Aircraft Icing Prediction Using ML and Onboard Weather Sensors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing aircraft icing detection systems are inadequate for unmanned aerial vehicles (UAVs) and small helicopters, as they lack human intervention to avoid or detect ice accumulation quickly, leading to potential loss of control and catastrophic failures due to ice buildup on rotor blades.
Innovation Solution
A machine-learning-based predictive ice detection system using outside air temperature and dew point sensors, combined with supervised and unsupervised learning algorithms, provides early warning and proactive navigation maneuvers to avoid icing conditions without the need for weather radar, minimizing power and weight requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine-learning-based predictive ice detection system is implemented, then early warning capability and prediction accuracy are improved, but system complexity and computational requirements increase
Solution Approach 1:
The system performs preliminary actions by continuously collecting and storing atmospheric data (temperature, humidity, pressure, dew point) before icing conditions occur. The machine learning model is pre-trained on historical data to recognize patterns that precede icing events, enabling early warning predictions before actual ice accumulation begins, thus improving detection accuracy while managing system complexity through proactive data gathering.
Solution Approach 2:
The patent introduces an intermediary machine learning model that acts as a mediator between raw sensor data and icing condition predictions. This intermediary layer processes atmospheric parameters through trained algorithms (random forest, support vector machines, neural networks) to generate predictive outputs, bridging the gap between simple sensor measurements and complex icing condition assessments without requiring direct complex physical modeling.
2Reliability
If multiple sensors and machine learning algorithms are used, then prediction reliability is improved, but power consumption and weight increase
Solution Approach 1:
The system applies partial action by selectively activating different machine learning algorithms and data processing levels based on environmental conditions and flight phases. Rather than continuously running all computational models at full capacity, the system adjusts processing intensity to match the actual risk level, maintaining high reliability when needed while reducing power consumption during low-risk periods through dynamic computational resource management.
Solution Approach 2:
The patent utilizes parameter changes by monitoring multiple atmospheric parameters (temperature, humidity, pressure, dew point) and adjusting prediction model activation based on combinations of these parameters. When parameters indicate high-risk conditions, the system increases computational activity; when parameters show safe conditions, it reduces processing, thus maintaining reliability through parameter-based triggers while optimizing power consumption through conditional computation.
3Loss of time
If stand-off distance prediction is increased, then avoidance time is improved, but measurement precision requirements increase
Solution Approach 1:
The system performs preliminary detection by identifying atmospheric conditions that precede actual icing by a measurable stand-off distance and time. By analyzing patterns in temperature, humidity, and pressure data before ice accumulation begins, the system provides early warning that enables avoidance maneuvers while maintaining acceptable precision through predictive rather than reactive measurement approaches.
Solution Approach 2:
The patent replaces direct mechanical/physical ice detection methods with machine learning-based predictive modeling. Instead of relying on physical ice sensors that require actual contact with ice, the system substitutes computational models that predict icing conditions based on atmospheric parameter patterns, thereby increasing stand-off distance and avoidance time while managing precision requirements through algorithmic pattern recognition rather than direct physical measurement.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
Systems and methods for machine-learning-based aircraft (102) icing (204, 212, 222) prediction use supervised and unsupervised learning (328) to process real-time environmental data, such as onboard measurements of outside air temperature and dew point (222), to predict a risk of icing (204, 212, 222) and determine whether to issue an icing (204, 212, 222) risk alert (312) to an onboard crew member or a remote operator, and/or to recommend an icing avoidance maneuver (214). The systems and methods can use reinforcement learning (512) to generate a confidence metric in the predicted risk of icing (204, 212, 222), to determine a time or distance to predicting icing (204, 212, 222), and/or to not issue an alert (312) or recommend a maneuver (214) in consideration of historical data in a "library of learning" and/or other flight (202) data such as airspeed, altitude, time of year, and weather conditions. The predictive systems and methods are low-cost and low-power, do not require onboard weather radar, and can be effective for use in smaller aircraft (102) that are completely icing-intolerant.