Predictable processing

Minerals processing software has stopped an alumina refinery’s liquor tanks from overflowing, improving production and safety along the way. Michael Mills writes.

Minerals processing plant operators often face difficult chal lenges when juggling produc tion demand, energy usage, maintenance needs and the volatile nature of the materi als.

Originally applied to the hydrocarbon processing field, Multi-variable Predictive Con trol (MPC) software has emerg ed as an excellent tool to help the plant operators manage this.

Processing control provider Apex Optimisation was con tracted to devise and install a solution for a large alumina refinery in Western Australia.

The plant’s caustic liquor tanks were overflowing at higher throughput rates, which posed a safety hazard to personnel and proved difficult to clean up.

The client’s first response was to lower throughput, but the associated production losses naturally made this a stop-gap solution only.

Apex senior consultant George la Grange told Aus tralian Mining that the project team quickly realised that MPC would be the best solution.

“There were no level mea surement devices on the tanks that were prone to overflow ing,” he said.

“It was therefore near impos sible to predict, from the control room, when a tank was over flowing or when it was about to.

“So the key constraint, the tank level, was not meas ured, meaning that any tra ditional control schemes would not be able to provide the level of regulation that was needed.”

According to la Grange, the software uses a dynamic mathe matic model to predict all the process variables, handles and constraints that could eventu ate when controlling the oper ation.

“A handle is anything that an operator would normally manipulate, such as steam flow through a heater or feed flow into a tank,” he said.

“These are parameters that an operator could move inde pendently without changing anything else in order to meet certain objectives.

“Constraints are measure ments you can’t exceed, such as the temperature at which a liquid will boil or a certain product specification.”

As an example, a milling circuit technician controlling particle size distribution would manipulate two handles, the mill throughput and speed as well as the grinding liquor fed through the material.

“The technology takes the handles and relates them to the constraints and targets and pre dicts how the process will respond to any handle changes,” la Grange said.

“It then uses this model to calculate the best way to meet the objectives.

“If the user changes a pro duct’s specifications to within certain limits, they may be pushing another factor outside its own limits.

“The operator will then have to make a second change to fix this, so the technology is designed to combine all of this into a single controller.”

However, la Grange said the technology was capable of more than keeping the process within safe boundaries.

“It is designed to gradually push a process towards a more optimal state with increased throughput and increased pro duction,” he said.

Unknown constraints

For the alumina refinery project, the team aimed to infer the plant and process constraints by look ing at other process measure ments, a technique often used in hydrocarbon processing appli cations.

Because the software is de signed to predict changes using measured variables, it is possi ble to infer an unmeasured con straint by using a known meas urement that has similarities.

According to la Grange, this had been successfully used in a number of previous applica tions commissioned at the refin ery.

“Apex often faces the chal lenges that emerged in this project, such as insufficient or inferior instrumentation or dif ficult-to-measure materials,” he said.

“However, in this instance there were no clear ‘substitute’ constraints that could simply be used to replace the unmea sured ones.”

The project team began interviewing the plant man agers, engineers and operators, in the hope of discovering what measurement could be used.

“Early on in the discus sions, the team realised that there would not be a single constraint that would infer whether a tank was overflow ing with total accuracy,” la Grange said.

“It was therefore important to identify a number of substi tutes.

“Having more than one con straint provided the applica tion with a hierarchy of safety nets based on estimated accu racy.”

Rapid implementation

MPC solutions typically take around six months to develop and install, but in this case the client needed it operational in a much quicker timeframe.

Apex adopted a streamlined process, described by la Grange as “akin to rapid prototyping,” which fast-tracked the design reviews and went straight into implementation.

According to la Grange, the technology has since met all of the customers goals and require ments.

“The application was imple mented and commissioned within two weeks, which is much quicker than a typical project,” he said.

“There have been no reports of any tanks overflowing, even though throughput has been pushed.

“This means that potential safety and environmental hazards have virtually been eliminated.

“Based on the subsequent financial benefits, the project had a payback period of less than a month.”

According to la Grange, the results in general can vary from plant to plant and depend on the objectives that were set.

“Typically, plants will want an increase in throughput and production,” he said.

“Because the user can set the constraints more intelli gently, they are able to grab every little opportunity that comes by.”

To keep up to date with Australian Mining, subscribe to our free email newsletters delivered straight to your inbox. Click here.