FEATURE / THERMAL INSPECTION Figure 9. Mode 1 data of the sub-optimal compaction tests. Course 2 failed to place the 1/4" tows at the far left and right side of the course due to insufficient compaction. consolidated enough to allow for heat transfer into those lower layers, revealing their geometry due to the changes in local thickness. The final cylinder was built to test the effects of suboptimal compaction on the thermal signature. On several crown patch layers the roller compaction was varied to as little as 25% of the design specifications. Because these patches were laid up along the length of the cylinder, rather than the circumference, the effects of this lower compaction was exacerbated by the roller conforming to a cylindrical surface (Figure 8). We collected mode-1 data and aligned it spatially to create Figure 9. The lower compaction settings produce a noticeable lack of adhesion between the course ply and the substrate, which also correlates with the darker/ colder regions in the thermal image due to lack of heat flow. We are still investigating the sensitivity of thermal signature to varying adhesion between prepreg layers. CONCLUSIONS We have shown that in situ thermal imaging has detected defects of interest in AFP without the need for visual inspection of each ply layup. We have also shown the benefit of a post layup line scan in understanding the final material state prior to curing. Both in situ and ex situ thermal 14 | SAMPE JOURNAL | imaging has the potential to create more efficient production environments, as well as enhancing the process development for new material systems and structural designs. The thermal system implemented in this study is also low cost, with the thermal camera procured for under $9,000. The next steps in our research will be to perform simple modifications to the ISAAC platform to allow the synchronous triggering of the thermal camera based on part coordinates. Currently part coordinates are inferred from knowing the starting point of the course and the time it takes to get to the end, which works for most cases. The ability to have a 1:1 comparison of thermal image to part coordinate will allow us to overlay our processed data into a three-dimensional model of the part. The second task will be to employ machine-learning algorithms for data reduction. The idea is to reduce the results to operator-friendly images and reports so that the operator can make a decision on repair necessity. We are also in the process of installing a small but powerful pre-processing computer (Intel NUC) on the AFP head of ISAAC itself to aid in data throughput. The on-board computer will handle the data reduction algorithms while saving the raw data simultaneously. This will enable faster data acquisitions and eliminate any data bandwidth limitations. J A N UA R Y/ F E B R UA R Y 2 0 2 0 w w w. s a m p e . o r ghttp://www.sampe.org