밸브의 내부 누설 현상은 밸브의 내부 부품의 손상에 의해 발생하며 배관 시스템의 사고와 운전정지를 일으키는 주요 요인이 다. 본 연구는 버터플라이형 밸브의 내부 누설에 따라 배관계에서 발생하는 음향방출 신호를 이용하여 배관 가동 중 실시간 누설 진단의 가능성을 검토하였다. 이를 위해 밸브의 작동 모드별로 측정한 시간영역의 AE 원시신호를 취득하였으며 이로부터 구축한 데이터셋은 데 이터 기반의 인공지능 알고리즘에 적용하여 밸브의 내부 누설 유무를 진단하는 모델을 생성하였다. 누설 유무진단을 분류의 문제로 정의 하여 SVM 기반의 머신러닝과 CNN 기반의 딥러닝 분류 알고리즘을 적용하였다. 데이터의 특징 추출에 기반한 SVM 분류 모델의 경우, 이 진분류 모델에서 구축된 모델에 따라 83~90%의 정확도를 나타냈으며, 다중 클래스인 경우 분류 정확도가 66%로 감소하였다. 반면, CNN 기반의 다중 클래스 분류 모델의 경우 99.85%의 분류 정확도를 얻을 수 있었다. 결론적으로 밸브 내부 누설 진단을 위한 SVM 분류모델은 다중 클래스의 정확도 향상을 위해 적절한 특징 추출이 필요하며, CNN 기반의 분류모델은 프로세서의 성능 저하만 없다면 누설진단과 밸브 개도 분류에 효율적인 접근방법임을 확인하였다.

Commercial split beam echosounder (ES70) installed on a krill fishing vessel was calibrated in order to utilize it in estimating biomass of Antarctic krill (Euphausia superba). The method of calibration was to analyze the difference between the bottom backscattering strength of the commercial split beam echosounder (i.e. ES70) and the scientific echosounder (i.e. EK60) at one of transects near South Shetland Islands designated by CCAMLR. 38 kHz and 120 kHz were used for the calibration, and krill swarm signal levels obtained from multi frequencies, was examined to verify the calibration result. The analysis result indicated possibility of calibration by bottom backscattering strength, since the proportion of krill swarm signals within 2 dB < SV 120 kHz-38 kHz < 12 dB (i.e. a common SV 120 kHz-38 kHz range of 38 kHz and 120 kHz to be an indicator of Antarctic krill) over the total acoustic signals were 26.95% and 92.04%, respectively before and after the calibration.

The hydro–acoustic technology has been widely used in not only South Korea but also many foreign countries for various scientific purposes. Unfortunately acoustic data especially collected from field surveys may contain noises caused by a variety of sources. Therefore, it is exceedingly important to eliminate noises when acoustic data are analyzed to derive quantitative results. This study introduced two methods for eliminating noises easily and effectively using post–processing software. Used acoustic data were collected on the Jinhae bay and Tongyeong coast of the South Sea in April 2015. The first method, that is the Wang’s method, placed emphasis on ‘erosion filter’ to eliminate only data samples contaminated by noises. The second method (Yamandu’s method) focused on the ‘resample by number of pings’ to remove pings contained noises. To substantiate the effectiveness of two methods, the mean Sv (Volume backscattering strength), mean height and depth of the fish schools detected were compared between before and after using the noise elimination methods. In the Wang’s method the mean Sv was decreased from –52.4 dB to –52.9 dB, and in the Yamandu’s method from –52.6 dB to –53.3 dB, indicating that noises were successfully eliminated. The mean height (1.5 m) and depth (19.0 m) were same between before and after using two methods showing that the shapes of fish schools were not changed.

To estimate weld quality of the resistance spot-welding, the acoustic emission features are investigated from the total acoustic emission signal at the single-spot weld. Typically, the resistance spot welding process consists of several stages: set-down of the electrodes, squeeze, current flow, forging, hold time, and lift-off. Various types of acoustic emission response corresponding to each stage can be separately analyzed by using back-propagation neural network classifier and wavelet transform technique. The presented machine learning results provide a validation for using back-propagation neural network and wavelet transform technique as a valuable insights into the resistance spot-welding process. Especially, a wavelet transform technique is demonstrated and the plots are very powerful in the recognition of the acoustic emission features

This study investigated tensile deformation of the stress aging heat-treated SM45C steel which are aging temperature at 250℃ and 300℃; aging time at 1 hour and, 3 hours; applied load at 300N and 400N by using an acoustic emission techniques (AEs). A signal processing technology is applied to evaluate an AE source characterization of different AE measurement systems DiSP & PCI-2. In this study, most suitable aging condition appeared at 250℃, 3 hours and 300N. But in cases of 250℃, 3 hours, 400N and 300℃, 3 hours, 400N conditions, yield load decreased compare to other conditions according to the over-aging phenomena. On the other hand, when arranged via AE amplitude results by K-means clustering pattern recognition of AE raw signals, tendency of signal strength appeared non-heat treatment condition, 'Class 1 < Class 2 < Class 3'; optimal condition, 'Class 3 < Class 2'; over-aging condition, 'Class 3 < Class 2 < Class 1'. This is judged by emitting a lot of AE energy when material causes plastic deformation because ductility increases on factor by over-aging phenomenon.

유리 원형 평판에서 힘의 세기가 1 dyne이고 면에 수직하게 작용하는 Heaviside계단 함수의 시간 의존성을 갖는 점 하중에 의한 진앙점에서 수직 변위를 이론적으로 계산하였다. 연필심 파괴시 방출되는 음향방출신호를 안정화회로가 부착된 Michelson 간섭계로 측정하여, 음향방출 발생원함수를 deconvolution방법을 이용하여 해석하였다. 연필심 파괴시 방출되는 음향방출 발생원을 파 전면에 약 0.7μsec의 지속시간이 갖는 dip부분과 약 0.5μsec인 계단 상승시간과 약 4.5N의 힘의 크기를 갖는 계단함수의 형태였다.

This paper proposes particle filter(PF) method using acoustic signal for localization of an underwater robot. The method uses time of arrival(TOA) or time difference of arrival(TDOA) of acoustic signals from beacons whose locations are known. An experiment in towing tank uses TOA information. Simulation uses TDOA information and it reveals dependency of the localization performance on the uncertainty of robot motion and senor data. Also, comparison of the PF method with the least squares method of spherical interpolation(SI) and spherical intersection(SX) is provided. Since PF uses TOA or TDOA which comes from measurement of external information as well as internal motion information, its estimation is more accurate and robust to the sensor and motion uncertainty than the least squares methods.