scholarly journals Dynamic Noise Reduction with Deep Residual Shrinkage Networks for Online Fault Classification

Sensors ◽  
2022 ◽  
Vol 22 (2) ◽  
pp. 515
Author(s):  
Alireza Salimy ◽  
Imene Mitiche ◽  
Philip Boreham ◽  
Alan Nesbitt ◽  
Gordon Morison
Keyword(s):  
Real World ◽  
Electromagnetic Interference ◽  
Fault Classification ◽  
Noise Levels ◽  
Time Frequency ◽  
Building Unit ◽  
Dynamic Noise ◽  
Shrinkage Methods ◽  
Unseen Data ◽  
Testing Accuracy

Fault signals in high-voltage (HV) power plant assets are captured using the electromagnetic interference (EMI) technique. The extracted EMI signals are taken under different conditions, introducing varying noise levels to the signals. The aim of this work is to address the varying noise levels found in captured EMI fault signals, using a deep-residual-shrinkage-network (DRSN) that implements shrinkage methods with learned thresholds to carry out de-noising for classification, along with a time-frequency signal decomposition method for feature engineering of raw time-series signals. The approach will be to train and validate several alternative DRSN architectures with previously expertly labeled EMI fault signals, with architectures then being tested on previously unseen data, the signals used will firstly be de-noised and a controlled amount of noise will be added to the signals at various levels. DRSN architectures are assessed based on their testing accuracy in the varying controlled noise levels. Results show DRSN architectures using the newly proposed residual-shrinkage-building-unit-2 (RSBU-2) to outperform the residual-shrinkage-building-unit-1 (RSBU-1) architectures in low signal-to-noise ratios. The findings show that implementing thresholding methods in noise environments provides attractive results and their methods prove to work well with real-world EMI fault signals, proving them to be sufficient for real-world EMI fault classification and condition monitoring.

scholarly journals Bearing Fault Classification Using Ensemble Empirical Mode Decomposition and Convolutional Neural Network

Electronics ◽  
2021 ◽  
Vol 10 (11) ◽  
pp. 1248
Author(s):  
Rafia Nishat Toma ◽  
Cheol-Hong Kim ◽  
Jong-Myon Kim
Keyword(s):  
Neural Network ◽  
Convolutional Neural Network ◽  
Empirical Mode Decomposition ◽  
Vibration Signal ◽  
Ensemble Empirical Mode Decomposition ◽  
Fault Classification ◽  
Original Signal ◽  
Time Frequency ◽  
Mode Decomposition ◽  
Reconstructed Signal

Condition monitoring is used to track the unavoidable phases of rolling element bearings in an induction motor (IM) to ensure reliable operation in domestic and industrial machinery. The convolutional neural network (CNN) has been used as an effective tool to recognize and classify multiple rolling bearing faults in recent times. Due to the nonlinear and nonstationary nature of vibration signals, it is quite difficult to achieve high classification accuracy when directly using the original signal as the input of a convolution neural network. To evaluate the fault characteristics, ensemble empirical mode decomposition (EEMD) is implemented to decompose the signal into multiple intrinsic mode functions (IMFs) in this work. Then, based on the kurtosis value, insignificant IMFs are filtered out and the original signal is reconstructed with the rest of the IMFs so that the reconstructed signal contains the fault characteristics. After that, the 1-D reconstructed vibration signal is converted into a 2-D image using a continuous wavelet transform with information from the damage frequency band. This also transfers the signal into a time-frequency domain and reduces the nonstationary effects of the vibration signal. Finally, the generated images of various fault conditions, which possess a discriminative pattern relative to the types of faults, are used to train an appropriate CNN model. Additionally, with the reconstructed signal, two different methods are used to create an image to compare with our proposed image creation approach. The vibration signal is collected from a self-designed testbed containing multiple bearings of different fault conditions. Two other conventional CNN architectures are compared with our proposed model. Based on the results obtained, it can be concluded that the image generated with fault signatures not only accurately classifies multiple faults with CNN but can also be considered as a reliable and stable method for the diagnosis of fault bearings.

scholarly journals Time-Frequency Feature Analysis

2021 ◽  
Author(s):  
Behnaz Ghoraani
Keyword(s):  
Pattern Recognition ◽  
Real World ◽  
Feature Detection ◽  
Classification Systems ◽  
Time Frequency ◽  
Analysis Methodology ◽  
Stationary Signals ◽  
Time Varying Frequency ◽  
Non Stationary Signal ◽  
Discriminative Pattern

Most of the real-world signals in nature are non-stationary, i.e., their statistics are time variant. Extracting the time-varying frequency characteristics of a signal is very important in understanding the signal better, which could be of immense use in various applications such as pattern recognition and automated-decision making systems. In order to extract meaningful time-frequency (TF) features, a joint TF analysis is required. The proposed work is an attempt to develop a generalized TF analysis methodology that exploits the benefits of TF distribution (TFD) in pattern classification systems as related to discriminant feature detection and classification. Our objective is to introduce a unique and efficient way of performing non-stationary signal analysis using adaptive and discriminant TF techniques. To fulfill this objective, in the first point, we build a novel TF matrix (TFM) decomposition that increases the effectiveness of segmentation in real-world signals. Instantaneous and unique features are extracted from each segment such that they successfully represent joint TF structure of the signal. In the second point, based on the above technique, two unique and novel discriminant TF analysis methods are proposed to perform an improved and discriminant feature selection of any non-stationary signals. The first approach is a new machine learning method that identifies the clusters of the discriminant features to compute the presence of the discriminative pattern in any given signal, and classify them accordingly. The second approach is a discriminant TFM (DTFM) framework, which is a combination of TFM decomposition and the discriminant clustering techniques. The developed DTFM analysis automatically identifies the differences between different classes as the distinguishing structure, and uses the identified structure to accurately classify and locate the discriminant structure in the signal. The theoretical properties of the proposed approaches pertaining to pattern recognition and detection are examined in this dissertation. The extracted TF features provide strong and successful characterization and classification of real and synthetic non-stationary signals. The proposed TF techniques facilitate the adaptation of TF quantification to any feature detection technique in automating the identification process of discriminatory TF features, and can find applications in many different fields including biomedical and multimedia signal processing.

MODELING AUDITORY PATHWAY FOR INTELLIGENT INFORMATION ACQUISITION

2004 ◽  
Vol 01 (04) ◽  
pp. 345-356
Author(s):  
HYUNG-MIN PARK ◽  
JONG-HWAN LEE ◽  
TAESU KIM ◽  
UN-MIN BAE ◽  
BYUNG TAEK KIM ◽  
...  
Keyword(s):  
Feature Extraction ◽  
Real World ◽  
Information Acquisition ◽  
Recognition Performance ◽  
Auditory Pathway ◽  
Blind Signal Separation ◽  
Top Down ◽  
Auditory Model ◽  
Time Frequency ◽  
Binaural Processing

An auditory model has been developed for an intelligent speech information acquisition system in real-world noisy environment. The developed mathematical model of the human auditory pathway consists of three components, i.e. the nonlinear feature extraction from cochlea to auditory cortex, the binaural processing at superior olivery complex, and the top-down attention from higher brain to the cochlea. The feature extraction is based on information-theoretic sparse coding throughout the auditory pathway. Also, the time-frequency masking is incorporated as a model of the lateral inhibition in both time and frequency domain. The binaural processing is modeled as the blind signal separation and adaptive noise canceling based on the independent component analysis with hundreds of time-delays for noisy reverberated signals. The Top-Down (TD) attention comes from familiarity and/or importance of the sensory information, i.e. the sound, and a simple but efficient TD attention model had been developed based on the error backpropagation algorithm. Also, the binaural processing and top-down attention are combined for speech signals with heavy noises. This auditory model requires extensive computing, and special hardware had been developed for real-time applications. Experimental results demonstrate much better recognition performance in real-world noisy environments.

Speech Enhancement Using Neuro-Fuzzy Classifier

2021 ◽  
pp. 164-181
Author(s):  
Judith Justin ◽  
Vanithamani R.
Keyword(s):  
Feature Extraction ◽  
Speech Enhancement ◽  
The Other ◽  
Objective Measures ◽  
Noise Levels ◽  
Fuzzy Classifier ◽  
Noisy Speech ◽  
Enhancement Technique ◽  
Time Frequency ◽  
Neuro Fuzzy

In this chapter, a speech enhancement technique is implemented using a neuro-fuzzy classifier. Noisy speech sentences from NOIZEUS and AURORA databases are taken for the study. Feature extraction is implemented through modifications in amplitude magnitude spectrograms. A four class neuro-fuzzy classifier splits the noisy speech samples into noise-only part, signal only part, more noise-less signal part, and more signal-less noise part of the time-frequency units. Appropriate weights are applied in the enhancement phase. The enhanced speech sentence is evaluated using objective measures. An analysis of the performance of the Neuro-Fuzzy 4 (NF 4) classifier is done. A comparison of the performance of the classifier with other conventional techniques is done for various noises at different noise levels. It is observed that the numerical values of the measures obtained are better when compared to the others. An overall comparison of the performance of the NF 4 classifier is done and it is inferred that NF4 outperforms the other techniques in speech enhancement.

scholarly journals A radio burst detection method based on the Hough transform

2020 ◽  
Vol 494 (2) ◽  
pp. 1994-2003
Author(s):  
Shifan Zuo ◽  
Xuelei Chen
Keyword(s):  
Parameter Space ◽  
Radio Burst ◽  
Hough Transform ◽  
Single Point ◽  
Fast Method ◽  
Noise Levels ◽  
Time Frequency ◽  
Straight Line ◽  
Fast Radio Burst ◽  
Truncation Threshold

ABSTRACT We present a simple and fast method for incoherent dedispersion and fast radio burst (FRB) detection based on the Hough transform, which is widely used for feature extraction in image analysis. The Hough transform maps a point in the time–frequency data to a straight line in the parameter space and points on the same dispersed f−2 curve to a bundle of lines all crossing at the same point, thus the curve is transformed to a single point in the parameter space, enabling an easier way for the detection of radio burst. By choosing an appropriate truncation threshold, in a reasonably radio quiet environment, i.e. with radio frequency interferences present but not dominant, the computing speed of the method is very fast. Using simulation data of different noise levels, we studied how the detected peak varies with different truncation thresholds. We also tested the method with some real pulsar and FRB data.

scholarly journals A Fault Diagnosis Method of Rolling Mill Bearing at Low Frequency and Overload Condition Based on Integration of EEMD and GA-DBN

2021 ◽  
Vol 2021 ◽  
pp. 1-13
Author(s):  
Jiang Ji ◽  
Chen Zhao ◽  
Yongqin Wang ◽  
Tuanmin Zhao ◽  
Xinyou Zhang
Keyword(s):  
Fault Diagnosis ◽  
Frequency Domain ◽  
Rolling Mill ◽  
Rolling Force ◽  
Low Frequency ◽  
Frequency Domain Analysis ◽  
Fault Classification ◽  
Low Speed ◽  
Time Frequency ◽  
Diagnosis Method

To solve the problems of difficult fault signal recognition and poor diagnosis effect of different damage in the same position in rolling mill bearing at low speed, a fault diagnosis method of rolling mill bearing based on integration of EEMD and DBN was proposed. The vibration signals in horizontal, axial, and vertical directions were decomposed and reconstructed by EEMD, and frequency domain analysis was carried out by using refined spectrum. Then, the signal's time-frequency domain index, rolling force, and torque component feature vector were input into genetic algorithm (GA) to optimize DBN model classification. In order to verify the effectiveness of the method, the experimental study was carried out on the two-high experimental rolling mill. The results show that EEMD combined with thinning spectrum can solve the problem of fault feature extraction well. Compared with time-frequency domain characteristic input, the prediction accuracy of DBN model is obviously improved. And the accuracy of GA-DBN model is higher, and the accuracy is 98.3%, and the time taken to diagnose is significantly reduced. Finally, the fault classification of different parts of bearings and the fault diagnosis of different damage in the same part are realized, which provides a good theoretical basis for the fault diagnosis of low-speed bearings and has important engineering significance.

scholarly journals OPTIMISING THE SMOOTHNESS AND ACCURACY OF MOVING AVERAGE FOR STOCK PRICE DATA

2018 ◽  
Vol 24 (3) ◽  
pp. 984-1003 ◽  
Author(s):  
Aistis RAUDYS ◽  
Židrina PABARŠKAITĖ
Keyword(s):  
Time Series ◽  
Real World ◽  
Stock Price ◽  
Weather Forecasting ◽  
Moving Average ◽  
New Method ◽  
Traffic Prediction ◽  
Real World Data ◽  
Price Series ◽  
Unseen Data

Smoothing time series allows removing noise. Moving averages are used in finance to smooth stock price series and forecast trend direction. We propose optimised custom moving average that is the most suitable for stock time series smoothing. Suitability criteria are defined by smoothness and accuracy. Previous research focused only on one of the two criteria in isolation. We define this as multi-criteria Pareto optimisation problem and compare the proposed method to the five most popular moving average methods on synthetic and real world stock data. The comparison was performed using unseen data. The new method outperforms other methods in 99.5% of cases on synthetic and in 91% on real world data. The method allows better time series smoothing with the same level of accuracy as traditional methods, or better accuracy with the same smoothness. Weights optimised on one stock are very similar to weights optimised for any other stock and can be used interchangeably. Traders can use the new method to detect trends earlier and increase the profitability of their strategies. The concept is also applicable to sensors, weather forecasting, and traffic prediction where both the smoothness and accuracy of the filtered signal are important.

scholarly journals Coherence and time-frequency analysis of impulse voltage and current measurements

2017 ◽  
Vol 26 (2) ◽  
pp. 118
Author(s):  
Jelena Dikun ◽  
Emel Onal
Keyword(s):  
Frequency Analysis ◽  
High Voltage ◽  
Electromagnetic Interference ◽  
Digital Processing ◽  
Noise Sources ◽  
Time Frequency Analysis ◽  
Time Frequency ◽  
Interference Level ◽  
Impulse Voltage ◽  
Short Time

The aim of this paper is to point out the advantages of the use of the time-frequency analysis in the digital processing of waveforms recorded in high voltage impulse tests. Impulse voltage tests are essential to inspect and test insulation integrity of high voltage apparatus. On the other hand, generated impulse currents are used for different test applications such as investigation of high current effects, electromagnetic interference (EMI) testing, etc. Obtained voltage and current waveforms usually have some sort of interferences originated from the different sources. These interferences have to be removed from the original impulse data in order to evaluate the waveform characteristics precisely. When the interference level is high enough, it might not be possible to distinguish signal parameters from the recorded data. Conventional filtering methods cannot be useful for some interference like white noise. In that case, time-frequency filtering methods might be necessary. In this study, the wavelet analysis, which is a powerful time-frequency signal processing tool, is used to recognize the noise of impulse current and voltage data. Thus, the noise sources can be determined by short time Fourier Transform, and a coherence approach is used to determine the bandwidth of noises.

scholarly journals A Unified Speech Enhancement System Based on Neural Beamforming With Parabolic Reflector

2020 ◽  
Vol 10 (7) ◽  
pp. 2218
Author(s):  
Tao Zhang ◽  
Yanzhang Geng ◽  
Jianhong Sun ◽  
Chen Jiao ◽  
Biyun Ding
Keyword(s):  
Speech Enhancement ◽  
Background Noise ◽  
Signal To Noise Ratio ◽  
Noise Levels ◽  
Parabolic Reflector ◽  
Time Frequency ◽  
Low Snr ◽  
Objective Metrics ◽  
Babble Noise ◽  
Fusion Framework

This paper presents a unified speech enhancement system to remove both background noise and interfering speech in serious noise environments by jointly utilizing the parabolic reflector model and neural beamformer. First, the amplification property of paraboloid is discussed, which significantly improves the Signal-to-Noise Ratio (SNR) of a desired signal. Therefore, an appropriate paraboloid channel is analyzed and designed through the boundary element method. On the other hand, a time-frequency masking approach and a mask-based beamforming approach are discussed and incorporated in an enhancement system. It is worth noticing that signals provided by the paraboloid and the beamformer are exactly complementary. Finally, these signals are employed in a learning-based fusion framework to further improve the system performance in low SNR environments. Experiments demonstrate that our system is effective and robust in five different noisy conditions (speech interfered with factory, pink, destroyer engine, volvo, and babble noise), as well as in different noise levels. Compared with the original noisy speech, significant average objective metrics improvements are about Δ STOI = 0.28, Δ PESQ = 1.31, Δ fwSegSNR = 11.9.

Sign in / Sign up

Export Citation Format

Share Document