• IEMEK Journal of Embedded Systems and Applications
• /
• v.19 no.3
• /
• pp.151-158
• /
• 2024

Although on-device artificial intelligence (AI) has gained attention to diagnosing machine faults in real time, most previous studies did not consider the model retraining and redeployment processes that must be performed in real-world industrial environments. Our study addresses this challenge by proposing an on-device AI-based real-time machine fault diagnosis system that utilizes continual learning. Our proposed system includes a lightweight convolutional neural network (CNN) model, a continual learning algorithm, and a real-time monitoring service. First, we developed a lightweight 1D CNN model to reduce the cost of model deployment and enable real-time inference on the target edge device with limited computing resources. We then compared the performance of five continual learning algorithms with three public bearing fault datasets and selected the most effective algorithm for our system. Finally, we implemented a real-time monitoring service using an open-source data visualization framework. In the performance comparison results between continual learning algorithms, we found that the replay-based algorithms outperformed the regularization-based algorithms, and the experience replay (ER) algorithm had the best diagnostic accuracy. We further tuned the number and length of data samples used for a memory buffer of the ER algorithm to maximize its performance. We confirmed that the performance of the ER algorithm becomes higher when a longer data length is used. Consequently, the proposed system showed an accuracy of 98.7%, while only 16.5% of the previous data was stored in memory buffer. Our lightweight CNN model was also able to diagnose a fault type of one data sample within 3.76 ms on the Raspberry Pi 4B device.

• Proceedings of the Korean Society of Precision Engineering Conference
• /
• 1996.11a
• /
• pp.439-443
• /
• 1996

A fuzzy teaming controller is experimentally designed to control the ball k beam system in this paper. Although most fuzzy controllers have been built just to emulate human decision-making behavior, it is necessary to construct the rule bases by using a learning method with self-improvement when it is difficult or impossible to get them only by expert's experience. The algorithm introduces a reference model to generate a desired output and minimizes a performance index function based on the error and error-rate using the gradient-decent method. In our balancing experiment of the ball ＆ beam system, this paper shows that the fuzzy control rules by learning are superior to the expert's experience.

• Journal of Korea Multimedia Society
• /
• v.21 no.10
• /
• pp.1182-1194
• /
• 2018

The Dynamic Adaptive Streaming over HTTP(DASH) is envisioned to evolve to meet an increasing demand on providing seamless video streaming services in the near future. The DASH performance heavily depends on the client's adaptive quality selection algorithm that is not included in the standard. The existing conventional algorithms are basically based on a procedural algorithm that is not easy to capture and reflect all variations of dynamic network and traffic conditions in a variety of network environments. To solve this problem, this paper proposes a novel quality selection mechanism based on the Deep Q-Network(DQN) model, the DQN-based DASH Adaptive Bitrate(ABR) mechanism. The proposed mechanism adopts a new reward calculation method based on five major performance metrics to reflect the current conditions of networks and devices in real time. In addition, the size of the consecutive video segment to be downloaded is also considered as a major learning metric to reflect a variety of video encodings. Experimental results show that the proposed mechanism quickly selects a suitable video quality even in high error rate environments, significantly reducing frequency of quality changes compared to the existing algorithm and simultaneously improving average video quality during video playback.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.19 no.2
• /
• pp.417-426
• /
• 2024

In general, the implementation of machine learning requires prior knowledge and experience with deep learning models, and substantial computational resources and time are necessary for data processing. As a result, machine learning encounters several limitations when deployed on embedded processors. To address these challenges, this paper introduces a novel approach where a genetic algorithm is applied to the convolution operation within the machine learning process, specifically for performing a selective convolution operation.In the selective convolution operation, the convolution is executed exclusively on pixels identified by a genetic algorithm. This method selects and computes pixels based on a ratio determined by the genetic algorithm, effectively reducing the computational workload by the specified ratio. The paper thoroughly explores the integration of genetic algorithms into machine learning computations, monitoring the fitness of each generation to ascertain if it reaches the target value. This approach is then compared with the computational requirements of existing methods.The learning process involves iteratively training generations to ensure that the fitness adequately converges.

• Convergence Security Journal
• /
• v.8 no.4
• /
• pp.81-89
• /
• 2008

This paper introduces u-learning service model based on context awareness. Also, it concentrates on agent-based WPAN technology, OSGi based middleware design, and the application mechanism such as context manager/profile manager provided by agents/server. Especially, we'll introduce the meta structure and its management algorithm, which can be updated with learning experience dynamically. So, we can provide learner with personalized profile and dynamic context for seamless learning service. The OSGi middleware is applied to our meta structure as a conceptual infrastructure.

• Journal of KIISE:Software and Applications
• /
• v.31 no.11
• /
• pp.1431-1438
• /
• 2004

Reinforcement learning learns policies for accomplishing a task's goal by experience through interaction between agent and environment. Q-learning, basis algorithm of reinforcement learning, has the problem of curse of dimensionality and slow learning speed in the incipient stage of learning. In order to solve the problems of Q-learning, new function approximation methods suitable for reinforcement learning should be studied. In this paper, to improve these problems, we suggest Fuzzy Q-Map algorithm that is based on online fuzzy clustering. Fuzzy Q-Map is a function approximation method suitable to reinforcement learning that can do on-line teaming and express uncertainty of environment. We made an experiment on the mountain car problem with fuzzy Q-Map, and its results show that learning speed is accelerated in the incipient stage of learning.

• Journal of The Korean Association of Information Education
• /
• v.12 no.2
• /
• pp.195-202
• /
• 2008

The programming learning by using a robot may be one of the most appropriate learning methods for enabling students to experience the creative learning of future society by avoiding the existing stereotyped style educational environment, and understand and improve algorithm which is the basic fundamental of mathematics and science. This study proposed four types of items of errors which may occur during robot programming by elementary school students, and made elementary school students in the fifth and sixth grades learn robot programming after developing the curriculum for the robot programming. Then, the study collected and classified errors that had occurred during the process of learning, and conducted a comparative analysis of computer-based programming language which had been previously studied. This study identified that robot programming in elementary school was shown superior to existing computer-based programming language as a creative learning method and tool through the field experience.

• Journal of Korea Multimedia Society
• /
• v.21 no.2
• /
• pp.188-198
• /
• 2018

With recent development of high-speed wide-area wireless networks and wide spread of highperformance wireless devices, the demand on seamless video streaming services in Long Term Evolution (LTE) network environments is ever increasing. To meet the demand and provide enhanced Quality of Experience (QoE) with mobile users, the Dynamic Adaptive Streaming over HTTP (DASH) has been actively studied to achieve QoE enhanced video streaming service in dynamic network environments. However, the existing DASH algorithm to select the quality of requesting video segments is based on a procedural algorithm so that it reveals a limitation to adapt its performance to dynamic network situations. To overcome this limitation this paper proposes a novel quality selection mechanism based on a Deep Q-Network (DQN) model, the DQN-based DASH ABR($DQN_{ABR}$) mechanism. The $DQN_{ABR}$ mechanism replaces the existing DASH ABR algorithm with an intelligent deep learning model which optimizes service quality to mobile users through reinforcement learning. Compared to the existing approaches, the experimental analysis shows that the proposed solution outperforms in terms of adapting to dynamic wireless network situations and improving QoE experience of end users.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.1
• /
• pp.334-349
• /
• 2022

The effectiveness of Wi-Fi networks is greatly influenced by the optimization of contention window (CW) parameters. Unfortunately, the conventional approach employed by IEEE 802.11 wireless networks is not scalable enough to sustain consistent performance for the increasing number of stations. Yet, it is still the default when accessing channels for single-users of 802.11 transmissions. Recently, there has been a spike in attempts to enhance network performance using a machine learning (ML) technique known as reinforcement learning (RL). Its advantage is interacting with the surrounding environment and making decisions based on its own experience. Deep RL (DRL) uses deep neural networks (DNN) to deal with more complex environments (such as continuous state spaces or actions spaces) and to get optimum rewards. As a result, we present a new approach of CW control mechanism, which is termed as contention window threshold (CW_Threshold). It uses the DRL principle to define the threshold value and learn optimal settings under various network scenarios. We demonstrate our proposed method, known as a smart exponential-threshold-linear backoff algorithm with a deep Q-learning network (SETL-DQN). The simulation results show that our proposed SETL-DQN algorithm can effectively improve the throughput and reduce the collision rates.

• Korean Journal of Applied Biomechanics
• /
• v.32 no.1
• /
• pp.1-8
• /
• 2022

Objective: This study proposes a methodology to analyze important variables that have a significant impact on the putting direction prediction using a machine learning-based putting direction prediction model trained with IMU sensor data. Method: Putting data were collected using an IMU sensor measuring 12 variables from 6 adult males in their 20s at K University who had no golf experience. The data was preprocessed so that it could be applied to machine learning, and a model was built using five machine learning algorithms. Finally, by comparing the performance of the built models, the model with the highest performance was selected as the proposed model, and then 12 variables of the IMU sensor were applied one by one to analyze important variables affecting the learning performance. Results: As a result of comparing the performance of five machine learning algorithms (K-NN, Naive Bayes, Decision Tree, Random Forest, and Light GBM), the prediction accuracy of the Light GBM-based prediction model was higher than that of other algorithms. Using the Light GBM algorithm, which had excellent performance, an experiment was performed to rank the importance of variables that affect the direction prediction of the model. Conclusion: Among the five machine learning algorithms, the algorithm that best predicts the putting direction was the Light GBM algorithm. When the model predicted the putting direction, the variable that had the greatest influence was the left-right inclination (Roll).