Multi-Robot System for Automated Fluorescent Penetrant Indication Inspection with Deep Neural Nets

Abstract Fluorescent Penetrant Inspection (FPI) is the most widely used Non Destructive Testing (NDT) method in the aerospace industry. FPI is currently a manual visual inspection process, which by means of fluorescent dye, aims to distinguish between relevant indications (associated with defects) and non-relevant indications (due to insufficient wash-off, dust or other non relevant factors). This NDT method is largely influenced by human factors due to its nature, introducing several challenges on inspection consistency and reliability. In this paper, a multi-robot inspection system is presented to automate the FPI process. The system autonomously performs image acquisitions of the part under inspection, guarantees full inspection coverage of the part, analyzes the images to recognize regions of interest (e.g., regions where fluorescent dye leaves certain linear characteristics), executes the wipe-off operation (enabling penetrant bleed-back process) as required by the FPI process, and subsequently distinguishes defects against other non-relevant indications by utilizing deep neural network models. This automated system has achieved an inspection accuracy comparable to a human inspector while providing benefits pertaining to consistency, reliability and productivity. A proof-of-concept system has been deployed in an aviation manufacturing environment, and experimental results have shown the system’s capacity to perform the FPI process and detect defects in aerospace components, hence enabling the automation of the entire FPI line.