Real-time expert system a real gold mine

ControlGlobal.com

Mine revenues jumped $7 million per year at Kemess Mines in British Columbia after installation of an expert system controlling the process, and pushed throughput by 5%.

By Greg Rasmussen, Mill Superintendent, Northgate Minerals Corp.

NORTHGATE MINERALS CORP. (Vancouver, BC, Canada) is a gold and copper mining company. Its principal assets are its Kemess

HAVING A BALL

Ball mills (shown here) reduces the mined material down to 150 microns, producing 300,000 ounces of gold per year.

South mine in north-central British Columbia, which produces 300,000 ounces per year of gold, and the adjacent Kemess North deposit, which contains an indicated resource of 5.4 million ounces of gold and 2 billion pounds of copper. The Kemess South mine has two mill lines, each consisting of a semi-autogenuous grinding (SAG) mill that grinds rock from 6 in. down to 1/2 in., and a ball mill that further reduces the material down to 150 microns. Powered by two 6,000-hp motors, each of the mills are 15.5 ft. long and 34-ft. in diameter and are controlled by an Invensys Foxboro I/A distributed control system (DCS).

In the past, the mills were not pushed that hard because operators, who also have many other jobs, were not able to focus on throughput. To overcome this, run the mills on a more consistent basis and stabilize our control strategy, the company decided to implement an expert system. We issued a request for proposals and selected mine process systems integrator Minnovex Technologies because the its people clearly understood the mining business and their references were excellent. Minnovex’ proposal also strongly emphasized operator buy-in, which anyone knows critical for a successful implementation of these types of systems.

Room for Improvement
Because operators did not have time to consistently monitor the operation of the mills, downtime was higher and throughput lower than it could have been. The mines operated with four different crews, each with their own individual procedures. Operators were usually able to make supervisory control adjustments but only about every 30 minutes under normal conditions and every 5–10 minutes in the case of upsets. The mills were producing 49,500 tons/day but we believed there was considerable room for improvement. Every 1% increase in throughput in the mills produces $1.5 million in incremental revenues, so the stakes were high.

Northgate began the implementation process with Minnovex by talking to the people that know the machinery and processes best, the mill’s operators. They were asked a number of questions including: How do you control the equipment? What control strategy do you use under various conditions? When do you make setpoint changes and how does the process behave in these situations? Northgate’s managers distilled their existing strategy into a flow chart of decision-making operations that everyone understood. The result was that on Minnovex’ very first trip to the plant, which involved two people and took five days, everyone--from operators to top management--agreed on the best operating practices for the mill. From the beginning the integrator had a very good idea of what the expert system would do and went back to their offices to create it.

Platform Reduces Development Time
Minnovex Expert Technology (MET) system consists of a toolkit and a repeatable and sustainable methodology designed for developing expert systems in the industries in which the company works. The toolkit was developed using Gensym’s G2 real-time expert system platform. An expert system is a software program that solves problems using information and reasoning techniques normally associated with a human expert. With such a system operators can assess, diagnose, and respond to unusual operating situations or seek to optimize operations. The MET system also provides a number of diagnostic and development tools so that system administrators can help ensure maintenance goals are met and continued growth are possible without extensive training or experience.

The G2 platform provides a structure that makes it easy to collect data in real time and determine what actions to take. Development time and effort are reduced because most of the application logic can be applied parametrically by changing the properties of or relationships between objects rather than by coding.

MET modules use the platform to optimize the operation of the mill lines by defining the state of the process through both measured and inferred values. MET then determines the operating actions that maximize throughput.

For example, suppose the mill was approaching an overload condition, with the power decreasing and load increasing. In this situation, the expert system employs fuzzy logic, an approach that emulates human reasoning. With fuzzy logic the rules of inference are approximate rather than exact.

LIFT AND SEPARATE

Kemess' flotation cells accept the output of the ball mills and separate the minerals from the crushed rock.

The control system feeds the instrument readings and the rate at which the load is changing to fuzzy-logic sets that are designed to encapsulate the experience of the operators in running the mill. At an 1,800 psi load, the expert system might have a 10% belief that the load is high. At a 2,000 psi load, it might have a 70% belief. At a 2,500 psi load the system may be completely certain (100%) that the load is high. The level of confidence that the load is high becomes the driver for the expert system’s actions, in this case, reducing the feed. The fuzzy logic employed within the expert system works just like a human operator by determining the magnitude of the response based on the confidence level in the original assumption.