E4 Lock Losses Investigation
(Preliminary report - June 1, 2001)
D. Chin, R. Gustafson, K. Riles (Michigan)
M. Ito (Oregon)
W. Butler (Rochester)
(with appreciated assistance from R. Adhikari, S. Penn, P.
Saulson)
Correlation with 1-3 Hz seismic noise
-
Commissioning experience before E4 indicated that large seismic noise in
the 1-3 Hz band contributed to difficult lock acquisition and retention.
-
Using the output of Ed Daw's band-limited rms seismic DMT monitor and Dave
Chin's lock transition DMT monitor, we confirmed a strong correlation
between 1-3 Hz seismic noise and lock livetime fraction over the course
of the 72-hour run.
-
Lock was consistently lost during the passage of trains early in the morning.
Steve Penn reported observing a dramatic rise in power at 1.2 Hz (BSC stack
mode) in the DC power on symmetric and antisymmetric port photodiodes as
one of these trains approached Monday morning. He suggests this indicates
overdamping on large optics, perhaps leaving us overly susceptible to seismic
noise at stack frequencies. A subsequent search by KR through data channels
for large and small optics coil currents at the moment lock was lost during
this train passage failed, however, to find any smoking guns (currents
saturating before lock loss), but it should be noted that only the end
mass coil current readouts were DC coupled.
Coil saturation after tidal drift
-
Because of other lock retention difficulties, straightforward saturation
of coils after long tidal drifts was less common than in the E2 run, but
here is one example of this phenomenon,
which should not be a problem once feed-forward tidal correction is implemented.
It should be noted that several of the end mass coil currents routinely
saturate during lock acquisition.
Apparent servo instability
-
The saturation of the ETMX LR coil in the preceding example was seen in
a large number of lock losses, including a class in which all three active
ETMX coils appear to undergo an unstable oscillation immediately preceding
lock loss. This plot shows a 3-minute
second trend of the six active end mass coils (1st column = X, 2nd column
= Y) for an example lock stretch Saturday morning, along with the transmitted
arm powers (3rd column). ETMX LR (ETMX_CAL) appears to saturate first at
about the time of lock loss. A magnified
view of the six coil currents confirms this guess. For reference, the
transmitted arm powers do not drop significantly until well after saturation
ensues, as shown in this plot with the
same time-zero offset but larger range. The source of the apparent instability
which causes an exponential growth of the ETMX coil currents with time
constant of about 5 ms is not yet known.
-
As a side note, we looked at all DMT triggers that were recorded in the
metadatabase at the time of lock loss. Although we didn't find any triggers
that gave us further insight into the lock loss cause, we did find two
unexpected and interesting triggers generated by Masahiro Ito's GlitchMon
monitor. These were caused by a transient
in the microphone in the X end station at the onset of coil saturation
and by another in the beam splitter OSEM
LR sensor at the first significant drop in X arm transmitted power. The
microphone transient may indicate acoustic noise pickup from the end mass
coil (switching) power supply as it approaches and reaches saturation,
or there may be a coupling through a common AC power circuit. The OSEM
sensor transient seems to indicate that the BS LR sensor is more exposed
than the other BS sensors to scattered laser light. Both types of transients
were observed to coincide with other lock losses too.
686 Hz oscillation in length servo
-
Rana Adhikari spotted an unusual oscillation in the arm length servo control
signals. This plot shows a gradual growth
in the envelope(min/max) of both signals, a growth that suddenly
comes to an end as the transmitted power in the X arm dramatically increases(!).
The turn-off in envelope growth and increase in X arm power turned out
to coincide with saturation of the ETMX UL coil (GDS-EX_TO2), as shown
in this plot, where the other two active
X end coils are unsaturated, but where the yaw and pitch suddenly change.
The natural interpretation is that after a long lock period where
tidal correction and residual coil matrix imbalance caused a drift away
from optimum ETMX mirror alignment, the saturation of an upper coil caused
a twisting of the mass that (temporalily) brought it back to better alignment.
-
What is not so easy to understand is the source of the original oscillation.
This
plot shows the power spectra of the two arm control signals and the
Mode Cleaner's fast control signal at a time when the envelope above was
growing. There is a strong (and new) peak at 686 Hz in the arm signals
that does not appear to originate from the input optics. This
plot shows the spectra of the differential mode control signal at 20-minute
intervals covering the absence of oscillation and at its full strength.
Information on locked stretches
-
The cumulative live-time fractions for arm locks during E4 were
X arm - 51%
Y arm - 49 %
Both arms - 48 %
-
List
of 2-arm lock stretches at least 1 minute long