Sensor Fusion for Additive Manufacturing In-situ Inspection

Fringe is an insitu profilometry sensor offered commercially by Additive Monitoring Inc or AMI (dba Phase3D).This product has been licensed to 16 User organizations and used for CRAD and CRADA programs with AFRL, DOE and NIST. Fringe reliably detects both melt layer and powder layer defects insitu on every layer from surface height signatures. It also is used to measure local layer thickness. In this Program we combine two insitu sensor types using sensor fusion to obtain improved improved Faw detection accuracy. The second insitu sensor used is EOS’ commercial Exposure OT thermography sensor, which is already installed on a large base of EOS LPBF printers. UDRI is our Research Institute partner and their EOS M290 printer has both Fringe and EOS Exposure OT insitu sensors installed. We derive a third, implicit insitu sensor signal for applied volumetric energy or VED. This is an implied measure of the local applied energy calculated using the local melt temperature (from Exposure OT) and the local layer thickness (derived from Fringe profilometry). UDRI will print two Gyroid test parts whose complex geometry is known to create local Flaws. These parts will be CT scanned to provide Flaw state “ground truth”.  In Phase I software will be written to extract co-located layer images from multiple data streams: melt and powder layer thickness,  melt and powder surfaces, thermography (melt temperature), VED and CT (ground truth). These images will be examined to identify and establish relationships and associated sensor fusion algorithms best able to detect real Flaws (as defined by CT data). This is a practical expedient given the short Phase I duration, to be greatly enhanced in Phase II as noted below. Flaw detection performance shall be measured (Flaw/No Flaw, True/False) and compiled using all insitu data streams both with and without sensor fusion algorithms. A confusion matrix and associated metrics (precision, recall, etc.) will be compiled for all single sensor and sensor fusion data to assess Flaw detection performance on the test part. This serves as an early objective measure of the value that the VED and sensor fusion concept. In Phase II a much larger Test database shall be developed. This data shall be analyzed using convolutional neural networks (CNN) much better able to establish reliable, accurate correlations between sensor signals and Flaw presence. This work positions the many users of EOS printers to complement their insitu inspection capabilities and Flaw detection capabilities by adding Fringe profilometry. Successful sensor fusion further enhances Users’ ability to find and understand Flaws that are created in their parts as they are made. This information helps Users develop and produce AM parts more reliably and more quickly. ???????