Prototyping package is blueprint for productivity

Rio Tinto improved coal yield and reduced water use in its Hunter Valley Coal Prep Plant by implementing radical control strategies on antiquated controllers.

Rio Tinto improved coal yield and reduced water use in its Hunter Valley Coal Prep Plant by implementing radical control strategies on antiquated controllers.

Rio Tinto's Hunter Valley Coal Preparation Plant (HVCPP) processes 15 million tonnes of coal per year.

Such high throughput means that there is potential for significant economic and environmental gains by optimising the major control loops and reducing variation.

However, the high throughput is a double-edged sword.

It also means that any downtime required to make such plant modifications will incur significant costs in lost production.

Rio Tinto partnered with Matrikon to establish a new approach to plant optimisation that has two major advantages over conventional approaches.

The control strategies could be proven to work before they were programmed into the plant hardware, removing the risk of extended plant downtime.

And the control strategies could be implemented substantially faster than would be possible using conventional techniques, even on hardware which is outdated and difficult to program.

This promised to keep disruptions to an absolute minimum.

HVCPP was already meeting industry standard performance metrics, but in the spirit of continuous improvement, Rio Tinto Planning and Improvement Superintendent Luke Dimech identified three control systems with room for improvement: dense-medium control, thickener control, receival throughput control.

In January 2006, Dimech engaged Matrikon to help optimise and tune these three control systems.

After investigating the control systems' performance, it was soon realised that the best results would come from some radical changes to the control algorithms, as there was little room for improvement using standard PID algorithms on these material recovery processes.

Their highly complex dynamics and hard-to-understand interactions, due to their recycle loops and nonlinear processing equipment meant that more sophisticated advanced control algorithms were called for.

Most existing coal prep plants, including this one, use outdated and rudimentary PLC controls that don't provide any control toolboxes more sophisticated than a standard PID loop.

It is time-consuming, difficult and downright risky to try and trial advanced control strategies using the existing controllers' limited programming interface.

Unless it could be proven that such radical changes to the controllers' algorithms would work, and work first time, there was little hope of getting senior management approval for modifying the process.

With this in mind, Matrikon and Rio Tinto came up with a prototyping method that would ensure 100% confidence in any new control strategy before modifications were made to the existing controllers' code.

This method also allowed implementation with virtually no plant downtime.

"The technique we used for prototyping, then implementing the plant modifications removed almost all the risk that this kind of work normally entails".

"Without Matrikon's novel approach we would have had to think twice about whether to go ahead at all with these modifications to our critical plant", said Dimech.

This approach was used on all three control systems, but this article will focus on the dense-medium control system.

Coal is readied for processing, or beneficiated, through treatment with a dense liquid medium.

Careful control of the density of this medium is important, as constant density is a prerequisite for the efficient separation of reject material from coal.

The more efficiently the reject material is separated from the coal, the better the coal quality and yield.

The dense-medium control is typically done by two control loops, controlling two splitter gates and one water valve.

In this strategy, the density is controlled (controller DC) by manipulating the over dense splitter gate.

To increase density, more over-dense medium is diverted to the dense medium circuit.

The level of the dense medium circuit's sump (controller LC) is controlled by manipulating both the water flow and the dense medium splitter gate position.

If the level is too low, then the dense medium is diverted to the sump to increase the level with extra water flow.

If the sump level is too high then the water flow is reduced to an operator-set minimum flow and the sump level is reduced by diverting flow away from the sump with the dense medium splitter gate.

The problem with this strategy is that as the sump level moves above and below the setpoint, the control system continuously oscillates between the two control modes.

Because the mode of the level control also has a significant effect on the density of the dense medium, the density oscillates in tandem with the level control action.

Decoupling the level control from the density control can remove the unwanted oscillations.

One way to do this is to exclusively control the level of the sump with the dense-medium splitter gate.

This leaves the water valve and the over-dense splitter gate to control the density.

This can be done by splitting the density controller output and biasing each output to reflect their true contribution to the density.

If these biases are correct, when over-dense medium is directed to the sump to increase density water addition will decrease accordingly to offset the volume impact of the over-dense medium.

This means density control won't affect sump level control.

Cleaning of the medium aids separation efficiency.

Experience indicated that optimum cleaning of the medium occurred when the gate was 30% open.

Using this knowledge, the dense-medium gate position signal was added as another input to the density-control algorithm.

This new input is used to bias the flow into the dense-medium circuit so the system is always biased toward working with the dense-medium gate in the position that gives optimum performance.

This new strategy sounds great in theory, but a lot of extra biases and tweaks were introduced that needed to be adjusted before the strategy could be commissioned.

Implementing a complex decoupling algorithm like this is fraught with risk and the prospect of costly plant downtime, especially when the existing control hardware is antiquated and unwieldy.

Matrikon's solution was to implement the strategy using ProcessACT, a software prototyping package, and its join-the-dots graphical interface.

This allowed us to quickly create and debug an advanced control strategy offline using advanced control toolboxes.

"Instead of translating the strategy into a proprietary programming language, we simply copied the block diagram, dropping prewritten blocks onto the screen and joining them together with arrows as appropriate".

"We could even simulate the control system offline by dropping in simulation blocks, further increasing confidence in the strategy".

"Once we were happy with the soft-controller that we had programmed on our laptop, Matrikon's OPC connectivity allowed us to hijack the PLC's I/O using nothing more than an Ethernet cable and 10 minutes' configuration".

"Once we had control of the I/O, we had control of the plant from our laptop, safe in the knowledge that if anything unexpected happened we could flick the software switch and put the old PLC controllers back in control", said Dimech.

"Connected to the plant, ProcessACT's modern user interface allowed us to quickly fine-tune and prove the new strategy on the actual plant".

"With our proven control strategy having eliminated the risk from our endeavour, it was time to tackle the task that no one wanted to do: implement it on the old-school PLC.

This was a painful experience; the programmer still wakes at night with a cold sweat when the interface invades his dreams".

"But when you know what your tuning constants and biases are and you are 100% confident that your control strategy definitely works on this specific plant, the task is much more manageable", Dimech concluded.

The reduced variation in density means Rio Tinto is now separating ash from coal more efficiently, giving a direct improvement to the bottom line.

Control system prototyping software with powerful advanced-control tools, easy-to-use interfaces, and connectivity options that can quickly hijack a PLC's I/O, mean that advanced control is no longer confined to new installations with state-of-the-art hardware, as was proved at Hunter Valley.

The approach of quickly proving advanced control strategies on the actual plant the strategies will be controlling, and then implementing those pre-tuned and fully commissioned strategies directly to a less friendly control platform minimised risk for all involved.

This new approach to process optimisation helped Rio Tinto to make many operational improvements, including improving its coal yield and reducing water consumption in a drought-prone area.

This could all be done safe in the knowledge they weren't risking any unplanned plant downtime.

Back to top