Bearing Datasets

These datasets include four distinct tests designed to validate an algorithm's ability to assess the health of bearings in rotating machines, focusing on condition monitoring. The targets in these tests indicate the moment when the condition of the bearing changes. The four tests consist of: synthetically generated signals, real-world measurements from a rotating machine at the Federal University of Uberl\u00e2ndia, and benchmark datasets from Case Western Reserve University (CWRU) and the HUST bearing dataset.Each test follows a similar structure: the system maintains a constant condition for a period, then transitions to a new condition, or returns to a previously observed one, and stays in that new state for another period before changing again. The aim of these tests is to evaluate the model's ability to correctly identify distinct conditions and detect the emergence of new conditions promptly, i.e., to determine if there is a significant delay in recognizing condition changes.Test 1 involves a synthetically generated signal by the finite element method with a proximity sensor, sample frequency of 1000 Hz, and duration of 3600 seconds, specifically designed to exhibit unbalance and misalignment variations at some specific times. The defect intensities vary across eight distinct conditions, with one of the conditions repeated to evaluate the clustering algorithm\u2019s memory capacity. Additionally, the test includes a run-up and run-down period during which the signal remains unstable.Test 2 is composed of a signal with a proximity sensor, sample frequency of 10,000 Hz, and duration of 1370 seconds, showcasing variations in speed. This signal was collected from a magnetic bearing bench located in the LMEst laboratory at the Federal University of Uberl\u00e2ndia (UFU).The machine operates under five different rotational speed conditions, including a baseline state. After reaching the fifth condition, it returns to two previously observed states.Test 3 comprises a public dataset provided by Case Western Reserve University (CWRU). It is a world-recognized standard dataset for bearing fault diagnosis. The signal contained in this dataset has a duration of 649 seconds. It was collected using an accelerometer sensor from rock drilling with sampling frequency of 12 kHz and 3 types of defects (Ball defect, Inner Race, and Outer Race) and 4 different intensities for each defect. The following describes all the conditions and the intensities:\u2022 Normal condition: Baseline\u2022 Ball defect:0.007 inches (0.178 mm): Condition #10.014 inches (0.356 mm): Condition #20.021 inches (0.533 mm): Condition #3\u2022 IR (Inner Race) defect:0.007 inches (0.178 mm): Condition #40.014 inches (0.356 mm): Condition #50.021 inches (0.533 mm): Condition #6\u2022 OR (Outer Race) defect:0.007 inches (0.178 mm): Condition #70.014 inches (0.356 mm): Condition #80.021 inches (0.533 mm): Condition #9Test 4 is composed of HUST, which is also a public dataset used as a benchmark for bearing fault classification. This dataset is from KG Bearing India and has a duration of 350 seconds and sampling frequency of 51,200. It contains a signal from a ball bearing collected with accelerometer sensor with artificially generated defects and different intensities. These defects are:\u2022 Normal condition: Baseline\u2022 IR (Inner Race) defect: Condition #1\u2022 OR (Outer Race) defect: Condition #2\u2022 Ball defect: Condition #3\u2022 IR and OR defects: Condition #4\u2022 IR and Ball defects: Condition #5\u2022 OR and Ball defects: Condition #6