Have you ever been delayed by a faulty train? Do you wonder why maintenance regimes fail to identify these faults earlier? What if existing onboard CCTV systems could be augmented to monitor component health?

What Papped does

Papped is an image based fault prediction system powered by computer vision that enables our users to deploy preventative maintenance regimes. Papped maximises the returns from maintenance time whilst minimising lost passenger hours by ensuring parts are serviced as soon as the system observes a problematic degradation as opposed to a fixed interval schedule and ensuring these defects do not progress into faults that impact passenger service.

How we built it

We built a demonstration of Papped observing the operation of train doors using recording equipment directed at the vestibule doors onboard an SNCF TGV train. Using frame markers and image stabilization software footage that simulated a fixed position CCTV camera pointed at external doors was produced. This footage was processed to detect edges and then straight vertical edges to produce a number of 'candidate edges' that may represent the door edge. Then by modeling the behaviour of the vertical image edges produced by the video frame in conjunction with the expected movement behaviour of a door we could track door position an accurately and smoothly.

This produces an estimate of the doors position which when taken in consideration with all of the other frames and a typical movement profile produced by historic data enables the identification of instances where door operation has become abnormal and the severity of this.

The individual events are stored by the node.js backend, parsed and consumed by an responsive web interface for engineers which includes footage of the problem.

Challenges

Reliable image processing poses the largest challenge and added value to the project. The CCTV footage captures a variety of conditions with obstructions to view from passengers, luggage and fixtures. Managing the lossy nature of the footage was accomplished through signal filtering, region prioritisation and dynamic smoothing of the position.

Onboard communication remains challenging throughout the industry. To achieve the maximum benefit of the system a status packet should be send on every actuation but the in reality this would occur whenever sufficient bandwidth is available.

The future for Papped

Papped's goal towards onboard deployment is to be packaged into a compact, low cost and low energy consumption node per cartridge that ties into existing WiFi enabled CCTV cameras and builds an onboard mesh network which forwards and collects near real time data to maintenance facilities whenever communication is available. The technology is flexible enough for application towards a multitude of moving parts whilst scaling to the lengths of new modern trains.

Share this project:
×

Updates