• 대한임베디드공학회논문지
• /
• 제19권3호
• /
• 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.

• 대한전기학회:학술대회논문집
• /
• 대한전기학회 1999년도 추계학술대회 논문집 학회본부 B
• /
• pp.524-526
• /
• 1999

In the construction of successful fuzzy models and/or controllers for nonlinear systems, the identification of a good fuzzy inference system is an important yet difficult problem, which is traditionally accomplished by a time-consuming trial-and-error process. In this paper, we propose a systematic identification procedure for complex multi-input single-output nonlinear systems with DNA coding method. A DNA coding method is optimization algorithm based on biological DNA as conventional genetic algorithms(GAs) are. The strings in the DNA coding method are variable-length strings, while standard GAs work with a fixed-length coding scheme. the DNA coding method is well suited to learning because it allows a flexible representation of a fuzzy inference system. We also propose a new coding method fur applying the DNA coding method to the identification of fuzzy models. This coding scheme can effectively represent the zero-order Takagi-Sugeno(TS) fuzzy model. To acquire optimal TS fuzzy model with higher accuracy and economical size, we use the DNA coding method to optimize the parameters and the number of fuzzy inference system. In order to demonstrate the superiority and efficiency of the proposed scheme, we finally show its application to a Duffing-forced oscillation system.

• 한국지능시스템학회:학술대회논문집
• /
• 한국퍼지및지능시스템학회 2000년도 춘계학술대회 학술발표 논문집
• /
• pp.47-50
• /
• 2000

In this paper, an optimal identification method using Multi-FNN(Fuzzy-Neural Network) is proposed for model ins of nonlinear complex system. In order to control of nonlinear process with complexity and uncertainty of data, proposed model use a HCM clustering algorithm which carry out the input-output data preprocessing function and Genetic Algorithm which carry out optimization of model. The proposed Multi-FNN is based on Yamakawa's FNN and it uses simplified inference as fuzzy inference method and Error Back Propagation Algorithm as learning rules. HCM clustering method which carry out the data preprocessing function for system modeling, is utilized to determine the structure of Multi-FNN by means of the divisions of input-output space. Also, the parameters of Multi-FNN model such as apexes of membership function, learning rates and momentum coefficients are adjusted using genetic algorithms. Also, a performance index with a weighting factor is presented to achieve a sound balance between approximation and generalization abilities of the model, To evaluate the performance of the proposed model, we use the time series data for gas furnace and the numerical data of nonlinear function.

• 한국위성정보통신학회논문지
• /
• 제9권4호
• /
• pp.104-110
• /
• 2014

본 논문에서는 무인항공기인 방송용 멀티콥터를 이용한 Full-HD급 이상 화질의 이미지를 효율적으로 전송하기 위해 이미지 압축 센싱 기법을 적용하고, Sparse 신호의 효율적 복원을 위해 Turbo 알고리즘과 Markov chain Monte Carlo (MCMC) 알고리즘의 복원 성능을 모의실험을 통해 비교 분석하였다. 제안된 복원 기법은 압축 센싱에 기반하여 데이터 용량을 줄이고 빠르고 오류 없는 원신호 복원에 중점을 두었다. 다수의 이미지 파일로 모의실험을 진행한 결과 Loopy belief propagation(BP) 기반의 Turbo 복원 알고리즘이 Gibbs sampling기반 알고리즘을 수행하는 MCMC 알고리즘 보다 평균 복원 연산 시간, NMSE 값에서 우수하여 보다 효율적인 복원 방법으로 생각된다.

• 전기학회논문지
• /
• 제61권5호
• /
• pp.744-752
• /
• 2012

In this study, the Polynomial-based Radial Basis Function Neural Networks is proposed as an one of the recognition part of overall face recognition system that consists of two parts such as the preprocessing part and recognition part. The design methodology and procedure of the proposed pRBFNNs are presented to obtain the solution to high-dimensional pattern recognition problems. In data preprocessing part, Principal Component Analysis(PCA) which is generally used in face recognition, which is useful to express some classes using reduction, since it is effective to maintain the rate of recognition and to reduce the amount of data at the same time. However, because of there of the whole face image, it can not guarantee the detection rate about the change of viewpoint and whole image. Thus, to compensate for the defects, Linear Discriminant Analysis(LDA) is used to enhance the separation of different classes. In this paper, we combine the PCA&LDA algorithm and design the optimized pRBFNNs for recognition module. The proposed pRBFNNs architecture consists of three functional modules such as the condition part, the conclusion part, and the inference part as fuzzy rules formed in 'If-then' format. In the condition part of fuzzy rules, input space is partitioned with Fuzzy C-Means clustering. In the conclusion part of rules, the connection weight of pRBFNNs is represented as two kinds of polynomials such as constant, and linear. The coefficients of connection weight identified with back-propagation using gradient descent method. The output of the pRBFNNs model is obtained by fuzzy inference method in the inference part of fuzzy rules. The essential design parameters (including learning rate, momentum coefficient and fuzzification coefficient) of the networks are optimized by means of Differential Evolution. The proposed pRBFNNs are applied to face image(ex Yale, AT&T) datasets and then demonstrated from the viewpoint of the output performance and recognition rate.

• 대한교통학회지
• /
• 제24권2호
• /
• pp.157-168
• /
• 2006

구간검지시스템에서 수집되는 통행시간 정보는 과거 개별차량의 검지기 통과시각(도착시각)을 기준으로 수집되는 특성이 있다. 따라서 구간검지시스템에 의해 수집되는 통행시간 정보는 도착시각기준이 아닌 출발시각기준으로 산출되어야 하고 현재시점(On-Line)에서 통행시간 추정 및 예측이 되어야 한다. 그러나 기존의 구간검지시스템을 이용한통행시간 추정 및 예측 연구들은 도착시각기준으로 수집되는 개별차량 통행시간 자료를 이용함으로 출발시각기준 On-Line 통행시간 정보 관련 연구를 체계적으로 접근하기 어려운 실정이다. 본 연구에서는 On-Line 출발시각기준의 통행시간 정보의 개념을 정립하고 이에 따른 시사점을 도출하였다 그리고 베이지안 추론을 이용하여 고속도로를 대상으로 한 출발시각기준의 On-Line 링크통행시간 추정 알고리즘을 개발하였다. 그 결과, 본 연구에서 개발한 알고리즘은 On-Line 통행시간 정보 질의 정확성과 신속성 측면에서 개선된 통행시간 대표값(평균값)을 추정하는 것으로 나타났다.

• 전기학회논문지
• /
• 제58권2호
• /
• pp.399-406
• /
• 2009

In this study, Polynomial Radial Basis Function Neural Network(pRBFNN) based on Fuzzy Inference System is designed and its parameters such as learning rate, momentum coefficient, and distributed weight (width of RBF) are optimized by means of Particle Swarm Optimization. The proposed model can be expressed as three functional module that consists of condition part, conclusion part, and inference part in the viewpoint of fuzzy rule formed in 'If-then'. In the condition part of pRBFNN as a fuzzy rule, input space is partitioned by defining kernel functions (RBFs). Here, the structure of kernel functions, namely, RBF is generated from HCM clustering algorithm. We use Gaussian type and Inverse multiquadratic type as a RBF. Besides these types of RBF, Conic RBF is also proposed and used as a kernel function. Also, in order to reflect the characteristic of dataset when partitioning input space, we consider the width of RBF defined by standard deviation of dataset. In the conclusion part, the connection weights of pRBFNN are represented as a polynomial which is the extended structure of the general RBF neural network with constant as a connection weights. Finally, the output of model is decided by the fuzzy inference of the inference part of pRBFNN. In order to evaluate the proposed model, nonlinear function with 2 inputs, waster water dataset and gas furnace time series dataset are used and the results of pRBFNN are compared with some previous models. Approximation as well as generalization abilities are discussed with these results.

• 한국정보통신학회논문지
• /
• 제6권7호
• /
• pp.1126-1132
• /
• 2002

In this paper, we propose a new scheme of motion tracking based on fuzzy inference (Fl) and wavelet transform (WT) from image sequences. First, we present a WT to segment a feature extraction of dynamic image . The coefficient matrix for 2-level DWT tent to be clustered around the location of Important features in the images, such as edge discontinuities, peaks, and corners. But these features are time varying owing to the environment conditions. Second, to reduce the spatio-temperal error, We develop a fuzzy inference algorithm. Some experiments are performed 0 testify the validity and applicability of the proposed system As a result, proposed method is relatively simple compared with the traditional space domain method. It is also well suited for motion tracking under the conditions of variation of illumination.

• 한국항해항만학회지
• /
• 제32권7호
• /
• pp.529-535
• /
• 2008

As a method to reduce collision accidents of ships at sea, this paper suggests an expert system for collision avoidance and navigation (hereafter "ESCAN"). The ESCAN is designed and developed by using the theory and technology of expert system and based on the information provided by AIS and RADAR/ARPA system. In this paper the ESCAN is composed of four(4) components; Facts/Data Base in charge of preserving data from navigational equipment, Knowledge Base storing production rules of the ESCAN, Inference Engine deciding which rules are satisfied by facts or objects, User System Interface for communication between users and ESCAN. The ESCAN has the function of real--time analysis and judgment of various encountering situations between own ship and targets, and is to provide navigators with appropriate plans of collision avoidance and additional advice and recommendation This paper, as a basic study, is to introduce the basic design and function of ESCAN.

• 대한전기학회:학술대회논문집
• /
• 대한전기학회 2000년도 하계학술대회 논문집 D
• /
• pp.3004-3006
• /
• 2000

In this paper, we propose an optimal design method of Fuzzy-Neural Networks model with parallel structure for complex and nonlinear systems. The proposed model is consists of a multiple number of FNN connected in parallel. The proposed FNNs with parallel structure is based on Yamakawa's FNN and it uses simplified inference as fuzzy inference method and Error Back Propagation Algorithm as learning rules. We use a HCM clustering and GAs to identify the structure and the parameters of the proposed model. Also, a performance index with a weighting factor is presented to achieve a sound balance between approximation and generalization abilities of the model. To evaluate the performance of the proposed model. we use the time series data for gas furnace and the numerical data of nonlinear function.