Hardware Failure Detection with Multimodal Fusion for Weak Signals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing rule-based hardware failure detection methods are insufficient and inaccurate, failing to cover all hardware issues and requiring manual handling, as they rely on strong signals related to serious failures and ignore weak signals.
Innovation Solution
A machine learning-based hardware failure detection model that utilizes multimodal feature fusion, combining hardware event logs and performance signals to detect failed components, employing pattern-level embedding for logs and frequency domain analysis for signals, with a training dataset refined through heuristic iterative procedures and low confidence ticket filtering.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If rule-based detection methods are used, then the detection process is simple to implement, but the detection accuracy and coverage are insufficient
Solution Approach 1:
The patent replaces rule-based detection mechanisms with a machine learning model that processes hardware event logs and performance signals. The model uses pattern-level embedding for logs and frequency domain analysis for signals, substituting manual rule configuration with automated learning-based detection that achieves superior accuracy while maintaining operational simplicity.
2Device complexity
If only strong signals are considered, then the detection method is easier to implement, but weak signals related to subtle hardware issues are ignored
Solution Approach 1:
The patent implements dynamic signal processing by transforming performance signals from time domain to frequency domain using Fast Fourier Transform. This dynamic transformation enables the detection system to capture both strong and weak signals across different frequency spectrums, allowing subtle hardware issues to be detected without overwhelming system complexity.
Solution Approach 2:
The patent adds a frequency domain dimension to the detection process by performing FFT on performance signals. This dimensional transformation allows the system to analyze signals from multiple perspectives (time and frequency domains simultaneously), improving detection coverage for both strong and weak signals while maintaining manageable complexity through structured processing.
3Device complexity
If manual handling is required for undetected hardware issues, then the detection system can be simpler, but the overall system efficiency decreases
Solution Approach 1:
The patent implements self-service through automated root cause inference that directly identifies failed hardware components based on detected patterns and signal analysis. The system automatically generates diagnostic conclusions without requiring manual intervention, enabling the detection system to handle all hardware issues autonomously and maintain high system efficiency.
4Measurement precision
If multimodal feature fusion is implemented, then the prediction accuracy improves, but the computational complexity increases
Solution Approach 1:
The patent segments the multimodal feature fusion process into distinct processing streams: one for hardware event logs using pattern-level embedding and another for performance signals using frequency domain analysis. Each stream is processed independently through specialized modules, then their features are fused at the decision level. This segmentation reduces computational complexity by avoiding redundant processing while maintaining high prediction accuracy through complementary feature integration.
Data Source
AI summary
The present disclosure proposes a method, apparatus and computer program product for performing hardware failure detection based on multimodal feature fusion. A set of hardware event logs of a machine may be obtained, the machine including multiple hardware components. A set of performance signals of the machine may be obtained, the set of performance signals being time-series data. At least one failed hardware component in the machine may be detected based on the set of hardware event logs and the set of performance signals.