Submarine ML Control Using Passive SONAR for Deterrent Avoidance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Submarines and submersibles face challenges in navigating underwater environments due to the inapplicability of surface ship navigation techniques, particularly in avoiding deterrents like trawlers, where detection and discrimination are difficult, and reliance on external data sources is limited.
Innovation Solution
A machine learning algorithm is trained onboard using sensor signals, such as SONAR, to determine relationships between sensor data and actions, enabling autonomous navigation away from deterrents by controlling buoyancy, rudder, and propulsors without external communication.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional surface ship navigation techniques are used, then navigation is simple and direct, but these techniques are precluded for submarines and submersibles operating underwater
Solution Approach 1:
The submarine/submersible trains its own machine learning algorithm using sensor data collected during normal operations. The system serves itself by autonomously learning navigation and deterrent avoidance strategies without external intervention, adapting to underwater constraints while maintaining operational simplicity
Solution Approach 2:
The system pre-trains the machine learning algorithm using historical sensor data and simulated scenarios before deployment. This preliminary training enables the submarine to quickly adapt to new environments and threats without requiring complex real-time decision-making systems
2Measurement precision
If active SONAR navigation is used, then position determination is accurate, but the submarine can be readily detected by deterrents
Solution Approach 1:
The patent replaces active SONAR (acoustic wave emission) with passive acoustic listening and machine learning-based position estimation. The system substitutes mechanical sound emission with algorithmic processing of ambient acoustic signals to determine position without revealing its presence
Solution Approach 2:
The machine learning algorithm acts as an intermediary between passive sensor inputs and navigation decisions. It processes ambiguous acoustic signals and sensor data to infer position and deterrent locations without requiring direct active SONAR contact, reducing detection risk while maintaining navigation accuracy
3Quantity of substance
If external data sources are used for training, then training data availability increases, but data transmission to submerged watercraft is problematic and of low bandwidth
Solution Approach 1:
The submarine collects and processes its own operational sensor data for training purposes. By serving itself, the system accumulates training data locally without relying on external data transmission, eliminating bandwidth limitations and information loss during data transfer
Solution Approach 2:
The system pre-processes and stores training data locally during operations before it is needed for algorithm refinement. This preliminary data accumulation ensures sufficient training data is available onboard without requiring subsequent data transmission from external sources
4Extent of automation
If machine learning algorithms are trained onboard, then autonomous control is achieved, but device complexity increases
Solution Approach 1:
The machine learning algorithm is pre-trained using simulated data and historical records before deployment. This preliminary training reduces the computational burden during onboard operation, as the algorithm only needs incremental learning from actual sensor data rather than complete training from scratch
Solution Approach 2:
The training process is segmented into phases: initial offline training with simulated data, incremental onboard training with actual sensor data, and periodic updates. This segmentation distributes computational complexity across time and locations, making onboard implementation feasible
Data Source
AI summary
A method of training a machine learning, ML algorithm to control a watercraft is described. The watercraft is a submarine or a submersible submerged in water. The method is implemented, at least in part, by a computer, comprising a processor and a memory, aboard the watercraft. The method comprises: obtaining training data including respective sets of sensor signals, related to respective deterrents, and corresponding actions of a set of communicatively isolated watercraft, including a first watercraft; and training the ML algorithm comprising determining relationships between the respective sets of sensor signals and the corresponding actions of the watercraft of the set thereof. A method of controlling a watercraft by a trained ML algorithm is also described.