Abstract

In the railroad industry, systematic health inspections of freight railcar bearings are required. Bearings are subjected to high loads and run at high speeds, so over time the bearings may develop a defect that can potentially cause a derailment if left in service operation. Current bearing condition monitoring systems include Hot-Box Detectors (HBDs) and Trackside Acoustic Detection Systems (TADS™). The commonly used HBDs use non-contact infrared sensors to detect abnormal temperatures of bearings as they pass over the detector. Bearing temperatures that are about 94°C above ambient conditions will trigger an alarm indicating that the bearing must be removed from field service and inspected for defects. However, HBDs can be inconsistent, where 138 severely defective bearings from 2010 to 2019 were not detected. And from 2001 to 2007, Amsted Rail concluded that about 40% of presumably defective bearings detected by HBDs did not have any significant defects upon teardown and inspection. TADS™ use microphones to detect high-risk bearings by listening to their acoustic sound vibrations. Still, TADS™ are not very reliable since there are less than 30 active systems in the U.S. and Canada, and derailments may occur before bearings encounter any of these systems.

Researchers from the University Transportation Center for Railway Safety (UTCRS) have developed an advanced algorithm that can accurately and reliably monitor the condition of the bearings via temperature and vibration measurements. This algorithm uses the vibration measurements collected from accelerometers on the bearing adapters to determine if there is a defect, where the defect is within the bearing, and the approximate size of the defect. Laboratory testing is performed on the single bearing and four bearing test rigs housed at the University of Texas Rio Grande Valley (UTRGV). The algorithm uses a four second sample window of the recorded vibration data and can reliably identify the defective component inside the bearing with up to a 100% confidence level. However, about 20,000 data points are used for this analysis, which requires substantial computational power. This can limit the battery life of the wireless onboard condition monitoring system. So, reducing the vibration sample window to conduct an accurate analysis should result in a more optimal power-efficient algorithm. A wireless onboard condition monitoring module that collects one second of vibration data (5,200 samples) was manufactured and tested to compare its efficacy against a wired setup that uses a four second sample window. This study investigates the root-mean-square values of the bearing vibration and its power spectral density plots to create an optimized and accurate algorithm for wireless utilization.

This content is only available via PDF.
You do not currently have access to this content.