Fuel Cell Stack Assembly Using ML Bleed-Down Fault Screening
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing fuel cell production methods rely on predetermined test limits for identifying faulty cells, leading to incorrect detection and inefficiencies in sorting out abnormal cells.
Innovation Solution
A machine learner is used to evaluate voltage curves during the 'bleed-down' phase, assigning fuel cells to fault-free or faulty groups based on learned behavior patterns, eliminating the need for predetermined test limits and improving detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If predetermined test limits are used to identify faulty fuel cells, then the production process is simple and fast, but the detection accuracy is low leading to incorrect sorting of fuel cells
Solution Approach 1:
The patent changes the evaluation parameter from static predetermined test limits to dynamic machine learning-based assessments. The machine learner analyzes multiple voltage curve parameters (voltage drop, slope, curvature) and continuously adapts its evaluation criteria based on learned patterns, transforming the detection system from rigid threshold-based to flexible pattern-based evaluation.
Solution Approach 2:
The patent replaces the mechanical/threshold-based test limit system with an intelligent machine learning system. Instead of using fixed predetermined limits, the system employs algorithms that automatically learn and adapt detection criteria from data, substituting rigid mechanical evaluation with adaptive intelligent evaluation.
2Reliability
If predetermined test limits are used for fuel cell testing, then the production process is efficient, but the reliability of fuel cell stack assembly is reduced due to incorrect detections
Solution Approach 1:
The patent implements feedback mechanisms where the machine learner continuously improves its detection accuracy by learning from results. The system analyzes outcomes of fuel cell assessments and uses this feedback to refine its evaluation models, creating a closed-loop system that progressively enhances reliability while maintaining production efficiency.
Solution Approach 2:
The patent performs preliminary machine learning-based assessment of fuel cells during production before final assembly. By evaluating and sorting fuel cells early in the production process using adaptive machine learning criteria, the system prevents faulty cells from reaching assembly, ensuring high reliability of the final fuel cell stack.
3Measurement precision
If traditional test limit comparison is used, then the production process is straightforward, but the number of incorrectly detected faulty cells is high
Solution Approach 1:
The patent creates a universal machine learning-based evaluation system that can assess different types of fuel cells and various fault conditions through a single adaptive platform. The machine learner is designed to handle multiple evaluation scenarios and can be applied across different production contexts, providing multi-functional detection capability.
Solution Approach 2:
The patent implements a self-improving detection system where the machine learner automatically refines its own evaluation criteria without requiring manual adjustment of test limits. The system serves itself by continuously learning from data and improving its detection accuracy autonomously, reducing the need for human intervention in optimizing detection parameters.
Data Source
AI summary
The presented invention relates to a production method (100) for producing a fuel cell stack. The production method (100) comprises producing (101) a number of fuel cells in a production line (201) by a plurality of production steps, determining (103) voltage values of a voltage of an individual fuel cell of the number of fuel cells from a start time, at which a fuel supply to the fuel cell is interrupted, to an end time, assigning (105) the voltage values to a first group, which describes a fault-free condition of the fuel cell, or to a second group which describes a faulty state of the fuel cell, by means of a machine learner, sorting out (107) fuel cells with voltage values assigned to the second group by the machine learner, and assembling (109) only fuel cells with voltage values assigned to the first group by the machine learner into a fuel cell stack.