회전 기계 고장 진단을 위한 적대적 순환 일관성 유지 학습기반 교차 도메인 적응 방법

장계봉,조성배 한국정보과학회 2022 정보과학회논문지 Vol.49 No.7

Research on data-based fault diagnosis models is being actively conducted in various industries. However, in the case of industrial equipment, various operating conditions occur, and it is difficult to secure sufficient training data. To solve this problem, a cross-domain adaptation technique can be utilized. In this study, we propose an adversarial consistency-maintaining transformation learning method that can maintain failure classification consistency even for the new untrained environmental data using the rotating body vibration data. The data generated through consistent learning creates a continuous invariant latent space between the new operating condition data distribution and the known data distribution and learns to maintain the failure classification performance through an adversarial learning network that shares the failure classification characteristic nformation. Therefore, the proposed method can provide a more stable and general classification performance by expanding the potential space to minimize the discrepancy between domain data. The experimental results of the proposed model showed about 88% accuracy for a real-machine dataset, and compared to the existing cross-domain adaptive learning methods, it showed a performance improvement of about 5-10%. According to the results of this study, it is expected to be an effective solution for the problem of equipment failure diagnosis at actual industrial sites. 다양한 산업분야에서 데이터 기반의 고장 진단 모델에 대한 연구가 활발히 이루어지고 있다. 그러나 산업 장비의 경우 다양한 운행 조건이 발생하고, 이에 따른 충분한 훈련 데이터 확보가 어렵다는 문제가 있다. 이러한 문제를 해결하기 위해 교차-도메인 적응 기법을 활용할 수 있다. 본 연구에서는 회전체 진동 데이터를 사용하여 학습되지 않은 새로운 환경 데이터에 대해서도 고장 분류 일관성을 유지할 수 있는 적대적 일관성 유지 변환 학습 방법을 제안한다. 일관성 유지 학습을 통해 생성된 데이터는 새로운 운행 조건 데이터 분포와 이미 알고 있는 데이터 분포 간의 연속적인 불변 잠재 공간을 생성하고, 고장 분류 특징 정보를 공유하는 적대적 학습 네트워크를 통해 고장 분류 성능을 유지하도록 학습한다. 따라서 제안된 방법은 도메인 데이터 간의 불일치를 최소화할 수 있는 잠재적 공간을 넓힘으로써 보다 안정적이고 일반적인 분류 성능을 확보할 수 있다. 제안한 모델의 실험결과는 약 88%내외의 성능 평가가 이루어졌으며, 기존 교차-도메인 적응 학습 기법들과 비교하여, 약 5~10%의 성능향상을 보였다. 이러한 연구 결과로 실제 산업현장에서 겪는 장비 고장 진단 문제의 효과적인 해결방안이 될 것으로 기대된다.

CO₂레이저 조사가 치아경조직에 미치는 영향에 관한 EPMA분석 및 주사전자현미경적 연구

장계봉,이정석,이찬영,이승종,Chang, Gye-Bong,Lee, Chung-Suck,Lee, Chan-Young,Lee, Seung-Jong 대한치과보존학회 1986 Restorative Dentistry & Endodontics Vol.12 No.1

The purpose of this study is to obtain the data of prestep in cavity preparation by observing changed conditions of laser irradiated dental hard tissues. The forty five extracted caries free human molars were divided into three groups and each group into five subgroups. A $CO_2$ laser irradiation was performed each subgroup of group I for one second with output power of 5 W, 10 Wand 20 W. On group II, laser irradiation was done ten times for 0.1 second duration using same power ratings as group I. On group III, laser irradiation was done 0.1 second, 0.2 second and 0.4 second with output power of 50 W. We investigated mineral contents and crater wall of obtained specimens, i.e., laser irradiated teeth, using computer controlled electron probe microanalyzer and scanning electron microscope. The following results were obtained: 1. Both calcium and phosphorus contents in laser irradiated enamel crater wall were increased, and magnesium content was decreased, but these trends were not statistically significant. 2. In laser irradiated dentin, change of mineral content was more significant. 3. In laser irradiated enamel and dentin, there were no significant differences on mineral content change due to irradiation condition and energy density. 4. In scanning electron microscopic study, enamel rods of the crater wall were fused and clefts were observed in parallel with the direction of enamel rod for all groups. 5. In laser irradiated dentin, irregular fusion and clefts were observed. In irradiated teeth with high power and short duration, the locally formed narrow crater wall was observed.

CO₂레이저 조사가 치아경조직에 미치는 영향에 관한 EPMA분석 및 주사전자현미경적 연구

장계봉,이정석,이찬영,이승종 大韓齒科保存學會 1986 Restorative Dentistry & Endodontics Vol.12 No.1

The purpose of this study is to obtain the data of prestep in cavity preparation by observing changed conditions of laser irradiated dental hard tissues. The forty five extracted caries free human molars were divided into three groups and each group into five subgroups. A CO_2 laser irradiation was performed each subgroup of group I for one second with output power of 5 W, 10 W and 20 W. On group Ⅱ, laser irradiation was done ten times for 0.1 second, duration using same power ratings as group Ⅰ. On group Ⅲ, laser irradiation was done 0.1 second, 0.2 second and 0.4 second with output power of 50 W. We investigated mineral contents and crater wall of obtained specimens, i.e., laser irradiated teeth, using computer controlled electron probe microanalyzer and scanning electron microscope. The following results were obtained: 1. Both calcium and phosphorus contents in laser irradiated enamel crater wall were increased, and magnesium content was decreased, but these trends were not statistically significant. 2. In laser irradiated dentin, change of mineral content was more significant. 3. In laser irradiated enamel and dentin, there were no significant differences on mineral content change due to irradiation condition and energy density. 4. In scanning electron microscopic study, enamel rods of the crater wall were fused and clefts were observed in parallel with the direction of enamel rod for all groups. 5. In laser irradiated dentin, irregular fusion and clefts were observed, In irradiated teeth with high power and short duration, the locally formed narrow crater wall was observed.

[지정토론] 치과의사전문의제도 개정(안)에 대한 고려사항

장계봉 대한치과의사협회 2003 대한치과의사협회지 Vol.41 No.2

치과용 아말감의 부식(腐蝕)에 관한 전기화학적 연구

張啓鳳 大韓齒科保存學會 1980 Restorative Dentistry & Endodontics Vol.6 No.1

ABSTRACTThe corrosion of silver amalgam is regarded as one of major causes in the failures of' dental amalgam restorations. To evaluate the corrosion resistance of dental amalgam alloys, electrochemical tests such as potential and polarization measurement were used widely. But these commonly used methods have not provided the sufficient informations on relative resistance of amalgam to corrosion.In this experiment, the corrosion currents were measured using electronic potentiostat to compare some commercial dental amalgam alloys. All alloys were triturated in a amalgamator and condensed into a mold described in A.D.A.specifiction No.1 to produce cylinder form specimens of 4mm diameter by 5mm long. After specimen kept for 1 week at 37"C, each specimen was embedded in epoxy resin. The surfaces of specililens were then polished luith a emery paper, diamond dust, and Al203. These specimens were immersed in artifical saliva kept at 37"C, and currents of each specimen were measured for 24 hours at 0. 0volt (SCE).The author obtained conclusions as follows : 1. High copper amalgam showed superior resistance against corrosion to conventional amalgam, but a pellet form of high copper amalgam seemed to be susceptible to corrosion. 2. In lathe-cut alloys, fine·cut had superior resistance against corrosion to regular-cut. 3. Non-zinc conventional amalgam alloys were more resistant to corrosion than that of zinc containing conventional amalgam alloys. 4. In both of high copper and conventional amalgams, predispensed forms tended to have better resisitance to corrosion than that of pellet forms.

C-LSTM 딥러닝기반 굴삭기 작업동작 분류모델

장계봉(Gyebong Jang),조성배(Sung-Bae Cho) 한국정보과학회 2019 한국정보과학회 학술발표논문집 Vol.2019 No.6

회전 기계 고장 진단을 위한 비지도 적대적 일관성 유지 변환 학습을 적용한 교차 도메인 적응 방법

장계봉(Gye-Bong Jang),조성배(Sung-Bae Cho) 한국정보과학회 2021 한국정보과학회 학술발표논문집 Vol.2021 No.6

제한된 데이터의 회전체 결함 진단을 위한 클래스 균형 샘플링기반 소수-샷 학습방법

김하율(Hayoul Kim),장계봉(Gyebong Jang),조성배(Sung-Bae Cho) 한국정보과학회 2021 한국정보과학회 학술발표논문집 Vol.2021 No.6

운전자 편향을 완화한 딥 러닝 기반 굴착기 부하 측정 시스템

박경원(Kyoung-Won Park),박용진(Yong-Jin Park),장계봉(Gye-Bong Jang) 한국자동차공학회 2023 한국자동차공학회 부문종합 학술대회 Vol.2023 No.5

Measuring the weight of excavation work of an excavator on a construction site is critical for both budget reduction and accurate progress tracking of construction projects. Typically, the existing weighing systems use sensors to measure the volume of work. However, precise measurement is difficult due to sensor errors or noise, and manual adjustments are necessary to correct minor errors. In addition, the sensors collected from different equipment and operators have different distributions, making it difficult to capture information for measuring weights. To overcome these limitations, this paper proposes a CNNLSTM-based deep learning algorithm that is effective for noisy data and can learn multivariate time series patterns. Additionally, it proposes a domain adaptation learning method that is robust to driver bias information. Through the proposed domain adaptation method using 22 sensors collected from two excavator operators, it is verified that weight measurement is possible without bias to the driver domain.