XXXI International Mineral Processing Congress 2024 Proceedings/Washington, DC/Sep 29–Oct 3 3971
pilot-scale ball mill. This finding underscores the potential
of the MonoRoll for energy conservation in comminution
processes.
CAHM DEM MODELING
Although DEM is a powerful tool for modeling and under-
standing the comminution processes in the CAHM tech-
nology, certain system simplifications and assumptions are
required to make the simulations tractable. These include:
• DEM does not account for mechanical losses of the
system.
• Heat transfer is not considered in the DEM model.
• The model does not take into account the moisture
content of the ore.
• Variations in ore competency are not considered.
• The model does not account for variations in ore
shape.
• The DEM model uses an average shape representa-
tion for rock particles with a truncated size distribu-
tion to limit the exponential increase in the number
of particles below a certain size.
• Smallest feed particle size is 6400 µm.
• Rocky models assume that the hammer roll has no
skewness, and its axis of rotation stays aligned.
DOE-Based Optimization of CAHM Geometry and
Operational Parameters
The optimization of the CAHM machine’s geometry and
operational parameters was conducted using a Design of
Experiments (DOE) approach. This approach allows for a
systematic exploration of the design space and helps iden-
tify the response parameters’ optimal combination. DOE
approaches start by defining the input parameters’ range to
be investigated. A space-filling design of virtual experiments
(DOE) method called the Latin Hypercube Sampling
(LHS) technique is then used to create initial training data
points. The training data points are then simulated, and a
surrogate model is fit to each response parameter.
The surrogate model used in this process is the Gaussian
Process Regression (GPR). This model provides a flexible,
non-parametric Bayesian framework for data analysis, and
is particularly suited for optimization of unknown, stochas-
tic, and expensive objective functions.
Finally, an optimization algorithm is utilized to find
the optimal combination of input parameters for a given
set of target parameters. The optimization algorithm used
in this process is the NSGA-II (Non-dominated Sorting
Genetic Algorithm II). This is a popular non-dominated
sorting-based multi-objective optimization algorithm that
is known for its better convergence near the true Pareto-
optimal front compared to other algorithms.
The decision-making process in the optimization is
guided by the Pareto-Front. This is a set of non-dominated
solutions, being chosen as optimal when no objective can
be improved without sacrificing at least one other objective.
DOE methodology allows us to target specific com-
binations of several geometric and operational parameters
across broad ranges we wish to study without requiring
that we run a specific simulation study for every possible
combination. This approach has proven effective in opti-
mizing both the geometric and operational parameters of
the CAHM machine. Using LHS, GPR, NSGA-II, and
Pareto-Front in this process ensures robust and efficient
optimization.
Optimization Criteria
The HPGRs were identified as comparison references for
evaluating CAHM’s performance. HPGR performance is
often measured in terms of specific energy (SE), which is
calculated as:
t t SE kW h
Throughput h
Power Consumption $=:
8
6kW@
D
B
However, specific energy only accounts for the throughput
performance of a machine and does not provide a mea-
surement for the comminution performance. This is par-
ticularly important when comparing CAHM and HPGR
where, given the same feed particle size distribution (PSD),
the product PSD of the two may be different.
To account for the efficiency of surface area libera-
tion, a new parameter called Comminution Energy (CE) is
introduced. It is calculated as:
CE kW h
m New Surface Area Generated h
m
Power Consumption kW@
2 2
$=
6
E
E
Comminution energy accounts for energy consumption
for grinding ore and generating new surface area. In the
geometry optimization process, both SE and CE were con-
sidered as optimization metrics simultaneously. This multi-
response approach ensures a comprehensive evaluation of
the machine’s performance, considering both throughput
and comminution efficiency.
One additional metric is gaining increased acceptance
across the industry in a way that helps to minimize the need
for matching particle size distribution curves to compare
performance: the size specific energy (SSE) is the energy
consumed to generate new mass of material below a certain
particle size. The SSE is often used to compare results below
pilot-scale ball mill. This finding underscores the potential
of the MonoRoll for energy conservation in comminution
processes.
CAHM DEM MODELING
Although DEM is a powerful tool for modeling and under-
standing the comminution processes in the CAHM tech-
nology, certain system simplifications and assumptions are
required to make the simulations tractable. These include:
• DEM does not account for mechanical losses of the
system.
• Heat transfer is not considered in the DEM model.
• The model does not take into account the moisture
content of the ore.
• Variations in ore competency are not considered.
• The model does not account for variations in ore
shape.
• The DEM model uses an average shape representa-
tion for rock particles with a truncated size distribu-
tion to limit the exponential increase in the number
of particles below a certain size.
• Smallest feed particle size is 6400 µm.
• Rocky models assume that the hammer roll has no
skewness, and its axis of rotation stays aligned.
DOE-Based Optimization of CAHM Geometry and
Operational Parameters
The optimization of the CAHM machine’s geometry and
operational parameters was conducted using a Design of
Experiments (DOE) approach. This approach allows for a
systematic exploration of the design space and helps iden-
tify the response parameters’ optimal combination. DOE
approaches start by defining the input parameters’ range to
be investigated. A space-filling design of virtual experiments
(DOE) method called the Latin Hypercube Sampling
(LHS) technique is then used to create initial training data
points. The training data points are then simulated, and a
surrogate model is fit to each response parameter.
The surrogate model used in this process is the Gaussian
Process Regression (GPR). This model provides a flexible,
non-parametric Bayesian framework for data analysis, and
is particularly suited for optimization of unknown, stochas-
tic, and expensive objective functions.
Finally, an optimization algorithm is utilized to find
the optimal combination of input parameters for a given
set of target parameters. The optimization algorithm used
in this process is the NSGA-II (Non-dominated Sorting
Genetic Algorithm II). This is a popular non-dominated
sorting-based multi-objective optimization algorithm that
is known for its better convergence near the true Pareto-
optimal front compared to other algorithms.
The decision-making process in the optimization is
guided by the Pareto-Front. This is a set of non-dominated
solutions, being chosen as optimal when no objective can
be improved without sacrificing at least one other objective.
DOE methodology allows us to target specific com-
binations of several geometric and operational parameters
across broad ranges we wish to study without requiring
that we run a specific simulation study for every possible
combination. This approach has proven effective in opti-
mizing both the geometric and operational parameters of
the CAHM machine. Using LHS, GPR, NSGA-II, and
Pareto-Front in this process ensures robust and efficient
optimization.
Optimization Criteria
The HPGRs were identified as comparison references for
evaluating CAHM’s performance. HPGR performance is
often measured in terms of specific energy (SE), which is
calculated as:
t t SE kW h
Throughput h
Power Consumption $=:
8
6kW@
D
B
However, specific energy only accounts for the throughput
performance of a machine and does not provide a mea-
surement for the comminution performance. This is par-
ticularly important when comparing CAHM and HPGR
where, given the same feed particle size distribution (PSD),
the product PSD of the two may be different.
To account for the efficiency of surface area libera-
tion, a new parameter called Comminution Energy (CE) is
introduced. It is calculated as:
CE kW h
m New Surface Area Generated h
m
Power Consumption kW@
2 2
$=
6
E
E
Comminution energy accounts for energy consumption
for grinding ore and generating new surface area. In the
geometry optimization process, both SE and CE were con-
sidered as optimization metrics simultaneously. This multi-
response approach ensures a comprehensive evaluation of
the machine’s performance, considering both throughput
and comminution efficiency.
One additional metric is gaining increased acceptance
across the industry in a way that helps to minimize the need
for matching particle size distribution curves to compare
performance: the size specific energy (SSE) is the energy
consumed to generate new mass of material below a certain
particle size. The SSE is often used to compare results below