View Paper

advertisement
Just In Time Mixture Proportioning
Ken W. Day
Consultant
ABSTRACT: The paper describes a system for automatically designing the next truck of concrete in a few
seconds, taking into account all relevant data available up to the time of batching. Such data should include
current test data on concrete and constituent materials, anticipated concrete temperature, required slump and
transport time. This paper presents the view that, in order to achieve minimum variability, the contents of
each truck of concrete need to be finally determined only a few minutes before that truck is batched, taking into account every piece of relevant information that can be made available at that point in time.
1 THE PROBLEM
The current pressure to reduce greenhouse gas
generation requires that the use of cement be minimized while maximizing the use of concrete. This
requires improved control to reduce the necessary
margin between minimum and mean strengths (and
durability etc) and maximum use of cement replacement materials.
There has been a tendency either to regard each
separate grade of concrete using replacement materials as a unique design, requiring its own separate
trial mixes and control history, or to incorporate an
effectiveness factor for the replacement material replacing cement. In fact the relative effectiveness of a
replacement material varies with both the strength
level and the replacement percentage. A better solution is to regard each different percentage combination of materials as a separate cement. The author’s
“Mix Tables” program(1) can then be used to establish a strength v water/cementitious ratio curve for
that cementitious combination. This paper then considers how a large number of such curves can be integrated into a control system.
The concept envisages a database frequently updated by information potentially contributed from
several different sources. The latest test result is not
necessarily the best current estimate of the property
in question. For example sand gradings depend on
sampling representative material and the laboratory
should maintain a cusum graph of sand specific surface and update the current value in the batching
system only when a significant change occurs. In
fact arranging for accurate current materials data is
the greatest remaining challenge not addressed in
this paper. This paper mainly addresses the problem
of feedback of current concrete test data and instantaneous mix revision using the latest input data.
2 DETECTING CHANGE
While cementitious material test data can enable
an immediate mix revision when a significant
change is encountered, only concrete test data can
provide really accurate adjustment. EN 206 and its
derivatives envisage evaluation of mix performance
over a period of months but what is required is a detection of any change in performance within a few
days. The author’s Multigrade, Multivariable,
Cusum QC program(1) provides such detection and is
displayed as a poster presentation at this symposium.
It involves cusum graphing of several other variables
in addition to strength and density on the same
screen (multivariable), so that strength variations can
be positively detected at an early stage through being confirmed and explained by matching variations
in other variables. Early detection is greatly enhanced by the multigrade technique enabling results
from all grades in use to appear on the same graphs.
This is accomplished by using the separate continually updated current mean value of each variable in
each grade as the cusum target for that variable. This
focuses the analysis on the detection of change rather than adherence to a pre-conceived target and
enables all the deviations from all grades to be combined in a cumulative sum.
DESIGNING AND ADJUSTING MIXES
The basis of the mix design and adjustment system is the author’s MSF (mix suitability factor)
technique(1) which involves the specific surface of
the aggregates modified for the cementitious content. This technique assumes that any change in aggregate (especially sand) grading can be compensated for by adjusting the sand % to maintain the
same overall specific surface. It also assumes that a
wide range of strengths can be accommodated in a
type of concrete having constant fresh concrete
properties, if the specific surface of the aggregates is
reduced as cementitious content is increased so as to
maintain a constant MSF.
Strength prediction uses the very old Feret equation, modified by two constants M and K:
Strength = M x 290 x (C/(C+W+A))2 +K
(Where C,W and A are the volumes of cement,
water and air)
As previously noted, each different combination
of cementitious materials is treated as a different
cement and the constants are determined by minimum sum of squared departures from the equation.
The constants are of course different for each different cement.
Having determined M and K values for each cementitious combination, it may be possible to adjust
the individual test results using these values so that
all results can be plotted on the same graph to detect
whether any results are anomalous and should be adjusted.
Each grade also requires a figure for its average
gain from 7 (or 3 in tropical countries) to 28days for
prediction purposes.
Of course strength and slump are not the only criteria of mix suitability. Especially for pumped or
highly workable concrete, segregation resistance is
also a factor and this may require a continuity of
grading, i.e. an absence of gaps in the aggregate
grading. Two alternative approaches have been devised for this:
1. The system enables the designer to input any desired set of figures constituting an ideal grading.
Two further numbers are to be input to constitute
inner and outer limits on the entered ideal. Grading points within the inner limit are ideal and any
outside the outer limit reject that grading. Points
between the two curves are totaled into a “Deviation Index” and the designer can specify a
limit on this number.
2. Approximately half of the aggregates will lie on
the finest 7 sieves so that, in a straight line grading, there will be approximately 7% of total aggregates on each of these sieves. A “Gap Index”
can be defined as the sum of the squares of percentage on each of the seven sieves minus 7.(See
Fig 1)
DESIGN SEQUENCE
1. The system assumes that a database will be set
up containing all available test data on aggregates, cementitious materials and concrete. The
data will need to be in time sequence and available for cusum analysis. The person in charge of
each material (more than one laboratory may be
involved) will be required to maintain a continually updated best estimate of current properties
that will be automatically accessed by the design
process. It is not as simple as using the last obtained test result for each property e.g. if a sand
sieve analysis indicates a change in grading, a
second sample should be taken to confirm it. On
the other hand, if a change in water requirement
is noted (after allowing for temperature variation) that should trigger an examination of the
possible causes.
2. Decide on the range of cementitious materials to
be available and on the combinations to be available. Enter these on the Cement screen, Fig2. In
a complete start-up situation it will be necessary
to base the calculation of the constants M and A
for each cementitious combination on trial mixes
but they should be revised from time to time
based on production test data. Before going to
the Just-in-Time process it is likely that a producer will have been operating the author’s “Mix
Tables” system (1) on several cementitious combinations.
3. Decide upon the range of products to offer. Go
to the Concrete Product screen (Fig3) and enter a
name for a range of mixes of different strengths
but the same required fresh concrete properties.
Enter the properties the range is to have including the cement group, the available aggregates
and any restrictions on composition. The same
range may require different constituents and
have a different cost at different plants.
ORDERING/SELECTION OF MIX
When an order is received (perhaps over the internet) the program will decide which product range
meets the requirements of the order and which plant
is to supply (availability, minimum cost). It will display (Fig 4) details of the proposed mix (not showing batch quantities) and the minimum price (which
may be different for different customers). The customer will inspect the proposed details and enter any
required changes in the “specified” column. The
price will be seen to change as such further requirements are specified. If the customer’s requirements
cannot be met by the product range in question, the
system will propose a change to a different range
and, in some cases, this may require human intervention.
Having agreed what is to be supplied, and at the
required delivery time, the plant operator keys “Design” and the intended batch quantities appear on his
screen. The system compares these to the quantities
used the last time this mix was batched and displays
a warning if any divergence exceeds the limits
which have previously been specified.
If there is no problem, the operator then keys
“batch” and the load is batched. The actual batch
quantities (including any batching error) then appear
alongside the intended batch quantities and the system calculates the anticipated yield, density and predicted strength from the actual batch. This data is
stored in the system and compared to any actual test
data later obtained on the particular truckload.
foot of this figure. This shows exactly how the mix
will be varied as this seventh requirement varies.
What this demonstrates is that, if the conditions at
the time of batching are known precisely, then the
mix to be supplied is also known precisely. It also
makes it quite clear that if a mix is designed assuming a given set of these parameters (as is traditional)
then if any of these parameters change, and the mix
is not adjusted, then the resulting concrete will not
have exactly the intended properties.
BATCH PLANT AND MATERIAL DATA
Although the system is valuable and effective if
operated solely on concrete test data, it becomes
much more powerful when combined with automatically integrated batch plant and material testing data.
Cusum graphs of such items as batching error and
sand specific surface transform the QC situation.
ANALYSIS OF RESULTS
Although designed as ranges of mixes, the subsequent test data will be analysed as individual mixes
but on a multigrade basis. Any variation in general
performance will be detected by multigrade, multivariable cusum graphs (Fig 6) but in addition, every
individual grade tested will have its own row in a table of statistically analysed data and these rows will
be arranged in order of departure from target
strength (Fig 7). Any under (or over) performing
grade will be immediately obvious and a (possibly
temporary) correction can be applied independently
of the revision of the M and A constants for the
overall range.
ACCEPTABILITY OF THE SYSTEM
It is to be anticipated that a system of this nature
will initially be viewed with scepticism and caution.
It is essential that it be initially operated in an overconservative manner until confidence in it is established. Two factors will be crucial in establishing
confidence. One is the accuracy with which the predicted strengths are confirmed by normal production
test data. The other is the standard deviation
achieved overall and in each grade.
The whole purpose of varying batch quantities
from truck to truck is to achieve lower variability in
the resulting concrete and the analysis system will
clearly display whether or not this is being achieved.
One objection that can be anticipated is that many
will object to approving a concrete mix which does
not have pre-defined quantities. The feature illustrated in Fig 5 has been devised to counter this attitude.
Seven principal requirements of the mix are shown
in a small table. Any six of these can be input and
the seventh can then be included in the table at the
FUTURE DEVELOPMENTS
The system program has yet to be finalized and at
this stage is only a demonstration of what is possible.
Combining all data from all cement groups into a
single strength curve (with different M and A values
for each cementitious combination) requires further
investigation. It has been suggested that a technique
involving neural networks (3) could be used. In neural networks, the mathematical relationships between the various variables are not specified. Instead, they learn from the examples fed to them. In
addition, they can generalize correct responses that
only broadly resemble the data in the learning phase
(4).
The details of the algorithm have been thoroughly
described by Lippmann (3) Several studies have
demonstrated that neural networks can build a more
accurate concrete behavior model than regression (4).
REFERENCES
1. Concrete Mix Design, Quality Control and
Specification by K. Day (3rd edition 2006,
Taylor and Francis).
2. Website www.kenday.id.au – contains many
papers on mix design, QC and specification
published over the years, also screencams
(videos) of the ConAd and other systems.
3. Lippmann, R. P. (1987) “An introduction to
computing with neural nets.” IEEE ASSP
Magazine, 4(2), 4-22.
4. Yeh, I-Cheng (2006), “Analysis of strength
of concrete using design of experiments and
neural networks,” Journal of Materials in
Civil Engineering, ASCE, Vol.18, No.4,
pp.597-604.
Download