Obviously, it is the ground wind that affects the touchdown accuracy the most. With perfect knowledge of the winds, all Snowflakes (as any other ADS) would land almost exactly on the target. Unfortunately, the only winds available onboard are the wind estimates, provided by the GNC algorithm in real-time during two phases, loitering and downwind descent. Of course, these estimates are not perfect, and the major problem is as follows. The last estimate on the wind magnitude comes at the end of the downwind leg, i.e. the construction of the final turn trajectory is based on this last update, obtained at an altitude of about 100m AGL. If the winds below 100m are the same as estimated at 100m then the system would land in close vicinity of the target, where the only source of error then would be an error in estimating winds at 100m. Of course, having the same winds throughout the last 100m of the drop is rarely the case (if ever at all).
By now this problem has been fixed having a networked target weather station broadcasting the ground winds to the Snowflake ADS in flight. However, for educational purposes, let us consider a couple of earlier drops from the standpoint of the Snowflake having only this last estimate at 100m. The key in this case is how accurate the estimates of the winds were at 100m and how much of a wind change the ADSs had to face on their final 100m to the ground. To this end, the data collected during the drops at YPG by Windpacks and wind tower were carefully analyzed, allowing detailed estimation of the cause of a downwind miss.
First, Fig.J shows the bird-eye view of two drops, a "bad" one (Fig.Ja) and a "good" one (Fig.Jb) with the time difference between them being slightly over one hour.
Figure J. Bird-eye trajectories of the two Snowflake drops (October of 2008) exhibiting different touchdown performance.
Figure K presents the data collected by the wind tower. First, one can observe the drift in the barometric pressure that the estimates of the ADS altitude are based upon. As seen, the onboard barometric pressure sensor settings have to be changed constantly to accommodate these changes in the ground pressure. Failure to do that results in the ADS thinking that it is higher above the ground level than it actually is (because the barometric pressure drops down as the sun heats the surface). Secondly, during the four sets of drops executed that day, the winds did change direction and speed. Starting with light variable-direction winds in the morning (below 1m/s) the ground winds became more consistent (in the wind direction) and stronger (up to about 4m/s)at noon.
Figure K. Wind tower data with vertical lines indicating Snowflake releases.
Next, Fig.L presents the data collected by the Windpack a couple of minutes before the first (in a set of two) ADS release. In this figure they are presented as recorded by the Windpack, where altitude is provided with respect to the mean sea level. It is seen that during the last 100m the winds do change and that is what causes touchdown errors. Figure M addresses this issue in more detail, providing the comparison of the Windpack data with that of the wind tower and Snowflake estimates. (In this figure, the Windpack altitude data is converted from MSL to AGL; and the Snowflake ADS altitude date is corrected to accommodate errors in altitude estimates, i.e. all data are altitude-synchronized.)
Figure L. Windpack data for the "bad" (a) and "good" (b) drops.
Figure M. Windpack data versus Snowflake estimate for the "bad" (a) and "good" (b) drops.
Figure N presents even more detailed analysis of the data shown in Fig.M, with concentration on the last two phases of the guided descent, final turn to the target and final approach. The left portion of each plot shows the difference between the tailwinds predicted by the Snowflake ADS and actual winds measured by the Windpack, and the crosswind component that the ADS had to overcome on its way down. On the right portion of each plot the vertical speed as recorded by the Snowflake ADS barometric altimeter is presented. Several calculated parameters shown on these plots represent a rough estimate of contribution that the winds, unaccounted for in the downwind direction, might have on overshooting the target.
Figure N. Downwind and cross wind components of unaccounted winds for the "bad" (a) and "good" (b) drops.
Consider Fig.Ma. As seen, the last estimate of the along-the-course wind component by the Snowflake autopilot occurred at about 93.6m corrected altitude AGL, and provided the value of 5.3m/s. However, the real winds suddenly die and amount to only about 1m/s all way down to the ground. Hence, there is a 4m/s difference between the last Snowflake estimate and the actual winds (Fig.Na) all way down. These unaccounted for tail winds that acted during the last 27s of the guided descent should have resulted in a large system overshoot of about 116m. Luckily, because of 15.5m error in altitude estimate the actual miss distance was smaller.
Now, consider Fig.Mb. Compared to the previous case, the wind estimate at the corrected 79m AGL is almost perfect (5.2m/s) and the winds do not change much during the final turn (Fig.Nb). This should have resulted in landing about 1.4m short. Again, because of ~8.2m error in altitude the actual touchdown point happed to be 5 m short of the intended point of impact. The data for these two Snowflake drops is consolidated in Table A. This table shows upwind components of the miss distance and then analyses the possible contribution of two sources of error, which are the altitude estimation error and winds unaccounted for during the last two phases of the flight. The last column shows the values of a miss distance that could be achieved if both errors were eliminated.
Table A. Estimates of possible causes for over- / under-shooting the target.
|Drop||Upwind component of a miss distance, m||Altitude error, m||Correction due to the altitude error, m||Correction due to unknown winds, m|
Estimated upwind miss distance, m
Once both sources of errors were eliminated (in the Snowflake-N version of the ADS) the exhibited performance matched these expectations (of having touchdown accuracy below 20m) fairly well.