VUES – A vision based Vehicle Underframe Examination System
Sits between the sleepers and examines the vehicle underframe in service conditions
Routine train inspection and maintenance is a continuous activity ensuring equipment is maintained to an acceptable safety standard. Preventing failure of expensive major components (engines, gearboxes, wheel sets, axle bearings) and safely maximising their service life is key to an efficient railway.
Many components are best monitored in near service conditions (train moving & fresh from service) rather than stationary in the depot (and cooled down). This cannot be done by humans.
Service intervals are established for components with known life-spans but cannot prevent unexpected failures. It is not feasible to inspect trains thoroughly every day, and not all warning signs of failures are visible to the human eye.
The solution is continuous automatic inspection based on hyperspectral imaging and computer vision algorithms. Specialist cameras, lights and software are used to detect faults and anomalies such as overheating components, damaged equipment and leaks of certain fluids.
Gobotix Vehicle Underframe Examination System – VUES, uses an array of hyperspectral cameras to detect and report changes and anomalies in the physical state of the underside of a railway vehicle as the vehicle passes over the system. The system gathers data of the visible components on the underside of the vehicle using a combination of cameras which are sensitive to a set of wavelengths of the electromagnetic spectrum (visible, thermal, ultraviolet).
Each vehicle is identified by reading its RFID tag and its history recalled. Then, unsupervised and supervised machine learning techniques are applied on the data to detect anomalies. VUES holds statistics of the measured signals as well as standard and safe operating ranges of components. VUES can identify changes to any component or visible area of interest and it is not limited to particular components. These areas can be labelled on one vehicle, tagged with a specific name, for example; “gearbox driveshaft flange bearing” and the temperature profile or appearance variance recorded over time. Once this labelling is complete for a type of vehicle, the data can be used for an entire fleet by type using the expected component heat/visible profiles. This is achieved by reference to image and temperature history records held in the system by vehicle type.
The data patterns learned by the system are used to identify anomalies. The anomaly is reported to maintenance engineers through the user interface and a service is scheduled to verify what problem the anomaly represents. The fault diagnosis information from the engineers is then entered into the system so that it can learn anomaly significance. If the same anomaly is detected in the future, it will be reported with a suggested list of possible problems.
As the system gathers more data, the need for routine service inspections by maintenance staff reduces since the diagnosis becomes more specific.
[wufoo username=”gobotix” formhash=”z1acut3v118o3yw” autoresize=”true” height=”639″ header=”show” ssl=”true”]