Chandra X-ray Center, Smithsonian Astrophysical Observatory, 60 Garden St., Cambridge, MA 02138, USA

The * Chandra X-Ray Observatory* was launched in July 1999,
and is thus in its fifth year on-orbit. The Monitoring
and Trends team at the * Chandra X-Ray Center* (CXC) is charged
with tracking observatory performance parameters to optimize
the mission's science return. We have built from scratch
an IDL-based system, called "dtrend" (derivative trending),
for visualizing and quantifying long-term trends. Data are input
from our databases of over 600 engineering mnemonics, averaged
on 5 minute intervals over the course of the entire mission.
Dtrend computes the mean, standard deviation, first derivative
and second derivative for each parameter. The derivatives are
then used to predict the next 6 month cycle. Output is presented
via web pages with statistical summary tables and graphics color-coded
to highlight threat level or potential problems.
This paper will discuss the algorithms and metrics
used to predict future behavior based on previous trends and how
the CXC can efficiently identify, track, and possibly curtail
problems to extend the length and quality of the * Chandra* science
mission.

Processing is done with IDL code to take advantage of built-in or readily available FITS I/O, statistical, and plotting routines. Table 1 lists the key steps involved. Note, default values are listed for all constant parameters, such as sigma clipping level and extrapolation time frame, in this description. The code gives control of these values to the user through keywords.

1 | Extract new data from MTA databases using DataSeeker |

(5 minute averages). | |

2 | Compute 1 hour averages (for faster run times and compressed |

storage), merge with previous data. | |

3 | Read merged data into IDL. |

4 | Apply filters (e.g. 3-sigma clipping, handle NaNs and missing data). |

5 | Look-up defined limits to color-code output. |

6 | Scatter plot each data column. |

7 | Apply smoothing and calculate derivatives. |

8 | Overplot smoothed curve (blue) and fit line (color-coded by limits). |

9 | Overplot 6 month extrapolation. |

10 | Plot derivative. |

11 | Overplot fit line (second derivative). |

12 | Produce html statistical summary page. |

Output is to World Wide Web pages for easy user access. See section 3.

3. Data Output/User Interface

The URL for SOT Trending is http://cxc.harvard.edu/mta/DAILY/mta_deriv.

Figure 1 shows example summary pages. On the left is the trending top level page. Here we present a table listing all the trended subsystems with links to the available analyses (total, daily minimum, daily maximum, and past quarter). The links are color-coded green, yellow, and red based on limit violations seen in the underlying pages to quickly identify the problem areas.

On the right is a subsystem summary and statistics page. Each link from the top level page expands to a statistical summary page. Here we list for each mnemonic the calculated mean, standard deviation, first derivative, and second derivative. These values are color-coded to easily identify the current or future problem areas. We also list units and a description of each mnemonic extracted from the limits look-up file for reference.

Figure 2 shows examples of our pop-up plotting windows. Each mnemonic links to a plot of the data. The top panel in each shows a scatter plot of the data with smoothed curve (blue) and fit line overplotted. Note the six month extrapolation plotted based on the second derivative. Any out-of-limit values are indicated with yellow or red colors.

There are several types of profiles commonly seen: On the left is a linear fit, , for simple cases or as a first step in cases not yet understood.

The right-hand figure shows that we do not have to plot only versus time. Here we show temperature versus sun angle, with time indicated by the color of the data points. It is clear that the EPHIN housing heats up most at forward-sun attitudes, but the problem is getting worse over time due to the deterioration and darkening of insulating materials. This profile may be best fit with a higher order polynomial or exponential decay model, such as or .

Other subsystems show more complicated structures, with multiple components. Solar array voltages, for instance, show an overall decreasing linear or exponential trend as well as seasonal sinusoidal variations. By carefully fitting both of these elements with something like , we can glean more information on the system's behavior and better estimate its future performance.

Overbeck, R. S. et. al. 2002, in ASP Conf. Ser., Vol. 281, Astronomical Data Analysis Software and Systems XI, ed. David A. Bohlender, Daniel Durand and T. H. Handley (San Francisco: ASP), 449

Spitzbart, B. D., Wolk, S. J. & Isobe, T. 2002, in Observatory Operations to Optimize Scientific Return III, ed. Quinn, Peter J., Proceedings of the SPIE, 4844, 476

Wolk, S. J. et. al. 2002, in ASP Conf. Ser., Vol. 281, Astronomical Data Analysis Software and Systems XI, ed. David A. Bohlender, Daniel Durand and T. H. Handley (San Francisco: ASP), 341

© Copyright 2004 Astronomical Society of the Pacific, 390 Ashton Avenue, San Francisco, California 94112, USA