Neural Network Impact Area Estimation for Ballistic Loads
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems struggle to accurately estimate the impact area of a smart load in real-time during a mission due to complex, non-linear dependencies on release and impact conditions, and existing models with six degrees of freedom are too resource-intensive for real-time applications on aircraft.
Innovation Solution
A processing system utilizing backpropagation neural networks to estimate the impact area and time of flight, reducing the number of vertices to 8 for simplified polygonal shapes, allowing for real-time calculations on board an aircraft, with a system architecture that includes input parameter management, impact area estimation, and time of flight calculation modules.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complex models with six degrees of freedom are used to predict load trajectory, then prediction accuracy is improved (error within a few metres), but computational resources required become too intensive for real-time applications on aircraft
Solution Approach 1:
The patent creates a simplified copy of the complex six-degree-of-freedom model by using neural networks to learn the essential relationships between input parameters and trajectory outcomes. This neural network copy reproduces the accurate prediction results of the complex model but executes much faster, enabling real-time applications on aircraft without requiring the intensive computational resources of the original model.
Solution Approach 2:
The patent transforms the complex physical model into a parameter-based neural network model. Instead of solving complex differential equations with six degrees of freedom, the system uses neural networks that process input parameters (release conditions, atmospheric data, target information) and directly output trajectory predictions. This parameter transformation maintains prediction accuracy while dramatically reducing computational complexity for real-time use.
2Measurement precision
If models with six degrees of freedom calculate polygonal impact areas with variable number of vertices, then prediction accuracy is improved, but device complexity increases making real-time execution difficult
Solution Approach 1:
The patent segments the impact area representation into a fixed structure of 8 vertices positioned at the intersections of circles centered at the release point. This segmentation approach simplifies the variable-vertex polygonal representations into a consistent 8-vertex format, reducing model complexity while maintaining sufficient accuracy for real-time impact area estimation on aircraft.
3Measurement precision
If offline simulators are used to calculate impact area, then calculation accuracy is improved, but calculation time becomes too long for real-time mission requirements
Solution Approach 1:
The patent performs preliminary action by training neural networks offline using extensive data from the complex six-degree-of-freedom model and experimental results. Once trained, these neural networks are deployed on aircraft to provide real-time impact area predictions. The heavy computational work is done in advance during training, while the actual real-time applications use the pre-trained networks for fast predictions.
Data Source
Figure 1
Figure 2~5
Figure 3
AI summary
A system and a method are here described for the estimation of the impact area of a ballistic or smart load, that can be launched from an aircraft as a function of data or signals indicative of the aircraft flight conditions upon release of the load and of predetermined impact conditions on the target, characterized by the estimation of a polygonal impact area defined by the coordinates of a central point and of a predetermined number of vertices by means of a plurality of corresponding neural networks.