XXXI International Mineral Processing Congress 2024 Proceedings/Washington, DC/Sep 29–Oct 3 499
For “ultrafines” samples, a 50-test experiment was
designed in randomized order to maximize the statistical
power of the experiment. Supplemental runs exploring the
design space and optimal conditions after the initial DOE
were also conducted.
JMP has a variety of regression methods available to
create models. Due to the presence of categorical variables
with greater than 2 levels the Least Squares fit regression
method was chosen to create the model for this separation
process. The model allows the user to view the estimated
responses as a function of different factor inputs. This allows
STET to estimate product grade and yield as function of
different machine settings like belt speed, gap, voltage, etc.
Once the design of experiment was completed the data
for the responses was entered into the table and the Least
Squares fit model was created. The next step for refining
the model is to evaluate the effect summary. The variables
and interaction terms which didn’t show a significant effect
were removed from the model.
Figure 3 is an example of an effects summary for a
model made during the testing of this material.
Once the model has been pared down to likely active
effects it is possible to make use of JMP’s ‘Prediction
Profiler’. The Prediction Profiler generates a visual rep-
resentation of the model that provides estimates of the
responses as a simultaneous function of different significant
factors and their effects that were included in the model.
When creating the design at the beginning of the experi-
ment the user enters the responses or outputs of the test-
ing they are interested in, JMP allows the user to rank the
relative importance of the responses and how they want
to manipulate the responses, e.g., maximize, minimize, or
match a target value. The ranking is then used to create
‘Desirability Functions’ for each response which provides
Table 4. Factors tested in design of experiment
Factors Levels Units
Belt Speed 15 65 Feet per second
Gap 400 500 Mil
Voltage 6 10 kV
Belt Material 1 2 3 4
Feed Port 1 2 3
Polarity Top Positive Top Negative
Figure 3. Schematic of machine identifying factors experimented with
For “ultrafines” samples, a 50-test experiment was
designed in randomized order to maximize the statistical
power of the experiment. Supplemental runs exploring the
design space and optimal conditions after the initial DOE
were also conducted.
JMP has a variety of regression methods available to
create models. Due to the presence of categorical variables
with greater than 2 levels the Least Squares fit regression
method was chosen to create the model for this separation
process. The model allows the user to view the estimated
responses as a function of different factor inputs. This allows
STET to estimate product grade and yield as function of
different machine settings like belt speed, gap, voltage, etc.
Once the design of experiment was completed the data
for the responses was entered into the table and the Least
Squares fit model was created. The next step for refining
the model is to evaluate the effect summary. The variables
and interaction terms which didn’t show a significant effect
were removed from the model.
Figure 3 is an example of an effects summary for a
model made during the testing of this material.
Once the model has been pared down to likely active
effects it is possible to make use of JMP’s ‘Prediction
Profiler’. The Prediction Profiler generates a visual rep-
resentation of the model that provides estimates of the
responses as a simultaneous function of different significant
factors and their effects that were included in the model.
When creating the design at the beginning of the experi-
ment the user enters the responses or outputs of the test-
ing they are interested in, JMP allows the user to rank the
relative importance of the responses and how they want
to manipulate the responses, e.g., maximize, minimize, or
match a target value. The ranking is then used to create
‘Desirability Functions’ for each response which provides
Table 4. Factors tested in design of experiment
Factors Levels Units
Belt Speed 15 65 Feet per second
Gap 400 500 Mil
Voltage 6 10 kV
Belt Material 1 2 3 4
Feed Port 1 2 3
Polarity Top Positive Top Negative
Figure 3. Schematic of machine identifying factors experimented with