Microcontroller Anomaly Detection via Current Signature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection methods for machines using electric current signals require powerful processors and significant memory resources, making them incompatible with microcontrollers, which are constrained in resources but have high integration capacity, limiting their implementation on embedded systems.
Innovation Solution
A method that learns normal electric current signatures on a microcontroller, comparing new signals to stored knowledge to detect anomalies, allowing for effective detection and alert generation, even on resource-constrained devices like microcontrollers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning techniques such as neural networks are used for anomaly detection, then detection rates are improved, but processor power and memory resources requirements increase significantly
Solution Approach 1:
The patent extracts only the essential features from electrical current signals (such as peak values, zero-crossing points, and waveform patterns) rather than using complex deep learning models. This feature extraction approach maintains high detection accuracy while significantly reducing computational requirements, allowing implementation on microcontrollers with limited resources.
Solution Approach 2:
The patent creates a simplified copy of the anomaly detection functionality that operates on extracted features rather than raw signal data. By working with reduced-dimensional feature representations (e.g., statistical parameters of current waveforms), the system achieves comparable detection performance to neural networks but with much lower computational overhead, enabling deployment on resource-constrained embedded devices.
2Measurement precision
If powerful processors and significant memory resources are used for anomaly detection algorithms, then detection accuracy is improved, but compatibility with microcontrollers is lost
Solution Approach 1:
The patent changes the operational parameters from complex neural network computations to simple feature comparison operations. By transforming the detection algorithm to work with extracted waveform parameters (amplitude, frequency, duty cycle) and comparing them against threshold values stored in microcontroller memory, the system maintains high accuracy while becoming fully compatible with microcontroller architectures.
Solution Approach 2:
The patent segments the anomaly detection process into distinct modular components: signal acquisition, feature extraction, feature comparison, and alert generation. Each segment can be independently implemented using simple microcontroller instructions, avoiding the need for complex integrated deep learning frameworks while maintaining overall system effectiveness.
3Reliability
If electrical current signatures are continuously recorded in knowledge base, then detection capability is improved, but memory usage increases
Solution Approach 1:
The patent extracts only the most discriminative features from electrical current signals (such as peak amplitudes, zero-crossing intervals, and waveform shapes) and stores only these simplified representations in the knowledge base. This feature extraction reduces the memory footprint significantly compared to storing raw signal data, while maintaining the ability to reliably detect anomalies through feature comparison.
Solution Approach 2:
The patent implements a selective recording mechanism that stores only the most representative and discriminative current signatures in the knowledge base, rather than recording all possible signals. By using techniques to identify and store only essential feature patterns, the system achieves sufficient detection capability with minimal memory usage, avoiding the excessive action of storing every possible signal variation.
Data Source
Figure 1
Figure 2
AI summary
One aspect of the invention relates to a method for monitoring (200) the operation of an electrically powered machine, the method (200) comprising a learning phase (F1) during which knowledge comprising electrical current signatures representative of the operation of the machine is constituted, and a monitoring phase (F2) during which the electrical current signals generated by the machine are compared to the knowledge so as to detect an anomaly in the operation of the machine.