Earticle

In this study, we explore the potential of leveraging machine learning techniques, specifically auto-encoders (AE), for the decoding of linear block codes. Our findings suggest that this approach can outperform the conventional ordered statistics decoding (OSD) method, especially in a Rayleigh fading channel environment. We have rigorously trained the AE under both additive white Gaussian noise and Rayleigh fading channel conditions to ensure robustness in its performance. The output of the AE is combined with the received vector in a suitable manner to perform OSD. Through our experiments, we demonstrate that this proposed decoding approach yields better results than the conventional OSD method in Rayleigh fading channel when we used (23,12) Golay code.

Reliable channel estimation is critical for achieving high-speed and energy-efficient communication in Unmanned Aerial Vehicle-Free Space Optical (UAV-FSO) systems, particularly under dynamic impairments such as atmospheric turbulence (AT) and pointing errors (PEs). This paper proposes a pilot-free channel estimation framework based on a Variational Autoencoder (VAE). The system employs Intensity Modulation/Direct Detection (IM/DD) with?-ary one-hot encoded symbols (). The VAE encodes noisy received signals into a 128-dimensional latent space and reconstructs the transmitted data, effectively learning the joint effects of AT, PEs, and AWGN. Unlike prior works that primarily consider boresight or Gaussian-jitter PEs, this study explicitly incorporates a Nakagami-modeled PE distribution, capturing UAV-induced beam misalignment under mobility, vibration, and turbulence coupling. Simulation results show that the proposed VAE significantly outperforms conventional estimators (LS, MMSE, LMMSE) and deep learning baselines (AE, DNN, CNN) across various turbulence strengths. Under strong turbulence and PEs, the VAE attains nearly two-fold lower MSE compared to CNN and DNN. In addition, evaluation on real turbulence-impaired datasets further validates robustness and generalization. The proposed pilot-free scheme delivers accurate channel estimation, reduced BER, and improved spectral efficiency, making it suitable for real-time adaptive UAV-FSO communication.

분산 환경에서 다양하게 사용되고 있는 IoT 장치는 의료·환경·교통·바이오·공공장소 등 사용 분야가 다양해지면 서 IoT 장치에서 송·수신되는 데이터의 중요도가 점점 증가하고 있다. 본 논문에서는 IoT 데이터의 신뢰성을 보장하기 위한 방법으로 수 많은 데이터들을 다양한 중요 속성별로 분류·처리하도록 오토인코더 기반의 IoT 연계 처리 기법을 제안한다. 제안 기법은 오토인코더 기반의 IoT 연계 처리를 위해서 IoT 데이터를 특성별로 블록체인으로 묶어 처리하 도록 IoT 데이터별로 상관관계 지수를 사용한다. 제안 기법은 IoT 데이터의 신뢰성을 보장하기 위해서 상관관계 지수 에 적용된 블록체인 기반의 n-계층 구조로 확장 운영한다. 또한, 제안 기법은 IoT 데이터의 상관관계 지수에 따라 IoT 수집 데이터에 가중치를 적용하여 IoT 데이터를 선택할 수 있을 뿐만 아니라 실시간으로 IoT 데이터의 무결성을 검증 하는 비용을 낮출 수 있다. 제안 기법은 n-계층 구조로 IoT 데이터를 확장할 수 있도록 IoT 데이터의 처리 비용을 유지한다.

IoT devices, which are used in various ways in distributed environments, are becoming more important in data transmitted and received from IoT devices as fields of use such as medical, environment, transportation, bio, and public places are diversified. In this paper, as a method to ensure the reliability of IoT data, an autoencoder-based IoT-linked processing technique is proposed to classify and process numerous data by various important attributes. The proposed technique uses correlation indices for each IoT data so that IoT data is grouped and processed by blockchain by characteristics for IoT linkage processing based on autoencoder. The proposed technique expands and operates into a blockchain-based n-layer structure applied to the correlation index to ensure the reliability of IoT data. In addition, the proposed technique can not only select IoT data by applying weights to IoT collection data according to the correlation index of IoT data, but also reduce the cost of verifying the integrity of IoT data in real time. The proposed technique maintains the processing cost of IoT data so that IoT data can be expanded to an n-layer structure.

With the growing reliance on deep learning-based image compression techniques, ensuring the security of compressed image data has become increasingly important. Traditional encryption methods operate directly on raw image pixels, often resulting in high computational costs and inefficiencies in real-time applications. In this paper, we propose a lightweight encryption method for securing compressed images within an autoencoder-based framework. Instead of encrypting the image itself, our approach focuses on shuffling the latent tensor using a randomly generated mixing order, which is then encrypted as the key. This method significantly reduces the size of encrypted data while maintaining strong security. Our experiments, conducted on the CIFAR100 dataset, demonstrate that even a few random mixing operations make the latent tensor and the decoded image unreadable, preventing unauthorized reconstruction of the original image. Moreover, the proposed method achieves substantial computational efficiency compared to conventional encryption methods such as ChaCha20, making it particularly suitable for time-sensitive applications, including real-time drone image transmission and surveillance.

Pervasive enhancement and required enforcement of the Internet of Things (IoTs) in a distributed massively multiplayer online architecture have effected in massive growth of Big-Data in terms of server over-load. There have been some previous works to overcome the overloading of server works. However, there are lack of considered methods, which is commonly applicable. Therefore, we propose a combing Sparse Auto-Encoder and Load-Balancing, which is ProGReGA for Big-Data of server loads. In the process of Sparse Auto-Encoder, when it comes to selection of the feature-pattern, the less relevant feature-pattern could be eliminated from Big-Data. In relation to Load-Balancing, the alleviated degradation of ProGReGA can take advantage of the less redundant feature-pattern. That means the most relevant of Big-Data representation can work. In the performance evaluation, we can find that the proposed method have become more approachable and stable.

GAN(Generative Adversarial Networks)이 합성 이미지 생성 및 기타 다양한 응용 프로그램에 현재 사용되고 있지만, 생성 모델을 제어하기가 어렵다. 문제는 생성 모델의 잠재 공간에 있는데, 이미지 생성과 관련하여 입력된 잠재 코드를 받아 특정 텍스트 및 신호에 따라 지정된 대상 속성이 향상되도록 하고 다른 속성은 크게 영향을 받지 않도록 하기 위해서는 상당한 제약이 요구된다. 본 연구에서는 이미지 생성 및 조작과 관련하여 변동 자동 인코더의 잠재 벡터 에 관해 특정 제약을 수반한 모델을 제안한다. 제안된 모델에 관해 TensorFlow의 변동 자동 인코더를 통해 실험한 결과 이미지의 생성 및 조작과 관련하여 비교적 우수한 성능을 갖는 것으로 확인된다.

Recent research shows that latent directions can be used to image process towards certain attributes. However, controlling the generation process of generative model is very difficult. Though the latent directions are used to image process for certain attributes, many restrictions are required to enhance the attributes received the latent vectors according to certain text and prompts and other attributes largely unaffected. This study presents a generative model having certain restriction to the latent vectors for image generation and manipulation. The suggested method requires only few minutes per manipulation, and the simulation results through Tensorflow Variational Auto-encoder show the effectiveness of the suggested approach with extensive results.

기계 장비의 진동 데이터는 필연적으로 노이즈를 포함하고 있다. 이러한 노이즈는 기계 장비의 유지보수를 진행하는데 악영향을 끼친다. 그에 따라 데이터의 노이즈를 얼마나 효과적으로 제거해주냐에 따라 학습 모델의 성능을 좌우한다. 본 논문에서는 시계열 데이터를 전처리 함에 있어 특성추출 과정을 포함하지 않는 Denoising Auto Encoder 기법을 활용하여 데이터의 노이즈를 제거했다. 또한 기계 신호 처리에 널리 사용되는 Wavelet Transform과 성능 비교를 진행했다. 성능비교는 고장 탐지율을 계산하여 진행했으며 보다 정확한 비교를 위해 분류 성능 평가기준 중 하나인 F-1 Score를 계산하여 성능 비교를 진행했다. 고장을 탐지하는 과정에서는 One-Class SVM 기법을 활용하여 고장 데이터를 탐지했다. 성능 비교 결과 고장 진단율과 오차율 측면에서 Denoising Auto Encoder 기법이 Wavelet Transform 기법에 비해 보다 좋은 성능을 나타냈다.

Vibration data of mechanical equipment inevitably have noise. This noise adversely af ects the maintenance of mechanical equipment. Accordingly, the performance of a learning model depends on how effectively the noise of the data is removed. In this study, the noise of the data was removed using the Denoising Auto Encoder (DAE) technique which does not include the characteristic extraction process in preprocessing time series data. In addition, the performance was compared with that of the Wavelet Transform, which is widely used for machine signal processing. The performance comparison was conducted by calculating the failure detection rate. For a more accurate comparison, a classification performance evaluation criterion, the F-1 Score, was calculated. Failure data were detected using the One-Class SVM technique. The performance comparison, revealed that the DAE technique performed better than the Wavelet Transform technique in terms of failure diagnosis and error rate.

Intelligent tutoring system enables users to effectively learn by utilizing various artificial intelligence techniques. For instance, it can recommend a proper curriculum or learning method to individual users based on their learning history. To do this effectively, user's characteristics need to be analyzed and classified based on various aspects such as interest, learning ability, and personality. Even though data labeled by the characteristics are required for more accurate classification, it is not easy to acquire enough amount of labeled data due to the labeling cost. On the other hand, unlabeled data should not need labeling process to make a large number of unlabeled data be collected and utilized. In this paper, we propose a semi-supervised learning method based on feedback variational auto-encoder(FVAE), which uses both labeled data and unlabeled data. FVAE is a variation of variational auto-encoder(VAE), where a multi-layer perceptron is added for giving feedback. Using unlabeled data, we train FVAE and fetch the encoder of FVAE. And then, we extract features from labeled data by using the encoder and train classifiers with the extracted features. In the experiments, we proved that FVAE-based semi-supervised learning was superior to VAE-based method in terms with accuracy and F1 score.

In this paper, we propose an automatic composition method using time series embedding of RNN Auto-Encoder. RNN Auto-Encoder can learn existing songs and can compose new songs from the trained RNN decoder. If one song is fully trained in the RNN Auto-Encoder, the song is embedded into the vector values of RNN nodes in the Auto-Encoder. If we train a lot of songs and apply a specific vector to the decoder of Auto-Encoder, then we can obtain a new song that combines the features of trained multiple songs according to the given vector. From extensive experiments we could find that our method worked well and generated various songs by selecting of the composition vectors.

항공 수요가 글로벌화와 함께 지속적으로 증가하고 있으며, Urban Air Mobility(UAM)의 도래로 도시 지역에서의 항공 이동이 중요한 이슈로 부상하고 있다. 이러한 배경 속에서 안전한 항공 시스템의 구축은 더욱 중요해지고 있다. 이상 징후 또는 비정상 작동을 탐지하기 위해 Auto Encoder(AE) 기반 이상 항적 탐지 연구가 이루어지고 있다. 항적은 환경에 영향을 받기 때문에 이상 항적 탐지 모형에 환경을 반영해야 한다는 어려움이 있다. 이를 해결하기 위해 본 연구는 환경 데이터와 궤적 데이터를 입력으로 사용하면서, 궤적 데이터만을 출력하는 Long Short Term Memory-Auto Encoder(LSTM-AE) 모형을 새롭게 제안한다. 일부 항적에 대한 Mean Absolute Error를 산출하여 모델의 재구성 능력을 확인하였다. 또한 전형적인 이상 항적인 착륙 복행 항적에 대한 Area under the ROC curve(AUC)를 비교하여 이상 항적 탐지 능력을 확인하였다. 이를 통해 일반적인 AE 모형, LSTM-AE 모형보다 제안 모형이 환경 변수에 대한 궤적의 패턴을 성공적으로 학습하며 이상 패턴 탐지 능력이 우수하다는 것을 확인하였다. 본 연구의 결과를 바탕으로 향후 더욱 다양한 환경 변수의 적용과 함께 더욱 정교한 모형 개발이 기대된다.

With the rise in aviation demand and the emergence of Urban Air Mobility, developing a safe aviation system in urban areas is becoming increasingly important. This study addresses the challenge of detecting anomalous flight trajectories, which can be influenced by environmental factors. We propose a novel Long Short-Term Memory-Auto Encoder (LSTM-AE) model that processes both environmental and trajectory data but only reconstructs trajectory data in its output. This approach was validated by assessing the average reconstruction error for specific trajectories. Additionally, the model's ability to identify anomalies was confirmed by evaluating the Area under the ROC curve (AUC) for typical anomalous trajectories, such as go-around maneuvers. Our findings indicate that the proposed LSTM-AE model effectively learns trajectory patterns in relation to environmental variables and shows enhanced anomaly detection capabilities compared to traditional AE and LSTM-AE models. These results contribute to the development of advanced models that incorporate a wider range of environmental factors, enhancing safety in urban air travel.

Visual place recognition is widely researched area in robotics, as it is one of the elemental requirements for autonomous navigation, simultaneous localization and mapping for mobile robots. However, place recognition in changing environment is a challenging problem since a same place look different according to the time, weather, and seasons. This paper presents a feature extraction method using a deep convolutional auto-encoder to recognize places under severe appearance changes. Given database and query image sequences from different environments, the convolutional auto-encoder is trained to predict the images of the desired environment. The training process is performed by minimizing the loss function between the predicted image and the desired image. After finishing the training process, the encoding part of the structure transforms an input image to a low dimensional latent representation, and it can be used as a condition-invariant feature for recognizing places in changing environment. Experiments were conducted to prove the effective of the proposed method, and the results showed that our method outperformed than existing methods.

딥러닝의 급속한 발전은 패턴인식 분야의 성능을 혁신했으며 몇몇 문제에서는 인간 수준을 넘어서는 결과들을 보여주고 있다. 데이타를 분류하는 패턴인식과 달리 본 논문에서는 주어진 몇개의 한국어 문장으로부터 비슷한 문장들을 생성하는 문제를 다룬다. 이를위해 생성모델 중의 하나인 Variational Auto-Encoder 기반의 모델을 한국어 생성에 맞게 개선하고 적용하는 방법들을 논의한다. 첫째, 교착어인 한국어의 특성상 띄어쓰기를 기준으로 단어 생성시 단어의 개수가 너무 많아 이를 줄이기 위해 조사 및 어미들을 분리할 필요가 있다. 둘째, 한국어는 어순이 비교적 자유롭고 주어 목적어 등이 생략되는 경우가 많아 기존의 단방향 인코더를 양방향으로 확장한다. 마지막으로, 주어진 문장들을 기반으로 비슷하지만 새로운 문장들을 생성하기 위해 기존 문장들의 인코딩된 벡터표현들로부터 새로운 벡터를 찾아내고, 이 벡터를 디코딩하여 문장을 생성한다. 실험 결과를 통해 제안한 방법의 성능을 확인한다.

Due to the revolutionary advances in deep learning, performance of pattern recognition has increased significantly in many applications like speech recognition and image recognition, and some systems outperform human-level intelligence in specific domains. Unlike pattern recognition, in this paper, we focus on generating Korean sentences based on a few Korean sentences. We apply variational recurrent auto-encoder (VRAE) and modify the model considering some characteristics of Korean sentences. To reduce the number of words in the model, we apply a word spacing model. Also, there are many Korean sentences which have the same meaning but different word order, even without subjects or objects; therefore we change the unidirectional encoder of VRAE into a bidirectional encoder. In addition, we apply an interpolation method on the encoded vectors from the given sentences, so that we can generate new sentences which are similar to the given sentences. In experiments, we confirm that our proposed method generates better sentences which are semantically more similar to the given sentences.

운전 중 감정 인식은 사고를 예방하기 위해 꼭 필요한 과제이다. 더 나아가 자율 주행 시대에서 자동차는 모빌리티의 주체로 운전자와의 감정적인 소통이 더욱 요구되고 있으며 감정 인식 시장은 점점 확산되고 있다. 이에 따라 본 연구 방안에서는 수집하기 비교적 용이한 데이터인 심리데이터와 행동 데이터를 이용해 운전자의 감정을 분류하는 인공지능 모델을 개발하고자 한다. 오토인코더 모델을 통해 잠재 변수를 추출하고, 이를 본 분류 모델의 변수로 사용하였으며, 이는 성능 향상에 영향을 미침을 확인하였다. 또한 기존 뇌파 데이터를 포함했을 때 보다 본 논문이 제시하는 프레임워크를 사용하였을 때 성능이 향상됨도 확인하였다. 최종적으로 심리 및 개인정보데이터, 행동 데이터만을 통해 운전자의 감정 분류 정확도 81%와 F1-Score 80%를 달성하였다.

Emotion recognition while driving is an essential task to prevent accidents. Furthermore, in the era of autonomous driving, automobiles are the subject of mobility, requiring more emotional communication with drivers, and the emotion recognition market is gradually spreading. Accordingly, in this research plan, the driver's emotions are classified into seven categories using psychological and behavioral data, which are relatively easy to collect. The latent vectors extracted through the auto-encoder model were also used as features in this classification model, confirming that this affected performance improvement. Furthermore, it also confirmed that the performance was improved when using the framework presented in this paper compared to when the existing EEG data were included. Finally, 81% of the driver's emotion classification accuracy and 80% of F1-Score were achieved only through psychological, personal information, and behavioral data.

In this paper, we propose a deep auto-encoder-based pipe leak detection (PLD) technique from time-series acoustic data collected by microphone sensor nodes. The key idea of the proposed technique is to learn representative features of the leak-free state using leak-free time-series acoustic data and the deep auto-encoder. The proposed technique can be used to create a PLD model that detects leaks in the pipeline in an unsupervised learning manner. This means that we only use leak-free data without labeling while training the deep auto-encoder. In addition, when compared to the previous supervised learning-based PLD method that uses image features, this technique does not require complex preprocessing of time-series acoustic data owing to the unsupervised feature extraction scheme. The experimental results show that the proposed PLD method using the deep auto-encoder can provide reliable PLD accuracy even considering unsupervised learning-based feature extraction.

As a reaction of progressively famous online networking, cyber bullying has risen as a significant issue distressing kids, teenagers and youthful grown-ups. Machine learning strategies make programmed identification of harassing messages in online networking conceivable, and this could develop a solid and safe web-based social networking environment. In this significant research zone, one basic issue is strong and discriminative numerical representation learning of instant messages. In this paper, we propose another representation learning strategy to handle this issue. Our strategy named Semantic-Enhanced Marginalized Demising Auto-Encoder (smSDA) is produced by means of semantic expansion of the famous profound learning model stacked demising auto-encoder. The semantic expansion comprises of semantic dropout clamor and sparsely limitations, where the semantic dropout commotion is planned in view of area learning and the word installing procedure. Our proposed strategy can abuse the concealed element structure of tormenting data and take in a powerful and discriminative representation of content. Exhaustive tests on two open cyber bullying corpora (Twitter and MySpace) are led, and the outcomes demonstrate that our proposed approaches beat other gauge content representation learning techniques.

본 논문에서는 저전력 무선 음향센서 모듈을 통한 데이터 수집과 딥 오토인코더를 이용한 데이터 분석을 통해 배관의 누출을 탐지하는 시스템을 제안한다. 데이터 통신량을 줄이기 위해서 푸리에 변환을 통해 음향센서 데이터 양을 약 1/800로 감소시키는 저전력 무선 음향센서 모듈을 구성하였고, 20kHz~100kHz 주파수 신호를 이용하여 가청 주파수 대역에서 발생하는 노이즈에 강인한 누출 탐지 시스템을 설계하였다. 또한, 데이터 양의 감소에도 배관 누출을 정확하게 탐지하도록 딥 오토인코더를 이용한 데이터 분석 기법을 설계하였다. 수치적인 실험을 통해, 본 논문에서 제안한 누출 탐지 시스템이 고주파 대역대의 노이즈가 섞인 환경에서도 99.94%의 높은 정확도와 Type-II error 0%의 높은 성능을 보이는 것을 검증하였다.

In this paper, we propose a pipe leak detection system through data collection using low-power wireless acoustic sensor modules and data analysis using deep auto-encoder. Based on the Fourier transform, we propose a low-power wireless acoustic sensor module that reduces data traffic by reducing the amount of acoustic sensor data to about 1/800, and we design the system that is robust to noise generated in the audible frequency band using only 20kHz~100kHz frequency signals. In addition, the proposed system is designed using a deep auto-encoder to accurately detect pipe leaks even with a reduced amount of data. Numerical experiments show that the proposed pipe leak detection system has a high accuracy of 99.94% and Type-II error of 0% even in the environment where high frequency band noise is mixed.

본 논문은 재생에너지 발전량 데이터를 기반으로 발전량 예측값을 도출하기 위한 연구를 수행하였다. 발전량 측정을 위해 재생에너지의 발전량 데이터, 환경데이터와 같은 다변량 변수를 측정하였다. 측정값의 안정성과 신뢰성 확보를 위해 ARIMA(Auto Regressive Integrated Moving Average) 모델을 이용한 전처리를 수행하였다.재생에너지의 발전량 예측을 위해 딥러닝 모형인 LSTM-Autoencoder 결합모델을 이용하여 발전량을 측정했고, 예측정확도를 높이기 위해 각 다변량 변수의 장기종속성 학습 방식을 이용하였다. 정제된 데이터를 활용한 환경변수들의 추세는 기존 데이터를 그대로 사용했을 때 보다 안정되었고, 상관관계 분석 결과를 반영하여 다변량 변수 중 상관성이 높은 변수만을 활용하여 재생에너지 발전량 예측 알고리즘을 개선하여 약 94%의 정확도가 측정되었다.

This paper conducted a study to derive the predicted value of power generation based on the renewable energy generation data. In order to measure the power generation, multivariate variables such as power generation data and environmental data of renewable energy were measured. In order to secure the stability and reliability of the measured values, preprocessing was performed using the ARIMA (Auto Regressive Integrated Moving Average) model. In order to predict the power generation of renewable energy, the power generation was measured using the LSTM-Autoencoder combination model, a deep learning model, and the long-term dependency learning method of each multivariate variable was used to increase the prediction accuracy. The trend of environmental variables using refined data was more stable when the existing data were used as it was, and the accuracy of about 94% was measured by improving the renewable energy generation prediction algorithm by using only the highly correlated variable among the multivariate variables by reflecting the results of the correlation analysis.

최근 보안 디바이스의 물리적 취약성을 찾을 수 있는 부채널 분석 분야에서 딥러닝을 활용한 연구가 활발히 진행되고 있다. 하지만, 최신 딥러닝 기반 부채널 분석 기술 연구는 템플릿 공격 등과 같은 프로파일링 기반 부채널 분석 환경에서 파형을 옳게 분류하기 위한 연구에 집중되어 있다. 본 논문에서는 이전 연구들과 다르게 딥러닝을 신호 전처리 기법으로 활용하여 차분 전력 분석, 상관 전력 분석 등과 같은 논프로파일링 기반 부채널 분석의 성능을 고도화할 수 있는 방법을 제안한다. 제안기법은 오토인코더를 부채널 분석 환경에 적합하게 변경하여 부채널 정보의 노이즈를 제거하는 전처리 기법으로, 기존 노이즈 제거 오토인코더는 임의로 추가한 노이즈에 대한 학습을 하였다면 제안하는 기법은 노이즈가 제거된 라벨을 사용하여 실제 데이터의 노이즈를 학습한다. 제안기법은 논프로파일링 환경에서 수행 가능한 전처리 기법이며 하나의 뉴런 네트워크의 학습만을 통해 수행할 수 있다. 본 논문에서는 실험을 통해 제안기법의 노이즈 제거 성능을 입증하였으며, 주성분분석 및 선형판별분석과 같은 기존 전처리 기법들과 비교하여 우수하다는 것을 보인다.

In side-channel analysis, which exploit physical leakage from a cryptographic device, deep learning based attack has been significantly interested in recent years. However, most of the state-of-the-art methods have been focused on classifying side-channel information in a profiled scenario where attackers can obtain label of training data. In this paper, we propose a new method based on deep learning to improve non-profiling side-channel attack such as Differential Power Analysis and Correlation Power Analysis. The proposed method is a signal preprocessing technique that reduces the noise in a trace by modifying Auto-Encoder framework to the context of side-channel analysis. Previous work on Denoising Auto-Encoder was trained through randomly added noise by an attacker. In this paper, the proposed model trains Auto-Encoder through the noise from real data using the noise-reduced-label. Also, the proposed method permits to perform non-profiled attack by training only a single neural network. We validate the performance of the noise reduction of the proposed method on real traces collected from ChipWhisperer board. We demonstrate that the proposed method outperforms classic preprocessing methods such as Principal Component Analysis and Linear Discriminant Analysis.

회전설비의 진동신호는 주변 환경에 의해 간섭을 받기 때문에 실험실 환경에서 수집된 진동신호는 실제 산업현장의 상황과 차이가 크며 이를 처리하는 과정 없이 신호를 분석하여 이상탐지 및 결함분류를 하는 경우 예측성능을 보장할 수 없을 뿐만 아니라 결과해석에 왜곡이 발생할 수 있다. 때문에 진동신호의 전처리 단계에서 반드시 잡음을 제거하는 과정이 동반되어야 한다. 본 논문에서는 전통적으로 잡음제거에 사용된 신호처리 필터 대신 딥러닝 비지도학습인 1D-CNN 기반의 오토인코더 모델을 이용하여 신호의 잡음을 제거하는 디노이징 모델을 적용하고 잡음이 제거된 신호에 대해 이상신호 및 결함모드를 분류하는 인공지능 기반의 예측정비 시스템을 제안한다.

Because the vibration signal of the rotating facility is interfered by the surrounding environment, the vibration signal collected in the laboratory environment differs greatly from the situation of the actual industrial site, and if abnormal detection and fault classification are performed by analyzing the signal without processing it, it is not only impossible to guarantee its predictive performance, but it can also cause distortion in the result analysis. Therefore, the process of removing noise must be accompanied at the pre-processing stage of the vibration signal. In this paper, instead of the signal processing filter traditionally used for noise removal, an artificial intelligence-based prediction maintenance system is proposed, which applies a dinoing model that eliminates noise of signals by using an autocoder model based on 1D-CNN, a deep learning non-instructor learning, and classifies abnormal signals and fault modes for signals with which noise has been removed.