"Double Festival" cargo transportation is busy
How to ensure that freight trains are always "in state"
Then it has to be mentioned
This pair of "fire eyes golden eyes"
Imagine relying on powerful AI technology
Tens of thousands of parts for a freight train
The inspection can be completed in less than ten minutes
▲TFDS truck fault detection system scene demonstration
This super powerful "AI car inspection" system
It is used by the Zhengzhou North Depot of China Railway Group
After AI training, HUAWEI CLOUD Pangu Big Model
TFDS system
▲The trackside detection station takes high-speed photos of the running trucks
In December 2022, the "AI vehicle inspection" system was officially put into trial. This section upgrades six TFDS detection stations located on trunk lines such as Beijing-Guangzhou Railway, Longhai Railway, and Beijing-Kowloon Railway, adopts "AI vehicle inspection" operation, and manually reviews the faults identified and warned by "AI vehicle inspection".
▲After applying the Pangu large model to train the "AI vehicle inspection" algorithm, TFDS identifies the job site
Compared with manual vehicle inspection, the method of "AI vehicle inspection" combined with manual review:
- The duration of assignments was reduced from an average of 17 minutes to 14 minutes;
- The number of workers was reduced from 4-5 to 2-3;
- The fault identification accuracy rate increased from 98.26% to 99.89%.
After 10 months of trial, the "AI vehicle inspection" system has realized the intelligent identification of more than 400+ faults mainly of railway truck stopping key faults, which has been greatly superior to manual operation, with higher efficiency and stronger reliability, and has been gradually promoted to many road bureaus across the country for trial.
What is the origin of this "AI vehicle inspection" system?
This system is called TFDS, that is, truck fault trackside image detection system. As long as a truck passes through the detection station, the electronic eye will shoot at high speed, dynamically collect more than 4,800 images such as underbody accessories and the side state of the car body, and transmit them to the dynamic inspection room in real time.
The traditional TFDS system requires dynamic inspectors to analyze each picture in time, find hidden dangers of vehicle failure, and feedback the pictures of fault parts to the front-line inspectors.
The "AI-trained" TFDS system can use "AI" technology to analyze the collected images in real time and automatically identify various types of railway wagon faults.
Why do you need to upgrade the TFDS of the Zhengzhou North Depot?
- "Too busy" here
The 80 inspection stations in the 5T testing workshop have to complete the inspection task of more than 40,000 trucks and more than 2.8 million pictures every day.
- Important issues settled
In 2021, the Truck Division of China Railway Group took the TFDS fault image intelligent recognition project as the first batch of scientific research plans of China Railway Group, and designated the 5T inspection workshop of Zhengzhou North Depot of Zhengzhou Bureau Group Company as the pilot unit of the project, and jointly researched and jointly promoted the TFDS fault image intelligent recognition project with Huawei and Huitie Technology Company.
Since then, the "AI vehicle inspection" system upgrade pilot "flowered" Zhengzhou North depot.
How powerful is the TFDS system after "AI training"?
- Experts and doctors jointly promote
In order to ensure the progress and effect of R&D, the 5T inspection workshop in this section has set up an experienced "expert" team to provide data from the aspects of fault classification and judgment methods based on typical fault cases in recent years, so as to improve the accuracy of the algorithm and reduce the false alarm rate.
- Algorithmic models are world-leading
The "AI training" of the TFDS system adopts the Huawei Cloud Pangu CV Model, the industry's largest visual pre-training model, and its training and inference speed are at the world's leading level.
At the same time, this "AI vehicle inspection" system based on Pangu large model automatically summarizes the characteristics of components according to a large number of data samples, automatically finds fault rules, and continuously improves the analysis effect in actual trials, so as to achieve gradual and fine identification of detailed features from the whole to the part and then to the fault details.
Another problem encountered with TFDS testing is the scarcity of failure samples. With the continuous improvement of the safety level of mainland railways, many faults have rarely occurred, and most people have never seen them, but once such faults occur, they are usually major failures. For AI models, small samples without samples will also restrict the improvement of model accuracy.
While Zhengzhou North Station simulated and manufactured a large number of fault samples in the section repair workshop, HUAWEI CLOUD generated a large number of training samples based on small-sample learning technology and sample generation technology to improve model quality.
For example, the fault of the rocking pillow disc is found only one fault sample nationwide, but through small sample learning, this fault can be correctly identified.
Video loading...
▲Huawei Developer Conference 2023 (Cloud) officially released the Pangu Railway model
In recent years, the railway sector has continuously increased investment in science and technology
The use of these new equipment and new technologies
The way people work in the field has been reshaped
Looking forward to more technological means in the future
It is applied to all aspects of railway transportation production
*This article is reprinted from: Central Plains Railway
Follow @HUAWEI CLOUD to learn more