Unchanging concepts - Concrete Mix Design, Quality Control and Specification - page 259

It is shown that the average number of results required to achieve detection
of a change is directly proportional to the standard deviation of those
results. Since early detection in turn enables a reduction in variability, a
self-intensifying cycle of variability reduction is commenced.
The questions of early-age and accelerated testing; of monitoring
batching performance; of analysing related variables such as slump, den-
sity, and temperature have been addressed elsewhere in this volume. For
the purpose of this investigation it was assumed that a continuous string
of test results is being received and converted into predicted 28-day results.
The relative efficiency of the different techniques in detecting a downturn
in such a string of results was examined.
Based on 40 years of plotting quality control charts for concrete, Day's
assumption is that the downturn is usually a sudden event or step change
rather than a gradually worsening trend. To highlight this point, a simulation
was conducted to automatically produce a string of 100 random but nor-
mally distributed results of any selected mean and standard deviation. It
then added a further 30 results of the same standard deviation but a lower
mean. This enabled examination of the performance of a control system in
respect of whether it raised false alarms during the initial stable period of
100 results and how long it took to detect the imposed change point at the
100-result mark. The results were automatically analysed by up to six differ-
ent detection systems at a time and the results reported as follows:
1. The number of results prior to a false alarm in the first 100 results, if
the number is 100, there were no false alarms.
2. The number prior to the first detection of change after the imposed
change point, if the number is 30, there was no detection.
The best detection system is not necessarily the one that shows the low-
est average number of results to give a detection. Any type of system can
be made more sensitive by narrowing its limits, at the cost of experiencing
more false alarms. It was not considered sufficient to find that one system
was extremely good at detecting changes but gave many false alarms, while
another gave few false alarms but was a poor detector. It is certainly of
interest to compare the relative severity of different national codes but the
authors' primary interest is in finding the most efficient way of detecting
a change. The exercise was therefore repeated after adjusting the nominal
specified strength so that each system gave similar false detection frequen-
cies when assessing the same concrete.
It was found to be important whether the adjustment was in the form of
a constant or that of a multiplier of the SD. The various national systems
often incorporate a fixed adjustment, for example, ACI 214 requires not
more than 1 in 100 results to be more than 500 psi (3.45 MPa) below the
specified strength and BS 5328 requires the running mean of four results to