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.

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.

No abstract

The purpose of this study is to characterize fracture modes in a concrete structure using an acoustic emission (AE) technique and a data-driven approach. To clarify the damage fracture process, the specimens, which are of reinforced concrete (RC) beams, undergo four-point bending tests. During bending tests, impulses occurring in the AE signals are automatically detected using a constant false-alarm rate (CFAR) algorithm. For each detected impulse, its acoustic emission parameters such as counts, duration, amplitude, risetime, energy, RA, AF are calculated and studied. The mean and standard deviation values of each of these parameters are computed in every 1-second AE signal and are considered as features demonstrating the damage status of concrete structures. The results revealed that as the damage level in concrete structures grows, these features also change accordingly which can be used to categorize the damage fracture stages. The study also carries out experiments to validate the efficiency of the proposed approaches in terms of visual and qualitative evaluations. Experimental results show that the proposed characterizing model is promising and outstanding with the classification performance in the experimental environment of over 82%.

No abstract

No abstract

No abstract

This paper proposes a reliable fault detection model for a pressure vessel under low pressure conditions. To improve the diagnostic performance, signals of different vessel health conditions are purified by eliminating noise so that signals of different categories are much more distinguishable. This de-noising technique uses a blind source separation (BSS) technique in which an initial noise-contaminated signal is separated into constituent sources. These individual sources are either device status-characterizing signal sources or interfering sources. Noise is removed and an unimpaired signal is regenerated from the characteristic sources. An abundant pool of heterogenous features are extracted from a refined signal to avoid omitting important fault-related properties. However, some of these fault features may be redundant and irrelevant, and they are likely to cause classification performance degradation. To choose the most discriminative fault-signatures from a plentiful feature set as well as reduce the dimensions of the feature input, this study proposes a new feature selection algorithm that associates a genetic algorithm (GA)-based discriminative feature analysis to a k-nearest neighbors (k-NN) classifier. The efficacy of involved techniques and the overall fault diagnostic model is examined in terms of visual and qualitative evaluations. Experimental results illustrated in this study justify that the proposed fault detection model is promising and outstanding compared to other state-of-the-art counterparts.