An Adaptive Machine Learning-based Velocity Prediction Control for Hybrid Electric Vehicles

Chigozie Uzochukwu Udeogu(치고지),Wansu Lim(임완수) 한국통신학회 2022 한국통신학회 학술대회논문집 Vol.2022 No.6

Performance comparison of deep learning model-based channel estimation and signal detection for OFDM systems

Chigozie Uzochukwu Udeogu,Angela Caliwag,Wansu Lim 한국통신학회 2022 한국통신학회 학술대회논문집 Vol.2022 No.2

Channel estimation and signal detection play key roles in ensuring the quality of end-to-end communication in orthogonal frequency-division multiplexing (OFDM) systems. Recently, many deep learning (DL) model-based estimation and detection approaches are being researched. These models have their advantages and disadvantages in channel estimation and signal detection for OFDM systems. To further open systematic research for real world applicability of DL in this area, this paper provides quantitative results of various DL models to compare both performance and reliability of these models to handle OFDM channels. Furthermore, simulation results show that DL scheme outperforms existing conventional schemes in terms of improving channel estimation and signal detection performance.

Remaining Useful Life Prediction for Supercapacitors Using an Optimized End-to-End Deep Learning Approach

Chigozie Uzochukwu Udeogu,안젤라,임완수 한국통신학회 2022 韓國通信學會論文誌 Vol.47 No.3

Remaining useful life (RUL) prediction for supercapacitors is particularly important to ensure the safety of the applied system and reduce the cost of operation. The existing RUL prediction method utilized health indicators (HIs) that are extracted by a conventional method. This method has the risk of dropping useful information in the supercapacitor data which leads to low accuracy because of poor quality features. To resolve this issue, this paper proposes an optimized end-to-end deep learning model for RUL prediction. Specifically, a genetic algorithm (GA) for automatic feature selection and long short-term memory (LSTM) network (GA-LSTM) for RUL prediction. GA is utilized for automatic feature extraction which ensures all important information in the supercapacitor data is considered during HI extraction. The combination of the best-selected features is used as the input to the LSTM model for final RUL prediction. Our proposed model achieved a root mean square error (RMSE) of 0.03 unlike the recurrent neural network, LSTM, and deep convolutional neural network with RMSE of 23.87, 0.51, and 0.38, respectively. When compared with other models, the overall results show that our model exhibits excellent performance for the RUL prediction of supercapacitors.

Synthetic Data Generation Using GAN for RUL Prediction of Supercapacitors

Miracle Udurume,Chigozie Uzochukwu Udeogu,안젤라,임완수 한국통신학회 2022 韓國通信學會論文誌 Vol.47 No.3

The remaining useful life (RUL) prediction of supercapacitors is an important part of supercapacitors management system. To accurately predict the RUL of supercapacitor, a large amount of capacity data is required which can be difficult to acquire due to privacy restrictions and limited access. Previous works have employed the use of deep learning models to synthetically generate data. However, a prerequisite ensuring the success of these models depends on their ability to preserve the temporal dynamics of the data. This paper presents a generative adversarial network (GAN) for synthetic data generation and a long short-term memory (LSTM) network for accurate RUL prediction. Firstly, the GAN model is employed for synthetic data generation and LSTM for RUL prediction. We show that the GAN model is capable of preserving the temporal dynamics of the original data and also prove that the generated data can be used to accurately carry out RUL prediction. Our proposed GAN model was able to achieve an accuracy of 85% after 500 epochs. The performance of the generated data set with the LSTM model achieved an RMSE of 0.29. The overall results show that synthetic data can be used to achieve excellent performance for RUL prediction.

Generative Adversarial Network with Face Alignment for Face Generation

Adib Kamali,Udurume Miracle,Udeogu Chigozie Uzochukwu,Angela Caliwag,Wansu Lim 한국통신학회 2021 한국통신학회 학술대회논문집 Vol.2021 No.11

Face generation is extensively conducted to increase the number of face images dataset. In face generation field, Generative Adversarial Network (GAN) have shown remarkable success in face image generations. However, most of the existing methods only generate face images from random noise, and cannot generate faces according to face alignment. This make GAN produce poor quality face images when using unaligned face image. In this paper, we propose face generation based on GAN with considering the face alignments. In detail, original face images which is not always aligned is fed to the face alignment module. Then the aligned face images is added noises. The aligned images with noise are then used as input for GAN based image generator. Generator and discriminator are trained to optimize the face generation model performance. Based on extensive experimental study, we present the analysis on face alignment and face generation result with and without considering face alignment.

Edge AI-based Brain-Computer Interface for Real-time Applications

Henar Mike Canilang,Chigozie Uzochukwu Udeogu,James Rigor Camacho,Erick Valverde,Angela Caliwag,Wansu Lim 대한인간공학회 2021 대한인간공학회 학술대회논문집 Vol.2021 No.11

Objective: This study aims to integrate brain computer interface (BCI) to edge AI devices for real-time EEG signal processing applications. For the specific implementation in this paper, we applied edge AI device-based EEG signal processing for emotion recognition. Background: The emergence of Electroencephalogram (EEG) based applications for intelligent applications is projected to have rapid advancements in the future. The BCI system enables efficient brain signal acquisition. Current intelligent convergence of EEG based applications includes brain signal processing integrated to deep learning models. It is expected that this convergence in intelligent EEG based applications will push through to on-device local processing such as edge AI devices for portability in state-of-the-art applications. The portability and practical usage of these systems in real-world applications could lead to the development and deployment of many other advanced embedded systems for EEG-based applications. Systems that can run locally on the edge without needing to be connected to a mobile network. Edge AI devices are the leading-edge computing platforms that process data locally to overcome the current constraints of IoT application. This paves way to the integration of edge-based processing as the computing paradigm to process and acquire EEG signals. Owing to the current research advancement for both EEG and edge applications, this paper aims to propose one of the many systematic applications of deploying edge-based EEG using a brain computer interface. Method: The input for this edge-based EEG signal processing is through the BCI interfaced to the edge AI device. The edge AI device deployed with a deep learning model for specific applications locally processes the acquired signal. These acquired signals are valuable for training deep learning models to realize practical applications at the edge. The processed EEG signals enable the system response of the system such as rapid emotion recognition. Results: Varying EEG signals were acquired in each of the BCI channels. These brain signals are segmented to different brain signal clusters such as Gamma waves (30㎐ to 100㎐), Beta waves (12㎐ – 30㎐), Alpha waves (7.5㎐ – 12㎐), Theta waves (4㎐-7.5㎐) and Delta waves (0.1㎐-4㎐) which have specific brain wave description. As for EEG emotion recognition applications, these wave signals are essential for efficient and accurate emotion recognition. The alpha, beta, and gamma waves are identified to be the most discriminative frequency ranges to identify emotion. Each of the EEG signal is classified for emotion recognition and identification such as 1) valence, 2) dominance, 3) arousal and 4) liking. High and low responses from these wave signals have corresponding positive, neutral, and negative emotions based on their neural patterns at parietal and occipital sites. Other applications can use the acquired EEG signals thus maximizing the possible application of edge-based EEG signal processing. Conclusion: The local processing of the EEG signal at the edge enables the edge-based EEG system application thus enabling system response and actuation. Edge EEG also enables local and cloud co-processing whereas this maximizes the benefits of the edge computing paradigm. With this co-processing capability, it enables an adaptive and portable real-time EEG signal processing which is a constraint to conventional EEG based emotion recognition system. Application: EEG is a physiological based emotion recognition which proves to be more accurate than conventional non-physiological emotion recognition. Also, with an edge-based EEG application, it enables portability and flexibility in terms of its deployment. This application aims to be a state-of-the-art innovation to existing physiological and non-physiological emotion recognition. Furthermore, this research paper implementation aims to emphasize the vast possible applications of edge-based EEG signal processing to bridge