Microcontroller Anomaly Detection Using Local TinyML Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Mobile electronic devices face challenges in efficiently monitoring performance and hardware resource utilization to detect anomalous events, leading to unnecessary resource consumption and battery drain due to the need for continuous data collection and transmission to the cloud.
Innovation Solution
Implementing a microcontroller with a Tiny Machine Learning (TinyML)-based machine-learning model that processes real-time sensor data locally, determining anomalous events without entering active mode, thus reducing resource and battery usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If continuous data collection and transmission to the cloud is implemented for anomaly detection, then detection capability is improved, but resource consumption and battery drain increase
Solution Approach 1:
The system segments the anomaly detection function by deploying a lightweight machine learning model directly on the mobile device's processor, separating the detection task from cloud-based processing. This allows local inference using sensor data without requiring continuous cloud communication, reducing energy consumption while maintaining detection capability.
Solution Approach 2:
The device performs self-diagnosis by using its own computational resources to run the machine learning model and analyze sensor data locally. The system serves its own anomaly detection needs without relying on external cloud services, enabling autonomous operation with minimal energy expenditure on data transmission.
2Reliability
If continuous data collection and transmission to the cloud is implemented for anomaly detection, then detection capability is improved, but device resources are consumed unnecessarily
Solution Approach 1:
The system segments the anomaly detection function by deploying a lightweight machine learning model directly on the mobile device's processor, separating the detection task from cloud-based processing. This allows local inference using sensor data without requiring continuous cloud communication, reducing energy consumption while maintaining detection capability.
Solution Approach 2:
The patent transforms the detection parameters by converting raw sensor data into feature representations that can be processed efficiently by the machine learning model. This parameter transformation enables accurate anomaly detection with reduced computational overhead, improving resource efficiency while maintaining detection accuracy.
3Power
If the device enters active mode for data processing, then processing capability is improved, but battery drain increases
Solution Approach 1:
The system applies partial action by activating the processor only when sensor data indicates potential anomalies or at scheduled intervals, rather than maintaining continuous active processing. The machine learning model processes data in a dormant state, activating full processing capability only when necessary, thereby reducing overall battery consumption while maintaining adequate detection power.
Data Source
AI summary
In one embodiment, a method performed by a microcontroller of an electronic device includes accessing one or more real-time sensor data associated with one or more sensors of the electronic device, determining, by a machine-learning model running on the microcontroller, that an anomalous event has occurred on the electronic device by processing the one or more real-time sensor data with the machine-learning model, and sending, upon the determination that the anomalous event has occurred, a notification regarding the anomalous event to an application running on the electronic device.