Wyniki wyszukiwania - Biblioteka Nauki

1

A new framework using deep auto-encoder and energy spectral density for medical waveform data classification and processing

100%

Karim A. M. , Güzel M. S. , Tolun M. R. , Kaya H. , Çelebi F. V.

|

tom Vol. 39, no. 1

148--159

EN

This paper proposes a new framework for medical data processing which is essentially designed based on deep autoencoder and energy spectral density (ESD) concepts. The main novelty of this framework is to incorporate ESD function as feature extractor into a unique deep sparse auto-encoders (DSAEs) architecture. This allows the proposed architecture to extract more qualified features in a shorter computational time compared with the conventional frameworks. In order to validate the performance of the proposed framework, it has been tested with a number of comprehensive medical waveform datasets with varying dimensionality, namely, Epilepsy Serious Detection, SPECTF Classification and Diagnosis of Cardiac Arrhythmias. Overall, the ESD function speeds up the deep auto-encoder processing time and increases the overall accuracy of the results which are compared to several studies in the literature and a promising agreement is achieved.

2

A fault detection method based on stacking the SAE-SRBM for nonstationary and stationary hybrid processes

100%

Huang L. , Ren H. , Chai Y. , Qu J.

International Journal of Applied Mathematics and Computer Science

|

2021

|

tom Vol. 31, no. 1

29--43

EN

This paper proposes a fault detection method by extracting nonlinear features for nonstationary and stationary hybrid industrial processes. The method is mainly built on the basis of a sparse auto-encoder and a sparse restricted Boltzmann machine (SAE-SRBM), so as to take advantages of their adaptive extraction and fusion on strong nonlinear symptoms. In the present work, SAEs are employed to reconstruct inputs and accomplish feature extraction by unsupervised mode, and their outputs present a knotty problem of an unknown probability distribution. In order to solve it, SRBMs are naturally used to fuse these unknown probability distribution features by transforming them into energy characteristics. The contribution of this method is the capability of further mining and learning of nonlinear features without considering the nonstationary problem. Also, this paper introduces a method of constructing labeled and unlabeled training samples while maintaining time series features. Unlabeled samples can be adopted to train the part for feature extraction and fusion, while labeled samples can be used to train the classification part. Finally, a simulation on the Tennessee Eastman process is carried out to demonstrate the effectiveness and excellent performance on fault detection for nonstationary and stationary hybrid industrial processes.

3

Gear pitting fault diagnosis using raw acoustic emission signal based on deep learning

84%

Li X. , Li J. , He D. , Qu Y.

|

tom Vol. 21, no. 3

403--410

EN

Gear pitting fault is one of the most common faults in mechanical transmission. Acoustic emission (AE) signals have been effective for gear fault detection because they are less affected by ambient noise than traditional vibration signals. To overcome the problem of low gear pitting fault recognition rate using AE signals and convolutional neural networks, this paper proposes a new method named augmented convolution sparse autoencoder (ACSAE) for gear pitting fault diagnosis using raw AE signals. First, the proposed method combines sparse autoencoder and one-dimensional convolutional neural networks for unsupervised learning and then uses the reinforcement theory to enhance the adaptability and robustness of the network. The ACSAE method can automatically extract fault features directly from the original AE signals without time and frequency domain conversion of the AE signals. AE signals collected from gear test experiments are used to validate the ACSAE method. The analysis result of the gear pitting fault test shows that the proposed method can effectively performing recognition of the gear pitting faults, and the recognition rate reaches above 98%. The comparative analysis shows that in comparison with fully-connected neural networks, convolutional neural networks, and recurrent neural networks, the ACSAE method has achieved a better diagnostic accuracy for gear fitting faults.

PL

Pitting kół zębatych stanowi jedno z najczęstszych uszkodzeń przekładni mechanicznych. Do wykrywania takich uszkodzeń stosuje się sygnały emisji akustycznej (AE), które, ze względu na niższą wrażliwość na hałas otoczenia, stanowią skuteczniejsze narzędzie diagnostyczne niż tradycyjne sygnały wibracyjne. Wykrywalność zużycia guzełkowatego (pittingu) kół zębatych przy użyciu sygnałów AE i splotowych sieci neuronowych jest jednak niska. Aby rozwiązać ten problem, w niniejszym artykule zaproponowano nową metodę diagnozowania uszkodzeń kół zębatych za pomocą surowych sygnałów AE, którą nazwano augmented convolution sparse autoencoder (konwolucją rozszerzoną z wykorzystaniem autoenkodera rzadkiego, ACSAE). Jest to metoda samouczenia jednowymiarowych splotowych sieci neuronowych realizowanego za pomocą autoenkodera rzadkiego. Metoda ta wykorzystuje teorię wzmocnienia do zwiększania adaptacyjności i odporności sieci. Metoda ACSAE pozwala na automatyczne wyodrębnianie cech degradacji bezpośrednio z oryginalnych sygnałów AE bez konieczności ich konwersji do domeny czasu i częstotliwości. Walidację metody przeprowadzono na podstawie sygnałów AE otrzymanych w badaniach kół zębatych. Analiza wyników badań pittingu kół zębatych wskazuje, że proponowana metoda pozwala na skuteczną detekcję tego typu uszkodzeń, przy wskaźniku wykrywalności powyżej 98%. Analiza porównawcza pokazuje, że metoda ACSAE cechuje się większą trafnością diagnostyczną w wykrywaniu błędów montażowych kół zębatych w porównaniu z sieciami neuronowymi w pełni połączonymi, splotowymi i rekurencyjnymi.

4

AutoCovNet: Unsupervised feature learning using autoencoder and feature merging for detection of COVID-19 from chest X-ray images

84%

Rashid N. , Hossain M. A. F. , Ali M. , Islam Sukanya M. , Mahmud T. , Fattah S. A.

|

2021

|

tom Vol. 41, no. 4

1685--1701

EN

With the onset of the COVID-19 pandemic, the automated diagnosis has become one of the most trending topics of research for faster mass screening. Deep learning-based approaches have been established as the most promising methods in this regard. However, the limitation of the labeled data is the main bottleneck of the data-hungry deep learning methods. In this paper, a two-stage deep CNN based scheme is proposed to detect COVID-19 from chest X-ray images for achieving optimum performance with limited training images. In the first stage, an encoder-decoder based autoencoder network is proposed, trained on chest X-ray images in an unsupervised manner, and the network learns to reconstruct the X-ray images. An encoder-merging network is proposed for the second stage that consists of different layers of the encoder model followed by a merging network. Here the encoder model is initialized with the weights learned on the first stage and the outputs from different layers of the encoder model are used effectively by being connected to a proposed merging network. An intelligent feature merging scheme is introduced in the proposed merging network. Finally, the encoder-merging network is trained for feature extraction of the X-ray images in a supervised manner and resulting features are used in the classification layers of the proposed architecture. Considering the final classification task, an EfficientNet-B4 network is utilized in both stages. An end to end training is performed for datasets containing classes: COVID-19, Normal, Bacterial Pneumonia, Viral Pneumonia. The proposed method offers very satisfactory performances compared to the state of the art methods and achieves an accuracy of 90:13% on the 4-class, 96:45% on a 3-class, and 99:39% on 2-class classification.