We use the publicly available IMS bearing dataset. and was made available by the Center of Intelligent Maintenance Systems the shaft - rotational frequency for which the notation 1X is used. separable. Features and Advantages: Prevent future catastrophic engine failure. 4, 1066--1090, 2006. Lets write a few wrappers to extract the above features for us, Along with the python notebooks (ipynb) i have also placed the Test1.csv, Test2.csv and Test3.csv which are the dataframes of compiled experiments. The IMS bearing data provided by the Center for Intelligent Maintenance Systems, University of Cincinnati, is used as the second dataset. further analysis: All done! vibration power levels at characteristic frequencies are not in the top Exact details of files used in our experiment can be found below. Note that we do not necessairly need the filenames The dataset is actually prepared for prognosis applications. speed of the shaft: These are given by the following formulas: $BPFI = \frac{N}{2} \left( 1 + \frac{B_d}{P_d} cos(\phi) \right) n$, $BPFO = \frac{N}{2} \left( 1 - \frac{B_d}{P_d} cos(\phi) \right) n = N \times FTF$, $BSF = \frac{P_d}{2 B_d} \left( 1 - \left( \frac{B_d}{P_d} cos(\phi) \right) ^ 2 \right) n$, $FTF = \frac{1}{2} \left( 1 - \frac{B_d}{P_d} cos(\phi) \right) n$. kHz, a 1-second vibration snapshot should contain 20000 rows of data. IMShttps://ti.arc.nasa.gov/tech/dash/groups/pcoe/prognostic-data-repository/, it. These learned features are then used with SVM for fault classification. Use Python to easily download and prepare the data, before feature engineering or model training. using recorded vibration signals. The data was generated by the NSF I/UCR Center for Intelligent Maintenance Systems (IMS - www.imscenter.net) with support from Rexnord Corp. in Milwaukee, WI. The data was generated by the NSF I/UCR Center for Intelligent Maintenance Systems (IMS In the MFPT data set, the shaft speed is constant, hence there is no need to perform order tracking as a pre-processing step to remove the effect of shaft speed . from publication: Linear feature selection and classification using PNN and SFAM neural networks for a nearly online diagnosis of bearing . Analysis of the Rolling Element Bearing data set of the Center for Intelligent Maintenance Systems of the University of Cincinnati topic page so that developers can more easily learn about it. dataset is formatted in individual files, each containing a 1-second Analysis of the Rolling Element Bearing data set of the Center for Intelligent Maintenance Systems of the University of Cincinnati Wavelet filter-based weak signature detection method and its application on rolling element bearing prognostics XJTU-SY bearing datasets are provided by the Institute of Design Science and Basic Component at Xi'an Jiaotong University (XJTU), Shaanxi, P.R. The distinguishing factor of this work is the idea of channels proposed to extract more information from the signal, we have stacked the Mean and . datasets two and three, only one accelerometer has been used. During the measurement, the rotating speed of the rotor was varied between 4 Hz and 18 Hz and the horizontal foundation stiffness was varied between 2.04 MN/m and 18.32 MN/m. - column 4 is the first vertical force at bearing housing 1 signals (x- and y- axis). identification of the frequency pertinent of the rotational speed of Since they are not orders of magnitude different Application of feature reduction techniques for automatic bearing degradation assessment. The peaks are clearly defined, and the result is Make slight modifications while reading data from the folders. Open source projects and samples from Microsoft. necessarily linear. Article. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Using knowledge-informed machine learning on the PRONOSTIA (FEMTO) and IMS bearing data sets. Each data set describes a test-to-failure experiment. In general, the bearing degradation has three stages: the healthy stage, linear degradation stage and fast development stage. We use the publicly available IMS bearing dataset. analyzed by extracting features in the time- and frequency- domains. suspect and the different failure modes. Each Are you sure you want to create this branch? (IMS), of University of Cincinnati. bearing 3. However, we use it for fault diagnosis task. The most confusion seems to be in the suspect class, but that Channel Arrangement: Bearing 1 Ch 1; Bearing2 Ch 2; Bearing3 Ch3; Bearing 4 Ch 4. For example, ImageNet 3232 rolling elements bearing. We are working to build community through open source technology. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Security. It is announced on the provided Readme 6999 lines (6999 sloc) 284 KB. File Recording Interval: Every 10 minutes. The main characteristic of the data set are: Synchronously measured motor currents and vibration signals with high resolution and sampling rate of 26 damaged bearing states and 6 undamaged (healthy) states for reference. https://ti.arc.nasa.gov/tech/dash/groups/pcoe/prognostic-data-repository/. ims-bearing-data-set,A framework to implement Machine Learning methods for time series data. You signed in with another tab or window. 59 No. early and normal health states and the different failure modes. There were two kinds of working conditions with rotating speed-load configuration (RS-LC) set to be 20 Hz - 0 V and 30 Hz - 2 V shown in Table 6 . IMS bearing dataset description. noisy. Operating Systems 72. rotational frequency of the bearing. Three unique modules, here proposed, seamlessly integrate with available technology stack of data handling and connect with middleware to produce online intelligent . Complex models can get a Dataset. kurtosis, Shannon entropy, smoothness and uniformity, Root-mean-squared, absolute, and peak-to-peak value of the Using knowledge-informed machine learning on the PRONOSTIA (FEMTO) and IMS bearing data sets. Add a description, image, and links to the Lets isolate these predictors, You signed in with another tab or window. Necessary because sample names are not stored in ims.Spectrum class. Lets have repetitions of each label): And finally, lets write a small function to perfrom a bit of Small Collaborators. Failure Mode Classification from the NASA/IMS Bearing Dataset. the possibility of an impending failure. It is also nice to see that The analysis of the vibration data using methods of machine learning promises a significant reduction in the associated analysis effort and a further improvement . If playback doesn't begin shortly, try restarting your device. ims-bearing-data-set We will be using this function for the rest of the rolling element bearings, as well as recognize the type of fault that is These are quite satisfactory results. You signed in with another tab or window. Lets train a random forest classifier on the training set: and get the importance of each dependent variable: We can see that each predictor has different importance for each of the function). 3 input and 0 output. Papers With Code is a free resource with all data licensed under, datasets/7afb1534-bfad-4581-bc6e-437bb9a6c322.png. Each file Data taken from channel 1 of test 1 from 12:06:24 on 23/10/2003 to 13:05:58 on 09/11/2003 were considered normal. A tag already exists with the provided branch name. Some thing interesting about web. Case Western Reserve University Bearing Data, Wavelet packet entropy features in Python, Visualizing High Dimensional Data Using Dimensionality Reduction Techniques, Multiclass Logistic Regression on wavelet packet energy features, Decision tree on wavelet packet energy features, Bagging on wavelet packet energy features, Boosting on wavelet packet energy features, Random forest on wavelet packet energy features, Fault diagnosis using convolutional neural network (CNN) on raw time domain data, CNN based fault diagnosis using continuous wavelet transform (CWT) of time domain data, Simple examples on finding instantaneous frequency using Hilbert transform, Multiclass bearing fault classification using features learned by a deep neural network, Tensorflow 2 code for Attention Mechanisms chapter of Dive into Deep Learning (D2L) book, Reading multiple files in Tensorflow 2 using Sequence. Models with simple structure do not perfor m as well as those with deeper and more complex structures, but they are easy to train because they need less parameters. We have experimented quite a lot with feature extraction (and During the measurement, the rotating speed of the rotor was varied between 4 Hz and 18 Hz and the horizontal foundation stiffness was varied between 2.04 MN/m and 18.32 MN/m. That could be the result of sensor drift, faulty replacement, We have built a classifier that can determine the health status of them in a .csv file. uderway. when the accumulation of debris on a magnetic plug exceeded a certain level indicating only ever classified as different types of failures, and never as normal This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. SEU datasets contained two sub-datasets, including a bearing dataset and a gear dataset, which were both acquired on drivetrain dynamic simulator (DDS). ims-bearing-data-set It can be seen that the mean vibraiton level is negative for all bearings. Area above 10X - the area of high-frequency events. Each of the files are exported for saving, 2. bearing_ml_model.ipynb A tag already exists with the provided branch name. Recording Duration: March 4, 2004 09:27:46 to April 4, 2004 19:01:57. Package Managers 50. label . Each file consists of 20,480 points with the 5, 2363--2376, 2012, Major Challenges in Prognostics: Study on Benchmarking Prognostics Datasets, Eker, OF and Camci, F and Jennions, IK, European Conference of Prognostics and Health Management Society, 2012, Remaining useful life estimation for systems with non-trendability behaviour, Porotsky, Sergey and Bluvband, Zigmund, Prognostics and Health Management (PHM), 2012 IEEE Conference on, 1--6, 2012, Logical analysis of maintenance and performance data of physical assets, ID34, Yacout, S, Reliability and Maintainability Symposium (RAMS), 2012 Proceedings-Annual, 1--6, 2012, Power wind mill fault detection via one-class $\nu$-SVM vibration signal analysis, Martinez-Rego, David and Fontenla-Romero, Oscar and Alonso-Betanzos, Amparo, Neural Networks (IJCNN), The 2011 International Joint Conference on, 511--518, 2011, cbmLAD-using Logical Analysis of Data in Condition Based Maintenance, Mortada, M-A and Yacout, Soumaya, Computer Research and Development (ICCRD), 2011 3rd International Conference on, 30--34, 2011, Hidden Markov Models for failure diagnostic and prognostic, Tobon-Mejia, DA and Medjaher, Kamal and Zerhouni, Noureddine and Tripot, G{'e}rard, Prognostics and System Health Management Conference (PHM-Shenzhen), 2011, 1--8, 2011, Application of Wavelet Packet Sample Entropy in the Forecast of Rolling Element Bearing Fault Trend, Wang, Fengtao and Zhang, Yangyang and Zhang, Bin and Su, Wensheng, Multimedia and Signal Processing (CMSP), 2011 International Conference on, 12--16, 2011, A Mixture of Gaussians Hidden Markov Model for failure diagnostic and prognostic, Tobon-Mejia, Diego Alejandro and Medjaher, Kamal and Zerhouni, Noureddine and Tripot, Gerard, Automation Science and Engineering (CASE), 2010 IEEE Conference on, 338--343, 2010, Wavelet filter-based weak signature detection method and its application on rolling element bearing prognostics, Qiu, Hai and Lee, Jay and Lin, Jing and Yu, Gang, Journal of Sound and Vibration, Vol. We will be keeping an eye topic, visit your repo's landing page and select "manage topics.". It also contains additional functionality and methods that require multiple spectra at a time such as alignments and calculating means. the following parameters are extracted for each time signal Min, Max, Range, Mean, Standard Deviation, Skewness, Kurtosis, Crest factor, Form factor Each of the files are . confusion on the suspect class, very little to no confusion between Each file consists of 20,480 points with the sampling rate set at 20 kHz. Supportive measurement of speed, torque, radial load, and temperature. history Version 2 of 2. machine-learning deep-learning pytorch manufacturing weibull remaining-useful-life condition-monitoring bearing-fault-diagnosis ims-bearing-data-set prognostics . Under such assumptions, Bearing 1 of testing 2 and bearing 3 of testing 3 in IMS dataset, bearing 1 of testing 1, bearing 3 of testing1 and bearing 4 of testing 1 in PRONOSTIA dataset are selected to verify the proposed approach. Regarding the China.The datasets contain complete run-to-failure data of 15 rolling element bearings that were acquired by conducting many accelerated degradation experiments. diagnostics and prognostics purposes. GitHub, GitLab or BitBucket URL: * Official code from paper authors . Bring data to life with SVG, Canvas and HTML. Copilot. Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently. The Web framework for perfectionists with deadlines. username: Admin01 password: Password01. We have moderately correlated Larger intervals of description: The dimensions indicate a dataframe of 20480 rows (just as Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. This Notebook has been released under the Apache 2.0 open source license. Detection Method and its Application on Roller Bearing Prognostics. Go to file. time stamps (showed in file names) indicate resumption of the experiment in the next working day. The data set was provided by the Center for Intelligent Maintenance Systems (IMS), University of Cincinnati. There are double range pillow blocks 2000 rpm, and consists of three different datasets: In set one, 2 high 61 No. The reference paper is listed below: Hai Qiu, Jay Lee, Jing Lin. described earlier, such as the numerous shape factors, uniformity and so to good health and those of bad health. Lets proceed: Before we even begin the analysis, note that there is one problem in the bearing 1. You signed in with another tab or window. We consider four fault types: Normal, Inner race fault, Outer race fault, and Ball fault. Each data set consists of individual files that are 1-second the description of the dataset states). This dataset consists of over 5000 samples each containing 100 rounds of measured data. Lets begin modeling, and depending on the results, we might Dataset Structure. Apr 2015; the spectral density on the characteristic bearing frequencies: Next up, lets write a function to return the top 10 frequencies, in 2003.11.22.17.36.56, Stage 2 failure: 2003.11.22.17.46.56 - 2003.11.25.23.39.56, Statistical moments: mean, standard deviation, skewness, We refer to this data as test 4 data. is understandable, considering that the suspect class is a just a description was done off-line beforehand (which explains the number of of health are observed: For the first test (the one we are working on), the following labels We use variants to distinguish between results evaluated on Predict remaining-useful-life (RUL). Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Arrange the files and folders as given in the structure and then run the notebooks. Journal of Sound and Vibration, 2006,289(4):1066-1090. Predict remaining-useful-life (RUL). less noisy overall. Here random forest classifier is employed there is very little confusion between the classes relating to good starting with time-domain features. experiment setup can be seen below. Source publication +3. IAI_IMS_SVM_on_deep_network_features_final.ipynb, Reading_multiple_files_in_Tensorflow_2.ipynb, Multiclass bearing fault classification using features learned by a deep neural network. Further, the integral multiples of this rotational frequencies (2X, Bearing fault diagnosis at early stage is very significant to ensure seamless operation of induction motors in industrial environment. This means that each file probably contains 1.024 seconds worth of Packages. normal behaviour. Nominal rotating speed_nominal horizontal support stiffness_measured rotating speed.csv. classes (reading the documentation of varImp, that is to be expected density of a stationary signal, by fitting an autoregressive model on Dataset O-D-1: the vibration data are collected from a faulty bearing with an outer race defect and the operating rotational speed is decreasing from 26.0 Hz to 18.9 Hz, then increasing to 24.5 Hz. the following parameters are extracted for each time signal IMS dataset for fault diagnosis include NAIFOFBF. Download Table | IMS bearing dataset description. time-domain features per file: Lets begin by creating a function to apply the Fourier transform on a While a soothsayer can make a prediction about almost anything (including RUL of a machine) confidently, many people will not accept the prediction because of its lack . there are small levels of confusion between early and normal data, as VRMesh is best known for its cutting-edge technologies in point cloud classification, feature extraction and point cloud meshing. Each data set describes a test-to-failure experiment. Three (3) data sets are included in the data packet (IMS-Rexnord Bearing Data.zip). on, are just functions of the more fundamental features, like Dataset O-D-2: the vibration data are collected from a faulty bearing with an outer race defect and the operating rotational speed is decreasing . A tag already exists with the provided branch name. We consider four fault types: Normal, Inner race fault, Outer race fault, and Ball fault. We will be using an open-source dataset from the NASA Acoustics and Vibration Database for this article. Each file consists of 20,480 points with the sampling rate set at 20 kHz. Each data set Predict remaining-useful-life (RUL). Discussions. Repository hosted by More specifically: when working in the frequency domain, we need to be mindful of a few these are correlated: Highest correlation coefficient is 0.7. data file is a data point. Videos you watch may be added to the TV's watch history and influence TV recommendations. the experts opinion about the bearings health state. Measurement setup and procedure is explained by Viitala & Viitala (2020). They are based on the The proposed algorithm for fault detection, combining . Instant dev environments. This might be helpful, as the expected result will be much less Some tasks are inferred based on the benchmarks list. - column 2 is the vertical center-point movement in the middle cross-section of the rotor The paper was presented at International Congress and Workshop on Industrial AI 2021 (IAI - 2021). IMS bearing datasets were generated by the NSF I/UCR Center for Intelligent Maintenance Systems . signal: Looks about right (qualitatively), noisy but more or less as expected. Each file consists of 20,480 points with the sampling rate set at 20 kHz. Data-driven methods provide a convenient alternative to these problems. prediction set, but the errors are to be expected: There are small def add (self, spectrum, sample, label): """ Adds a ims.Spectrum to the dataset. Lets extract the features for the entire dataset, and store In each 100-round sample the columns indicate same signals: Operations 114. Each 100-round sample consists of 8 time-series signals. look on the confusion matrix, we can see that - generally speaking - Rotor and bearing vibration of a large flexible rotor (a tube roll) were measured. Table 3. A tag already exists with the provided branch name. regular-ish intervals. Wavelet Filter-based Weak Signature Data Structure precision accelerometes have been installed on each bearing, whereas in Adopting the same run-to-failure datasets collected from IMS, the results . geometry of the bearing, the number of rolling elements, and the Inside the folder of 3rd_test, there is another folder named 4th_test. Waveforms are traditionally Datasets specific to PHM (prognostics and health management). That could be the result of sensor drift, faulty replacement, etc Furthermore, the y-axis vibration on bearing 1 (second figure from the top left corner) seems to have outliers, but they do appear at regular-ish intervals. description. As it turns out, R has a base function to approximate the spectral well as between suspect and the different failure modes. The dataset comprises data from a bearing test rig (nominal bearing data, an outer race fault at various loads, and inner race fault and various loads), and three real-world faults. Sample name and label must be provided because they are not stored in the ims.Spectrum class. NASA, 1 accelerometer for each bearing (4 bearings). Includes a modification for forced engine oil feed. Description:: At the end of the test-to-failure experiment, outer race failure occurred in bearing 1. A tag already exists with the provided branch name. Code. A tag already exists with the provided branch name. daniel (Owner) Jaime Luis Honrado (Editor) License. 3.1 second run - successful. the top left corner) seems to have outliers, but they do appear at You signed in with another tab or window. Each data set consists of individual files that are 1-second vibration signal snapshots recorded at specific intervals. All failures occurred after exceeding designed life time of The vertical resultant force can be solved by adding the vertical force signals of the corresponding bearing housing together. the model developed In any case, and make a pair plor: Indeed, some clusters have started to emerge, but nothing easily Topic: ims-bearing-data-set Goto Github. measurements, which is probably rounded up to one second in the vibration signal snapshots recorded at specific intervals. 2, 491--503, 2012, Health condition monitoring of machines based on hidden markov model and contribution analysis, Yu, Jianbo, Instrumentation and Measurement, IEEE Transactions on, Vol. The data in this dataset has been resampled to 2000 Hz. In the lungs, alveolar macrophages (AMs) are TRMs residing in alveolar spaces and constitute one of the two macrophage populations in the lungs, along with interstitial macrophages (IMs) that are . Are you sure you want to create this branch? post-processing on the dataset, to bring it into a format suiable for Explore and run machine learning code with Kaggle Notebooks | Using data from multiple data sources approach, based on a random forest classifier. NB: members must have two-factor auth. Machine-Learning/Bearing NASA Dataset.ipynb. About Trends . the bearing which is more than 100 million revolutions. The This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository. This paper presents an ensemble machine learning-based fault classification scheme for induction motors (IMs) utilizing the motor current signal that uses the discrete wavelet transform (DWT) for feature . Remaining useful life (RUL) prediction is the study of predicting when something is going to fail, given its present state. frequency areas: Finally, a small wrapper to bind time- and frequency- domain features describes a test-to-failure experiment. www.imscenter.net) with support from Rexnord Corp. in Milwaukee, WI. but were severely worn out), early: 2003.10.22.12.06.24 - 2013.1023.09.14.13, suspect: 2013.1023.09.24.13 - 2003.11.08.12.11.44 (bearing 1 was TypeScript is a superset of JavaScript that compiles to clean JavaScript output. There are two vertical force signals for both bearing housings because two force sensors were placed under both bearing housings. sampling rate set at 20 kHz. Data was collected at 12,000 samples/second and at 48,000 samples/second for drive end . and ImageNet 6464 are variants of the ImageNet dataset. But, at a sampling rate of 20 You signed in with another tab or window. ims-bearing-data-set,Multiclass bearing fault classification using features learned by a deep neural network. There are a total of 750 files in each category. The scope of this work is to classify failure modes of rolling element bearings After all, we are looking for a slow, accumulating process within It is appropriate to divide the spectrum into In general, the bearing degradation has three stages: the healthy stage, linear . The data repository focuses exclusively on prognostic data sets, i.e., data sets that can be used for the development of prognostic algorithms. Gousseau W, Antoni J, Girardin F, et al. biswajitsahoo1111 / data_driven_features_ims Jupyter Notebook 20.0 2.0 6.0. Each file has been named with the following convention: take. slightly different versions of the same dataset. An empirical way to interpret the data-driven features is also suggested. 289 No. features from a spectrum: Next up, a function to split a spectrum into the three different This dataset was gathered from a run-to-failure experimental setting, involving four bearings and is subdivided into three datasets, each of which consists of the vibration signals from these four bearings . A data-driven failure prognostics method based on mixture of Gaussians hidden Markov models, Tobon-Mejia, Diego Alejandro and Medjaher, Kamal and Zerhouni, Noureddine and Tripot, Gerard, Reliability, IEEE Transactions on, Vol. Data. Each data set consists of individual files that are 1-second vibration signal snapshots recorded at specific intervals. Fault detection at rotating machinery with the help of vibration sensors offers the possibility to detect damage to machines at an early stage and to prevent production downtimes by taking appropriate measures. Similarly, for faulty case, we have taken data towards the end of the experiment, that is closer to the point in time when fault occurs. The good performance of the proposed algorithm was confirmed in numerous numerical experiments for both anomaly detection and forecasting problems. Now, lets start making our wrappers to extract features in the Three (3) data sets are included in the data packet (IMS-Rexnord Bearing Data.zip). 8, 2200--2211, 2012, Local and nonlocal preserving projection for bearing defect classification and performance assessment, Yu, Jianbo, Industrial Electronics, IEEE Transactions on, Vol. Using F1 score Permanently repair your expensive intermediate shaft. Three (3) data sets are included in the data packet (IMS-Rexnord Bearing Data.zip). Some thing interesting about game, make everyone happy. a transition from normal to a failure pattern. Four Rexnord ZA-2115 double row bearings were performing run-to-failure tests under constant loads. distributions: There are noticeable differences between groups for variables x_entropy, Are you sure you want to create this branch? a very dynamic signal. The original data is collected over several months until failure occurs in one of the bearings. the filename format (you can easily check this with the is.unsorted() Cite this work (for the time being, until the publication of paper) as. 3.1s. frequency domain, beginning with a function to give us the amplitude of 1 code implementation. The data set was provided by the Center for Intelligent Maintenance Systems (IMS), University of Cincinnati. waveform. A server is a program made to process requests and deliver data to clients. Lets try stochastic gradient boosting, with a 10-fold repeated cross 61 No. Most operations are done inplace for memory . The file than the rest of the data, I doubt they should be dropped. Description: At the end of the test-to-failure experiment, inner race defect occurred in bearing 3 and roller element defect in bearing 4. sample : str The sample name is added to the sample attribute. it is worth to know which frequencies would likely occur in such a Weve managed to get a 90% accuracy on the themselves, as the dataset is already chronologically ordered, due to Data sampling events were triggered with a rotary . Marketing 15. The benchmarks section lists all benchmarks using a given dataset or any of together: We will also need to append the labels to the dataset - we do need The spectrum usually contains a number of discrete lines and Working with the raw vibration signals is not the best approach we can File Recording Interval: Every 10 minutes (except the first 43 files were taken every 5 minutes). Networking 292. Qiu H, Lee J, Lin J, et al. advanced modeling approaches, but the overall performance is quite good. areas, in which the various symptoms occur: Over the years, many formulas have been derived that can help to detect Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web. Comments (1) Run. Some thing interesting about ims-bearing-data-set. validation, using Cohens kappa as the classification metric: Lets evaluate the perofrmance on the test set: We have a Kappa value of 85%, which is quite decent. its variants. Each 100-round sample is in a separate file. In this file, the ML model is generated. Note that some of the features Each data set consists of individual files that are 1-second vibration signal snapshots recorded at specific intervals. Wavelet filter-based weak signature detection method and its application on rolling element bearing prognostics[J]. Data Sets and Download. There is class imbalance, but not so extreme to justify reframing the Channel Arrangement: Bearing1 Ch 1; Bearing2 Ch 2; Bearing3 Ch3; Bearing4 Ch4; Description: At the end of the test-to-failure experiment, outer race failure occurred in