Term Memory (LSTM) and Generalized Unscented Transformation-Unscented Kalman Filter (GenUT-UKF). In this hybrid approach, LSTM captures nonlinearities in battery dynamics from the trained battery data, while GenUT-UKF suppresses noise in the estimation due to unseen data with different scenarios: i) ambient temperature variations and ii) drive cycle variations and model uncertainty. To analyze its efficacy, the performance of the proposed hybrid algorithm is compared with LSTM and LSTM-UKF. The results show that LSTM GenUT-UKF achieves superior SoC estimation, with an average accuracy improvement of 42.35% over LSTM and 0.0816% over LSTM-UKF. The comparison of robustness of LSTM GenUT-UKF with LSTM and LSTM-UKF shows that it outperforms LSTM with average improved accuracy of 99.14% and 98.02% and LSTM-UKF with 0.2341% and 0.1802% against drive cycle uncertainty and ambient temperature variation, respectively. For model uncertainty, LSTM and LSTM-UKF with improved average accuracy of 89.7% and 0.16%, respectively. Finally, the proposed algorithm is implemented on embedded hardware to demonstrate its practicability.

Precise measurement of State of Charge (SoC) in Li-ion battery is crucial for effective utilization of energy storage system. However, it is challenging with physics-based model due to computation complexity while data-driven model often lacks generalization for unseen data. To improve the robustness of data-driven model, hybrid algorithms are employed. This paper focuses on combining Temporal Fusion Transformer (TFT)-Long Short-Term Memory (LSTM) to improve the generalization of state-of charge estimation of unseen drive cycles. By leveraging the attention mechanism feature, temporal feature extraction of TFT along with memory capability of LSTM TFT-LSTM model is developed with optimal hyperparameters using Optuna framework. This model is used to estimate the SoC during the discharging cycle of a Turnigy Graphene Li-Ion Battery (5000mAh, 65C) for the LA92 drive cycle at an ambient temperature of 40 degrees C.Additionally, uncertainty analysis is performed against US06 drive cycle. Finally, the performance of TFT-LSTM is compared with LSTM model. The results demonstrate that TFT-LSTM achieves better performance compared to LSTM with Root Mean Square Error (RMSE) of 1.1962 for unseen US06 drive cycle. These findings demonstrate the robustness of hybrid TFT-LSTM against unseen drive cycle and highlight its potential under realistic conditions with variability.

In recent years, carbon emissions are increasing worldwide due to excessive usage fossil fuels. To overcome these emissions, lithium-ion (Li-ion) batteries have become more prominent alternative. Li-ion batteries are used as primary component of energy storage systems for sustainable energy in response to rising global carbon gases. Battery Management System (BMS) in Electric Vehicles (EVs) is an important aspect and is indicated by two parameters called State of Charge (SoC) and State of Health (SoH). Of these two, SoC value is related to energy distribution, charging and discharging of batteries. Hence Estimating SoC value is of high important in BMS for optimum usage of batteries. Recent trends in Artificial Intelligence and Deep Learning provides a way for new developments in algorithms for estimating SoC. At the same time, the use of programmable devices like Field Programmable Gate Arrays (FPGAs) for data processing applications provides acceleration in time and optimal use of hardware. Pynq boards which are Zynq dependent and one variant of FPGAs are capable of executing python programs directly on hardware. This paper focuses on developing different DNN architectures for estimating SoC of a Li-ion battery of an EV and realizing on Pynq Z2 board.

Electric Vehicles (EV) have gained popularity in recent years to reduce the amount of greenhouse gas emissions and utilize renewable energy sources more effectively. Fast charging of Lithium-Ion batteries (Li-Ion) in EV is a serious issue affecting battery life. The main objective of the proposed work is to develop a Hybrid Electro-Thermal Model (H-ETM) using multiple model approach to generate an optimal charging profile to enhance State-of-Health (SoH) of Li-Ion based on Multi-Objective Genetic Algorithm (MOGA). The hybrid model is developed by integrating four local models based on a multi-model approach to improve accuracy. For the dataset collected from real battery, a single model over the entire SoC range can provide terminal voltage accuracy of ±10 mV and the proposed multi-model approach yields an improved accuracy of ±5 mV. Further, the optimal current profiles under varying weight coefficients for charging time and temperature rise are generated using the proposed H-ETM. The suggested strategy's Pareto fronts are used as references to alter charging current rate to further satisfy diversified user demands, particularly for charging speed and temperature fluctuations in different charging applications. The proposed method provides more feasibility to select optimal charging patterns based on the requirements of the user by taking the trade-off between charging time and internal battery temperature rise while maintaining the constraints in state-of-charge, charging current, internal temperature rise, and charging time

Estimation of State of Charge (SoC) with higher accuracy is very essential for range prediction, optimal discharging of Lithium-ion batteries, etc. Physics-based models are commonly employed for SoC estimation to achieve higher accuracy. However, it is challenging due to the need of precise initial SoC. To address this issue, data-driven approach is used to develop SoC prediction model. The effectiveness of data driven models strongly relies on reliable data and optimal model hyperparameters. The major uncertainties influencing data driven model prediction are aleatoric (data) and epistemic (model). In this paper, Long Short-Term Memory (LSTM) model was developed with optimal hyperparameters using Bayesian algorithm to estimate the SoC of the discharging cycle of Turnigy Graphene Li-Ion Battery with specifications of 5000mAh, 65C for LA92 drive cycle at ambient temperature of 25 degrees C. Further the uncertainty analysis was performed against the developed model with battery data:-voltage, current and temperature against ambient temperature variations and model hyperparameters uncertainties. The results show that the LSTM model predicts the SoC with Root Mean Square Error (RMSE) of 0.5033 and R-2 of 0.9994. The robust measures such as RMSE, Mean Absolute Deviation (MAD), and Interquartile Range (IQR) were evaluated against aleatoric uncertainties and epistemic uncertainties. These fruitful results of the proposed work help to develop generalized robust model in future for real-time data with variability.

Accurate estimation of State of Health (SoH) and Internal Resistance (IR) is essential for optimized performance and longevity, particularly in the new-era of electric vehicles and renewable energy storage devices. The traditional methods struggle with battery data complexity in dynamic situations to enhance the above parameters. This paper presents a data-driven model with hybrid architecture that combines CNN and LSTM models which revolutionize the battery management and focusing on the SoH and IR estimation in dynamic environments, CNN-LSTM models excel by extracting spatial features and learning temporal dependencies within the battery datasets. This increases accuracy and enables real-time monitoring, diagnosis, and optimization of battery health. This paper also shows the execution of LSTM with and without sampling of the datasets for efficient runtime and produces accurate results, this marks a significant leap in the battery management and enabling smarter resource utilization for sustainable energy solutions.

Sophisticated control schemes play a crucial role in enhancing the State-of-Health (SOH) of Lithium-ion (Li-ion) batteries by generating an optimal charging profile with operating constraints such as charging time and battery lifespan. However, the development of model-based control strategies with constraints for highly critical systems like battery is challenging due to its complex behavior. In this paper, a hybrid digital twin-based Model Predictive Control (MPC) is proposed to generate an optimal charging profile for a 3.6V, 18Ah LiFePO4battery. The hybrid digital twin is developed by combining physics-based Electro-Thermal Model (ETM) and linear regression model. The optimal control problem is formulated by imposing constraints on input current and batteries internal states and MPC is designed. The demonstrated simulation results show the performance of the proposed control scheme by satisfying the objective function of minimizing charging time and battery temperature to enhance SOH of battery.

This paper aims at design of Enhanced PID (PIDE) controller that eliminates kick based on adaptive fuzzy scheme for the higher order point -kinetic model of the pressurized water reactor (PWR) to assure stable smooth power operations under load following operating conditions, sudden external disturbances and model uncertainties. The dynamics of PWR type of nuclear reactor is significantly changing with power levels, hence fixed weighted multi-modelbased gain scheduled PID controller (FW -GSPID) and adaptive fuzzy based gain scheduled enhanced PID controller (AF-GSPIDE) are proposed with appropriately identified input variables to accommodate the scheduling of controller parameter as function of power level with ability to reject the kicks expected in the control signal. In order to regulate the reactor core power, the reactor model is linearized and the PID controller parameters are optimally tuned for different power levels with these multi-models. Most of the reported works on fuzzy based control schemes uses error and change in error as input variables for the controller design whereas the novelty in the proposed AF-GSPIDE is that the controller parameters K P , K I and K D are tuned based on process variable i.e. neutron density (n r ) and rate of change of process variable (dn r ) to ensure the smooth control under sudden changes in the load disturbances and parameter variations. With proper choice of membership functions and formulation of rule base along with newly identified inputs, the simulation results shows the improved potential of the proposed scheme at different power levels in the presence of uncertainties and external disturbances.

Battery is one of the major components of electric vehicles, which highly influences the performance of Electric Vehicles (EVs). However, enhancing the life of a Lithium-ion (Li-ion) battery is a challenging task because they have a high risk of fire hazards due to the electrochemical properties of Li-ion. Various key factors that have a significant influence on the health of the battery include a number of charge-discharge cycles, temperature, voltage, and current profiles. Hence, a highly efficient Battery Management System (BMS) with an optimal charging facility is needed. The main objective of the proposed work is to generate Multiple Hybrid Artificial Intelligence (MHAI)-based optimal charging current profiles for Li-ion batteries with minimum charging time and temperature rise in order to enhance the State Of Health (SOH). In this work, an open-source dataset of Li-ion-18650 from the National Aeronautics and Space Administration (NASA) was used. Firstly, the battery charging profile range from 0 to 100% is divided into 4 groups (0-25%, 26-50%, 51-75%,76-100%), and four hybrid AI models are developed and validated. to find the optimal charging current in each of the 4 regions instead of using a single AI model for the entire charging profile. For model development, temperature, maximum chargeable capacity, and charging time are considered as outputs, and charging voltage and current are taken as inputs. Long Short Term Memory (LSTM), Random Forest (RF), and Coulomb Counting (White Box Model) are used to develop models to predict temperature, maximum chargeable capacity, and charging time, respectively. Finally, Particle Swarm Optimization (PSO) is used to find the optimal current value for the developed models to minimize both the charging time and temperature rise using the weighted-sum method of the MultiObjective Particle Swarm Optimization (MOPSO) technique. The results show the feasibility of the proposed approach.

A bioreactor is a specialized vessel which has the provision of cell cultivation under a sterile environment and the control of the environmental parameters enhances growth. A variety of products related to the food, beverage, and pharmaceutical industries are produced using bioprocesses. Because of the complex dynamics of bioprocesses, controlling them is a difficult and delicate undertaking. Additionally, because of their high nonlinearity, modelling and parameter estimates are made even more challenging. The difficulty of this endeavour is increased by the dearth of online measurements for the biomass and substrate concentrations. The design of an effective controller for any process requires an efficient model. A combination of more than one type of model in a hybrid form can give a better performance. The first principles model is coupled with the data obtained from the real-time bioreactor setup to yield a hybrid model. The process parameters are estimated utilizing a recurrent neural network approach. The developed hybrid model is tested with the real-time bioreactor response and found to be satisfactory.

This article addresses a multilayer neural network (MNN)-based optimal adaptive tracking of partially uncertain nonlinear discrete-time (DT) systems in affine form. By employing an actor-critic neural network (NN) to approximate the value function and optimal control policy, the critic NN is updated via a novel hybrid learning scheme, where its weights are adjusted once at a sampling instant and also in a finite iterative manner within the instants to enhance the convergence rate. Moreover, to deal with the persistency of excitation (PE) condition, a replay buffer is incorporated into the critic update law through concurrent learning. To address the vanishing gradient issue, the actor and critic MNN weights are tuned using control input and temporal difference errors (TDEs), respectively. In addition, a weight consolidation scheme is incorporated into the critic MNN update law to attain lifelong learning and overcome catastrophic forgetting, thus lowering the cumulative cost. The tracking error, and the actor and critic weight estimation errors are shown to be bounded using the Lyapunov analysis. Simulation results using the proposed approach on a two-link robot manipulator show a significant reduction in tracking error by 44% and cumulative cost by 31% in a multitask environment.

Introduction: Water scarcity and water pollution are two major issues in India. Circular economy-based wastewater treatment technology provides the most sustainable solutions for solving these issues. In this paper, a novel multi-objective decentralized controller (MODC) is proposed for benchmarking a multi-input multi-output (MIMO) activated sludge wastewater treatment plant (WWTP) to achieve maximum effluent quality with minimum cost. WWTPs with conventional control schemes consume more energy to achieve the desired effluent quality. Methods: In this study, a MIMO model is developed for the activated sludge process (ASP) from a physics-based model, and relative gain array (RGA) analysis are carried out to determine the interaction between the loops to identify a suitable control scheme for the MIMO process. In addition, a multi-objective decentralized control problem is formulated to achieve the conflicting multiple objectives of improving effluent quality and minimizing operational costs by efficient usage of energy. Results and discussion: The desired quality and cost reduction are verified by comparing the integral square error (ISE) and control effort (CE) values of a closed-loop WWTP. A multi-objective evolutionary algorithm (MOEA), namely, the non-dominated sorting genetic algorithm (NSGA)-II, successfully solves the multi-objective control problem. NSGA-II provides several optimal solutions in the Pareto front. In order to demonstrate the feasibility of the proposed controller, three optimal solutions are selected from the Pareto-optimal front, and their closed-loop performances are evaluated qualitatively and quantitatively for both servo and regulatory operations. Improving the quality of effluent enhances active sludge production, which in turn increases the methane production in the anaerobic digester.

Industries use soft sensors for estimating output parameters that are difficult to measure on-line. These parameters can be determined by laboratory analysis which is an offline task. Now a days designing Soft sensors for complex nonlinear systems using deep learning training techniques has become popular, because of accuracy and robustness. There is a need to find pertinent hardware for realizing soft sensors to make it portable and can be used in the place of general purpose PC. This paper aims to propose a new strategy for realizing a soft sensor using deep neural networks (DNN) on appropriate hardware which can be referred as embedded soft sensor (ESS). The work focuses on developing an ESS for estimating lactose concentration in a simulated and experimental bioreactor using DNN and realizing it on the Zynq based System on Chip (SoC). Deep neural network is developed for the process with certain number of hidden layers. The model parameters of the process is represented at input layer and lactose concentration is considered at output layer. The performance of the ESS has been observed with the number of hidden layers and different activation functions. Then the optimized neural network is chosen for realizing on hardware. Comparison is made among the values obtained from hardware realization, software simulation and laboratory analysis. Output analysis shows that the values obtained through hardware realization are closer to the values obtained through laboratory analysis. From the results it can be concluded that Deep learning provides a better way, alternative to traditional techniques for realizing ESS on hardware. From the proposed work, it can be shown that if any sensor is unavailable for measuring any parameter then this ESS can be used to measure the values. Since this ESS is realized on reconfigurable hardware like SoC, it can be portable and flexible to measure values.

An online adaptive deep neural network (DNN) scheme has been introduced for the tracking control of a nonlinear bioprocess with uncertain internal dynamics. First, a detailed controllability analysis is conducted for the Lutein bioprocess to represent the bioprocess as a nonlinear system in affine form. Next, a controller consisting of a DNN-based function approximator is designed for the nonlinear Lutein production bioprocess. It is demonstrated that closed-loop tracking control of a bioprocess for a desired yield profile is possible only with two inputs. The set point trajectory to yield maximum Lutein production is shown by the proposed online adaptive deep NN controller. The proposed controller exhibits self-learning capability under closed loop condition, due to the online learning phase. In other words, no explicit offline learning phase is required and online learning is preferred due to lack of a priori training data for approximating complex nonlinear functions. Simulation results are provided to confirm the performance of the proposed approach.

Soft Senors provides an alternative way for measuring the process variables which cannot be measured online. Deep learning training techniques has become popular for designing Soft sensors for complex nonlinear systems because of accuracy and robustness. The work presented in this paper was to design a Soft Sensor to estimate bioprocess variables like lactose and ethanol concentrations in the bioreactor using deep neural networks(DNN).Here an unsupervised learning has been used to pre-train the network and supervised learning was used for training the network. self-organizing Maps(SOM) was used for pre-training the network and error back propagation algorithm was used for training the network. The Design of soft sensor using such Combination of algorithms results in better performance in terms of estimated values and measurement error when compared with other methods estimation. Furthermore, the performance of the soft sensor has been observed with different hidden layers and it was concluded that with three hidden layers the soft sensor gives accurate results when compared to that of the single layer.

An online learning neural network (NN) based adaptive scheme has been proposed for tracking the power level of higher order point kinetic pressurized water reactor (PWR) with unknown dynamics under local and global load following conditions and emergency conditions. The PWR type of nuclear reactors are linear parameter varying (LPV) systems whose parameters vary with power, ageing effects and changes in nuclear core reactivity with fuel burn up. The drastic changes in plant parameters should be accommodated in a reactor control system through on-line identification to ensures safe operation for the power plant which demands for adaptive control system. To track the demand changes which happens during the load following and emergency conditions considered, the PID controller parameters needs to be varied. However, the proposed NN controller is based on feedback linearisation of nonlinear discrete time system with unknown internal dynamics by using a multilayer neural network acting as function approximator. Online weight tuning algorithm based on a modified delta rule and projection algorithm are used for the update of the weights of the proposed neural network controller along with conventional PID controller. Simulation studies with the proposed controller on a non linear PWR core shows that the proposed algorithm exhibits improved performance in terms of lesser ISE/ IAE/ ITAE over using PID controller under load following conditions. This new proposed power level controller has self learning capability and it needs the measurement of all the state variables and the desired trajectory.(c) 2021 Elsevier Ltd. All rights reserved.

Bioreactor plays a significant role in many industries such as pharmaceuticals, food products, etc. as these processes depend on the microorganisms. High biomass yield can generally be achieved by operating the bioreactor in fed-batch mode, with an effective model and a suitable advanced control scheme. Modeling a fed-batch bioreactor is a challenging task due to its nonlinear and dynamic behavior. In this work, a hybrid model is developed based on the experimental data collected from a bioreactor that describes the dynamic behavior of aerobic fed-batch cultures of Escherichia coli (E. coli). The biomass profile obtained from hybrid model with GA based feed profile input is used as the desired set point for the Model Predictive Controller (MPC). The parameters of MPC are tuned using Chicken Swarm Optimization (CSO) algorithm. The controller thus designed to obtain maximum biomass concentration uses a predictive model and dynamically updates the feed profile. The real-time automation strategy developed by the authors using LabVIEW (Laboratory Virtual Instrument Engineering Workbench) platform is capable of controlling the key variables such as temperature, Dissolved Oxygen (DO), pH, and antifoam simultaneously during fermentation. The implementation of this optimally tuned controller with optimal set point profile improves the biomass concentration significantly during the fed-batch operation of the bioreactor.

In this paper, a multi-layer neural network (MNN) based online optimal adaptive regulation of a class of nonlinear discrete-time systems in affine form with uncertain internal dynamics is introduced. The multi-layer neural networks (MNN)-based actor-critic framework is utilized to estimate the optimal control input and cost function. The temporal difference (TD) error is derived from the difference between actual and estimated cost function. The MNN weights of both critic and actor are tuned at every sampling instant as a function of the instantaneous temporal difference and control policy errors. The proposed approach does not require the selection of any basis function and its derivatives. The boundedness of the system state vector and actor and critic NN weights are shown through Lyapunov theory. Extension of the proposed approach to MNNs with more hidden layers is discussed. Simulation results are provided to illustrate the effectiveness of the proposed approach.