Hybrid Intelligent Predictive Maintenance Model for Multiclass Fault Classi
No Thumbnail Available
Date
2021
Journal Title
Journal ISSN
Volume Title
Publisher
Research Square
Abstract
Data recorded from monitoring the health condition of industrial equipment are often high-dimensional,
nonlinear, nonstationary and characterised by high levels of uncertainty. These factors limit the efficiency of
machine learning techniques to produce desirable results when developing effective fault classification
frameworks. This paper sought to propose a hybrid artificial intelligent predictive maintenance model based
on Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN),
Principal Component Analysis (PCA) and Least Squares Support Vector Machine (LSSVM) optimised by the combination of Coupled Simulated Annealing and Nelder-Mead Simplex optimisation algorithms
(ICEEMDAN-PCA-LSSVM). Here, ICEEMDAN was first employed as a denoising technique to decompose
signals into series of Intrinsic Mode Functions (IMFs) of which only relevant IMFs containing the relevant
fault features were retained for signal reconstruction. PCA was then employed as a dimension reduction
technique through which the resulting set of uncorrelated features extracted served as input for LSSVM for
classifying various fault types. The proposed technique is compared with three established methods (Linear
Discriminant Analysis (LDA), Support Vector Machine (SVM) and Artificial Neural Network (ANN)) with
multiclass classification capabilities. The various techniques were tested on an experimental UCI machine
learning benchmark data obtained from multi-sensors of a hydraulic test rig. The results from the analysis revealed that the proposed ICEEMDAN-PCA-LSSVM technique is versatile and outperformed all the
compared classifiers in terms of accuracy, error rate and other evaluation metrics considered. The proposed
hybrid technique drastically reduced the redundancies and the dimension of features, allowing for the efficient
consideration of relevant features for the enhancement of classification accuracy and convergence speed.
Description
This article is published by Research Square and is also available at DOI: https://doi.org/10.21203/rs.3.rs-600110/v1