AdaBoost-LSTM 기반 도메인 적응 기술을 이용한 항공기 엔진의 잔여 유효 수명 예측

서승환,황정우 한국통신학회 2024 韓國通信學會論文誌 Vol.49 No.2

Along with the advent of high-quality deep learning algorithms, several methods have been published for the domain adaptation (DA) problem on remaining useful life. Most of them are unsupervised DA methods and popular adversarial approaches are known to have best performance among them. But, we have found out that adversarial approaches have an unstable problem that is the performance critically depends on the starting weights of the deep-learning networks. Furthermore, unsupervised DA methods could get a limited performance improvement if domain shift is larger than some extent. This paper proposes a supervised DA method based on AdaBoost with Long Short-Term Memory (LSTM) as base estimators. The proposed approach is effective when target domain data is much smaller than source domain data. On a publically accessible dataset, the proposed methodology is tested, and when compared to previous unsupervised domain adaption prediction methods, it reaches state-of-the-art prediction performance.

Prediction of the remaining useful life of rolling bearings by LSTM based on multidomain characteristics and a dual-attention mechanism

Huaiqian Bao,Lijin Song,Zongzhen Zhang,Baokun Han,Jinrui Wang,Junqing Ma,Xingwang Jiang 대한기계학회 2023 JOURNAL OF MECHANICAL SCIENCE AND TECHNOLOGY Vol.37 No.9

This study proposes a framework for bearing remaining useful life (RUL) prediction that uses multidomain features and a dual-attention mechanism (DAM). First, sparsity measures are introduced as new feature parameters to comprehensively and accurately extract the degradation features of bearings. Second, a long short-term memory network integrated with DAM is applied for RUL prediction. DAM simultaneously applies the attention mechanism to the time steps and feature dimension to increase the attention to important information and enhance the prediction performance of the network. Third, a pseudo-normalization method is proposed to solve the problem of unknown bearing test data in actual working conditions under the premise of retaining the original data characteristics and RUL prediction accuracy as much as possible. Lastly, the proposed framework is experimentally proven on public datasets and compared with other methods to prove its feasibility and effectiveness.

An improved regularized particle fi lter for remaining useful life prediction in nuclear plant electric gate valves

Ren-yi Xu,Hang Wang,Min-jun Peng,Yong-kuo Liu 한국원자력학회 2022 Nuclear Engineering and Technology Vol.54 No.6

Accurate remaining useful life (RUL) prediction for critical components of nuclear power equipment is animportant way to realize aging management of nuclear power equipment. The electric gate valve is oneof the most safety-critical and widely distributed mechanical equipment in nuclear power installations. However, the electric gate valve's extended service in nuclear installations causes aging and degradationinduced by crack propagation and leakages. Hence, it is necessary to develop a robust RUL predictionmethod to evaluate its operating state. Although the particle filter(PF) algorithm and its variants can dealwith this nonlinear problem effectively, they suffer from severe particle degeneracy and depletion, whichleads to its sub-optimal performance. In this study, we combined the whale algorithm with regularizedparticle filtering(RPF) to rationalize the particle distribution before resampling, so as to solve theproblem of particle degradation, and for valve RUL prediction. The valve's crack propagation is studiedusing the RPF approach, which takes the Paris Law as a condition function. The crack growth is observedand updated using the root-mean-square (RMS) signal collected from the acoustic emission sensor. Atthe same time, the proposed method is compared with other optimization algorithms, such as particleswarm optimization algorithm, and verified by the realistic valve aging experimental data. The conclusion shows that the proposed method can effectively predict and analyze the typical valve degradationpatterns

철강 도금로의 예지보전을 위한 열화 기반 잔존수명 분석

신준호,김창욱 한국융합학회 2019 한국융합학회논문지 Vol.10 No.12

Smart factory, a critical part of digital transformation, enables data-driven decision making using monitoring, analysis and prediction. Predictive maintenance is a key element of smart factory and the need is increasing. The purpose of this study is to analyze the degradation characteristics of a galvanizing kettle for the steel plating process and to predict the remaining useful life(RUL) for predictive maintenance. Correlation analysis, multiple regression, principal component regression were used for analyzing factors of the process. To identify the trend of degradation, a proposed rolling window was used. It was observed the degradation trend was dependent on environmental temperature as well as production factors. It is expected that the proposed method in this study will be an example to identify the trend of degradation of the facility and enable more consistent predictive maintenance. 제조산업 분야의 디지털트랜스포메이션의 일환인 스마트공장은 데이터 기반으로 모니터링 및 분석 그리고 예측을 통해서 의사결정 방식을 획기적으로 변화시키고 있다. 특히 설비에 대한 예지보전은 스마트공장의 핵심적인 요소로서 필요성이 증대되고 있다. 본 연구의 목적은 철강 도금공정의 예지보전을 위해 도금로 설비의 열화 특성을 고려한 잔존수명 분석과 예측모델을 산출하는 것이다. 상관성 분석, 다중회귀 분석, 주성분회귀 분석 그리고 시간의 경과에 따른 열화의 추이 파악을 위하여 이동회귀 방식을 제안하여 진행하였다. 그 결과 도금로 열화는 생산성 인자들과 주된 의존적 관계가 있으며, 특히 환경 온도 인자들의 영향성이 열화의 추이 변화에 관계가 있음을 추론할 수 있었다. 예측된 잔존수명을 활용하여 도금로 교체가 필요한 시점을 사전에 알려주는 예지보전을 구현하였다. 향후 설비의 열화 추이 파악에 본 연구에서 수행한 방안이 적절한 사례가 되어 보다 정합성 있는 예지보전 구현이 가능해지기를 기대한다.

Particle Filter 알고리즘 기반 이산 웨이블릿 변환 기법의 경험적 모델 설계를 통한 리튬이온 배터리의 잔여 수명 예측

김재원(Jaewon Kim),박진형(Jinhyeong Park),권상욱(Sanguk Kwon),신승화(Seunghwa Sin),김범종(Bumjong Kim),김종훈(Jonghoon Kim) 한국자동차공학회 2022 한국 자동차공학회논문집 Vol.30 No.3

Electric vehicles(EVs) are being commercialized as practical alternatives to the Zero emission vehicle. Lithium-ion batteries(LIBs), the main source of energy for EVs, are one of the components that affect the economy and the safety of EVs due to LIB’s high energy density and long lifespan. However, LIBs can be problematic in terms of fading capacity and reduced life due to prolonged charging/discharging. Therefore, accurate remaining-useful-life(RUL) prediction is essential. In this paper, the discharge capacity of the lithium-polymer battery pack and the nickel manganese cobalt(NMC) battery were extracted through current signals. By using discrete wavelet transform(DWT), it is possible to induce capacity regeneration, such as the noise of the NMC battery, by compression and decomposition. To provide accurate batteries’ replacement time, RUL is implemented based on particle filter(PF). The result shows that the RUL prediction of the decomposed signals with noise improved by about 5% compared to the raw signal data. Therefore, in this paper, it was proposed that using decomposed signals can be an advantage in terms of data storage space and the effect of reduced RUL time.

잔여 유효 수명 예측 모형과 최소 수리 블록 교체 모형에 기반한 비용 최적 예방 정비 방법

주영석,신승준,Choo, Young-Suk,Shin, Seung-Jun 한국산업경영시스템학회 2022 한국산업경영시스템학회지 Vol.45 No.3

Predicting remaining useful life (RUL) becomes significant to implement prognostics and health management of industrial systems. The relevant studies have contributed to creating RUL prediction models and validating their acceptable performance; however, they are confined to drive reasonable preventive maintenance strategies derived from and connected with such predictive models. This paper proposes a data-driven preventive maintenance method that predicts RUL of industrial systems and determines the optimal replacement time intervals to lead to cost minimization in preventive maintenance. The proposed method comprises: (1) generating RUL prediction models through learning historical process data by using machine learning techniques including random forest and extreme gradient boosting, and (2) applying the system failure time derived from the RUL prediction models to the Weibull distribution-based minimum-repair block replacement model for finding the cost-optimal block replacement time. The paper includes a case study to demonstrate the feasibility of the proposed method using an open dataset, wherein sensor data are generated and recorded from turbofan engine systems.

Remaining useful life prediction of planet bearings based on conditional deep recurrent generative adversarial network and action discovery

Jun Yu,Zhenyu Guo 대한기계학회 2021 JOURNAL OF MECHANICAL SCIENCE AND TECHNOLOGY Vol.35 No.1

Planet bearing vibration signals are complex and non-stationary owing to the time-varying vibration transmission path from fault point to accelerometers. In addition, due to the influence of speed and load variation, planetary gearboxes often operate under varying conditions, which causes the acquired vibration signals to be nonlinear. These seriously degrade the performance of the RUL prediction methods. Moreover, there are only few training samples for the RUL prediction of planet bearings. To address these problems, an RUL prediction method of planet bearings using conditional deep recurrent generative adversarial network (C-DRGAN) and action discovery (AD) is presented. First, gated recurrent unit neural network is integrated with conditional generative adversarial network to construct the C-DRGAN, which extracts fault features from nonlinear and non-stationary signals so as to realize the RUL prediction of planet bearings under small samples and varying conditions. Then, the training algorithm based on AD is employed to train the C-DRGAN to enhance convergence speed and reduce training time. Finally, a multiple linear regression classifier is utilized to predict the RUL of planet bearings. The performance of the presented method is validated through an accelerated fatigue life experiment of planet bearings. Experimental analyses demonstrate that this method possesses strong processing adaptability to nonlinear and non-stationary signals and obtains excellent performance under small samples.

Enhancing aircraft engine remaining useful life prediction via multiscale deep transfer learning with limited data

LIU QIAN,Zhang Zhiyao,GUO PENG,WANG YIFAN,Liang Junxin 한국CDE학회 2024 Journal of computational design and engineering Vol.11 No.1

Predicting the remaining useful life (RUL) of the aircraft engine based on historical data plays a pivotal role in formulating maintenance strategies and mitigating the risk of critical failures. None the less, attaining precise RUL predictions often encounters challenges due to the scarcity of historical condition monitoring data. This paper introduces a multiscale deep transfer learning framework via integrating domain adaptation principles. The framework encompasses three integral components: a feature extraction module, an encoding module, and an RUL prediction module. During pre-training phase, the framework leverages a multiscale convolutional neural network to extract distinctive features from data across varying scales. The ensuing parameter transfer adopts a domain adaptation strategy centered around maximum mean discrepancy. This method efficiently facilitates the acquisition of domain-invariant features from the source and target domains. The refined domain adaptation Transformer-based multiscale convolutional neural network model exhibits enhanced suitability for predicting RUL in the target domain under the condition of limited samples. Experiments on the C-MAPSS dataset have shown that the proposed method significantly outperforms state-of-the-art methods.

리튬이온 배터리의 상태지표와 용량 감소 예측을 위한 Bidirectional LSTM 및 SoH 기술

박정언(Jeong-Eon Park),정종문(Jong-Moon Chung) 한국자동차공학회 2023 한국자동차공학회 부문종합 학술대회 Vol.2023 No.5

Predicting the state-of-health (SoH) and remaining useful life (RUL) of a vehicle’s battery is very important for battery manufacturers and customers. To predict SoH and RUL of a battery, various data-driven models have been studied. There are many battery health indicators used to represent battery performance degradation, but most studies have not considered them. Therefore, in this paper, a model for predicting the battery aging state is constructed by selecting health indicators representing battery aging in charge/discharge experiments and using this data as input to a bidirectional long-short term memory (LSTM) deep learning model. This model successfully predicts the battery capacity degradation curve and results in a root-mean-squared error (RMSE) as low as 0.0051.

Fault state detection and remaining useful life prediction in AC powered solenoid operated valves based on traditional machine learning and deep neural networks

Utah, M.N.,Jung, J.C. Korean Nuclear Society 2020 Nuclear Engineering and Technology Vol.52 No.9

Solenoid operated valves (SOV) play important roles in industrial process to control the flow of fluids. Solenoid valves can be found in so many industries as well as the nuclear plant. The ability to be able to detect the presence of faults and predicting the remaining useful life (RUL) of the SOV is important in maintenance planning and also prevent unexpected interruptions in the flow of process fluids. This paper proposes a fault diagnosis method for the alternating current (AC) powered SOV. Previous research work have been focused on direct current (DC) powered SOV where the current waveform or vibrations are monitored. There are many features hidden in the AC waveform that require further signal analysis. The analysis of the AC powered SOV waveform was done in the time and frequency domain. A total of sixteen features were obtained and these were used to classify the different operating modes of the SOV by applying a machine learning technique for classification. Also, a deep neural network (DNN) was developed for the prediction of RUL based on the failure modes of the SOV. The results of this paper can be used to improve on the condition based monitoring of the SOV.