Each Apollo mission
prior to Apollo 11 was plagued with about 20,000 defects
apiece. [Ralph Rene]
Is that a lot? Were they crucial systems? Mr. Rene either doesn't
know or doesn't wish to disclose it. This is where Mr. Rene might
have benefitted from some actual engineering experience.
The command module alone contained two million individual parts.
Even if those 20,000 defects had been confined to the command module
alone, that's one defective part for every 100 parts. Adding in the
service module, the lunar module, and the Saturn V launch vehicle, the
more accurate figure is one defect in 300 parts. Since this was a
highly complex experimental spacecraft, this is not an unusually high
number.
But of course a per-part measurement isn't exactly correct. Some
of the defects were simply anomalies in performance, or procedural
problems. The issue here is not design flaws or procedural errors but
rather a meaningful context in which to evaluate a reported number of
defects. Non-engineers are alarmed by what appears to them to be a
very large number of defects. But since those unfamiliar with the
rigors of design for manned spaceflight and the procedures for
reporting and disposing of defect reports are not equipped to put that
statistic in a meaningful framework, this represents a kind of
deception on Mr. Rene's part. He too is likely not to know whether
this constitutes a large or small number of defects for a product of
this type and size. But since alarmism fits well into his agenda,
Mr. Rene is not served by a thorough investigation into these alleged
defects.
One must also consider the nature of the engineering development
process. When the product has reached an advanced stage of design,
the quality control engineers will begin to examine the design and
suggest methods for testing. When components of the product become
available they are subjected to "unit testing" that may produce a
limited number of defect reports. Not until the prototypes are fully
assembled and submitted to flight test can the real process of quality
control begin. At this point the focus of the engineers shifts from
active design to reactive treatment of discovered defects.
If you plot defect reports against the development timeline, you
discover that this pattern of testing produces the maximum number of
defect reports shortly before deployment. That number drops off just
as dramatically to the point of deployment.
Just one defect could
have blown the whole thing. [Ralph Rene]
Only if it were a defect in a critical system for which there was
no redundancy, the lunar module ascent engine for example. Implying
that no defects of any kind were permitted during a successful flight
is very naive.
Defect reports, or "chits" as they are known in the industry, come
in many types and levels of severity. The most serious are "show
stoppers" which directly impact the safety of the crew and the
reliability of the spacecraft. Leaks in the RCS fuel system, for
example, would be show-stoppers. But the majority of chits are not
for critical items. In fact as many as half the chits may be what are
known informally as CYAs, (for "cover your ass").
Since quality control engineers are often held just as responsible
as design and manufacturing engineers for defects in products, quality
controllers tend to prolifically write chits so that in the event of
future failure they can prove to their managers that they did their
job. Often these CYA chits are nothing more than differences in the
interpretation of the product's written specifications. But since
little or nothing can or ought to be done about the CYA issues, many
languish in defect tracking systems as "open" issues when in fact the
designers have no intention of addressing them.
The defect reporting mechanism is also used to introduce requests
for additional functionality or enhancements. Failure to address
these defects does not necessary diminish the safety or functionality
of the spacecraft. These chits basically start out saying, "It would
be nice if ..." Since these observations frequently illuminate
deficiencies in the original specifications, it's worth paying
attention to them. But it's wrong to classify them as something that
must be corrected before a safe flight.
Understanding that Apollos 7, 8, 9, and 10 were essentially flight
tests of the Apollo hardware prior to operational deployment, we
realize that the purpose of these flights was to discover defects.
The unmanned flight tests (Apollos 4, 5, and 6) established the basic
spaceworthiness of the Apollo spacecraft. At this point it was
determined that the spacecraft were capable of ferrying astronauts
safely to space and back to earth, although they were not yet capable
of executing a lunar landing mission. But since flight tests are
conducted specifically in order to discover defects, it is not
surprising that a large number of defects was discovered.
The operational flights too encountered in-flight defects. Apollo
13 is the notable example, but we can examine the defect lists for
successful flights as well.
The readiness reviews that precede each manned space launch do not
require that all defects be corrected prior to launch, only those
defects which directly impact the mission success or safety of the
crew. It is common to waive responsibility for items that have
redundant backups or established safety margins, or which are not
mission critical. Although airlines don't appreciate this being
generally known, hardly a commercial flight takes to the air without
some number of components having been "red-tagged" as unusable. And
if our automobiles were subject to the same rigorous inspection and
acceptance criteria as a manned space launch, the number of
"show-stoppers" might astound us.
|