Engine Catalyst Diagnostics Using Support Vector Machine
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing engine catalyst diagnostic methods face challenges in achieving high accuracy with modest training requirements and reduced computational needs, often requiring extensive calibration and windowing functions that focus on specific operating conditions, leading to trade-offs in diagnostic performance.
Innovation Solution
The use of a support vector machine (SVM) for catalyst monitoring, which simplifies calibration and reduces computational requirements by applying parameter readings to generate classification outputs, indicating degraded catalyst performance based on a threshold percentage of classification outputs, and enabling non-intrusive monitoring with reduced windowing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If model based approaches or fuzzy logic are used for catalyst diagnostics, then diagnostic accuracy can be improved, but calibration requirements and training data needs increase
Solution Approach 1:
The patent replaces complex model-based diagnostic systems and fuzzy logic approaches with a neural network-based system. This substitution maintains high diagnostic accuracy while significantly reducing calibration requirements and training data needs, as the neural network is trained offline and requires minimal online calibration.
Solution Approach 2:
The patent changes the approach from extensive online calibration to offline training with modest data requirements. By pre-training the neural network offline and using it for real-time diagnostics with minimal calibration, the system achieves high accuracy without the complex calibration procedures required by traditional methods.
2Measurement precision
If extensive windowing functions are applied to consider only selected operating conditions, then diagnostic precision under specific conditions improves, but data analysis time and computational complexity increase
Solution Approach 1:
The patent performs preliminary classification of operating conditions offline during the neural network training phase. This preliminary action creates a structured framework that enables fast real-time diagnostics without requiring extensive windowing functions during actual operation, thus reducing data analysis time while maintaining precision.
Solution Approach 2:
The patent implements a dynamic approach where the neural network adapts to different operating conditions through its learned representations during offline training. This dynamic capability allows the system to handle various operating conditions efficiently in real-time without requiring static windowing functions, reducing computational complexity and analysis time.
3Measurement precision
If comprehensive training data is collected for model based approaches, then diagnostic accuracy improves, but computational requirements and calibration effort increase
Solution Approach 1:
The patent performs the computationally intensive training process offline before deployment. By completing comprehensive training with extensive data in advance, the system achieves high diagnostic accuracy while minimizing online computational requirements, as the trained neural network requires only inference operations during real-time diagnostics.
Solution Approach 2:
The patent creates a trained neural network model that captures the diagnostic knowledge from comprehensive training data. This copied model can then be deployed with minimal computational resources, as it has already learned the complex patterns from the training data during the offline phase, reducing real-time computational requirements while maintaining accuracy.
Data Source
AI summary
Embodiments for predicting catalyst function are disclosed. One example embodiment includes applying a set of parameter readings for a given sample to a support vector machine to generate a classification output, recording a plurality of classification outputs for a plurality of successive samples over a first duration, and indicating catalyst degradation if a threshold percentage of the classification outputs indicates degraded catalyst performance. In this way, catalyst degradation may be indicated using a simplified model that does not require extensive calibration.