China Railway Zhengzhou Group Speeds up Train Operations with Huawei's TFDS solution

We are experiencing a new wave of enterprise transformation that will further integrate the real world with the continuously extending digital world. Information technology (IT), coupled with artificial intelligence (AI), is creating a new and broader future for us.

Traditional industries have to integrate the digital with the real so as to build modernized industrial systems and transform them into high-end, intelligent, and green industries. In recent years, digital technologies have been widely used by the transportation industry to create a more affordable, intelligent, and all-round better service and travel experience for the general public.

Traditional Operations Need Intelligent Analysis

Founded on March 11, 1949, China Railway Zhengzhou Group is a state-owned large railway transport company. Zhengzhou sits at the center of China's railway network, connecting north, south, east, and west of the country.

The railway sections of the Beijing-Guangzhou Railway, Lanzhou-Lianyungang Railway, and Beijing-Kowloon Railway, which are administered by Zhengzhou North Rolling Stock Depot, are the busiest among all sections of each line. There are 80 inspection workstations in the 5T inspection workshop, and inspectors need to check more than 2.8 million pictures of over 40,000 freight trains each day. This means each inspector has to handle 15,000 pictures on average per day. However, freight trains are continuously speeding up while remaining overloaded, which challenges train operation security.

Therefore, Zhengzhou Group urgently needs some means of intelligent analysis to improve the accuracy and efficiency of railway fault identification.

Intelligence Boosts Quality and Efficiency

The freight train business department of China State Railway Group chose TFDS as the pioneering project of its first batch of scientific research programs. The 5T inspection workshop of Zhengzhou North Rolling Stock Depot, Huawei, and Huitie Technology were chosen to cooperate on the project.

The 5T inspection workshop instantly set up a team composed of experienced leaders and staff, who guided Huawei through fault classification and judgment, all based on the typical fault data they had accumulated over the course of many years. This helped Huawei improve the algorithm accuracy and lower the false alarm rate. Huawei also established a professional team comprising 20 experts with PhDs in algorithms. Huawei's chief scientists in the AI domain were also selected to join so that they could provide guidance over the course of the algorithm's development.

Huawei Pangu Railway Model ensures that Huawei's TFDS solution is built on advanced technology. As the world's largest CV training model based on 3 billion parameters, the Pangu Railway Model can vastly shorten algorithm training periods, accelerate algorithm iteration, and improve accuracy. Based on the Model, the TFDS solution can automatically summarize characteristics of components and find out the causes of faults. It can continuously optimize how it analyzes through a deep learning network and massive data samples. It can also pinpoint over 95% of train models as well as various types of faults by following the rule of analyzing overall faults first, then local faults, and finally fault characteristics. This includes 307 types of faults that are specified in the official regulations and procedures of freight train usage and maintenance, alongside over 100 types of faults that are visible to the TFDS. Along with high identification performance, TFDS is capable of making warnings about all key faults and greatly shortening algorithm training periods.

During the cold winter, when all is heated by coal, it is essential to ensure an adequate amount of coal transport and smooth traffic along trunk lines and hubs. However, in November 2022, China's rail transport was severely affected by the COVID-19 pandemic and suffered from a reduced labor force. Despite that, the 5T inspection workshop of Zhengzhou North Rolling Stock Depot still managed to make sure the depot ran normally by using Huawei's TFDS solution to safeguard it. "The pandemic made coal transport harder this winter. However, this system meant we could roll with the punches. We will apply it across other lines in the railway sections we run," said the director of the workshop.

Compared with purely manual operations, the solution enhances working efficiency by 200%, improves the fault identification rate to 99.3%, and speeds up train inspection. Inspectors now have much lighter workloads under a brand-new smart mode of production and organization. To sum up, the workshop has made the following achievements:

Wider scope of identification, no failed alarms against faults

Huawei's TFDS solution can identify images taken by two types of detection devices, including -2 and -3. It can also identify all the key faults of more than 95% of train models without failed alarms.

Cluster deployment, on-demand deployment of computing power

Zhengzhou Group has piloted centralized deployment on a small scale and has seen 30% higher resource utilization.

200% higher working efficiency

95% of fault-free images can be filtered out and the average number of false alarms per train is reduced to below four through AI-powered identification. Inspectors now check fewer images. Huawei's TFDS solution can work around the clock with high precision, considerably relieving stress on inspectors. During the pandemic, in spite of having half the original on-duty workforce, Zhengzhou Group could still identify fault images efficiently to ensure the normal operation of the depot.

China's railways hold a leading position in both network scale and equipment maturity. Huawei's TFDS solution helps railways digitalize traditional operations. Thanks to the solution, Zhengzhou Group can now overhaul trains with a higher intelligence level and save operational costs by tens of millions of yuan. With manual inspection replaced by intelligent identification, the Group can proactively avoid train driving risks and ensure smooth train operations.