10 D.P. Rohe Fig. 1.12 Geometry showing the 1075 measurement point locations colored per scan. White elements denote the space between scan patches 5–10 alignment points. A coordinate system transformation was then computed by the SLDV software using the locations of the alignment points in the current coordinate system and the locations in the part coordinate system. This process allowed efficient re-alignment of the laser heads and mirrors with the test article for each scan. For the top scans, the lasers were arranged on tripods; therefore, the two step alignment-and-transform procedure described in Sect. 1.3.1.2 was followed for those scans. In hindsight it would have been prudent to build a second stand from the 95-mm optical rails in order to bypass the alignment portion as was done for the bottom scans. A large number of measurement points were to be scanned for this test, so the most efficient way to define them in the laser software was to import the measurement points from an external geometry file. These points then needed to be reduced for each scan to those points within the lasers’ field of view and on the side facing the laser heads. Points were also removed if the angle of incidence between any of the laser beams and the surface were too high. Figure 1.12 shows the measurement geometry color-coded by scan. Even with the efficient bottom alignment strategy, this test was long running due to the sequential scanning of over 1000 points as well as labor-intensive setting up and aligning the lasers for all of the scans. Testing was performed over 5 days, so it would be very difficult to go back and retake data if some data acquisition parameter needed to be changed (e.g. forgetting to apply a window, bandwidth too low, etc.). This is contrary to testing with a large number of accelerometers where once the gauges are adhered to the surface data can be taken and retaken rather easily if there are enough channels available. 1.3.2.3 Data Analysis Similarly to the conical structure test, the laser data was collected into a single scan file and exported to analyze in MATLAB. Over 6000 FRFs were measured by the laser system, accounting for the over 1000 scan points, two shaker inputs, and three measurement directions per point. It became clear when performing this analysis that the SMAC algorithm began to struggle with a data set of this size. When all 6000CFRFs were included in the SMAC analysis, the correlation coefficient—a key parameter indicating the presence of a mode in the SMAC algorithm—was approximately unity for the entire bandwidth. Essentially, SMAC was identifying a mode at every frequency line. When the set of FRFs was reduced by only keeping one of every 5 or 25 measurement points, the correlation coefficient began to look more reasonable, as shown in Fig. 1.13. The frequency and damping parameters extracted from the reduced analysis could then be imported into a full analysis to extract the full mode shapes. Extracted mode shapes were somewhat noisy which is attributed to the large angle of incidence and relatively poor surface properties of the test article, especially on the darker surface near the tip of the nose. Two example mode shapes are shown in Fig. 1.14. 1.4 Lessons Learned and Lingering Deficiencies A number of lessons were learned from the testing of the two test articles described in this report. Large, complex test articles will often require multiple scans to measure all parts of the test article, so developing an efficient method of repositioning the laser head is likely the most important step to making SLDV measurements of large structures practical. Because multiple
RkJQdWJsaXNoZXIy MTMzNzEzMQ==