Railway structure hazard predictor

• US 5,529,267 A
• Filed: 07/21/1995
• Issued: 06/25/1996
• Est. Priority Date: 07/21/1995
• Status: Expired due to Fees

• Current Assignee: Union Switch & Signal, Inc. (Hitachi, Ltd.)
• Original Assignee: Union Switch & Signal, Inc. (Hitachi, Ltd.)
• Inventors: Colbaugh, Michael E., Boyle, C. Franklin, Giras, Theo C.
• Primary Examiner(s): Le, Mark T.
• Application Number:
  US08/505,745
• Time in Patent Office: 340 Days
• Field of Search: 246/120, 246/121, 246/122 R, 246/246, 246/49, 246/473, 246/167, 246/218, 246/219, 246/249, 246/270 R, 73/786, 73/788, 73/800, 359/109, 385/12, 385/13
• US Class Current: 246/120
• CPC Class Codes: B61L 1/06 actuated by deformation of rail; actuated by vibration in rail
• CPC Class Codes: B61L 23/047 Track or rail movements

Abstract

A hazard predictor that processes both rail and superstructure measurements to predict some potentially hazardous conditions on a railway structure. Measurement is collected in real time with the aid of fiber optic sensor based linear array mesh, and processed with a neural network. Sensors placed under the rail and sensors placed laterally of the rail provide data collection in real time both during occupied and unoccupied periods. In some embodiments the measurement data is compressed into two signatures which can be represented as two vectors. The collinearity of the vectors and the angle between the vectors are utilized to interpret the data as to track conditions. The angle between the descriptors can be used to predict the severity of degradation of the structure. The predictor can be used to manage maintenance of the structure and interface with existing railway signalling equipment to provide traffic management.

Claims

1. A railway hazard predictor for monitoring railway track having rail mounted on a structure comprising:

at least one rail sensor mounted to sense physical characteristics imposed on said rail from a railway vehicle occupying said rail;

at least one alignment sensor mounted intermediate said rail and said structure to sense relative strain between said rail and said structure independent of said physical characteristic sensed by said rail sensor;

collection means for real time collecting data from said rail sensor and said alignment sensor;

processing means for analyzing said data and producing a rail signal indicative of rail integrity and a structure signal indicative of structure integrity; and

a neural network for reducing said rail signal and said structural signal into a train presence and misalignment detection.

2. The railway hazard predictor of claim 1 wherein said processing means includes means for providing a presence vector and an alignment vector;

and

said neural network is sensitive to the angle between said presence vector and said alignment vector.

3. The railway hazard predictor of claim 2 wherein said neural network considers collinearity of said presence vector and said alignment vector as indication of a normal mode.

4. The railway hazard predictor of claim 1 wherein said processing means is at least one of spatial or spatiotemporal.

5. The railway hazard predictor of claim 1 wherein said neural network is a Kohonen net using a winner-take-all strategy to build a feature map for classifying said data.

6. The railway hazard predictor of claim 1 wherein said neural network is a back-propagated neural network using a sequence of spatial data.

7. The railway hazard predictor of claim 1 wherein said rail sensor and said alignment sensor are fiber optic sensors;

said alignment sensor is mounted to sense lateral strain between such rail and a portion of such structure; and

said rail sensor is mounted to sense vertical loading on such rail.

8. The railway hazard predictor of claim 1 wherein said processing means includes means for providing a presence vector and an alignment vector;

said neural network is sensitive to the angle between said presence vector and said alignment vector;

said processing means is at least one of spatial or spatiotemporal; and

said neural network is a Kohonen net using a winner-take-all strategy to build a feature map for classifying said data.

9. The railway hazard predictor of claim 1 wherein said processing means includes means for providing a presence vector and an alignment vector;

said neural network is sensitive to the angle between said presence vector and said alignment vector;

said processing means is at least one of spatial or spatiotemporal; and

said neural network is a back propagated neural network using a sequence of spatial data.

10. A railway hazard predictor for monitoring track having rail mounted on a structure comprising:

a sensor unit having at least one fiber optic rail sensor mounted to sense vertical loading on such rail;

at least one fiber optic alignment sensor mounted to sense lateral strain between such rail and a portion of such structure; and

output means for outputting the data from said rail sensor and said alignment sensors to processing means for predicting structural conditions.

11. The railway hazard predictor of claim 10 wherein said at least one fiber optic rail sensor is mounted beneath a portion of such rail;

and

said at least one fiber optic alignment sensor is mounted to sense strain in a plane generally perpendicular to the strain sensed by said fiber optic rail sensor.

12. The railway hazard predictor of claim 11 wherein said at least one fiber optic alignment sensor includes two fiber optic alignment sensors mounted on opposite sides of such rail.

13. The railway hazard predictor of claim 12 wherein said at least one rail sensor and said two alignment sensors are of length l;

wherein said at least one rail sensor and said two alignment sensors are connected into a single fiber optic output; and

said at least one rail sensor and said two alignment sensors are connected to said output through optical delay loops having respective lengths that are integer multiples of l.

14. A method for monitoring railway track having rail mounted on a structure to determine a hazard condition comprising:

sensing physical characteristics imposed on said rail from a railway vehicle occupying said rail;

sensing relative strain between said rail and said structure independent of said physical characteristic;

collecting said physical characteristics and said relative strain as real time data;

processing said data and producing rail signal indicative of rail integrity and structure data indicative of structure integrity; and

reducing said rail signal and said structural signal into a train presence and misalignment detection.

15. The method of claim 14 wherein said processing includes providing a presence vector and an alignment vector;

and

said reducing is sensitive to the angle between said presence vector and said alignment vector.

16. The method of claim 15 wherein said reducing considers collinearity of said presence vector and said alignment vector and indication of a normal mode.

17. The method of claim 14 wherein said processing is at least one of spatial or spatiotemporal.

18. The method of claim 14 wherein said reducing is by a Kohonen net using a winner-take-all strategy and builds a feature map for classifying said data.

19. The method of claim 14 wherein said reducing is by back-propagated neural network using a sequence of spatial data.

20. The method of claim 14 wherein said sensing of physical characteristics uses a fiber optic sensor mounted to sense vertical loading of such rail;

and

said sensing strain senses lateral strain between such rail and a portion of such structure.

21. The method of claim 14 wherein processing includes providing a presence vector and an alignment vector;

said reducing is sensitive to the angle between said presence vector and said alignment vector;

said processing is at least one of spatial or spatiotemporal; and

said reducing uses a Kohonen net using a winner-take-all strategy and builds a feature map for classifying said data.

22. The method of claim 14 wherein said processing includes providing a presence vector and an alignment vector;

said reducing is sensitive to the angle between said presence vector and said alignment vector;

said processing is at least one of spatial or spatiotemporal; and

said reducing includes back-propagation of a neural network using a sequence of spatial data.