L-best Inference Algorithm for Automated Multi-Failure Diagnosis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing diagnosis methods for complex systems require user interpretation to identify multiple simultaneous failures, which is skill-dependent and inefficient, and are computationally intensive, leading to increased diagnosis time and unnecessary component replacements.
Innovation Solution
The implementation of an L-best inference class algorithm combined with marginal inference, using heuristic techniques to automatically analyze symptoms and provide ranked lists of failure sets, reducing computational costs and improving troubleshooting efficiency by selecting appropriate inference techniques based on symptom data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If L-best inference class algorithm is used to automatically identify multiple failure sets, then diagnosis accuracy and automation are improved, but computational time and complexity increase
Solution Approach 1:
The patent segments the computational problem by dividing failure sets into different sizes (single failures, double failures, triple failures, etc.). The algorithm processes each segment separately using L-best inference, allowing automated identification of multiple failure combinations without requiring exhaustive computation of all possible failure sets simultaneously.
Solution Approach 2:
The patent applies partial action by limiting the search to L-best (top L) inference results for each failure set size rather than computing all possible failure combinations. This provides sufficient automation for practical diagnosis purposes while avoiding the computational burden of exhaustive enumeration of all failure sets.
2Reliability
If exhaustive search of failure sets is performed, then complete diagnosis coverage is achieved, but computational intensity increases significantly
Solution Approach 1:
Instead of performing exhaustive search of all possible failure sets, the patent computes L-best inference results for failure sets of different sizes (single, double, triple failures). This partial approach achieves sufficient diagnosis completeness for practical purposes while dramatically reducing computational intensity compared to exhaustive enumeration of all possible failure combinations.
Solution Approach 2:
The patent segments the exhaustive search problem into manageable chunks by processing failure sets of different sizes separately (k=1 for single failures, k=2 for double failures, etc.). Each segment is solved using L-best inference, providing reliable diagnosis coverage without the computational burden of a single exhaustive search.
3Productivity
If marginal inference class algorithm is used, then computational efficiency is improved, but ability to identify multiple simultaneous failures requires user interpretation
Solution Approach 1:
The patent dynamically adjusts the diagnosis approach by computing L-best inference results for multiple failure set sizes (k=1, 2, 3, etc.) and automatically presenting the results in a structured format. This dynamic computation maintains high diagnostic speed while automatically identifying multiple simultaneous failures, eliminating the need for users to manually interpret marginal inference results.
Solution Approach 2:
The patent provides feedback by presenting ranked failure sets with their probabilities in a structured output format. The system automatically computes and presents multiple failure combinations (single, double, triple failures) with their likelihoods, enabling users to quickly identify the most probable failure scenarios without requiring advanced interpretation skills.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A system/method of diagnosing combinations of failures in a system includes receiving symptom data (116) including information relating to observed or detected symptoms in a system. The system/method generates (D4a, D4b, D5) failure data (118) including information relating to at least one most probable failures in the system based on the symptom data, and processes (D9) the failure data and the symptom data using an L-best inference (e.g. a Ranked Algorithm (RA))technique in order to generate failure set data (120), the failure set data including information relating to at least one most probable combination of the failures that explain the symptoms.