Nonlinear Regression Network Kernel Selection for Fault Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional condition monitoring techniques, such as linear regression models, are insufficient in accurately representing the nonlinear relationships between sensor inputs and outputs, leading to inadequate fault detection in machinery.
Innovation Solution
An incremental learning method for building a nonlinear regression network by selecting and adding kernels from a library, such as linear, Gaussian, polynomial, or sigmoid kernels, and refining the network until a predetermined regression error is minimized or reduced, using a cost function to optimize kernel selection and replacement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a linear regression model is used to estimate expected sensor values, then the model is simple and computationally efficient, but it is insufficient to accurately represent nonlinear relationships between sensor inputs and outputs
Solution Approach 1:
The patent transforms the linear regression model into a nonlinear regression network by changing the functional parameters from linear to nonlinear kernels. The system selects and combines multiple kernels (linear, Gaussian, polynomial, sigmoid) with different parameters to accurately represent nonlinear relationships while maintaining computational efficiency through incremental learning and cost function-based selection.
Solution Approach 2:
The patent creates a composite regression network by combining multiple kernel functions (linear, Gaussian, polynomial, sigmoid) into a unified nonlinear regression model. Each kernel contributes different characteristics to the overall model, similar to how composite materials combine different substances to achieve superior properties. The cost function evaluates and selects the optimal combination of kernels for accurate nonlinear relationship representation.
2Measurement precision
If multiple kernels are added to the regression network to improve accuracy, then the estimation precision improves, but the computational complexity and time increase
Solution Approach 1:
The patent performs preliminary action by pre-defining a library of candidate kernels with known characteristics before the actual regression task. During incremental learning, the system evaluates each candidate kernel using a cost function and selects the most suitable ones in advance, avoiding the need to develop complex kernels from scratch during the learning process. This reduces computational time while maintaining accuracy.
Solution Approach 2:
The patent applies partial action by selectively adding only the necessary kernels to the regression network based on the specific problem requirements and cost function evaluation. Instead of using all possible kernels, the system incrementally adds kernels until the regression error reaches an acceptable level, avoiding unnecessary computational complexity while achieving sufficient accuracy for the given application.
3Ease of manufacture
If a fixed regression network structure is used, then the model is simple to implement, but it cannot adapt to new data or changing conditions
Solution Approach 1:
The patent transforms the fixed regression network into a dynamic, adaptive system through incremental learning. The network structure can be modified by adding, removing, or replacing kernels based on new data and changing conditions. The cost function continuously evaluates performance and guides structural adjustments, allowing the model to adapt while maintaining implementation simplicity through systematic kernel management.
Solution Approach 2:
The patent implements self-service by enabling the regression network to automatically adjust its own structure through the cost function evaluation mechanism. The system independently determines which kernels to add or remove based on performance metrics, without requiring manual reconfiguration. This self-adjusting capability provides adaptability while keeping the implementation process automated and simple.
Data Source
AI summary
A method for identifying a potential fault in a system includes obtaining a set of training data. A first kernel is selected from a library of two or more kernels and the first kernel is added to a regression network. A next kernel is selected from the library of two or more kernels and the next kernel is added to the regression network. The regression network is refined. A potential fault is identified in the system using the refined regression network.