Jaeyoung Kim

Time-Varying and Multiresolution Envelope Analysis and Discriminative Feature Analysis for Bearing Fault Diagnosis

This paper presents a reliable fault diagnosis methodology for various single and multiple combined defects of low-speed rolling element bearings. This method temporally partitions an acoustic emission (AE) signal and selects a portion of the signal, which contains intrinsic information about the bearing failures. This paper then performs frequency analysis for the selected time-domain AE signal by using multilevel finite-impulse response filter banks to obtain the most informative subband signals involving abnormal symptoms of the bearing defects. It does this by using a 2-D visualization tool that represents the percentage of the Gaussian-mixture-model-based residual component-to-defect component ratios via time-varying and multiresolution envelope analysis (TVMREA). Then, fault signatures in the time and frequency domains are extracted in the informative subband signals. Since all the extracted fault features may not be equally useful for diagnosis, the proposed genetic algorithm (GA)-based discriminative feature analysis (GADFA) selects the most discriminative subset of fault signatures. In experiments, single and multiple combined bearing defects under various conditions are used to validate the effectiveness of this fault diagnosis scheme using TVMREA and GADFA. Experimental results indicate that this reliable fault diagnosis methodology accurately identifies bearing failure type across a variety of conditions. In addition, GADFA outperforms other state-of-the-art feature analysis techniques, yielding 7.3%-46.6% performance improvements in average classification accuracy.

A Hybrid Feature Selection Scheme for Reducing Diagnostic Performance Deterioration Caused by Outliers in Data-Driven Diagnostics

In practice, outliers, defined as data points that are distant from the other agglomerated data points in the same class, can seriously degrade diagnostic performance. To reduce diagnostic performance deterioration caused by outliers in data-driven diagnostics, an outlier-insensitive hybrid feature selection (OIHFS) methodology is developed to assess feature subset quality. In addition, a new feature evaluation metric is created as the ratio of the intraclass compactness to the interclass separability estimated by understanding the relationship between data points and outliers. The efficacy of the developed methodology is verified with a fault diagnosis application by identifying defect-free and defective rolling element bearings under various conditions.

High-Performance and Energy-Efficient Fault Diagnosis Using Effective Envelope Analysis and Denoising on a General-Purpose Graphics Processing Unit

This paper proposes an effective envelope analysis-based methodology for machinery condition monitoring and validates its efficacy by identifying bearing failures with 1-s acoustic emission (AE) signals sampled at 1 MHz. The proposed condition monitoring methodology of low-speed bearings consists of denoising to improve the signal-noise ratio of the acquired AE signal by employing a soft-thresholding technique with adaptively estimated positive and negative noise levels and an effective envelope analysis to detect the periodic impacts of the AE signals inherent in bearing defects by utilizing a 2-D visualization technique based on the improved residual frequency component-to-peak ratios. Despite the fact that the proposed method shows satisfactory performance for bearing condition monitoring, its computational complexity limits its use in real-time applications. To improve the performance and reduce the energy consumption of the proposed method, this paper proposes an efficient parallel implementation of the proposed method on a general-purpose graphics processing unit (GPGPU) by exploiting the memory hierarchy and the massive parallelism inherent in the proposed method. Experimental results indicate that the proposed GPGPU-based approach achieves an at least 68.9× speed improvement compared to the same sequential implementation on well-known Texas Instruments digital signal processors (TI DSPs). In addition, the proposed GPGPU approach reduces the energy consumption by at least 66% compared to TI DSPs.

An FPGA-Based Multicore System for Real-Time Bearing Fault Diagnosis Using Ultrasampling Rate AE Signals

The demand for online fault diagnosis has recently increased in order to prevent severe unexpected failures in machinery. To address this issue, this paper first proposes a comprehensive bearing fault diagnosis algorithm, which consists of fault signature extraction through time-frequency analysis and one-against-all multiclass support vector machines in order to make reliable decisions. In addition, acoustic emission (AE) signals sampled at 1 MHz are used for the early identification of bearing failures. Despite the fact that the proposed fault diagnosis methodology shows satisfactory classification accuracy, its computation complexity limits its use in real-time applications. Therefore, this paper also presents a high-performance multicore architecture, including 64 processing elements operating at 50 MHz in a Xilinx Virtex-7 field-programmable gate array device to support online fault diagnosis. The experimental results indicate that the multicore approach executes 1339.3x and 1293.1x faster than the high-performance Texas Instrument (TI) TMS320C6713 and TMS320C6748 digital signal processors (DSPs), respectively, by exploiting the massive parallelism inherent in the bearing fault diagnosis algorithm. In addition, the multicore approach outperforms the equivalent sequential approach that runs on the TI DSPs by substantially reducing the energy consumption.

Envelope analysis with a genetic algorithm-based adaptive filter bank for bearing fault detection

This paper proposes a fault detection methodology for bearings using envelope analysis with a genetic algorithm (GA)-based adaptive filter bank. Although a bandpass filter cooperates with envelope analysis for early identification of bearing defects, no general consensus has been reached as to which passband is optimal. This study explores the impact of various passbands specified by the GA in terms of a residual frequency components-to-defect frequency components ratio, which evaluates the degree of defectiveness in bearings and finally outputs an optimal passband for reliable bearing fault detection.

Implementation of image processing and augmented reality programs for smart mobile device

We implemented real-time image processing program using OpenCV library for Apples iPhone4 smart mobile phone. Our image processing program can do various operations such as thresholding, adaptive thresholding, edge detection and contour detection. Convenient user interface was developed using Objective-C. Also, we implemented augmented reality program on iPhone4. ARToolKitPlus by Wagner was analyzed for each stage of the library before the implementation. Also, we experimented on augmented reality with smart phone by using VRToolKit which is implemented by Loulier. We confirmed that our created earth model with OpenGL ES is augmented smoothly in real-time. This research will be a basis for developing smart phone programs such as face recognition, marker recognition and augmented reality program.

Bearing Fault Diagnosis under Variable Speed Using Convolutional Neural Networks and the Stochastic Diagonal Levenberg-Marquardt Algorithm

This paper presents a novel method for diagnosing incipient bearing defects under variable operating speeds using convolutional neural networks (CNNs) trained via the stochastic diagonal Levenberg-Marquardt (S-DLM) algorithm. The CNNs utilize the spectral energy maps (SEMs) of the acoustic emission (AE) signals as inputs and automatically learn the optimal features, which yield the best discriminative models for diagnosing incipient bearing defects under variable operating speeds. The SEMs are two-dimensional maps that show the distribution of energy across different bands of the AE spectrum. It is hypothesized that the variation of a bearing’s speed would not alter the overall shape of the AE spectrum rather, it may only scale and translate it. Thus, at different speeds, the same defect would yield SEMs that are scaled and shifted versions of each other. This hypothesis is confirmed by the experimental results, where CNNs trained using the S-DLM algorithm yield significantly better diagnostic performance under variable operating speeds compared to existing methods. In this work, the performance of different training algorithms is also evaluated to select the best training algorithm for the CNNs. The proposed method is used to diagnose both single and compound defects at six different operating speeds.

Reliable bearing fault diagnosis using Bayesian inference-based multi-class support vector machines

This letter presents a multi-fault diagnosis scheme for bearings using hybrid features extracted from their acoustic emissions and a Bayesian inference-based one-against-all support vector machine (Bayesian OAASVM) for multi-class classification. The Bayesian OAASVM, which is a standard multi-class extension of the binary support vector machine, results in ambiguously labeled regions in the input space that degrade its classification performance. The proposed Bayesian OAASVM formulates the feature space as an appropriate Gaussian process prior, interprets the decision value of the Bayesian OAASVM as a maximum a posteriori evidence function, and uses Bayesian inference to label unknown samples.

Incipient fault diagnosis in bearings under variable speed conditions using multiresolution analysis and a weighted committee machine

Incipient defects in bearings are traditionally diagnosed either by developing discriminative models for features that are extracted from raw acoustic emission (AE) signals, or by detecting peaks at characteristic defect frequencies in the envelope power spectrum of the AE signals. Under variable speed conditions, however, such methods do not yield the best results. This letter proposes a technique for diagnosing incipient bearing defects under variable speed conditions, by extracting features from different sub-bands of the inherently non-stationary AE signal, and then classifying bearing defects using a weighted committee machine, which is an ensemble of support vector machines and artificial neural networks. The proposed method also improves the generalization performance of the neural networks to enhance their classification accuracy, particularly with limited training data.

A Massively Parallel Approach to Real-Time Bearing Fault Detection Using Sub-Band Analysis on an FPGA-Based Multicore System

The fact that rolling element bearing faults have an amplitude-modulating effect on their characteristic frequencies calls for sub-band analysis to determine an optimal sub-band signal that contains intrinsic information about bearing faults. In this regard, it is significant to accurately assess the presence of a bearings abnormal symptoms. Hence, a bearing abnormality index (BAI) that properly quantifies how much information a sub-band signal contains about bearing faults is presented. Additionally, to facilitate real-time sub-band analysis based on the BAI, a massively parallel approach is introduced, where the approach involves the use of the multicore system. Likewise, the multicore system supports high-performance computing by exploiting 128 processing elements operating at 200 MHz in a Xilinx Virtex-7 field-programmable gate array (FPGA) device.