EPE Association: EPE 2022 - DS1l: Application of Artificial Intelligence to Power Electronics and Drive Systems

EPE 2022 - DS1l: Application of Artificial Intelligence to Power Electronics and Drive Systems
You are here: EPE Documents > 01 - EPE & EPE ECCE Conference Proceedings > EPE 2022 ECCE Europe - Conference > EPE 2022 - Topic 10: > EPE 2022 - DS1l: Application of Artificial Intelligence to Power Electronics and Drive Systems

   [return to parent folder]

   Deep-Learning fault detection and classification on a UAV propulsion system
By Pierre-Yves BRULIN [View]
[Download]

Abstract: A fault detection and identification method using a Deep-Learning classification method is used to identify several faults that may occur on a UAV propulsion system. Training is performed from a dataset acquired from a simplified multiphysics simulation of the system which allows for the generation of large datasets of modular, interconnected and scalable components of various sizes and performances. We aim to provide a model able to identify faults occurring on a propulsion system using a reduced set of input signals.

   Inductor Design Optimization Using FEA Supervised Machine Learning
By David CAJANDER [View]
[Download]

Abstract: An optimal inductor design methodology using dimensioning models derived from Finite Element Analysis (FEA) supervised Artificial Neural Networks (ANN) is presented. The efficiency of such trained ANN dimensioning models in terms of compromise between precision and computing time is demonstrated for the cylindrical inductor topology with air and magnetic material core including saturation.

   Online Islanding Detection scheme for Grid Connected Distributed Generation Systems
By Mohammed Ali KHAN [View]
[Download]

Abstract: Data aggregation in smart grids is a key component for emergency responses during abnormalities in the grid. To efficiently utilize the aggregated data, and achieve fast identification of these abnormalities, this paper develops an online islanding detection approach. The development of the technique is realized with an online learning algorithm implemented using the large-scale support vector machine (LaSVM). The algorithm adopts a classification problem for islanding detection in grid-connected systems by considering a set of independent variables and unknown variables. The independent variables are related to the known islanding events in the grid-connected system, and the unknown variables are related to the dynamics of the grid operating in real-time. The proposed approach solves this problem by training the known and unknown variables and identifying new instances through sequential minimal optimization. The training and validation results provided indicate 99.8 \% accuracy for islanding detection under standard operating conditions of the grid-connected system.

   SNNFT: Sequential Neural Network-Fuzzy Thermal Early Warning System for Lithium-ion Batteries
By Chaoyu DONG [View]
[Download]

Abstract: Due to the promotion of electric vehicles and new energy sources, lithium-ion batteries have been widely used. However, temperature has a great influence on the performance and safety of lithium-ion batteries during operation. Therefore, it is very important to predict the temperature of lithium-ion batteries and implement thermal early warning. In order to solve this problem, this paper designed a Sequential neural network-fuzzy thermal early warning system (SNNFT). First, the SNNFT uses a denoising autoencoder to eliminate the noise in real-time measurement. Then it combines the long short-term memory network and the temporal convolutional network that can handle the time series problem well to realize the accurate prediction of the lithium-ion battery temperature. And the SNNFT applies interpretable adaptive network-based fuzzy inference system model to build thermal early warning system. Complete experiments are conducted to verify the reliability advantages.

EPE 2022 - DS1l: Application of Artificial Intelligence to Power Electronics and Drive Systems
You are here: EPE Documents > 01 - EPE & EPE ECCE Conference Proceedings > EPE 2022 ECCE Europe - Conference > EPE 2022 - Topic 10: > EPE 2022 - DS1l: Application of Artificial Intelligence to Power Electronics and Drive Systems
	[return to parent folder]

	Deep-Learning fault detection and classification on a UAV propulsion system By Pierre-Yves BRULIN	[View] [Download]
Abstract: A fault detection and identification method using a Deep-Learning classification method is used to identify several faults that may occur on a UAV propulsion system. Training is performed from a dataset acquired from a simplified multiphysics simulation of the system which allows for the generation of large datasets of modular, interconnected and scalable components of various sizes and performances. We aim to provide a model able to identify faults occurring on a propulsion system using a reduced set of input signals.

	Inductor Design Optimization Using FEA Supervised Machine Learning By David CAJANDER	[View] [Download]
Abstract: An optimal inductor design methodology using dimensioning models derived from Finite Element Analysis (FEA) supervised Artificial Neural Networks (ANN) is presented. The efficiency of such trained ANN dimensioning models in terms of compromise between precision and computing time is demonstrated for the cylindrical inductor topology with air and magnetic material core including saturation.

	Online Islanding Detection scheme for Grid Connected Distributed Generation Systems By Mohammed Ali KHAN	[View] [Download]
Abstract: Data aggregation in smart grids is a key component for emergency responses during abnormalities in the grid. To efficiently utilize the aggregated data, and achieve fast identification of these abnormalities, this paper develops an online islanding detection approach. The development of the technique is realized with an online learning algorithm implemented using the large-scale support vector machine (LaSVM). The algorithm adopts a classification problem for islanding detection in grid-connected systems by considering a set of independent variables and unknown variables. The independent variables are related to the known islanding events in the grid-connected system, and the unknown variables are related to the dynamics of the grid operating in real-time. The proposed approach solves this problem by training the known and unknown variables and identifying new instances through sequential minimal optimization. The training and validation results provided indicate 99.8 \% accuracy for islanding detection under standard operating conditions of the grid-connected system.

	SNNFT: Sequential Neural Network-Fuzzy Thermal Early Warning System for Lithium-ion Batteries By Chaoyu DONG	[View] [Download]
Abstract: Due to the promotion of electric vehicles and new energy sources, lithium-ion batteries have been widely used. However, temperature has a great influence on the performance and safety of lithium-ion batteries during operation. Therefore, it is very important to predict the temperature of lithium-ion batteries and implement thermal early warning. In order to solve this problem, this paper designed a Sequential neural network-fuzzy thermal early warning system (SNNFT). First, the SNNFT uses a denoising autoencoder to eliminate the noise in real-time measurement. Then it combines the long short-term memory network and the temporal convolutional network that can handle the time series problem well to realize the accurate prediction of the lithium-ion battery temperature. And the SNNFT applies interpretable adaptive network-based fuzzy inference system model to build thermal early warning system. Complete experiments are conducted to verify the reliability advantages.