Professional Documents
Culture Documents
ABSTRACT
Rio Tinto Yarwun refinery has developed a multi-variable predictive controller (MPC) for optimising our
bauxite milling circuit. The controller was initially developed prior to commissioning Yarwun Refinery’s
major expansion in 2012 in response to the greatly increased milling capacity utilisation requirements of
the expanded refinery. Configured using Honeywell’s Profit platform, the application has proved an
effective way to control mill rates and operation so that the plant can be operated up to constraints at all
times.
In addition to delivering higher milling throughput, the MPC has been further developed to deliver
additional layers of optimisation. This has included optimisation between bauxite mills, assisting in
refinery volume management, and optimising the use of different bauxite grades.
1. INTRODUCTION
Yarwun refinery has an extensive advanced
process control application deployment
incorporating 13 multi-variable predictive
controllers (MPCs). This paper outlines a case
study of the development, maintenance, and
benefits of one of these MPCs which is used
for optimising and controlling the bauxite ball
mills. First developed in 2011/12 to address
an expected production bottleneck from a
refinery expansion – the bauxite mill MPC has
proved a robust control solution and a flexible
foundation for further optimising the area.
2. PROJECT IMPLEMENTATION Figure 1. Mill Process Overview
2.1 Background The Yarwun Stage 2 expansion lifted refinery
The first stage of Yarwun Refinery had a nameplate to 3.4 MTPA but only added one
nameplate of 1.4 MTPA and was supported by additional identical bauxite mill. This required
two closed circuit bauxite ball mills. Each mill mill utilisation and throughput per mill to be
is feed with bauxite at a controlled rate from a increased substantially from historic rates (for
bauxite stockpile by conveyor. The bauxite is example, 0.7 MTPA alumina per mill to 1.13
wet ground with digestion liquor and grinding MTPA alumina per mill). Due to concern that
media. The overflow of the mill is pumped this may become a refinery bottleneck, a
across two vibratory wet screens. The screen project was initiated to build an MPC controller
undersize material reports to a tank which is with the objective of maximising mill rates.
pumped to the slurry storage area while the Figure 2 shows relevant equipment during
screen oversize material is recycled to the feed stage 2 construction – the existing and new
chute of the mill. This arrangement for one bauxite mills can be seen in the foreground
mill is shown in Figure 1. with the bauxite stockpiles. The new and old
digestion units are to the left and right of the
photo respectively.
progressed following a typical MPC
development methodology.
Data collection and site survey were
completed in December 2011 with a main
deliverable being a Functional Design
Specification. This phase included reviewing
process flows, objectives and operating
performance; specification of control system
calculations and tuning of required base level
regulatory control strategies.
Detailed design was completed in February
2012. Deliverable being a controller platform
Figure 2. Yarwun Refinery Oct 2010
with initial seed models and step testing plan.
2.2 Project Scope Step testing was completed in 3 shifts in late
Project scope was intentionally limited to one February 2012. Traditionally, model
mill with focus on delivery time and addressing development is done from either first principles
the main optimisation concern – bauxite or open loop step testing. These methods can
throughput. Optimisation of rates between be time consuming and, in the case of open
Mills (e.g. power consumption etc.) was loop step testing, expose the process to large
considered but excluded from scope as it was disturbances and out of control operation.
not a primary concern in the context of the This project was the first time Yarwun Refinery
refinery expansion. The objectives of the MPC has used Honeywell Profit Stepper software
controller are summarised as below: for model identification. This software allows
• Process Objectives: an MPC to operate with seed models under
closed loop control while perturbing the
1. When running Mill-limited, maximise process with numerous small manipulated
the throughput of the mill variable movements. Process response is
captured to allow model identification and
2. When not Mill-limited, must match refinement. The advantages were that it could
process demand (Slurry storage tank be completed quickly, with limited input from
at 90%) operators or disturbance to the process.
This simple example is mirrored across the • Mill feed rate of increase when the
refinery where it has often been found that manipulated variables have a rate of
operating strategy changes can often be change limit.
achieved by subtle changes in existing MPC
optimisation settings.
5. BENEFITS
3.3 Mill Prioritisation
The MPC has proved over time to be a reliable
A multitude of reasons exist for wanting to and robust method for controlling and
favour rates through one mill relative to optimising Yarwun Refinery’s bauxite ball
another. These could include: mills. When initially commissioned the mill
• Referencing rates away from a mill with an MPC was credited with achieving a capacity
operational issue that doesn’t have online increase of 7.5%. Despite the complex
measurement dynamics of a recycling process, the model
was able to optimise the mill within the
• Trying to preferentially consume from a constraints.
bauxite stock pile.
High utilisation rates approaching 100%
• Favouring a mill that has more reliable (Figure 3) and broad acceptance by plant
instrumentation. operators have been achieved.
The MPC allows this type of biasing to be Some manipulated Variables (MVs) limits ae
easily achieved by simply changing the relative sometimes set incorrectly and can
optimisation values of manipulated variables unnecessarily constrain the operation of the
on each mill. Operator interfaces were equipment and as a result are dropped from
developed to allow flexibility in manipulating the model. The ongoing challenge is the
relative Mill optimisation settings so that rates setting/maintaining of constraint limits that are
through one or another mill can be favoured to real.
take advantage of some unmeasured variable.
Figure 3. Bauxite Mill MPC utilisation (12
months)
6. CONCLUSION
The bauxite milling MPC has been a
successful control change for Yarwun
Refinery. Benefits have included higher
throughput capacity, flexibility and reliability.
Automation of model identification (i.e. use of
Honeywell Profit Stepper software) was
effective for reducing model identification time
and allow closed loop control while performing
identification.
Rework and redesign can be avoided with
greater consideration for future optimisation
objectives and how controller structure will
achieve these.