Advantech Medical AI System
Thermal & Acoustic Improvement
Improving cooling efficiency and reducing system fan noise of a Medical AI Computer
Spring 2023 MAE 156B Senior Design Project
University of California, San Diego
This project is sponsored by Advantech
Advantech USM-500 Medical Computer Front & Rear
OVERVIEW
Background
AI has revolutionized medical imaging diagnostics, allowing medical professionals to supplement operating procedures in real time. However, AI image processing is resource intensive so computers with powerful processing chips are needed to provide real-time metrics. Computations for these processing tasks are mainly done by computer processing units (CPU) and graphical processing units (GPU). These components generate high amounts of heat during operation and cooling them with heat sinks and air convection alone produces high system fan noise.
Objective
The primary objective of this project is to improve the thermal and acoustic performance of a medical AI computer system through the use of a liquid cooling system. These improvements will aim to maximize cooling efficiency while minimizing system fan noise, thus minimizing its impact on the surrounding operating room environment and increasing system longevity. As a secondary objective, an anti-leak system will be developed to prevent and protect from liquid coolant leaks, further increasing system reliability and longevity.
Final Design
Isometric View
Top-down View
Description of Final Design
The design solution to tackle the primary and secondary objectives is to utilize a liquid cooling system with water blocks for both the CPU and GPU, while also manufacturing an external case to store the larger cooling components. The USM-500 utilizes an Intel LGA-115x socket CPU and RTX NVIDIA Quadro A4500 GPU for its main AI computing purposes, so planning was done around those essential components to cool. The project employed the Bykski Metal/POM GPU Water Block and the Bykski CPU-XPR-CI CPU Water Cooling Block to replace the existing air cooling fans and heat sinks on the existing GPU and CPU in the USM-500. The external case outside was manufactured with PMMA acrylic and is of similar size to the USM-500 at 320 x 309.5 x 138.65 mm. It also contains two 240mm Corsair radiators, four Noctua 120mm low-noise fans, one 800 L/h centrifugal water pump, a 50mm x 190mm or 370 mL reservoir, clear ⅜ in. ID x ½ in. OD PrimoFlex soft tubing, and LEAKSHIELD device for leak protection measures.
Summary of Performance Results
Initial Target Goals
Reduce CPU and GPU temperatures by 10°C, respectively
Reduce overall system fan noise by 10 dBA
Performance Results
CPU and GPU temperatures were reduced by 12-16°C and 46.1-48.2°C, respectively
System fan noise was decreased by 8.1-19.6 dBA
Rendering performance was increased by 71 FPS-195 FPS in HeavyLoad
Coefficient of Performance: 0.01323-0.00745 °C/W
LEAKSHIELD successfully protects from small (pinhole) leaks