Microcontroller Anomaly Detection Using Incremental PCA
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection methods for microcontrollers in physical systems are inefficient due to high memory usage and complex computations, especially when producing models from data streams with limited memory capacity, and may not effectively distinguish normal from abnormal data.
Innovation Solution
A method involving incremental computation of principal components using covariance-free analysis and modified Gram-Schmidt orthonormalization to create a simple anomaly detection model with a user-defined detection threshold, which can be implemented by a microcontroller, allowing for real-time monitoring and alert generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional anomaly detection methods (Gaussian mixture model, Z-score) are used, then anomaly detection capability is achieved, but memory occupation and computational complexity increase significantly
Solution Approach 1:
The patent extracts only the essential features needed for anomaly detection by computing principal components incrementally and storing only the orthonormal base and detection threshold, discarding redundant data and complex model structures that consume excessive memory and computational resources
Solution Approach 2:
The patent changes the parameter representation from storing complete data samples or complex probability distributions to storing only the orthonormal base vectors and a scalar detection threshold, fundamentally reducing the memory footprint while maintaining detection capability
2Reliability
If complete learning data is stored for model production, then accurate anomaly detection model is produced, but memory capacity is exceeded
Solution Approach 1:
The patent segments the learning process into incremental updates, processing data in small batches and maintaining only compressed representations (orthonormal base) rather than storing complete learning datasets, enabling model production within limited memory constraints
Solution Approach 2:
The patent creates a compressed copy of the learning data in the form of an orthonormal base that captures the essential variance of the data without requiring storage of the original complete dataset, enabling accurate anomaly detection with minimal memory usage
3Reliability
If complex machine learning algorithms are implemented, then comprehensive anomaly detection is achieved, but implementation on microcontroller becomes difficult
Solution Approach 1:
The patent replaces expensive, complex machine learning algorithms with simple, computationally inexpensive operations (vector projections and threshold comparisons) that can be easily implemented on resource-constrained microcontrollers, sacrificing some algorithmic complexity for practical deployability
Solution Approach 2:
The patent substitutes complex statistical computations (Gaussian mixture models, covariance calculations) with simpler linear algebra operations (orthonormalization, vector projections) that are more suitable for implementation on embedded microcontroller systems with limited floating-point capability
Data Source
AI summary
According to one aspect, a computer-implemented method can be used for producing an anomaly detection model. The method includes obtaining a learning data stream from a physical system, incrementally computing principal components of the learning data stream, performing orthonormalization of the principal components computed so as to obtain an orthonormal base representing the learning data stream, and producing the anomaly detection model including the orthonormal base and a detection threshold defined by a user. The anomaly detection model can then be applied to a physical system.