Unchanging concepts - Concrete Mix Design, Quality Control and Specification - page 261

Table 11.1 National criteria as in national codes
AC1214

AS3600 BS 5328 N Cusum

False alarm frequency 52.36 93.6 46.81 70.74

Average detection delay 1.75 12.90 2.64 4.11

Maximum average detection delay 8.06 20.15 7.26 10.54

Adjusted (by Constant Margin) to comparable false alarm frequency
Adjustment in char strength (MPa) 1.75 -0.60 6.50 NA

False alarm frequency 63.8 64.5 77.8 71.2

Maximum average detection delay 17.6 20.3 22.5 16.0

alarm frequency during the process of selecting the deduction margin and
detection limit. This was done as a separate exercise using the techniques
of this investigation in which a large range of margins and limits were tried
(in sets of six) to find the most efficient combination.
The basic techniques embodied in the national codes (individual result
limit and limits for running means of 3, 4, 5, and 30) were also separately
examined (Table 11.1). This was necessary because some of the combina-
tions were optional and also to avoid concluding that the code incorporating
the largest number of individual criteria (ACI 214) was necessarily the best.

11.3.2 Visual cusum
In the initial stages of the investigation, hundreds of graphs of the run
of 130 results were examined. It was noted that the basic cusum graph
almost invariably showed a quite distinct downturn at the exact point of
the artificial downturn, even when the drop ratio was so small that the
numerical system detection efficiency was poor.
It should be noted however that this is far from the same thing as
concluding that the detection efficiency of the basic cusum is almost per-
fect. The technique looks better in retrospect than it does in genuine use.
Examination of the overall 130 result trial tells nothing of how many of
the small false downturns in the cusum graph might have been mistaken
for the real downturn or for how long an operator might have regarded
the real downturn as such a false one. So, while the keen and experienced
operator using cusum graphing will already have acted before the detec-
tion system provides a signal, the less experienced operator will be glad of
the confirmation provided by the system and the less keen operator will be
prodded into action.
What is clear is that, on looking back after concluding that a downturn
has occurred, the basic cusum graph will show exactly when that downturn
occurred. This is very valuable information because the same logic applies
to any other variable for which a cusum graph is drawn, and therefore it is
usually easy to match up cause and effect.