Advanced control systems and techniques are steadily getting smarter and quicker, but there’s still room for improvement
By Russell A. Carter, Managing Editor
Fuzzy logic, Boolean operations, neural network models, genetic algorithms, model predictive control. These are the concepts that form the lingua franca of the engineers and scientists who design expert process control systems—a common language that allows them to develop products and techniques capable of boosting mineral plant throughput and recovery performance through process optimization. It’s also a language that’s probably foreign to the average mill operator sitting in a control room, scanning a bank of screens that graphically display plant status and trends, but its vocabulary defines the highly sophisticated control techniques that enable average operators to consistently keep a plant running at expert-operator levels.
Given the vast process-control knowledge base—drawing upon decades of expert system development—and the computer processing power available to almost anyone today, the upper-single-digit plant-performance improvements commonly reached by process optimization systems may not seem all that impressive. However, achieving large performance gains in a mine plant can be a tough proposition. A group of authors from Utah, USA-based expert-system developer KnowledgScape described the problem succinctly in a paper presented at a 2009 technical conference. Noting that once an expert system has been installed in a plant and the optimization process runs its course, plant performance usually doesn’t improve significantly thereafter, they explain that “…by in large, expert systems are more static than they are adaptive. Static rules, taking into account basic and fundamental relationships of grinding and flotation have proven themselves time and time again to be capable of significantly improving average plant performance. However, mineral processing unit operations are generally thought of as being nonlinear and complex. Nonlinear in this context means that the future state of the plant is dependent upon the current state. So if SAG mill bearing pressure is at 600 and the feed rate is increased 50 t/h, the dynamic response will be very different than the response if the bearing pressure had been at 700. The concept of complex means that once a series of system changes are made, reversing those changes does not return you to the same place you originally started from. Clearly this phenomenon, while not understood by many, contributes to the difficulty of controlling and optimizing typical flow sheets as well as contributing to the performance plateau we’re currently on.”1
The solution, they suggest, lies in “data mining”—extracting useful information from large databases. But more on that later.
Model the Process, or Model the Operator?
Advanced process control systems for mineral applications are offered in two basic flavors: model predictive control (MPC) and expert systems. The differences between the two types of systems are mainly that:
• MPC uses a model of the process; an expert system uses a model of the operator.
• MPC is predictive; an expert system is algebraic.
• MPC is closed-loop control; an expert system is open-loop control.
• MPC is algorithm-based; an expert system is rules-based.2
The advantages of one type of system over another in mine plant applications has been a topic of debate for as long as they’ve been competing in the market. Some systems are hybrids; for example, ABB’s expert control strategy includes techniques such as variable gain, multivariable fuzzy rule blocks, neural networks and MPC. And many are modular; Metso Cisa’s OCS process control software, for example, comprises a number of modules integrated into a single structure, including:
• A “fuzzy expert” module, a real-time expert system with an inference engine, crisp and fuzzy logic reasoning and a knowledge base.
• A soft sensor module using adaptive predictive models and a filter estimator for on-line self-tuning of the models. Users can develop new models in Visual Basic.
• An optimizer module, with a constrained SQP algorithm.
• A neural network module.
• A vision module such as VisioFroth for froth flotation plants, VisioRock for SAG and AG milling or crushing and screening plants, VisioPellet for pelletizing plants, etc.
Control systems are not a substitute for a plant DCS or PLC system; they are primarily a high-level supervisory authority that supplies set points to the lower layer of control systems. The general path prescribed by these products—measure, control, then optimize—is illustrated by the diagram below, and by a recent example of success applying this approach to a mill operated by a long-established, technically savvy European base-metals producer.
ABB reported that it had successfully applied its Expert Optimizer system to the zinc flotation circuits of Boliden’s Garpenberg copper-zinc concentrator in Sweden. The objectives of the effort were twofold: First, to stabilize the process, which is subject to external disturbances such as ore-quality changes. Second, to maximize zinc revenue by improving zinc recovery and concentrate grade.
According to ABB, successful optimization of the circuit required three main steps. Process behavior had to be predicted using a reliable dynamic model of the flotation circuit and cost functions and operating constraints then had to be implemented, to produce a robust MPC application. Among the variables that could be manipulated were cell level control, air supply, froth level and reagents.
After these steps were taken, a measurement program began in late 2008 and was completed in mid-2009. The results, according to ABB, showed that the concentrator’s efficiency (directly corresponding to revenue) was at least one percentage unit higher when MPC was used, compared with the existing manual control strategy. In addition, zinc concentrate grade was higher and more consistent, while the recovery rate remained constant.
Solving Basic Problems First
Regardless of how sophisticated and smart they may be, advanced process control strategies will not correct inherent design or flowsheet problems. Potential customers for advanced control systems products need to be aware of some basic considerations and requirements as well as the opportunities afforded by these systems, according to Phillip Thwaites, Process Control Group manager for Xstrata Process Support, who addressed the topic at the AutoMining 2008 conference held in Santiago, Chile. In order to gain the most value from advanced control solutions, Thwaites suggested operators first determine:
• Is your feed stable?
• Are your instruments calibrated and performing?
• Are you aware of wireless instruments (including vibration)?
• Is your control system up-to-date and stable?
• Are you in manual or auto control?
• Are your operators acting on alarms or are they nuisance?
• Do you understand and accept your process variability?
• Are you operating within the design targets and process constraints (pumps, cyclones, roasters, furnaces, etc.)?
• Are you using your surge capacity, or running tight level control?
• Are you at optimum and are the controls robust?
• Are you benefiting from asset management systems?
• Are failure/fault detection systems implemented?
• Can you make the same product for less energy consumption?
Although optimization of specific plant processes, i.e., flotation circuit performance, concentrate grade, recovery, etc., has been a prime focus for advanced control methods, energy consumption/conservation is gaining importance as another optimization objective. Sometimes advanced control strategies can accomplish both. In a recent issue of the XPS Bulletin newsletter,3 Thwaites de-
scribed how these twin goals were achieved at an Xstrata nickel concentrator.
Grinding, he noted, is one of the most important and expensive processes in the mineral processing industry, often accounting for around half of the total operating costs in base metal concentrators. Con-
sequently, this has led to a focus on development and implementation of grinding control and optimization strategies. In addition, optimizing a flotation circuit can significantly reduce reagent costs, while maximizing recoveries. The benefit of this in the overall metal extraction process can be considerable.
As an example of these benefits, Thwaites pointed to a presentation by Eduardo Nuñez, an XPS Process Control engineer, at the Canadian Mineral Processors 2009 conference in Ottawa. Nuñez described a new grinding control strategy implemented at Xstrata Nickel’s Strathcona mill, which optimized the grinding operation by maximizing the throughput while maintaining product quality for flotation. The implemented control strategy has the added benefit of improving the stability of the rougher flotation circuit and improving energy efficiency of the rod and primary ball mills. According to Nuñez, “The implementation of a new grinding control strategy led to energy efficiency increases of 7.1% and 7.5% in the rod and ball mills, respectively.”
Another XPS project, this one involving Xstrata Alloys’ Eland platinum concentrator in South Africa, highlighted the fact that major gains can be achieved by optimizing a standard control system before turning to advanced control techniques. As explained by XPS staff members Jocelyn Bouchard and Martin Émond,4 the Eland concentrator was commissioned in Novem-
ber 2007. In 2008, XPS Process Mineralogy carried out an ore characterization program and a mixed collector lab program. An improvement in Pt/Pd recoveries was identified by finer secondary grinding, and by the use of the mixed collectors. XPS Process Control then followed up by conducting an optimization program encompassing the entire mill operation.
Process control improvements were key objectives of the program and early results demonstrated how leveraging from current assets can bring significant improvements through overall reduction of variability, revealing what can be achieved in a standard (PLC or DCS) control system, with early-stage process interventions results providing steadier operation of the primary ball mill circuit. Following implementation of changes identified by the optimization program at the Eland mill, significant and sustained reductions of standard variation were observed for key process variables: primary ball mill feed rate (72% reduction), rougher feed density (58% reduction) and rougher feed flow rate (47% reduction).
The authors point out that optimizing the plants assets not only involved proper utilization of circuit capacity, but also “leveraging” from positive effects on downstream processing stages. The importance of proper grinding control, for example, went beyond issues of energy costs and state of liberation. The stability of operation and optimizing the density and size distribution of the grinding circuit product required careful consideration, as they extensively influenced flotation performance.
Most of the larger process-control solution providers that focus on serving the mining sector can provide a full range of control solutions for a wide variety of process applications. Outotec, for example, has for many years assisted customers in optimizing their grinding circuits by providing particle size analyzers and MillSense mill charge analyzers. It has steadily extended its portfolio and now offers an advanced grinding solution—incorporating a comprehensive measure/stabilize/optimize approach—along with enhanced control strategies for flotation.
Its latest flotation control product is CellStation, a local/field integrated controller for both air and level control that is expandable and connectable into a plant-wide DCS using Profibus DP (Ethernet), eliminating any need for wiring of 4–20 mA signals. The CellStation unit is delivered fully operational, only requiring minimal configuration in the field, and can therefore provide a control solution without involving expensive third-party system integrators.
CellStation, according to Outotec, is designed for controlling flotation machines with one to three separate air feeds and a common slurry level with one or two output discharge valves. Outotec’s EXACT-Level expert system functionality adapts to the process changes and stabilizes the slurry level in the entire bank of flotation cells. The functioning of the controller is defined by parameter settings only and thus there is no need for custom programming.
The company says CellStation control loops and parameters are easy to configure from the plant control system or from a local panel to observe immediate effect on the circuit. Process data is available for process historians and databases over the Profibus link. Trends of the process data are also available for operators to observe on a local panel on the plant floor.
The EXACT system, according to the company, monitors the entire flotation line and effectively compensates for process disturbances before cell levels are affected. Its independence from flotation line configuration and capability for joint operation with traditional PID controllers allow easy implementation to existing and new concentrators. An automatic tuning component identifies process parameters on-line, making use and maintenance of the system simple.
Mining the Data
As mentioned earlier in this article, when an advanced control system is installed it’s fairly typical for the plant to achieve an initial boost in productivity through process optimization, followed by a performance plateau. Engineers and software designers at expert-systems developer Knowledge-
Scape are interested in finding out whether expert-system strategies can be improved to wring even more improvement out of a given system.
The answer may lie in closer scrutiny of the vast amount of data generated by modern process control systems, the volume of which is likely to grow significantly as these systems become capable of handling more inputs and variables. However, rarely—if ever, does all this data receive close analysis—particularly in real time, and this represents a missed opportunity to devise more effective control strategies for grinding and flotation circuits, for example. Data mining may provide the means to extract value from this raw data.
The data mining process, according to the SME conference paper presented by Knowledgescape authors Lynn B. Hales, Michael L. Hales, Clark D. Burbidge and Dustin T. Collins, consists of several steps:
• Data cleaning—removing noise, inconsistent data, outliers;
• Data integration—combining data from multiple sources in a time-unifying manner;
• Data transformation—changing according to rules that unify the data or make it suitable for data mining operations;
• Data mining—the process of applying intelligent methods to the data to extract meaningful data patterns;
• Pattern evaluation—sorting through the discovered patterns to identify those that represent knowledge of the underlying processes; and
• Knowledge presentation—visualization processes to present the mined knowledge to the user.
The objectives of the process are twofold, according to the authors: 1) discovery—“tell me something I don’t already know; and 2) prediction—the use of a discovery to predict the future. For those skeptics who question how statistical analysis, employing methods such as linear and nonlinear stepwise regression, auto-regressive moving averages, analysis of variances, etc., differs from data mining, they explain that “data mining is statistical analysis that is highly automated and rigorously applied to all available data in a completely exhaustive manner.”
They also note that the algorithms used in the statistical analysis are quite different and much improved from just a few years ago, and conclude that, “…just as expert control promised improved grinding and flotation performance 20 years ago, tightly integrated and focused data mining enhancements to present expert system technologies promise further improvements in the years to come.”
References
1. L.B. Hales, M.L. Hales, C. Burbidge, D. Collins, “How to Increase Plant Performance with Artificial Intelligence and Expert Systems,” Preprint No. 09-074, SME Annual Meeting & Exhibit, 2009, Denver, Colorado.
2. Greg Martin and Steve McGare, “Optimizing the Cement Production Process Using Model Predictive Control,” Pavilion Technologies, May 2002.
3. Eduardo Nuñez, “Increasing Energy Efficiency: How Process Control Can Help,” XPS Bulletin, Issue No. 2, 2009.
4. Jocelyn Bouchard and Martin Émond, “Asset Optimization: Xstrata Alloys Eland Concentrator,” XPS Bulletin, Issue No. 3, 2009.
