After the completion of ProBaNNt, the decision on how to deal with any given munitions item will no longer be based on a situational ad-hoc-analysis of a single expert, but on a holistic assessment of many past EOD operations supported by artificial intelligence (AI). The overall aim of the project is to:

For this purpose, an easy to use EOD support software tool will be developed. The tool will not replace EOD experts or undermine their decision-making, but instead strengthen their position by providing additional unprecedented analytical capacity. In order to achieve this, the entire process from data acquisition, munitions assessment, risk assessment and the ultimate selection of an EOD method needs to be revisited. During ProBaNNt the following challenges will be addressed:

EOD Database

Thousands of EOD datasets were recorded by expert companies, but remain largely unused. During ProBaNNt these data will be merged into an EOD database, a data management structure will be established and a standard operating procedure for the recording and subsequent handling of such data will be defined. The project consortium possesses a sufficiently large dataset to perform a large data evaluation. The partners will seek contact with relevant authorities and other EOD companies to extend the dataset and to increase representativeness across the industry and diverse environmental conditions. Data provided by external partners will be anonymized.

Data Acquisition

The project will feature field campaigns for the acquisition of new data. ROVs and AUVs will be equipped with high-resolution optical cameras and high-resolution sonar/acoustic cameras to perform visual and acoustic mapping of pre-determined munitions sites. These recordings will be used to subsequently generate photogrammetric/acoustic 3D reconstructions of munitions items to improve decision-making during clearance campaigns. As EOD depends on a comprehensive situational assessment, (e.g. potential presence of other munitions, sediment properties, degree of object burial in the seafloor), a characterization of the surrounding environment of the munitions object has to be performed. Data will be acquired using sediment characterization, side scan sonar and magnetometer mapping

Fig. 1:

Example object area mapping design with an AUV that is equipped with optical camera and magnetometers (© GEOMAR).

Bayesian Neural Network (BNN)

A Bayesian Neural Network (BNN) will be set-up and trained, based on the data gathered in A) and B), to serve as an analytic algorithm for an EOD support tool. For this purpose, critical parameters of munitions, clearance technologies and environmental conditions of past EOD operations will be identified. The BNN provides the ability to identify and analyse complex interrelationships. This way, previously unknown correlations between parameters will become obvious and suggestions for required adaptations of munitions clearance operations can be made. The network will be used to perform probabilistic assessment of risk paths towards unplanned detonation and leakage of hazardous substances into the sea. Ultimately, both a risk value for the given munitions item and a recommendation for EOD procedures will be generated.