A statistical comparison of neuroclassifiers and feature selection methods for gearbox fault diagnosis under realistic conditions

Pacheco, Fannia; Oliveira, José Valente de; Sánchez, René–Vinicio; Cerrada, Mariela; Cabrera, Diego; Li, Chuan; Zurita, Grover; Artés, Mariano

doi:10.1016/j.neucom.2016.02.028

Cited by 57 publications

(25 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…A ranking can be obtained by counting the number of times that a method was a winner in the pairwise comparison. See [46,47] for further details. Here, we are using the usual α = 0.05 and the Wilcoxon test as posthoc.…”

Section: On Different Scenarios For Dealing With the Unbalanced Data Setmentioning

confidence: 99%

Exploiting Generative Adversarial Networks as an Oversampling Method for Fault Diagnosis of an Industrial Robotic Manipulator

Cabrera

Sánchez

et al. 2020

Applied Sciences

Self Cite

View full text Add to dashboard Cite

Data-driven machine learning techniques play an important role in fault diagnosis, safety, and maintenance of the industrial robotic manipulator. However, these methods require data that, more often that not, are hard to obtain, especially data collected from fault condition states and, without enough and appropriated (balanced) data, no acceptable performance should be expected. Generative adversarial networks (GAN) are receiving a significant interest, especially in the image analysis field due to their outstanding generative capabilities. This paper investigates whether or not GAN can be used as an oversampling tool to compensate for an unbalanced data set in an industrial manipulator fault diagnosis task. A comprehensive empirical analysis is performed taking into account six different scenarios for mitigating the unbalanced data, including classical under and oversampling (SMOTE) methods. In all of these, a wavelet packet transform is used for feature generation while a random forest is used for fault classification. Aspects such as loss functions, learning curves, random input distributions, data shuffling, and initial conditions were also considered. A non-parametric statistical test of hypotheses reveals that all GAN based fault-diagnosis outperforms both under and oversampling classical methods while, within GAN based methods, an average accuracy difference as high as 1.68% can be achieved.

show abstract

Section: On Different Scenarios For Dealing With the Unbalanced Data Setmentioning

confidence: 99%

Exploiting Generative Adversarial Networks as an Oversampling Method for Fault Diagnosis of an Industrial Robotic Manipulator

Cabrera

Sánchez

et al. 2020

Applied Sciences

Self Cite

View full text Add to dashboard Cite

show abstract

“…Different methods for the feature selection have been mentioned in the published references such as multidimensional scaling [76], independent component analysis [77], latent semantic indexing [78], and partial least square [79], and PCA [80]. The related literature review showed that PCA is one of the best statistical techniques for feature reduction [81][82][83][84][85]. In this study, PCA was used to extract the principal components (PCs), to be used as predictors in order to make an ANN model more effective in predicting the energy output of Iranian tea production.…”

Section: Principal Component Analysismentioning

confidence: 99%

Integration of principal component analysis and artificial neural networks to more effectively predict agricultural energy flows

Nikkhah

Rohani

Rosentrater

et al. 2019

Env Prog and Sustain Energy

View full text Add to dashboard Cite

There are some studies regarding the prediction of agricultural energy flows using artificial neural networks (ANNs). These models are quite sensitive to correlations amongst inputs, and, there are often strong correlations amongst energy inputs for agricultural systems. One potential method to remediate this problem is to use principal component analysis (PCA). Therefore, the purpose of this research was to predict energy flows for a specific agricultural system (Iranian tea production) via a novel methodology based on ANNs, and using principal components as model inputs, not raw data. PCA results showed that the first and second components could account for more than 99% of variation in the data, thus the dimensions of the data set could be decreased from six to two for the prediction of energy flows for Iranian tea production. Using these principal components as inputs, an ANN model with 2–15–1 structure was determined to be optimal for energy flow modeling of this system. To conclude, the results of this study highlighted that the use of PC as ANN inputs improved ANN model prediction through reducing its complexity and eliminating data colinearity. Many agricultural systems could benefit from using this methodology for energy modeling. © 2019 American Institute of Chemical Engineers Environ Prog, 38:e13130, 2019

show abstract

“…With the tremendous deployment of different kinds of sensors and actuators, the Internet of Things (IoT) emerges as an advanced method to connect devices and collect the status data [1]. Aided by the use of a large amount of operation data, the data-driven fault diagnosis is considered as a modern technique in Industry 4.0 and has become a research hotspot in recent years [2][3][4][5]. In the area of urban rail transit, the significant increase of the line mileage and the passenger throughput leads to a high capacity utilization of the existing infrastructure [6].…”

Section: Introductionmentioning

confidence: 99%

Fault Diagnosis of Railway Point Machines Using the Locally Connected Autoencoder

et al. 2019

View full text Add to dashboard Cite

Data-driven fault diagnosis is considered a modern technique in Industry 4.0. In the area of urban rail transit, researchers focus on the fault diagnosis of railway point machines as failures of the point machine may cause serious accidents, such as the derailment of a train, leading to significant personnel and property loss. This paper presents a novel data-driven fault diagnosis scheme for railway point machines using current signals. Different from any handcrafted feature extraction approach, the proposed scheme employs a locally connected autoencoder to automatically capture high-order features. To enhance the temporal characteristic, the current signals are segmented and blended into some subsequences. These subsequences are then fed to the proposed autoencoder. With the help of a weighting strategy, the seized features are weight averaged into a final representation. At last, different from the existing classification methods, we employ the local outlier factor algorithm to solve the fault diagnosis problem without any training steps, as the accurate data labels that indicate a healthy or unhealthy state are difficult to acquire. To verify the effectiveness of the proposed fault diagnosis scheme, a fault dataset termed “Cu-3300” is created by collecting 3300 in-field current signals. Using Cu-3300, we perform comprehensive analysis to demonstrate that the proposed scheme outperforms the existing methods. We have made the dataset Cu-3300 and the code file freely accessible as open source files. To the best of our knowledge, the dataset Cu-3300 is the first open source dataset in the area of railway point machines and our conducted research is the first to investigate the use of autoencoders for fault diagnosis of point machines.

show abstract

A statistical comparison of neuroclassifiers and feature selection methods for gearbox fault diagnosis under realistic conditions

Cited by 57 publications

References 33 publications

Exploiting Generative Adversarial Networks as an Oversampling Method for Fault Diagnosis of an Industrial Robotic Manipulator

Exploiting Generative Adversarial Networks as an Oversampling Method for Fault Diagnosis of an Industrial Robotic Manipulator

Integration of principal component analysis and artificial neural networks to more effectively predict agricultural energy flows

Fault Diagnosis of Railway Point Machines Using the Locally Connected Autoencoder

Contact Info

Product

Resources

About