Prediction of Failure in Lubricated Surfaces Using Acoustic Time–Frequency Features and Random Forest Algorithm

Shevchik, Sergey; Saeidi, Fatemeh; Meylan, Bastian; Wasmer, Kilian

doi:10.1109/tii.2016.2635082

Cited by 61 publications

(40 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Although Zhang et al [30] proposed a few-shot learning strategy for bearing fault diagnosis, since the training and testing sets are of the same class, its essence is still a variety of transfer learning methods based on prior knowledge and fine-tuning. 6. Figure 6a,b shows healthy bearings, Figure 6c,d shows bearings from the artificial damage set, and Figure 6e,f from the natural damage dataset.…”

Section: Case Studymentioning

confidence: 99%

“…Therefore, accurate prediction and diagnosis of various bearing failures in real industrial scenarios is of great significance. In the past few years, a large number of traditional signal processing and machine learning methods have been applied to bearing fault detection, including wavelet transform (WT), Fourier transform, empirical mode decomposition (EMD) [1,2], principal component analysis (PCA) [3], SVM [4], k-nearest neighbor [5], and random forest [6]. Ren [7] proposed a 3-D waterfall spectrum in combination with reassigned wavelet scalogram method to solve non-linear and non-stationary vibration signal, while Yan [8] proposed a novel multiscale morphology analysis method, which can preserve signal details and has a good performance in detecting the defects in bearing.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Few-Shot Rolling Bearing Fault Diagnosis with Metric-Based Meta Learning

Wang

Kong

et al. 2020

Sensors

View full text Add to dashboard Cite

Fault diagnosis methods based on deep learning and big data have achieved good results on rotating machinery. However, the conventional deep learning method of bearing fault diagnosis is mostly based on laboratory artificial simulation data, and there is an error with actual fault data, which will reduce the generalization performance of the deep learning method. In addition, labeled data are very precious in real industrial environment. Due to expensive equipment and personnel safety issues, it is difficult to obtain a large amount of high-quality fault labeling data. Therefore, in this paper, we propose a metric-based meta-learning method named Reinforce Relation Network (RRN) for diagnosing bearing faults with few-shot samples. In the proposed method, a 1D convolution neural network is used to extract fault features, and a metric learner is used to predict the similarity between samples under different transfer conditions. Label smoothing and the Adabound algorithm are utilized to further improve the performance of network classification. The performance of the proposed method is verified on a dataset which contains artificial damage and natural damage data. The comparison studies with other methods demonstrate the superiority of the proposed method in the few-shot scenario.

show abstract

Section: Case Studymentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Few-Shot Rolling Bearing Fault Diagnosis with Metric-Based Meta Learning

Wang

Kong

et al. 2020

Sensors

View full text Add to dashboard Cite

show abstract

“…The RF algorithm for data classification is a combination of simple but various predictors, which reduces the computational complexity with great ease and achieves better results (Zhu et al 2018) The RF algorithm is primarily modified version Bagging since it employs the regression tree process as the training algorithm (Shevchik et al 2017).…”

Section: Random Forest Algorithmmentioning

confidence: 99%

Seizure episodes detection via smart medical sensing system

Shah

Fan

Ren

et al. 2018

J Ambient Intell Human Comput

View full text Add to dashboard Cite

Cyber-physical systems (CPS) consist of seamless network of sensors and actuators integrated with physical processes related to human activities. The CPS exploits sensors and actuators to monitor and control different physical process that can affect the computations of the devices. This paper presents the monitoring of physical activities exploiting wireless devices as sensors used in medical cyber-physical systems. Patients undergoing epileptic seizures experience involuntary body movements such as jerking, muscle twitching, falling, and convulsions. The proposed method exploits S-Band sensing used in medical CPS that leverage wireless devices such as omni-directional antenna at the transmitter side, four-beam patch antenna at the receiver side, RF signal generator and vector signal analyzer that perform signal conditioning by providing amplitude and raw phase data. The method uses wireless monitoring and recording system for measurement and classification of a clinical condition (epileptic seizures) versus normal daily routine activities. The data acquired that are perturbations of the radio signal is analyzed as amplitude, phase information, and statistical models. Extracting the statistical features, we leverage various machine learning algorithms such as support vector machine, random forest, and K-nearest neighbor that classify the data to differentiate patient's various activities such as press-ups, walking, sitting, squatting, and seizure episodes. The performance parameters used in three machine learning algorithms are accuracy, precision, recall, Cohen's Kappa coefficient, and F-measure. The values obtained using five performance parameters provide the accuracy of more than 90%.

show abstract

“…In addition to this, it has been found that low and high frequencies are associated with different lubricant conditions [28]. Peak frequencies can vary depending on wear mechanism, with adhesive wear emission suggested to occur around 1.1 MHz and abrasive wear between 0.25-1 MHz [31,32], while stick slip results in dominant AE frequencies around 10 kHz [20].Time-frequency analysis can also be applied to AE signals, with evolving PSD spectra indicating different stages wear [19] and wavelet analysis detecting the onset of scuffing [33,34] and other wear states [35] [36].Reviewing this literature it is apparent that the instantaneous RMS value of the acquired AE signal has been the most widely used parameter to correlate with friction and wear [17-19, 22, 23, 37, 38]. However, recently questions have been raised regarding the suitability of relying on this approach alone [1,39].…”

mentioning

confidence: 99%

“…Time-frequency analysis can also be applied to AE signals, with evolving PSD spectra indicating different stages wear [19] and wavelet analysis detecting the onset of scuffing [33,34] and other wear states [35] [36].…”

mentioning

confidence: 99%

Using acoustic emission to characterize friction and wear in dry sliding steel contacts

Geng

Puhan

Reddyhoff

2019

Tribology International

View full text Add to dashboard Cite

Acoustic emission (AE) was recorded during tribological tests on 52100 steel specimens under different loads. AE signals were transformed to the frequency domain using a Fast Fourier Transform and parameters such as power, RMS amplitude, mean frequency, and energy were analyzed and compared with friction coefficient and wear volume measurements. Results show that certain acoustic frequencies reflect friction while others reflect wear. If frequencies are chosen optimally, AE and friction signals are highly correlated (Pearson coefficients >0.8). SEM and Raman analysis reveal how plastic deformation and oxide formation affect friction, wear and AE simultaneously. AE recordings contains more information than conventional friction and wear volume measurements and are more sensitive to changes in wear mechanism. This all demonstrates AE's potential as a tool to monitor tribological behavior.waveform can correlate with the quantity of wear particles, as does the AE energy [14]. Moreover, the amplitude and length of a train of AE waves can be associated with individual slip events during the stick-slip [15]. In addition to this, the AE event location can determin mode I crack propagation in fretting fatigue [16]. The instantaneous RMS amplitude has been found to correlate with CoF [17][18][19], distinguish between different stages in CoF evolution [20] and even predict CoF based on a power law relation [18]. AE RMS can also be sensitive to applied load, velocity and mechanical properties of sliding components [20]. The integrated AE RMS can be related to frictional work and wear under different sliding speeds [18] and loads [21] and can detect the sliding speed threshold for accelerated wear [18]. The integrated AE RMS [22,23] may also have a direct correlation with the wear volume and wear rate [24].In the frequency domain, AE parameters such as energy and median frequency (MDF) can be used, with the former being correlated to CoF [25,26] and wear volume [27] and the latter being correlated to CoF [26,28] and frictional work [29]. The amplitude of individual AE frequency components can also be excited by specific wear mechanisms, which can be determined based on power spectral density (PSD) and auto-covariance techniques [30]. In addition to this, it has been found that low and high frequencies are associated with different lubricant conditions [28]. Peak frequencies can vary depending on wear mechanism, with adhesive wear emission suggested to occur around 1.1 MHz and abrasive wear between 0.25-1 MHz [31,32], while stick slip results in dominant AE frequencies around 10 kHz [20].Time-frequency analysis can also be applied to AE signals, with evolving PSD spectra indicating different stages wear [19] and wavelet analysis detecting the onset of scuffing [33,34] and other wear states [35] [36].Reviewing this literature it is apparent that the instantaneous RMS value of the acquired AE signal has been the most widely used parameter to correlate with friction and wear [17-19, 22, 23, 37, 38]. However, recently questions h...

show abstract

Prediction of Failure in Lubricated Surfaces Using Acoustic Time–Frequency Features and Random Forest Algorithm

Cited by 61 publications

References 28 publications

Few-Shot Rolling Bearing Fault Diagnosis with Metric-Based Meta Learning

Few-Shot Rolling Bearing Fault Diagnosis with Metric-Based Meta Learning

Seizure episodes detection via smart medical sensing system

Using acoustic emission to characterize friction and wear in dry sliding steel contacts

Contact Info

Product

Resources

About