Relay Failure Prediction Using Load-Aware Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting the failure of elementary relays are inadequate, leading to premature replacement and unnecessary wear, as they rely on general "B10 values" without considering specific load conditions, resulting in early failures and inefficient maintenance.
Innovation Solution
A machine learning-based approach that monitors and predicts relay failures by analyzing measurable relay and load sizes, using a database of various switching cycles and failure data to determine characteristic values, which indicate the probability of failure, allowing for timely replacement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If general B10 values are used for relay replacement decisions, then relay reliability is improved through standardized testing, but maintenance efficiency deteriorates due to premature replacements and unnecessary switching cycles
Solution Approach 1:
The patent transforms the single B10 value parameter into multiple condition-specific parameters (B10 values for different loads, temperatures, and operating conditions). This allows the system to select the appropriate B10 value based on actual operating conditions, preventing premature replacements while maintaining reliability standards.
Solution Approach 2:
The patent performs preliminary determination of B10 values for various operating conditions during the design and testing phase. This preliminary action creates a lookup table of condition-specific reliability data that can be quickly applied during operation, eliminating the need for real-time complex analysis while improving maintenance decision accuracy.
2Reliability
If relays are replaced based on minimum B10 values across all loads, then reliability is ensured under worst-case conditions, but resource efficiency deteriorates due to unnecessary replacements and wasted switching cycles
Solution Approach 1:
The patent applies different B10 values to different operating conditions (loads, temperatures, environments) rather than using a single conservative value for all cases. This local quality approach matches the actual stress conditions to appropriate reliability data, preventing unnecessary replacements while ensuring reliability where needed.
Solution Approach 2:
The patent converts the potentially harmful effect of using conservative minimum B10 values into a benefit by systematically analyzing different operating conditions. The worst-case B10 values become the basis for a comprehensive condition-specific database, transforming a conservative approach into a precise, condition-aware maintenance strategy.
3Measurement precision
If condition-specific B10 values are determined for multiple loads and temperatures, then maintenance precision is improved, but device complexity increases due to multiple tests and data management requirements
Solution Approach 1:
The patent segments the operating conditions into distinct categories (different loads, temperatures, environments) and determines B10 values for each segment separately. This segmentation allows the complex testing to be organized into manageable, standardized test cases that can be systematically stored and retrieved.
Solution Approach 2:
The patent creates a universal B10 value database that serves multiple functions: reliability prediction, maintenance scheduling, and condition monitoring. This multi-functional database structure handles various operating conditions through a unified system, reducing the need for separate complex systems for each test condition.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a method (200) for predicting the failure of an elementary relay (100), wherein the method (200) comprises the following steps: acquiring (201) measurable relay and load parameters (310) of the elementary relay (100) over a number of switching cycles of the elementary relay (100); determining (202) characteristic values (330) of the elementary relay (100) based on a weighting of the acquired relay and load parameters (310) according to a machine learning algorithm (321), wherein the characteristic values (330) indicate a functional state of the elementary relay (100) for the respective switching cycle; and determining (203) a failure probability (370) of the elementary relay (100) based on the determined characteristic values (330).