Jinpin KG 4208-G4 Improves Efficiency of Railway Fault Detection

 

Background:

A Beijing-based railway application and safety inspection equipment company, founded in 1998, has long been dedicated to software development for railway transportation and integration of safety inspection equipment systems. The company has strong technical capabilities and rich engineering implementation experience in large-scale information management system development, system integration, intelligent video analysis, machine vision detection, and other fields. They have built an integrated business chain from products to solutions and consulting services in the railway transportation field. The company is a professional system developer and consulting service provider in the railway transportation industry and is one of the few in China that provides comprehensive solutions in the railway transportation business. The company's railway vehicle information software products are in an absolute leading position in China's railway transportation industry and have multiple patents based on industrial vision technology for dynamic detection of operating vehicles, lines, and equipment.

 

Challenge:

The top of the locomotive and the pantograph inspection system use multiple high-speed and high-resolution cameras to capture images of the major components of the locomotive's top pantograph and insulators while scanning the panoramic view of the locomotive's top. The captured images are transmitted to the server and automatically analyzed and detected through image recognition to identify worn-out sliding plates, as well as common faults of pantographs, such as wear and tear, cracks, and missing parts, to generate automatic alerts. Inspection personnel can inspect and confirm the abnormal alarm areas on the locomotive's top by browsing the images.

 

First, a high-definition video file is generated by capturing a panoramic view of the top of the locomotive using high-speed and high-resolution cameras, and then the high-definition video is converted into high-definition images through professional transcoding software. Image recognition algorithms are used to identify the images, and then the faulty points are found. The CPU of the server plays a critical role in image recognition, and the CPU's computing performance directly affects the efficiency of fault diagnosis. As the camera's resolution becomes higher, the generated video becomes larger, and the files that need to be processed become larger, and CPU processing performance has become a bottleneck in the system.

 

Solution:

Based on the customer's pain point of insufficient computing power and a deep understanding of the application, Jinpin Computing Technology recommended the use of the Jinpin KG4208-G4 high-performance computing platform, equipped with 2 Intel Xeon E5-2650v4 processors and 128GB of high-speed cache, and 2 Nvidia GTX 2080TI GPU accelerators, to use the parallel processing power of thousands of CUDA cores of the GPU to improve work efficiency.

 

Jinpin KG4208-G4 GPU Computing Server:

Product Features:

Supports 4 Nvidia GPU accelerators (Tesla/Geforce)

Supports 2 Intel Xeon E5-2600V4 processors

Supports up to 16 DDR4 memories, up to 2TB.

Outputs 448 TFLOPS of deep learning training performance

 

Result:

After adopting the Jinpin KG4208-G4 high-performance computing platform, the image classification recognition work was transferred from CPU processing to GPU processing. Since the number of CUDA cores is 100 times that of CPU cores, and one computer supports 4 GPU processors, the parallel work of thousands of processors achieves amazing processing speed.

The CPU classification recognition after video collection needs to wait for several hours, while the GPU classification recognition is shortened to minutes. According to customer feedback, using GPU for image recognition improves performance by 65 times compared to using CPU for recognition, directly improving work efficiency, and receiving high praise from the leadership of a railway bureau.