The sensitivity of various features that are characteristic of a machine defect may vary considerably under different operating conditions. Hence it is critical to devise a systematic feature selection scheme that provides guidance on choosing the most representative features for defect classification. This paper presents a feature selection scheme based on the principal component analysis (PCA) method. The effectiveness of the scheme was verified experimentally on a bearing test bed, using both supervised and unsupervised defect classification approaches. The objective of the study was to identify the severity level of bearing defects, where no a priori knowledge on the defect conditions was available. The proposed scheme has shown to provide more accurate defect classification with fewer feature inputs than using all features initially considered relevant. The result confirms its utility as an effective tool for machine health assessment.

PCA-based feature selection scheme for machine defect classification

The vibration/acoustic signals of rolling element bearings contain rich information of health conditions for bearing maintenance. However, due to the complexity of measured signals, the status information is hard to be well extracted. This paper explores the envelope analysis technique for fault diagnosis of rolling element bearings, and proposes a new method, called wavelet packet envelope manifold (WPEM) approach, to extract the intrinsic envelope structure for well identification of the specific characteristic frequency. The WPEM method performs a manifold learning algorithm on a high-dimensional subband envelope feature space without the optimal band selection. In particular, the proposed method can be implemented by the following procedures. First, the wavelet packet transform is carried out to decompose the vibration signal into a series of subband signals at particular time-frequency subspaces. Then, the envelopes of these signals are achieved by the Hilbert transform to construct a high-dimensional WPE matrix. Finally, the manifold learning is addressed on the WPE matrix to learn the new WPEM feature. The WPEM integrates the envelope information of different subband frequency contents in a nonlinear way and exhibits the merit of in-band noise suppression, which is superior to the traditional enveloping methods based on a selected frequency band. The effects of the parameters on the WPEM performance are also discussed in this paper. The effectiveness is confirmed by practical applications to bearing fault diagnosis.

Wavelet Packet Envelope Manifold for Fault Diagnosis of Rolling Element Bearings

First, an improved local mean decomposition (LMD) method called differential rational spline-based LMD (DRS) is developed for signal decomposition. Differential and integral operations are introduced in LMD, which can weaken the mode mixing problem. Meanwhile, an optimized rational spline interpolation is proposed to calculate the envelope functions aiming to reduce the large errors caused by moving average in the traditional LMD. A series of product functions (PFs) is obtained after the application of the proposed DRS-LMD. Then, Kullback–Leibler (K–L) divergence is adopted to select main PF components that contain most fault information. The machine fault can be easily identified from the amplitude spectrum of the selected PF component. The effectiveness of the proposed DRS-LMD and K–L strategy is tested on simulated vibration signals and experimental vibration signals. Results show that the proposed method can increase the decomposition accuracy of the signals and can be used to detect early faults on the gears and rolling bearings.

Early Fault Diagnosis of Rotating Machinery by Combining Differential Rational Spline-Based LMD and K–L Divergence

Failure modes, mechanisms and effect analysis on temperature redundant sensor stage

The increasing complexity of manufacturing machines and the continued demand for high productivity have led to growing applications of sensor networks to enable more reliable, timely, and comprehensive information gathering from the machines being monitored. An effective and efficient utilization of sensor networks requires new sensor designs that enable adaptive event-driven information gathering based on the condition of the machines, as well as a coordinated information distribution adjusted to the available communication bandwidth of the network. This paper investigates several fundamental aspects regarding the architectural design of a sensory node controller (SNOC). The SNOC is the key element in a large-scale sensor network that coordinates the operation of individual sensors and the communication among various sensing clusters to realize distributed intelligent sensing. A parametric SNOC design that dynamically adjusts the power supply and the data-acquisition procedure to reduce the overall energy consumption of the sensor network is presented. Considerations on both the hardware and software aspects of the design to achieve energy efficiency are described, and analytical formulations are derived. Simulation results for a sensor network consisting of 40 SNOCs, each coordinating eight physical sensors, have shown that the design is able to reduce the energy consumption by about 43%, as compared to traditional techniques. A prototype SNOC was designed and implemented, based on the platform of a commercially available microcontroller, and experimentally tested for its ability to dynamically adjust the power consumption. The study has provided a concrete input to the design optimization and experimental realization of an SNOC-based sensor network for machine-system monitoring.

Architectural design of a sensory node controller for optimized energy utilization in sensor networks

This paper presents the design, optimization, and implementation of a virtual instrument (VI), which is an essential part of an integrated bearing condition monitoring system. The VI is designed using the graphical programming language LabVIEW and is capable of performing on-line measurement functions, including data acquisition, display, and analyses in the time and frequency domains, as well as data archiving. The issues of data length selection and VI real-time capability have been investigated to optimize the VI operation and improve data-processing efficiency. The VI system presents an effective and user-friendly human-machine interface for on-line bearing condition monitoring, which is of critical importance to real-time fault diagnosis and intelligent manufacturing.

C. Wang

Papers

A virtual instrumentation system for integrated bearing condition monitoring