Image Data Stream Switching for GPU Failure Safety
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems in endoscopic applications face challenges in ensuring failure safety, particularly in avoiding image data stream tears during surgeries, as graphic processor units (GPUs) may overheat, reducing computing capacity and making it difficult to detect malfunctions quickly.
Innovation Solution
A method that creates a first checking data stream from the image data stream input and a second checking data stream from the output, diverted downstream of the image processing unit, allowing for comparison to detect deviations and switch to a backup path if the GPU fails, ensuring uninterrupted image processing without relying on GPU computing power.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a monitoring module is used to monitor the GPU, then failure safety is improved, but the GPU must reduce its output when overheating, reducing computing capacity
Solution Approach 1:
The monitoring function is segmented from the GPU itself and placed in a separate monitoring module. This allows the monitoring of critical parameters (temperature, clock frequency) to occur independently, so the GPU does not need to reduce its output for monitoring purposes. The segmentation enables the monitoring system to detect failures without impacting GPU performance.
Solution Approach 2:
A monitoring module acts as an intermediary between the GPU and the control system. This intermediary monitors the GPU's critical parameters and can trigger switching to a backup path when failures are detected, without requiring the GPU to reduce its computing capacity. The intermediary handles the monitoring and decision-making, allowing the GPU to operate at full capacity.
2Reliability
If additional tasks are transmitted to the GPU for monitoring, then malfunction detection is improved, but the GPU must process additional data, increasing processing time
Solution Approach 1:
The monitoring function is extracted from the GPU's processing tasks. Instead of transmitting additional monitoring tasks to the GPU, a separate monitoring module performs the monitoring by observing critical parameters directly. This extraction eliminates the need for the GPU to process additional monitoring data, thus avoiding increased processing time while maintaining effective malfunction detection.
Solution Approach 2:
The monitoring module performs self-service monitoring by directly observing the GPU's critical parameters without requiring the GPU to execute additional monitoring tasks. The monitoring system independently assesses the GPU's health status through parameter observation, eliminating the time overhead of additional processing tasks.
3Measurement precision
If the GPU processes additional monitoring data, then monitoring accuracy is improved, but the image processing speed is reduced
Solution Approach 1:
The monitoring function is segmented into a separate module that operates independently from the image processing pipeline. This segmentation allows the monitoring module to achieve high measurement precision by dedicating its resources to parameter observation, while the GPU maintains full image processing speed without the burden of additional monitoring computations.
Data Source
AI summary
A camera controller (1) with an image data stream input (2) and an image data stream output (4), wherein a first image data stream path (3) and a second image data stream path (5) are formed. An image processing unit (6) is arranged in the second image data stream path (5). A switching unit (7) is controlled by a comparison unit (8). The comparison unit (8) compares a first checking data stream (9), which was diverted from an image data stream (1) upstream of the image processing unit (6), and a second checking data stream (10), which was diverted from the image data stream downstream of the image processing unit (6). The image data stream paths are switched via the switching unit (7) if the deviation of the second checking data stream (9) from the first checking data stream (10) exceeds a threshold value.