Neural Network aided burning Rate Determination of Energetic Materials

A Convolutional Neural Network (CNN) model was trained using Machine Learning (ML) tools with the aim of automating the analysis of high-speed recordings of solid rocket propellant combustion tests. To obtain sufficient training data for such a CNN, simplified regression images were simulated. Then different model constructions with different input, hidden and output layers were created, trained and evaluated regarding its scoring. The model with the smallest mean square deviation from the expected value was then fed with real data and shows a high accuracy as expected. The model is able to determine different regression lines in different sections with changing slopes, which allows to determine their slopes per section and in the overall average. From this, the burn rate can be calculated automatically with the place and time conditions from the calibration files of the corresponding measurement.