Huawei Cloud launched the Pangu Railway Model at the ‘Reshaping Industries with AI” conference, which took place in Dongguan, China.

The trailblazing Pangu Model 3.0 and Ascend AI cloud services offer a state-of-the-art AI solution which will transform fault identification across the rail freight industry whilst supporting critical safety standards, increasing efficiency and cutting labour intensity.

Pangu Model 3.0 provides a system of pre-trained models which are quickly changeable in order to support specific scenario needs whilst handling complex challenges to an array of industries. Making use of huge data sets alongside machine learning algorithms, the AI solution will work with weather forecasting, drug development, within the mining industry and for the rail industry identifying train faults.

Significant challenges face the rail industry in train fault identification with the current approach, seeing a great deal of manual inspection via the Train Freight Detection System, which is labour-intensive, costly and inefficient.

Introducing Huawei’s Pangu Railway Model will support the difficulties arising from this approach by utilising AI to automate the process, which in turn reduces the likelihood of human error, creating increased levels of safety in a more efficient manner.

Currently, Train Freight Detection Systems are located between tracks on rail networks and work as a train passes the detection station, allowing the system to calculate the train’s speed via magnetic steel sensors attached to the wheels.

The TFDS then adjusts its snapshot frequency in order to match the train’s speed, allowing one image to be captured every few milliseconds. The pictures are then uploaded to a central server in the main for manual fault detection due to technical limitations.

In China, a typical depot sees over 40,000 trains operate across 800 lines each day, which is a vast amount of manual checks to be completed by the TFDS system.

80 pictures are taken by the system for each freight carriage, of which there are roughly 50 per train and this leads to thousands of pictures taken for every train. The vast number of images taken means that those inspecting the pictures have approximately 10 minutes to carry out the necessary checks, looking for faults and cracks in undercarriages.

The Pangu Railway Model provides accurate identification of 67 types of freight cars and can discover 430 types of faults on both the railway and freight cars. The AI solution can scan millions of images captured using TFDS before filtering out 95% of those which are free of faults.

These innovative solutions provide train inspectors with more time to view the remaining images, which brings improvements in accuracy and efficiency.

Zhang Pingan, Huawei Executive Director and CEO of Huawei Cloud said:”The capability of the Pangu Railway large model is continuously improved. Now, we can accurately identify 67 types of trucks running on the live network and more than 430 types of faults. The fault detection omission rate is zero, and the screening rate of fault-free images is as high as 95%.”

The groundbreaking Model also makes use of continuous learning, which improves performance. Abnormalities found can be manually labelled and added to the model, which adds to its performance. Further to this, the Pangu Railway Model can support simultaneous detection across distributed services, which reduces the time needed for detection substantially and can be completed in less than 8 minutes.