Railway vehicle applications such as public address system or track fault monitoring require position information, often delivered by GPS, that might not always be available, for example in tunnels or valleys. Therefor Televic and Flanders Make developped a low-cost, accurate position estimation system that runs on a railway certified platform and delivers an estimate of the vehicle position within a range of 5 meters, even when the GPS signal is missing during 2 minutes.
Televic develops, manufactures and installs top end high-tech communication systems for specific niche markets. Televic Rail, one of its companies, has developped together with Flanders Make an accurate and reliable localization system for estimating the train position/speed using on-board measurements. The information of this localization system can be used in the following applications:
- A public address system (Passenger Information Systems division) that accurately locates the train for the announcement when the train arrives in stations.
- A Track fault monitoring system (Mechatronics division) that accurately indicates the location of detected track faults.
GPS positioning provides a good solution when sufficient number of non-obstructed satellite signals are available. Usually a minimum of four signals is necessary to achieve a correct position estimation. When GPS fails because of e.g. tunnels, valleys, etc., dead reckoning become necessary. Dead reckoning (DR) is the process of calculating the current position of the vehicle based on the knowledge of previous position and other available quantities such as accelerations, speeds and angular rate. These physical quantities are permanently available because measured in the body frame, usually by a system called Inertial Measurement Unit (IMU). In the framework of the Mechatronics 4.0 vis-traject, Flanders Make and Televic Rail developed a dead reckoning algorithm able to fusion GPS data, acceleration and rotation rate measured by an IMU and speed radar data. The developed algorithm is also able to cope with different sampling rates at which these signals are available, ranging from 1 Hz to 1 kHz. Kalman filtering has been selected to realize the process of data fusion.
Here below, we briefly describe the tests carried out on Lommel Proving Grounds, the developed algorithm used for position estimation and demonstrate that a positioning error below 5 m can be achieved, even when the GPS data is missing during 2 minutes.
Test and data
As an alternative to measurements using trains, it has been decided to perform tests using a car on which the target measurement systems were installed. In order to carry out these measurements in controlled and safe environment, the Lommel Proving Ground (LPG) has been chosen. The LPG includes a wide range of road types and events, but also allow for constant speed tests, various turn radius, etc.
The car was equipped with the speed radar and the GPS/IMU module in order to get the data needed in the algorithm. A high-end GPS system was also installed in order to have a continuous reference position measurement and to assess the real performance of the fusion algorithm. A number of tests in different conditions (i.e. various speeds, grounds, turning radius) have been carried out on tracks 5, 10 and 16. For measurement on track 10 the GPS module has been disabled each loop at a fixed point for 120s.
Data of different nature are available for the dead reckoning algorithm. They all carry useful information that can help keeping on locating the vehicle in absence of GPS information. These data are speed (from the speed radar), accelerations from the accelerometer and angular rotation from the gyros. If the GPS information is available (often at lower rate), the data can be used to get a more accurate location and to have the position information at the rate corresponding to the highest data rate.
In this context, we have used a Kalman filter (KF) to fusion these different types of data. In the case of a KF, this is done simultaneously by means of two models: one expressing the dynamics of the internal variables (the dynamics model) and another expressing how to link the measurement with these variables (the measurement model).
Basically when a data sample is received, it is fused into the global state of the KF (update phase) and then a prediction step is applied in order to determine what will be the next system state. The advantage of the KF is that the measurement model can be different at each time therefore allowing data to have different rates. We can for instance design a measurement model involving only accelerations or all data simultaneously. The global strategy of the DR system is to use the data actually available and to select a different measurement model depending which data is available.
For the validation of the algorithm the focus has been put mainly on the test driven on track 10 where the conditions are close to railway conditions and where the GPS module was disabled for a period of 120 seconds.
Tests show that he average error is not null even in presence of GPS signal. The main reason is that the low-end GPS and high-end GPS (reference) system do not deliver the same position, meaning that in general we are limited by the accuracy of the GPS (supposed the system delivering the true position). During the time the GPS is disabled, the error does not grow too fast. The error is stable between 2 m and 6 m.
Flanders Make and Televic Rail cooperated to develop an accurate and reliable localization system for estimating the train position/speed using on-board measurements. This system includes GPS, IMU and radar information. It uses a Kalman filter based fusioning which enables to retrieve an accurate position estimate for railway vehicle, even when some signal is missing for a limited period of time.
Based on experimental data, acquired with an instrumented vehicle on Lommel Proving Grounds, it was demonstrated that the developed system achieves a positioning error below 5 m, even when the GPS data is missing during 2 minutes. Televic Rail is currently working towards an implementation of the developed algorithm on a railway certified embedded platform.
These results have been achieved in the framework of the Mechatronics 4.0 vis-traject, which is funded by VLAIO. In this vis-traject, Sirris, Flanders Make and iMinds transfer innovative mechatronic solutions to concrete, industrial cases of companies in the following application domains: sensor architectures, advanced control design, sensors and algorithms for condition monitoring and machine diagnosis, vision and scan-based methods for characterization of product features.