POOR vehicle dynamic performance and ride quality may occur at track locations that do not exceed track geometry or safety standards, such as curve entry or exit, special trackwork, and track misalignments that promote yaw instability or hunting. Poor ride quality may not be an indicator of unsafe operation, but may point to an area of track or a vehicle that needs maintenance to prevent further degradation. Conversely, track geometry locations that exceed track geometry or safety standards do not often cause poor ride quality or poor vehicle performance.
To improve and advance current track geometry inspection practices and maintenance procedures, Transportation Technology Center Inc (TTCI) developed a track inspection method known as performance-based track geometry (PBTG). Trained neural networks within PBTG relate the complex dynamic relationships that exist between vehicles and track geometry to vehicle performance. They also identify track segments that may generate unwanted vehicle responses.
PBTG is now in use with three North American railways and an overseas one, and is also suitable for use by a transit system to optimise track and fleet maintenance. Onboard accelerometers and a PBTG neural network identify track locations that need work without direct measurement of the track geometry. This allows monitoring of track condition between scheduled track geometry measurements.
PBTG can also identify cars whose performance is beginning to deteriorate. If all cars in the fleet are equipped with PBTG accelerometers, it is possible to build a database of information to monitor the condition of both the cars and the track over time.
PBTG also uses measured track geometry and the PBTG neural network to predict vehicle performance on the track. This helps identify locations in the track likely to cause poor ride quality or other issues related to vehicle performance, which is how North American freight railways are currently applying PBTG.
To support the Transit Cooperative Research Programme, TTCI conducted research to develop methods for evaluating track geometry that account for transit vehicle performance and passenger ride quality using a combination of modelling techniques with PBTG and Nucars. These studies will help to determine improvements in track geometry and track maintenance practices to be developed in Phase II of the project.
Nucars simulations and data collected on transit systems are being used to train PBTG neural networks to evaluate the model's ability to predict ride quality. The goal is to ascertain the ability of Nucars and PBTG to properly predict vehicle performance and ride quality and to identify track geometry locations that need maintenance using the initial ride quality measurements and the measured track geometry inputs.
TTCI partnered with Dallas Area Rapid Transit (Dart) to carry out this research, with Dart providing a test vehicle for TTCI to perform characterisation and ride quality tests. All testing was performed on the Dart light rail network which has a variety of track structures and a wide range of operating speeds.
Dart's Kinki Sharyo super light rail vehicle (SLRV), which has three sections and can accommodate up to 150 seated and standing passengers, was selected and fully characterised for testing. The data obtained from the characterisation studies was used to develop a Nucars model representing the vehicle, while the characterised vehicle was equipped with accelerometers and various displacement transducers to collect passenger ride quality data. Track geometry measurements were collected within two weeks of the ride quality measurements and used as comparisons with predictions from the Nucars model and for future PBTG neural network training.
The ride quality and track geometry comparison was undertaken to determine if there was a correlation between them. Locations on the Red Line that had ride quality issues were identified from the ride quality test. The top part of Figure 1 shows the accelerations measured under the driver's seat in the leading cab of the SLRV in this area, while the bottom section of Figure 1 shows the track geometry measured in the same area. The 2-second peak-to-peak value is approximately 0.35g. In the area where this occurs, there is a deviation in the lateral alignment.
Figure 2 shows the frequency content of the acceleration data and lateral alignment of the track geometry. There are peaks at approximately 1Hz and 1.65Hz in the lateral vehicle response. In the lateral alignment of the track geometry in this area there is also a 1Hz peak corresponding to a wavelength of 28.35m. In addition there is a 1Hz response of the vehicle that correlates to the 1Hz frequency content of the lateral alignment of the track.
It is possible to identify track geometry that can cause ride quality issues, such as the lateral deviations with the 28.35m wavelength, which cause a dynamic response in the vehicle. It is important to note that although these track geometry deviations do not exceed any safety criteria, they can affect passenger ride quality. To identify the track geometry issues that affect ride quality, it is imperative to take track geometry and ride quality measurements at the same time.
Hunting may be triggered by a combination of lateral deviation, speed, and wheel/rail interaction, so will be important in the next phase of this project to investigate the potential triggers in more detail.
Vehicle model
The results of the test so far show that it should be possible to identify the effect of track geometry deviations on vehicle ride quality response during Phase II of the project. However, some work is still required to improve the vehicle model to predict this response correctly. Identifying the influence of the following factors on vehicle response will be important to accurately model and determine track geometry triggers:
• wheel/rail interface, including profile shapes and contact geometry
• vehicle speed, and
• understanding and identifying rigid body vibration modes of the vehicle.
After all the issues have been investigated, the track geometry and ride quality data collected during Phase I at Dart will be used to train neural networks to predict ride quality. The validated Dart vehicle Nucars model will be used to run simulations at different speeds to generate additional neural network training data. The neural networks will then be used to predict ride quality over measured track not used in the training, while the neural network output will be compared with Nucars simulation predictions and measured ride quality to determine the accuracy of the neural network predictions.
If neural networks are determined to be a viable option for predicting ride quality, a different vehicle on a different transit system will be selected for further investigation.
