Narrative:

After pushback and starting both engines; the left generator tripped off. I attempted 1 reset; which did not correct the problem. Concurrently; the first officer was running the after start checklist. He initially turned the APU off; then immediately canceled the APU shutdown when the generator problem surfaced. We went through the MEL drill: calling dispatch; talking to maintenance; disconnecting the generator per the MEL; and informing flight attendants and passenger about the status of the aircraft. After completing these items; we continued our taxi believing that we had met the applicable MEL dispatch requirements; with both of us convinced the APU was running. Shortly after takeoff; the first officer realized that the APU; in fact; was not running; despite the APU selector being in the 'on' position. He immediately restarted the APU and we continued to our destination without further incident. Probability suggests that the APU was actually not running prior to takeoff. Assuming that to be the case; how could 2 experienced pilots believe otherwise? I think there are several contributing factors. First; addressing MEL items; especially those that require pilots to follow (M) procedures that may be new or confusing to them; while parked on an active taxiway is one of the most distracting and time sensitive tasks we face. Second; pilots tend to assume that a system; once turned on and operating normally; continues in that state until purposely turned off or some direct indicator appears that indicates a 'fault' has occurred. To the best of our recollection; we never received any APU fault EICAS message. Third; operating the APU; or any system for that matter; in an unusual sequence (canceling the APU shutdown versus the normal 'on then off;' 'off then on' sequence or as mandated by an abnormal or maintenance checklist) may very well mask indications that 'normal' operations would produce and 'normal' habit patterns would identify. I; and most pilots I assume; use the 'trust but verify' method when starting the APU. I now will use that same philosophy whenever the shutdown cancellation feature is used as well.

Google
 

Original NASA ASRS Text

Title: A B757 CREW DEPARTED WITH THE APU TURNED OFF AFTER DISCONNECTING A MALFUNCTIONING ENG GENERATOR FOLLOWING ENG START BUT PRIOR TO TKOF.

Narrative: AFTER PUSHBACK AND STARTING BOTH ENGS; THE L GENERATOR TRIPPED OFF. I ATTEMPTED 1 RESET; WHICH DID NOT CORRECT THE PROB. CONCURRENTLY; THE FO WAS RUNNING THE AFTER START CHKLIST. HE INITIALLY TURNED THE APU OFF; THEN IMMEDIATELY CANCELED THE APU SHUTDOWN WHEN THE GENERATOR PROB SURFACED. WE WENT THROUGH THE MEL DRILL: CALLING DISPATCH; TALKING TO MAINT; DISCONNECTING THE GENERATOR PER THE MEL; AND INFORMING FLT ATTENDANTS AND PAX ABOUT THE STATUS OF THE ACFT. AFTER COMPLETING THESE ITEMS; WE CONTINUED OUR TAXI BELIEVING THAT WE HAD MET THE APPLICABLE MEL DISPATCH REQUIREMENTS; WITH BOTH OF US CONVINCED THE APU WAS RUNNING. SHORTLY AFTER TKOF; THE FO REALIZED THAT THE APU; IN FACT; WAS NOT RUNNING; DESPITE THE APU SELECTOR BEING IN THE 'ON' POS. HE IMMEDIATELY RESTARTED THE APU AND WE CONTINUED TO OUR DEST WITHOUT FURTHER INCIDENT. PROBABILITY SUGGESTS THAT THE APU WAS ACTUALLY NOT RUNNING PRIOR TO TKOF. ASSUMING THAT TO BE THE CASE; HOW COULD 2 EXPERIENCED PLTS BELIEVE OTHERWISE? I THINK THERE ARE SEVERAL CONTRIBUTING FACTORS. FIRST; ADDRESSING MEL ITEMS; ESPECIALLY THOSE THAT REQUIRE PLTS TO FOLLOW (M) PROCS THAT MAY BE NEW OR CONFUSING TO THEM; WHILE PARKED ON AN ACTIVE TXWY IS ONE OF THE MOST DISTRACTING AND TIME SENSITIVE TASKS WE FACE. SECOND; PLTS TEND TO ASSUME THAT A SYS; ONCE TURNED ON AND OPERATING NORMALLY; CONTINUES IN THAT STATE UNTIL PURPOSELY TURNED OFF OR SOME DIRECT INDICATOR APPEARS THAT INDICATES A 'FAULT' HAS OCCURRED. TO THE BEST OF OUR RECOLLECTION; WE NEVER RECEIVED ANY APU FAULT EICAS MESSAGE. THIRD; OPERATING THE APU; OR ANY SYS FOR THAT MATTER; IN AN UNUSUAL SEQUENCE (CANCELING THE APU SHUTDOWN VERSUS THE NORMAL 'ON THEN OFF;' 'OFF THEN ON' SEQUENCE OR AS MANDATED BY AN ABNORMAL OR MAINT CHKLIST) MAY VERY WELL MASK INDICATIONS THAT 'NORMAL' OPS WOULD PRODUCE AND 'NORMAL' HABIT PATTERNS WOULD IDENT. I; AND MOST PLTS I ASSUME; USE THE 'TRUST BUT VERIFY' METHOD WHEN STARTING THE APU. I NOW WILL USE THAT SAME PHILOSOPHY WHENEVER THE SHUTDOWN CANCELLATION FEATURE IS USED AS WELL.

Data retrieved from NASA's ASRS site as of January 2009 and automatically converted to unabbreviated mixed upper/lowercase text. This report is for informational purposes with no guarantee of accuracy. See NASA's ASRS site for official report.