<< Return to Webinar Files
Webinar Presentation
HTML version of the presentation
Image descriptions are contained in brackets. [ ]
Smart Sensors and Infrastructures for Transportation
(July 29, 2021)
Rail Neutral Temperature Estimation using Impulse Vibration and Machine Learning
Presenter: Yuning Wu
Presenter’s Org: Department of Civil & Environmental Engineering, University of Utah
T3 webinars are brought to you by the Intelligent Transportation Systems (ITS) Professional Capacity Building (PCB) Program of the U.S. Department of Transportation (USDOT)’s ITS Joint Program Office (JPO). References in this webinar to any specific commercial products, processes, or services, or the use of any trade, firm, or corporation name is for the information and convenience of the public, and does not constitute endorsement, recommendation, or favoring by the USDOT.
[The slides in this presentation contain the logo of the University of Utah.]
Slide 1: Rail Neutral Temperature Estimation using Impulse Vibration and Machine Learning
Presenter: Yuning Wu (yuning.wu@utah.edu)
Infrastructure Sensing & Experimental Mechanics (iSEM) Laboratory Department of Civil & Environmental Engineering, University of Utah
In corporation with Dept. of Civil & Environmental Engineering, University of Illinois, Urbana-Champaign
ITS Professional Capacity Building Program
T3e Webinar
Smart Sensors and Infrastructures for Transportation
[This slide contains the U.S. DOT triskelion.]
Slide 2: Outline
- 1. Introduction: Motivation & Literature Review
- 2. Experimental and Numerical Studies
- Preliminary Experimental Observations
- On the Existence of ZGV modes in Rails
- Sensitivity to temperature and axial load
- ML-RNT Predictive Tool Development
- Framework Development
- Field Data Collection
- System Performance Evaluation
- Conclusion
Slide 3: Motivation
Simple facts
- U.S. ranks 1st by length of railroad networks (160,141 miles by 2017)
- The continuous welded rail (CWR) has been widely adopted (106,500 miles)
- The U.S. rail networks transported over 1,621.8 million tons of commodity in 2017
- There are potential increasing demands from railroad industry for safety & efficiency
[This slide contains two images: (1) a photo a train preparing to load oil from the Bakken formation on a railroad and (2) a chart of railroads moving more crude oil by originated carloads and terminated carloads.]
Slide 4: Motivation
Due to lack of expansion joints, CWR develops internal tensile/compressive stresses when the rail temperature is below or above the stress-free temperature.
[This slide contains three images: (1) a collage of photos of railroad tracks. (2) a photo of a 2002 Crescent City Amtrak derailment with a caption reading “2002 Crest City Amtrak derailment 4 deaths, 142 injuries, $8.3M.” (3) a photo of a 2012 Northbrook UP derailment with a caption reading “2012 Northbrook UP derailment 2 deaths, 31 cars derailed, bridge collapsed, $5.3M.”]
Slide 5: Motivation
Federal Railroad Administration (FRA) safety statistics report the “track alignment irregularities (buckled/sun kink)” as one of the leading causes of train accident.
Specific Causes: |
Total |
Type of Accident |
Reportable Damage |
Casualty |
Cnt |
% |
Coll |
Der |
Other |
Amount |
% |
kld |
Nonf |
T207- Detail fracture- shelling/head check |
267 |
7.4 |
2 |
265 |
- |
89,306,643 |
13.4 |
2 |
5 |
T109- Track alignment irreg(buckled/sunkink) |
152 |
4.2 |
1 |
150 |
1 |
85,063,378 |
12.8 |
2 |
3 |
T220- Transverse/compound fissure |
208 |
5.8 |
- |
206 |
2 |
44,927,561 |
6.7 |
0 |
0 |
T110- Wide gage (defective/missing crossties) |
555 |
15.5 |
1 |
552 |
2 |
42,904,761 |
6.4 |
0 |
7 |
T001- Roadbed settled or soft |
109 |
3 |
2 |
105 |
2 |
34,211,301 |
5.1 |
0 |
16 |
T213- Joint bar broken (compromise) |
9 |
0.3 |
- |
9 |
- |
25,851,789 |
3.9 |
0 |
132 |
T221- Vertical split head |
148 |
4.1 |
1 |
147 |
- |
24,930,317 |
3.7 |
0 |
3 |
T202- Broken base of rail |
120 |
3.3 |
- |
120 |
- |
24,579,199 |
3.7 |
0 |
3 |
T299- Other rail and joint bar defects |
46 |
1.3 |
- |
45 |
1 |
24,044,945 |
3.6 |
0 |
3 |
T102- Cross level track irreg.(not at joints) |
86 |
2.4 |
- |
84 |
2 |
22,277,255 |
3.3 |
0 |
0 |
T201- Bolt hole crack or break |
51 |
1.4 |
1 |
50 |
- |
20,177,698 |
3 |
0 |
1 |
T002- Washout/rain/slide/etc. dmg -track |
31 |
0.9 |
- |
28 |
3 |
18,648,983 |
2.8 |
1 |
11 |
T111- Wide gage(spkies/other rail fasteners) |
163 |
4.5 |
- |
163 |
- |
16,795,744 |
2.5 |
0 |
1 |
T210- Head and web sep(outside jt bar limit) |
133 |
3.7 |
- |
132 |
1 |
16,727,662 |
2.5 |
1 |
3 |
Over 200M reportable damage and nearly 100% derailment rate over the past two decades.
Thermal buckling prevention is high priority industry goal.
Slide 6: Research Problem
Industrial demand - to facilitate the thermal stress management within CWR structure, the knowledge of the stress-free temperature or Rail Neutral Temperature (RNT) is critical.
Item # |
Phenomena / Technique |
Work / Investigator / Exp. Publication |
|
1 |
Rail uplift (VERSE) |
Samavedam and Kish 1987 & 1995 |
Static deformation |
2 |
Strain measurement (Wheatstone bridge) |
Harrison et al. 1999, Liu et al. 2018 |
Static deformation |
3 |
Strain measurement (DIC) |
Knopf & Rizos et al. 2020 |
Static deformation |
4 |
Strain measurement (Hole-drilling) |
Zhu & Lanza di Scalea 2017, Harrington et al. 2017 |
Static deformation |
5 |
D’stressen |
Read & Shust 2007 |
Vibration |
6 |
Vibration (Modal frequency, mode shape) |
Boggs & Murray 1994, Koob 2005, Damljanoviฤ & Weaver 2006 |
Vibration |
7 |
Vibration (Video imaging) |
Sefa Orak & Rizzo et al. 2018 |
Vibration |
8 |
Acoustoelastic effect (Rayleigh wave, flexural modes, birefringence, horizontal shear waves) |
Egle & Bray 1979, Man & Paroni 1996, Gokhale & Hurlebaus 2008, Hurley 2014, Albakri & Tarazaga 2018 |
Ultrasound |
9 |
Diffuse ultrasound |
Turner et al. 2011 |
Ultrasound |
10 |
Nonlinear ultrasonic guided waves |
Bartoli et al. 2010, Nucera & Lanza di Scalea 2014a & 2014b |
Ultrasound |
11 |
Electromechanical impedance method |
Phillips et al, 2012, Zhu & Lanza di Scalea, 2016 |
Local stiffness |
12 |
Nonlinear solitary waves |
Nasrollahi & Rizzo 2018 & 2019 |
Local stiffness |
13 |
Magnetic (MAPS-SFT, metal magnetic memory, magnetic Barkhausen noise) |
Read 2010, Wegner 2007, Zhang et al. 2011
Shu et al. 2016 |
|
14 |
Piezospectroscopy |
Kim & Yun 2018 & 2019 |
|
Research Problem โ to determine RNT or longitudinal thermal load in a nondestructive and nondisruptive manner without the need of reference measurement at zero stress state with a reasonable accuracy (±10℉).
Source: https://safetydata.fra.dot.gov/OfficeofSafety
Slide 7: Experimental & Numerical Studies
- Impulse vibration test has been widely used in experimental modal analysis, structural dynamics testing, impact echo test, and so on.
- Why not try the impulse vibration test on rails?
Impact and detection within/close to the same cross section
- Sample 0 - 3 ft rail sample
- Sample 1 - 60 ft rail sample
- Sample 2 โ CWR
*Siemens Webinar “Understand the basics and mathematics behind modal analysis”
**D. Feng et al “Experimental validation of cost-effective vision-based structural health monitoring,” MSSP 2017.
*** NDT.net
[This slide contains five images: (1) a photograph of impulse vibration testing with the caption “Source: Siemens*.” (2) a photograph of impulse vibration testing with the caption “Source: Feng**.” (3) a chart of the impact test with the caption “Source: ***.” (4) a graphic showing how an impactor works. (5) a photo of an impact test.]
Slide 8: Experimental & Numerical Studies
Vibration spectrum from rail samples
Sample 0
3 ft rail sample
Sample 1
60 ft rail sample
Sample 2
CWR with full track setup
[This slide contains three images: (1) a photo of an impact test on the track with a graph readout of the Sample 0 test on a graph. (2) a photo of a rail sample with the readout of the Sample 1 test on a graph. (3) a photo of a CWR with full track setup with the Sample 2 test readout on a graph.]
Slide 9: Experimental & Numerical Studies
Existence of ZGV Modes in Rails
Are we generating the Zero-group Velocity modes in the rails? Note ๐_๐=∂ω/∂๐
- Hypothesis 1: There exists Zero-group Velocity (ZGV) mode(s) in free rails, and the specific mode(s) can be generated and detected by impulse vibration test.
- Hypothesis 2: The Zero-group Velocity (ZGV) frequencies in rails will be affected by both temperature and axial load.
*Prada et al “Local vibration of an elastic plate and zero-group velocity Lamb modes” JASA 2008
[This slide contains a series of graphs of zero-group velocity modes in rails.]
Slide 10: Experimental & Numerical Studies
Existence of ZGV Modes in Rails
- Semi-Analytical Finite Element (SAFE) analysis to calculate dispersion curves of rails
- Waveguide’s cross section is discretized by finite elements
- The displacements along the wave propagation is assumed as an analytical harmonic exponential function, ๐^(โ๐ω๐ก)
[This slide contains four images: (1) a graph showing frequency(kHz) versus wavenumber(1/m). (2) an image pf two cross-sections of a rail, one solid black, one solid white. (3) a screenshot of math equations. (4) a three-dimensional graph labeled “Propagation plane”]
Slide 11: Experimental & Numerical Studies
Existence of ZGV Modes in Rails
Semi-Analytical Finite Element (SAFE) analysis to calculate dispersion curves of rails provides a strong evidence that ZGV modes exist in free rails. (Hypothesis 1)
Note ๐๐=∂ω/∂๐
[This slide contains two graphs: (1) a pair of graphs of semi-analytical finite element analysis to calculate dispersion curves of rails labeled “zoom-in k-figure” and (2) a pair of graphs of semi-analytical finite element analysis to calculate dispersion curves of rails labeled “group velocity.”]
Slide 12: Experimental & Numerical Studies
Eigenmode analysis - predict the behavior of Zero-group Velocity (ZGV) frequencies
- Hypothesis 2: The Zero-group Velocity (ZGV) frequencies in rails will be affected by both temperature and axial load.
- Simulations of rails when subjected temperature variation and axial load
- Perform data collection when rail is subjected to temperature variations
- Two-step model creation
[This slide contains two images: (1) a graphic of the two-step model and (2) three charts of frequency by group velocity.]
Slide 13: Experimental & Numerical Studies
Eigenmode analysis - predict the behavior of Zero-group Velocity (ZGV) frequencies
Behaviors of confined CWRs subjected to temperature and longitudinal load were analyzed using eigenmode analysis
Modest extent of variation in local vibration modal frequencies introduced by longitudinal load
Largest extent of variation in local vibration modal frequencies introduced by the rising temperature. (Hypothesis 2)
[This slide contains two graphs: (1) a plot graph of resonance frequency shifts due to longitudinal force. (2) a plot graph of resonance frequency shifts due to rail temperature.]
Slide 14: ML-RNT Predictive Tool
Framework Development
[This slide contains four images: (1) a grouping of photographs of people working on the railroad labeled “Field data collection.” (2) grouping of graphics of thermal load model, wave mode analysis, and dispersion curves labeled “Numerical modelling.” (3) a graphic of network architecture labeled “machine learning with augmented field data.” (4) a graphic of a flowchart about the ML-RNT Predictive Tool.]
Slide 15: ML-RNT Predictive Tool
Data Collection & Analysis โ Revenue-service site
[This slide contains six images: (1) a photo of maintenance workers working on instrumentation, (2) a photo of maintenance workers working on instrumentation with a data logger in the foreground, (3) a photo of a maintenance worker welding a section of rail, (4) a photo of a cut section on the railroad, (5) a photo of a maintenance worker cutting on the railroad, and (6) a close-up photograph of the welded rail.]
Slide 16: ML-RNT Predictive Tool
ML-RNT Predictive Tool
[This slide contains five images: (1) a photo of a thermocouple and strain gauge, (2) a photo of the gauge on a railroad, (3) a line graph of temperature from August 1-4, (4) a graph of amplitude by frequency, (5) a graph of amplitude by frequency.]
Slide 17: ML-RNT Predictive Tool
Regression Model for RNT estimation
A regression model was proposed:
- Input: rail vibration modal parameters
- Output: Rail neutral temperature (RNT)
A mathematical formulation to relate CWR modal features ƒ_k^๐ to RNT is given by
This all represents a simple regression model in the form of ๐ฆ=๐๐ฅ+๐, where independent (RNT and T) and dependent (ƒ_k^๐) variables are related to each other through unknown coefficients (๐บ_๐^๐, ๐บ_๐^σ, and ƒ_๐^๐
๐๐) that need to be determined through a regression analysis.
ƒ_1^๐ |
= |
ƒ_1^๐
๐๐ |
+ |
๐_1^๐ |
๐_1^๐ |
โ๐ธ๐ผ(๐โ๐
๐๐)
๐โ๐
๐๐) |
ƒ_2^๐ |
ƒ_2^๐
๐๐ |
๐_2^๐ |
๐_2^๐ |
โฎ |
โฎ |
โฎ |
โฎ |
ƒ_(๐โ1)^๐ |
ƒ_(๐โ1)^๐
๐๐ |
๐_(๐โ1)^๐ |
๐_(๐โ1)^๐ |
ƒ_๐^๐ |
ƒ_๐^๐
๐๐ |
๐_๐^๐ |
๐_๐^๐ |
↓ |
↓ |
↓ |
↓ |
↓ |
measured multiple modal features on rail temperature |
measured multiple modal features on RNT |
thermal load sensitivity |
temperature sensitivity |
thermal stress |
Slide 18: ML-RNT predictive system
ML-RNT prediction results
Use the resonance frequency of the ZGV mode at 76 kHz and rail temperature as input to the NN
Model performance |
Measured frequency (Hz) |
Model frequency (Hz) |
Model error (Hz) |
76628.6 |
76628.4 |
0.2 |
76588.6 |
76580.3 |
8.3 |
76548.6 |
76547.9 |
0.7 |
76691.8 |
76690.4 |
1.4 |
76657.8 |
76658.4 |
-0.6 |
76600.3 |
76604.0 |
-3.7 |
76654.6 |
76654.7 |
-0.1 |
76614.6 |
76614.7 |
-0.1 |
76582.6 |
76577.0 |
5.5 |
76571.8 |
76571.0 |
0.8 |
ML-RNT prediction using 76kHz |
Measured RNT (℃) |
Estimated RNT (℃) |
Estimation error (℃) |
32.59 |
31.99 |
-0.6 |
32.49 |
21.37 |
-11.12 |
33.03 |
32.67 |
-0.36 |
32.70 |
29.97 |
-2.73 |
32.50 |
31.60 |
-0.90 |
32.92 |
36.45 |
3.53 |
31.89 |
33.68 |
1.79 |
31.72 |
32.89 |
1.16 |
31.67 |
24.99 |
-6.67 |
31.5 |
31.18 |
-0.32 |
The error of RNT estimation is well bounded within ± 5 ℃ range on vibrational data collected at the revenue-service site.
[This slide contains a line graph of Test Error and prediction error.]
Slide 19: Summary
- The contactless sensing technique can effectively promote the local vibration modes in CWRs, which are distinctive from the ones obtained in lab.
- The proposed sensing technique was deployed for field data collection over a wide range of temperature and thermal stress levels.
- Numerical models were developed to understand and predict the CWR vibrational behavior under the influence of temperature and RNT.
- An excellent agreement (discrepancies less than 0.01%) between model and experimental results were obtained.
- The performance of the developed RNT predictive tool was evaluated using field measurements as input. And the proposed framework can support an estimation accuracy of ± 5 ℃, when measurement or model noise is low.
Slide 20: Acknowledgement
This work was supported by the U.S. National Academy of Sciences Rail Safety IDEA program and partially funded by the startup package at the University of Utah.
The field data collection was coordinated and supported by BNSF Railway Company and Utah Transit Authority.
[This slide contains four images: (1) the logo of the National Academies of Sciences, Engineering, and Medicine, (2) the logo of University of Utah, (3) the logo of BNSF Railway, and (4) the logo of UTA.]
Slide 21: References
- Zhu, Xuan. Non-destructive and Semi-destructive methods for Thermal Stress Measurement in the Continuous Welded Rails. Diss. UC San Diego, 2016.
- USDOT FRA safety data website https://safetydata.fra.dot.gov
- Kish, A., and Samavedam,G., “Longitudinal force measurement in continuous welded rail from beam column deflection response,” American railway engineering association (1987).
- Read, D., and Shust, B, “Investigation of prototype rail neutral temperature measurement system,” Railway Track and Structures, 103(6), 19-21(2007).
- Xuan, Z., and Lanza di Scalea, F., “Thermal Stress Measurement in Continuous Welded Rails Using the Hole-Drilling Method,” Experimental Mechanics, 57, 165-178(2017).
- Knopf, K., Rizos, D. C., Qian, Y., and Sutton, M., “A non-contacting system for rail neutral temperature and stress measurements: Concept development,” Structural Health Monitoring, 20(1), 84-100(2020).
- Gokhale, S., and Hurlebaus, S, “Monitoring of the stress-free temperature in rails using the acoustoelastic effect,” AIP Conference Proceedings, 975(1), 1368-1373(2008).
- Nucera, C., and Lanza di Scalea, F., “Nondestructive measurement of neutral temperature in continuous welded rails by nonlinear ultrasonic guided waves,” The Journal of the Acoustical Society of America, 136(5), 2561-2574(2014).
- Boggs, T. P., “Determination of Axial Load and Support Stiffness of Continuous beams by Vibration Analysis,” AAR Research Report, 1994.
- Xuan, Z., and Lanza di Scalea, F., “Sensitivity to axial stress of electro-mechanical impedance measurements,” Experimental Mechanics, 56(9), 1599-1610(2016).
- Nasrollahi, A., and Rizzo, P., “Numerical Analysis and Experimental Validation of an Nondestructive Evaluation Method to Measure Stress in Rails,” ASME J Nondestructive Evaluation, 2(3), 031002-0310014(2019).
- . Helen, W., “Using Magnetic Barkhausen Noise Technology and Finite Element Method to Study the Condition of Continuous Welded Rails on the Darwin-Alice Springs Line,” Journal of Civil Engineering and Architecture, 5(7) (2011).
- Feng, C., and Paul, D.W., “The effect of load on guided wave propagation,” Ultrasonics, 47(1-4), 111-122 (2007).
Slide 22: Thanks!
Any questions?
Yuning Wu (yuning.wu@utah.edu)
Infrastructure Sensing & Experimental Mechanics (iSEM) Laboratory
Department of Civil & Environmental Engineering, University of Utah
↑ Return to top