On-Orbit, Non-Destructive Surface Surveillance and Inspection with Convolution Neural Network

Abstract

In this paper, the concept for on-orbit, non-destructive Infrared survey and inspection of the surface defects on an inter-planetary human module with large surface area and power capabilities, for long flight duration is derived. Automated Probe with thermal imaging camera is used to capture 2D thermal images at that position during rendezvous around the human module. Thermal imaging datasets are classified under binary classification problem and Custom CNN with TensorFlow Architecture is developed. The test accuracy obtained at initial stage of development is about 92%. Converted 2D high resolution grey thermal images are segmented to measure cracks by mapping the pixels. Upon identification of fault position, on-board crew is alerted and original designer is updated, to address the problem remotely. Thereby, an effort has been done herein to significantly reduce the crew EVA spent in survey for surface faults during mission in harsh space environment and the corresponding pre-mission training requirements.