Revisiting the Application of Machine Learning in Micrometoroid and Orbital Debris Protection and Risk Assessment

At the 2012 Hypervelocity Impact Symposium we demonstrated the utility of artificial neural networks (ANNs) for predicting the outcome of micrometeoroid and orbital debris (MMOD) impact on a Whipple shield at hypervelocity [1]. We established that machine learning (ML) techniques like ANN are well suited to high dimensionality problems like MMOD impact for which the impact mechanics and material behaviours are highly complex. Indeed, when trained on a database of more than 1000 hypervelocity impact experiments, the ANN had a higher perforation/non-perforation classification accuracy than the state-of-the-art, semi-analytical ballistic limit equation [2] (92% vs. 71%), and could replicate phenomenological features in the shatter regime after [3] that were not considered in the BLE due to their complexity. Follow-on studies demonstrated that other ML techniques such as support vector machines (SVMs) could provide comparable results to the ANN, albeit with different strengths and weaknesses [4], and that these ML models could be utilized to both identify their areas of predictive uncertainty and improve the shield design process [5].