Automated Fault Classification via Kernel Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current fault detection systems in complex environments are labor-intensive and costly to develop, requiring human intervention and specific knowledge of each platform, leading to delayed detection of equipment failures and increased indirect costs due to secondary damage and unscheduled maintenance.
Innovation Solution
An automated system and method for extracting discriminatory features from time series data using the Kernel Split Find algorithm, which segments data into uniform blocks, identifies candidate kernels, and applies Random Forest variable importance analysis to develop a fault classification system without human intervention, applicable to multiple systems with different sensor suites.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual fault detection systems are used with human intervention, then detection accuracy can be maintained, but development time and costs increase significantly
Solution Approach 1:
The system performs automatic feature extraction and fault detection without requiring manual human intervention. The algorithm autonomously processes sensor data, extracts discriminatory features, and identifies faults, enabling the system to serve itself and eliminating the time-consuming manual development process while maintaining detection accuracy
Solution Approach 2:
The patent replaces manual mechanical analysis processes with an automated computational algorithm. The Kernel Split Find algorithm and Random Forest classifier substitute human experts' manual feature extraction and fault diagnosis work, dramatically reducing development time while preserving or improving detection precision through systematic automated analysis
2Measurement precision
If platform-specific fault detection systems are developed, then detection accuracy for each platform is optimized, but the process becomes labor-intensive and costly
Solution Approach 1:
The patent develops a universal fault detection system that can be applied across multiple different platforms and sensor configurations. The algorithm is designed to work with various sensor suites and platform types without requiring separate custom development for each, reducing overall system complexity while maintaining platform-specific detection accuracy through adaptive feature extraction
Solution Approach 2:
The system adapts to different platforms by automatically adjusting its feature extraction parameters and analysis methods based on the specific sensor data characteristics of each platform. This dynamic parameter adaptation allows a single system design to handle multiple platforms effectively, reducing development complexity while preserving detection precision
3Loss of time
If automated algorithms are used for fault detection, then development time is reduced, but the system requires translation of control systems data into diagnostic indicators
Solution Approach 1:
The patent extracts discriminatory features directly from raw control systems sensor data using the Kernel Split Find algorithm. By automatically identifying and extracting the most relevant features without requiring manual translation or preprocessing into traditional diagnostic indicators, the system reduces development time while managing data translation complexity through automated feature selection
Data Source
AI summary
A method and system for automatically developing a fault classification system from time series data. The sensors need not have been intended for diagnostic purposes (e.g., control sensors). These methods and systems are functionally independent of knowledge related to a particular equipment system, thereby allowing seamless application to multiple systems, regardless of the suite of sensors in each system. Because this algorithm is totally automated, substantial savings in time and development cost can be achieved. The algorithm results in a classification system and a set of features that might be used to develop alternative classification systems without human intervention.