Airplane Flight Event Detection Using Self-Trained Sensor Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting airplane flight events in portable electronic devices require significant manual effort for training data generation and adaptation to new environments, limiting their reliability and robustness.
Innovation Solution
A method utilizing two models, one for ambient pressure and another for motion or sound data, where the first model generates training data for the second, allowing self-supervised or distantly supervised training and adaptation to various environments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual data annotation is used to train statistical models for detecting airplane flight events, then the model can be trained to detect events with sufficient reliability, but huge manual effort is required
Solution Approach 1:
The system uses the first model (trained on ambient pressure data) to automatically generate training data for the second model (detecting own motion or ambient sound). The first model's event detections serve as self-generated labels, eliminating the need for manual annotation while maintaining detection reliability across multiple event types
Solution Approach 2:
The detection system is divided into two specialized models: a first model for ambient pressure-based event detection and a second model for motion/sound-based event detection. Each model focuses on specific sensor data types, allowing them to be trained and optimized independently with targeted training data generation
2Adaptability or versatility
If the statistical model is improved to operate in new environments, then detection accuracy increases, but new training data needs to be generated involving significant manual effort
Solution Approach 1:
When deploying to new environments, the system automatically generates its own training data by collecting sensor data and using the first model's event detections as labels. This self-generated training data enables the second model to adapt to new environments without requiring manual annotation efforts
Solution Approach 2:
The system performs preliminary training data generation by using the first model to detect events and create labeled training data before fully training the second model for specific environments. This preliminary action prepares the system for rapid adaptation to new deployment scenarios
3Device complexity
If a single model is used to detect all airplane flight events, then the device complexity is reduced, but the measurement precision and reliability decrease
Solution Approach 1:
The system segments the detection task into two specialized models that process different sensor data types independently. The first model processes ambient pressure data while the second model processes own motion or ambient sound data, allowing each model to achieve high precision for its specific sensor type without the complexity of a universal model
Solution Approach 2:
The system achieves multi-functionality by having two models that can detect different sets of airplane flight events. The first model detects events based on pressure changes while the second model detects events based on motion or sound patterns, providing comprehensive event detection coverage across multiple event types
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach reduces manual effort and enhances the robustness and adaptability of portable electronic devices in detecting airplane flight events, improving their reliability across different environments and scenarios.
Implementation Method 1
A first module for sensing ambient pressure
Implementation Method 2
a second module for sensing own motion or ambient sound
Data Source
AI summary
A portable electronic device is controlled, for example to selectively activate a flight mode, based on first event data and/or second event data. The first event data is generated by operating a first model for detection of airplane flight events, AFEs, on first sensing data representing ambient pressure. The second event data is generated by operating a second model for detection of AFEs on second sensing data representing own motion or ambient sound. A control method in the portable electronic device generates training data comprising groups of time-aligned data samples from the first event data and the second sensing data, generates an updated second model by use of the training data, and replaces the second model by the updated second model. The control method facilitates adaptation of the portable electronic device to detect airplane flight events in new environments and enables improved robustness when detecting AFEs.