Real-time fault diagnosis and trend prediction of rolling bearings based on resampling dynamic time warping and time-domain indicator analysis

A sudden fault in a rolling bearing (RB) results in a large amount of downtime, which increases the cost of operation and maintenance. In this paper, a real-time diagnosis and trend prediction method for RBs is proposed. In this method, a novel resampling dynamic time warping (RDTW) algorithm is presented and two new time-domain indicators (NTDIRs) called TALAP and TRCKT are defined, which can describe the wear degree and trend of an RB inner ring wear fault (IRWF). TALAP and TRCKT are proposed by comprehensively considering the stability and sensitivity of existing time-domain indicators (TDIRs). First, RDTW is used to align the healthy vibration signal with the fault vibration signal. Then, the residual signal that can be used to monitor the running condition is obtained. TALAP and TRCKT of the residual signal are calculated to judge the degree of wear. When the wear limit is reached, a fault alarm is sent out and the downtime needed for replacement can be accurately indicated. The experimental results show that the method can perform accurate diagnosis and trend prediction of inner ring wear faults of RBs.

Residual-guided In-loop Filter Using Convolution Neural Network

The block-based coding structure in the hybrid video coding framework inevitably introduces compression artifacts such as blocking, ringing, and so on. To compensate for those artifacts, extensive filtering techniques were proposed in the loop of video codecs, which are capable of boosting the subjective and objective qualities of reconstructed videos. Recently, neural network-based filters were presented with the power of deep learning from a large magnitude of data. Though the coding efficiency has been improved from traditional methods in High-Efficiency Video Coding (HEVC), the rich features and information generated by the compression pipeline have not been fully utilized in the design of neural networks. Therefore, in this article, we propose the Residual-Reconstruction-based Convolutional Neural Network (RRNet) to further improve the coding efficiency to its full extent, where the compression features induced from bitstream in form of prediction residual are fed into the network as an additional input to the reconstructed frame. In essence, the residual signal can provide valuable information about block partitions and can aid reconstruction of edge and texture regions in a picture. Thus, more adaptive parameters can be trained to handle different texture characteristics. The experimental results show that our proposed RRNet approach presents significant BD-rate savings compared to HEVC and the state-of-the-art CNN-based schemes, indicating that residual signal plays a significant role in enhancing video frame reconstruction.

A Novel Fault Detection and Identification Framework for Rotating Machinery Using Residual Current Spectrum

A novel framework of model-based fault detection and identification (MFDI) for induction motor (IM)-driven rotating machinery (RM) is proposed in this study. A data-driven subspace identification (SID) algorithm is employed to obtain the IM state-space model from the voltage and current signals in a quasi-steady-state condition. This study aims to improve the frequency–domain fault detection and identification (FDI) by replacing the current signal with a residual signal where a thresholding method is applied to the residual signal. Through the residual spectrum and threshold comparison, a binary decision is made to find fault signatures in the spectrum. The statistical Q-function is used to generate the fault frequency band to distinguish between the fault signature and the noise signature. The experiment in this study is performed on a wastewater pump in an existing industrial facility to verify the proposed FDI. Two faulty conditions with mathematically known and mathematically unknown faulty signatures are experimented with and diagnosed. The study results present that the residual spectrum demonstrated to be more sensitive to fault signatures compare to the current spectrum. The proposed FDI has successfully shown to identify the fault signatures even for the mathematically unknown faulty signatures.

Data-Driven State Prediction and Sensor Fault Diagnosis for Multi-Agent Systems with Application to a Twin Rotational Inverted Pendulum

When a multi-agent system is subjected to faults, it is necessary to detect and classify the faults in time. This paper is motivated to propose a data-driven state prediction and sensor fault classification technique. Firstly, neural network-based state prediction model is trained through historical input and output data of the system. Then, the trained model is implemented to the real-time system to predict the system state and output in absence of fault. By comparing the predicted healthy output and the measured output, which can be abnormal in case of sensor faults, a residual signal can be generated. When a sensor fault occurs, the residual signal exceeds the threshold, a fault classification technique is triggered to distinguish fault types. Finally, the designed data-driven state prediction and fault classification algorithms are verified through a twin rotational inverted pendulum system with leader-follower mechanism.

SVM-Based Bearing Anomaly Identification with Self-Tuning Network-Fuzzy Robust Proportional Multi Integral and Smart Autoregressive Model

In this paper, the combination of an indirect self-tuning observer, smart signal modeling, and machine learning-based classification is proposed for rolling element bearing (REB) anomaly identification. The proposed scheme has three main stages. In the first stage, the original signal is resampled, and the root mean square (RMS) signal is extracted from it. In the second stage, the normal resampled RMS signal is approximated using the AutoRegressive with eXternal Uncertainty (ARXU) technique. Moreover, the nonlinearity of the bearing signal is solved using the combination of the ARXU and the machine learning-based regression, which is called AMRXU. After signal modeling by AMRXU, the RMS resampled signal is estimated using a combination of the proportional multi-integral (PMI) technique, the variable structure (VS) Lyapunov technique, and a self-tuning network-fuzzy system (SNFS). Finally, in the third stage, the difference between the original signal and the estimated one is calculated to generate the residual signal. A machine learning-based classification technique is utilized to classify the residual signal. The Case Western Reserve University (CWRU) dataset is used to evaluate anomaly identification performance of the proposed scheme. Regarding the experimental results, the average accuracy for REB crack identification is 98.65%, 97.7%, 97.35%, and 97.67%, respectively, when the motor torque loads are 0-hp, 1-hp, 2-hp, and 3-hp.

The Multi-Model Large Ensemble Archive as a climate noise generator: opportunities and outlooks for Observational Large Ensemble construction

<p>The Multi-Model Large Ensemble Archive (MMLEA) is a collection of CMIP5-generation single model initial condition large ensembles (SMILEs) and thus provides estimates of internal variability from several independently developed coupled climate models. Work is underway to determine whether these simulations provide a range of historical regional climate variability suitable for statistically increasing the observed temperature sample.  Alternative sequences of historical temperature can be constructed by combining a forced signal with estimates of internal climate noise; prior studies have used the forced response from one SMILE in concert with observational noise resampling to form an “observational large ensemble” (McKinnon et al. 2018). Analogous to a SMILE, an observational large ensemble can be used to statistically contextualize monthly to half-yearly extreme events, such as the persistently mild Siberian winter of 2020, and to develop additional extended hot or cold spell storylines to explore in future projections of regional climate.</p><p>In this study, an alternative approach to constructing an observational large ensemble of European surface air temperature over the historical period (1950-2014), made possible by the MMLEA, is explored. Rather than relying on forced response and internal variability, components not well-defined in the single realization of observed climate, the constructed circulation analogue method of dynamical adjustment is employed to separate temperature anomalies related to atmospheric circulation (“dynamic noise") from a more thermodynamically driven residual signal. The approach is advantageous because it can be applied in a similar manner to single realizations from both models and observations. Here, dynamic noise is computed by dividing each of the seven CMIP5-generation SMILEs in half and empirically estimating the component of temperature associated with interannual sea level pressure variability in one half of the SMILE using circulation analogues from members in the other half. Because ensemble means can be computed in SMILEs, it is possible to use the relationship between unforced temperature and unforced sea level pressure anomalies to construct dynamic noise. In observations, weekly-averaged analogues are assessed as a means to increase the size of the analogue pool such that the separation between dynamic noise and thermodynamic residual signal occurs in a manner more similar to that computed in the SMILEs.</p><p>The extent to which dynamic noise fields from different SMILEs are distinguishable from each other and from observational estimates is determined via spectral and spatial pattern analyses. To avoid introducing regional model bias into dynamic noise estimates, a mosaic approach will be taken; noise estimates from different models are mosaiced such that observed statistical properties are maintained at each grid point of the European domain. Upon validation, SMILE-derived dynamic noise and observational thermodynamic residual signal estimates are combined into a 50-member European observational large ensemble and evaluated via a multi-month extreme temperature frequency metric against the observational large ensemble developed by McKinnon et al. (2018). Anomalously persistent hot and cold spells found in the European observational large ensemble are further compared to events in out-of-sample future projections of climate from the CMIP6 archive.</p>

Local diurnal variations of the geomagnetic field during magnetically quiet conditions

<p>The Earth’s magnetic field as measured from ground-based magnetometers is composed of a variety of fields generated by diverse sources, spanning a broad amplitude and frequency spectrum. Long-term variable sources induce smooth changes, whereas short-term variable sources are able to induce rapid spikes in the geomagnetic field. An important aspect of Space Weather research is to understand the contribution and impact of each of these sources. In particular, knowing the amplitude and frequency of steady-like sources, like diurnal variations, enables us to determine the impact of sudden and hazardous events such as solar storms. The basic approach to this challenge is to identify the quiet magnetic field information within the recorded time-varying signal.<br>In this work, we examine the variance of the magnetically quiet diurnal and semi-diurnal components of the geomagnetic field, as recorded by ground-based magnetic observatories of the INTERMAGNET network. These variations are extracted by applying appropriately designed digital filters on the geomagnetic field time series. The residual signal is analysed in terms of local time and seasonal variations for selected locations under quiet magnetic conditions. This approach allows us to evaluate the applicability of the introduced filtering method. The obtained results improve our understanding of the driving sources of quiet currents such as the Sq current and the variations of their distributions with respect to regular solar irradiance variations. They will also contribute to a better extraction and description of the remaining/residual signal related to solar wind stimuli (e.g. ICMEs, CIRs) causing magnetic storms.</p>

Intelligent fault diagnosis algorithm for fiber optic current transformer

Affected by environmental factors, the performance of fiber optic current transformer (FOCT) will deteriorate over a long period of time. Intelligent fault diagnosis algorithm of Long-Short Term Memory (LSTM) combing with Support Vector Machine (SVM) is an effective way to deal with FOCT failures. According to the characteristics of LSTM, a signal prediction model in FOCT based on LSTM is proposed by analyzing the historical data. The residual signal can be obtained by the prediction signal and the observed signal. Set the residual threshold to determine whether the FOCT has fault. With the residual signal characteristics, a fault diagnosis model based on SVM is established. By analyzing the residual signal and extracting features, the diagnostic network can realize the pattern recognition and system fault diagnosis. Experiments demonstrate that the drift deviation fault, the ratio deviation fault and the fixed deviation fault can be diagnosed with an accuracy of 94.5%.