TOWARDS THE DESIGN OF A DATA MART FOR BUSINESS INTELLIGENCE IN MINERAL PROCESSING Francis Kofi Andoh-Baidoo Department of Computer Information Systems & Quantitative Methods College of Business Administration University of Texas Pan American 1201 W University Drive, Edinburg TX 78539 (956) 665-3397 andohbaidoof@utpa.edu ABSTRACT The process manager in the mineral processing industry typically relies on various data sources to support decisions to control and improve the plant s performance. The fragmented nature of the various data sources hinders the effectiveness of the process manager s decision making. While industries such as financial and healthcare have made tremendous investments in business intelligence applications to enhance decision, the mineral processing industries lack such investments. In this paper we argue that investments in business intelligence applications can be made to support the process manager s decision making activities. First, we identify and describe some of the various transactional data available to the process manager in a mineral processing company in Ghana. We then design a data mart and explain how online analytical processing and reporting systems can be developed from the data mart to provide business intelligence and thereby enhance the process manager s decision making. Keywords: Business intelligence, data mart, mineral processing, data mining, decision making INTRODUCTION Process managers in the mineral processing industry rely on data from diverse sources to improve the performance of the different process units and overall plant performance. For instance, in a typical mineral processing plant, the metallurgical laboratory performs laboratory analysis on various combinations of reagents and ores. This information forms the foundation upon which the plant s performance may be measured. During the plant operations, ore and pulp samples are collected from the plant and sent to the chemical laboratory for analysis. The gold, sulfur and carbon content are some of the parameters that are reported from the chemical laboratory to assist the process manager make decisions to improve plant performance. The data are reported per shift and daily. A typical mineral processing plant also has a distributed control system (DCS) or a variant process control and instrumentation system that provides real time information on the various units of the plant including pump speed, tank levels, valve openings and pulp temperatures. The process manager in consultation with the instrumentation engineers may change some of these parameters to enhance the plant s performance. Thus, the data from the chemical laboratory and the DCS are transactional data from the plant that are available to the process manager.
The performance of a process manager is measured by how he or she is able to use all these fragmented information to control the plant s parameters and thereby improve the plant s performance. In many instances, changes made in one unit have rippled effects on many other units throughout the plant; When a poor decision leads to wrong process parameter changes, it takes time to address the issues and in most cases gold is thrown away as tailings. Much as the change needs to be right, it also needs to be made quickly. What makes the process manager s decision making more challenging is the vast amount of data that needs to be analyzed in that short time span to ensure that effective decision is made. Hence, the fragmented nature of the source data is a great challenge for the process manager s decision making. The transactional data from the different sources can be combined into a data warehouse or data mart that can serve as a data source for business intelligence to enhance the process manager s decision making. In this paper, we propose the design of a data mart for mineral processing. We then explain how online analytical processing (OLAP) and reporting systems can be used to provide a multidimensional view of the historical data and relations between the processing variables from the data mart. The historical data can be used to provide business intelligence and thereby enhance the process manager s decision making. This research demonstrates the applicability of business intelligence in the mineral processing industry. Thus, for practice our main contribution is to demonstrate how various industries beyond financial and healthcare can take advantage of business intelligence using mineral processing as a case study. From a research perspective, we contribute to the build model type of information systems design science research (Andoh- Baidoo et al. 2007; March et al. 1995). This paper could be useful for engineering management research especially on the issue of decision making. With the increase in processing power, it is important that all industries take advantage of the available technologies to manage processes and enhance decision making and competitiveness (Karr et al. 1996). BACKGROUND The decision making of the process manager is complex because of the massive amount of information that needs to be analyzed. Decision support systems (DSS) are provided to assist decision makers make effective decisions. For the purposes of the discussion in this paper, decision support systems include all forms of information systems and technologies designed to assist one or more decision makers in making a specific decision or choosing a specific course of action (Scott Morton, 1984). One of the challenges of DSS is decision quality (Kasper et al. 2006). Further, Wright et al. (1998) note that 55% of the practitioners that were interviewed were dissatisfied with the capabilities of the software systems they use to support their modeling activities to support decision making. However, in recent years, new tools/technologies such as data warehousing, OLAP, data mining and web have influenced decision support systems (Shim et al. 2002). Business elligence and Technologies Just as the technologies used to provide decision support have changed over the time so has the name. Now both in practice and research, business intelligence is the common terminology used in references to applications that provide decision support capabilities in decision making and
especially those that provide analytics (Watson et al. 2007). The term was coined in the 1990s by Howard Dressner, who then was an analyst at the Gartner group. The term business intelligence has been defined as a broad category of applications, technologies, and processes for gathering, storing, accessing and analyzing data to help business users make better decisions (Watson 2009, p.487). The Gartner s Group survey of CIOs reports that business intelligence has been the number one technology priority since 2007 (The Gartner Group 2009). The business intelligence processes can be broken into two main processes: getting data in and getting data out (Watson 2009). The getting data in process involves the structuring of the data warehouse architecture based on the business requirements and existing transactional systems, transforming the transaction data to match the warehouse architecture and transporting the data into the data warehouse or data mart. Unlike a traditional database, a data warehouse is specifically designed to provide decision-oriented support for organizations. Typically a data warehouse stores large amount of an organization s operational data. This large historical data can then be queried to support organizational decision-making. A data warehouse is characterized as being subject-oriented, integrated, time-varying, non-volatile collection of data for use in organizational decision making (Inmon 1992). Thus, data can be collected on specific subject from various sources. A data mart is a small data warehouse with a focus on specific aspects of an organization such as a particular functional area, geographic region or division within the organization (Larson 2008; Watson 2009). These data marts may be fed by a large data warehouse (Watson 2009). The getting data in process poses more challenge than the getting data out process and makes up about 80 percent of both time and effort of the entire business intelligence activity (Watson 2009). Extraction, transformation and loading (ETL) of source data to the data mart or warehouse occurs during the getting data in process. Several vendors including Oracle and Microsoft provide technologies to ease the ETL activities. The focus of the getting data out process is to extract data for the decision maker by performing online analytical processing (OLAP) on the data and generating reports using reporting systems or other applications including data mining. Decision makers can query the data warehouse about specific activities in the past. OLAP tools are designed to provide dimensional analysis of historical data to support decision making. OLAP as a software technology enables knowledge workers and or decision makers (analysts, managers, and executives) gain insight into data by providing fast, consistent, interactive access to a wide variety of possible views of information that has been transformed from raw data to reflect the real dimensionality of the enterprise as understood by the decision-maker (cf. Shim et al. 2002, p.115). Vendors such as Brio, Cognos, Microsoft, Oracle and Hyperion provide front end query and analysis tools on the data mart. OLAP cube (multidimensional model) can be built from the data warehouse or data mart to support flexible roll-up and drill down analyses. Further, data mining tools using artificial and statistical techniques can be used on the data in the warehouse to provide sophisticated data analysis.
MINERAL PROCESSING Mineral processing is the study, design, implementation and improvement of processes that reduce particle size (grinding) and separate valuable mineral products from ores using techniques based upon material properties such as density, surface chemistry, magnetic susceptibility and electrical conductivity (Amankwah et al. 2005; Karr et al. 1996). Procedures for treating free milling gold ore while environmentally friendly poses challenges because it places specific requirements on the particle size after grinding and the addition of grinding aids (Amankwah et al. 2005). Treating non-free milling sulfides and oxidizes ore is more difficult because it requires more circuits and more reagents and poses more environmental challenges. Thus, effective treatment of such ore is very desirable (Karr et al. 1996). Mineral Processing at Ghana Gold Company In this paper, we consider two plants that are designed to treat sulfide/transition materials (gold embedded in sulfide matrix) at the Ghana Gold Company. For these plants, ore treatment involves crushing, milling, flotation, biological oxidation/roasting, leaching in cyanide, carbon adsorption, desorption, electrowining and smelting (Lindström et al. 1992). The mineral processing activities start with the delivery of gold ore to a stockpile which is fed into a jaw crusher. Crushed ore is fed to a milling circuit through vibrating feeders. The crushed ore is milled in a steel ball mill. The milling action liberates free gold, which can be concentrated in gravity concentrators (Amankwah et al. 2005). The less liberalized pulp is sent for flotation. This involves addition of reagents that allow the gold particles to attach to air bubbles, float onto columns and then collected into tanks. The flotation concentrate is oxidized at the BIOX. Sulfur and iron oxidizing bacteria in reactor (thiobaccillus thiooxidants and thiobaccillus ferrooxidants) tanks oxidize the sulfides and liberate occluded gold in the process to render it amenable to leaching (Lindström et al. 1992). Washed product from the BIOX is sent to leach tanks (tanks that contain pulp ore and dilute sodium cyanide). Activated carbon (freshly prepared carbon with large pores of surface areas greater than the size of a football pitch) is pumped counter current to the flow of the BIOX product to absorb the dissolved gold solution onto the carbon pores. The gold-loaded carbon is pumped into a stripping column (where dilute sodium hydroxide and sodium cyanide solution are heated to about 110 C) to desorp the gold adsorbed into an eluate tank. The gold rich eluate is pumped through electrowining cells in which the gold is recovered from the eluate by electroplating onto steel wool cathodes. Gold loaded onto steel wool cathodes is calcined in an electric oven and smelted in a diesel furnace to produced ore gold ingots. TRANSACTIONAL DATA IN THE MINERAL PROCESSING PLANT The input, processing and output parameters for each of the processing units in the plant are recorded. For example the output from the crusher serves as input feed for the mills. Each mill has specific ore size requirement which therefore places a size requirement of crushed ore. Thus, average size of input ore, average size of output ore and throughput are some of the parameters that are recorded each shift and each day over the life of the plant. These data are transactional data for the crusher section. Similarly, the different processing units such as the milling, flotation, roasting/biox, CIL and Electro winning provide transactional data. Other
transactional data are those provided by the DCS. Example of such data are tons milled, tons crushed, mill density, average tank level and average pump speed. Data Requirements The requirements management activity revealed that the process manager is typically interested in the following: Tons of Ore crushed per ore type, per shift, per plant Tons of Ore milled per ore type, per shift, per plant Recovery of Gold in the flotation circuit per shift, per plant Recovery of the BIOX circuit per shift, per plant Recovery of the CIL circuit per shift, per plant Tons of Flotation concentrate per shift per shift supervisor, per plant Tons of BIOX concentrate per shift, per plant Plant rolls to country Shift rolls into day, which rolls into month, which rolls into quarter, which rolls into year Shift supervisor rolls into plant manager which rolls into country THE DATA MART FOR MINERAL PROCESSING AT GHANA GOLD COMPANY The data model includes dimensions, facts, measures and attributes. In the following we define these terms and present examples of these parameters. Measures A measure is numeric quantity expressing some aspect of the organization s performance. The information represented by this quantity is used to support or evaluate the decision making and performance of the organization. A measure can also be called a fact (Larson 2008, p. 32). Table 1 is a list of the measures identified for mineral processing activities at Ghana Gold Company. The process manager is interested in measuring the recovery of some of the processing unit (eg., Flotation, BIOX, CIL). The recovery can be computed as: 100 * (Head Grade Tail Grade)/Head Grade. The plant data includes the head and tail grades for all the processing units. These data will serve as the measures in the data mart and subsequently used for OLAP.
Table 1. Mineral Processing Measures Measures Head Grade* at Crusher (Gold) BIOX Feed Grade (Gold) Head Grade Crusher (Sulfur) BIOX Feed Grade (Sulfur) Tail Grade Crusher (Gold) BIOX Feed (Tons) Tail Grade Crusher (Sulfur) BIOX Tail Grade (Gold) Tons Crushed BIOX Tail Grade (Sulfur) Head Grade at Mill (Gold) BIOX Tail (Tons) Head Grade at Mill (Sulfur) BIOX Reactor Temperature ( o C) Tail Grade at Mill (Gold) CIL Tails Grade (Gold) Tail Grade at Mill (Sulfur) CIL Tails Grade (Sulfur) Tons Milled CIL Tails (Tons) Flotation Concentrate Grade (Gold) Carbon Consumption (Tons) Flotation Concentrate Grade (Sulfur) Cyanide Consumption (Tons) Flotation Concentrate (Tons) Gold won (Oz) Flotation Tails Grade (Gold) Flotation Reagents Consumption (Tons) Flotation Tails Grade (Sulfur) Steel Ball consumption (Tons) Flotation Tails (Tons) * All the Head and Tail Grades are measures in grams/ton (g/t) Dimensional Tables and Hierarchies A dimension is categorization used to spread out an aggregate measure to reveal its constituent parts (Larson 2008, p. 32). Dimensional tables are parameters by which the measures are needed. For instance the process manager may be interested in a measure such as flotation tails per period of time. Here time will be the dimension. A dimension may form part of a larger structure having many levels. Such a structure is referred to as hierarchy. A hierarchy is a structure made up of two or more levels of related dimensions (Larson 2008, p. 37). Time for instance forms a hierarchy at many levels as it can be rolled up or drilled down. Thus, data collected over shifts can be rolled up over a month or year. Similarly, the data collected over a year can be drilled down to a particular shift on a particular day. A dimension also has attributes. An attribute is an additional piece of information pertaining to a dimension member that is not the unique identifier or the description of the member (Larson 2008, p. 37). Time Dimension The different management levels have interest in the plant performance across time. The performance of the plant is measured by shift, day, month, quarterly and yearly. Hence time is an important dimension in the mineral processing environment. Ore Source Dimension A specific ore source or a combination of several sources may require different ore treatment and generate different parameters. Hence, the process manager may require analysis of the different parameters per ore type. Hence ore type is a dimension for the data mart architecture. An ore may not necessarily be associated with a plant because an ore can be transported to be treated in another plant or another country. However the Ore type is associated with country where it is sourced and therefore Ore type forms a hierarchy with country.
Plant Dimension The mineral processing company has different plants in different locations across the globe. The firm may be interested in the plant performance across locations. Hence plant is an important dimension for business intelligence in mineral processing. Plant is associated with country hence plant and country form a hierarchy. Personnel Dimension The personnel dimension is used here to represent different kinds of personnel. This dimension can also be considered a high level hierarchy where lower levels are possible. For instance, the specific personnel in high and lower authority at the different plants may have influence on the plant performance. Hence the organization may be interested in understanding the plant performance per the personnel dimension. Some specific personnel may be Plant manager, Metallurgical superintendent, the Section metallurgist, Shift supervisor, and Plant Operator. Table 2 is a list of the dimension tables identified with the corresponding online transaction processing systems where the dimension is located. Dimension Ore Type Plant Shift Supervisor Process Manager Shift Day Month Quarter Year Table 2. Mineral Processing Dimensions OLTP Field Ore Corporate database Personnel database Personnel database Fact Table A fact table includes measures and attributes. Some of the measures for the Production fact table, ProductionFact (see figure 1) includes head grade, concentrate and tail grades for processing units such as the crushing, mill, flotation, CIL and electro winning. Figure 1 is a data mart schema of the mineral processing activities. Once the data has been transferred successfully into the data warehouse or data mart, OLAP and other business intelligence applications can be used to develop reports and analytics on the data to support decision making. Tools and applications for accessing data marts and data warehouses include: SQL queries, published reports, drillable/olap reports, Microsoft Excel, Executive information systems/dashboards/scorecards, data visualization, and data mining/predictive analytics (Watson 2009). Specifically the OLAP may provide analytics on the data requirements while the data mining may be used to develop relations that the decision makers are not even aware of.
ProductionFact Gold Electro won Tons Milled Tons Crushed BIOX Feed (Tons) CIL Feed (Tons) Head Grade Flotation (Gold) (3,2) Head Grade Flotation (Sulfur) (3,2) Tail Grade Flotation (Gold) (3,2) Tail Grade Flotation (Sulfur) (3,2) OreType (FK) ShiftNumber (FK). DimPlant PlantNumber (PK) PlantName CountryCode (FK) NChar(2) Figure 1. Data Mart for Mineral processing plants DimOreType OreTypeCode (PK) OreTypeName CountryCode (FK) DimShift ShiftNumber (PK) ShiftName DimPersonnel ShiftSupervisorID ShiftSupervisorName PlantMangerID ShiftNumber (FK) CountryCode (FK) DimCountry CountryCode (PK) CountryName NChar(2) Nvarchar(10) Nvarchar(10) NChar(2) NChar(2) Decision tree induction for instance may be used to examine variables that influence gold recovery at the different units of the plant. It can also be used to predict the tonnage of ore milled in a specific period such as shift, day, month, quarter or year. The Clustering technique can be used to cluster ore types into groups based on some similarity characteristics. Association rules can be developed on activities that occur together. For instance, poor flotation recovery may occur anytime milling is poor. Some of the predictions may be scientific in that it may be relations that are already known. However, the data mining techniques can be used to develop new knowledge from the data that may not have been known from the literature. The metallurgical laboratory test results and other industry data may serve as the source for the key indicator parameters (Larson 2008) such as desired recovery at the different units. Thus plant performance over a period of time can be compared with the key indicators. The process manger can then use the data to make decisions to improve plant performance, which can subsequently improve productivity and general corporate profitability.
DISCUSSION AND CONCLUSION We have presented a data mart that can used to develop a business intelligence application to assist process managers to make effective decisions and improve plant performance. For other gold processing companies, the data types and specific tables in the OLTP where the dimensions and attributes could be identified may be different. We identified and described the different data sources that process managers use to support decision making. A data model of relevant dimensional and fact table that can be used to provide online analytical processing (OLAP) for business intelligence was presented. We have demonstrated that transactional data can be transferred into a data warehouse or data mart environment from where business intelligence applications can be employed to perform analysis to support decision making. The kind of business intelligence project that a particular organization may develop depends on the size, needs and other factors. Running an effective business intelligence system to generate the desired goals will also be influenced by building the requisite skills among the developers as well as users of the systems. Nevertheless, mineral processing organizations cannot rely on intuition and transactional data to provide the necessary decision making needed to remain competitive especially in the information age where the process engineer is presented with tons of information from diverse systems. Future research seeks to develop and test a more comprehensive model that can be deployed in the organization. The application of the concept in other industries is a valid extension of the current work. Obviously there are additional dimensional tables that may not have been captured in the data model presented in this paper. REFERENCES Amankwah, R.K., Kahn, A.U., Pickles, C.A., and Yen, W.T. "Improved grandability and gold liberation by microwave pretreatment of a free-milling gold ore," Mineral Processing and Extractive Metallurgy (114) 2005, pp 30-36. Andoh-Baidoo, F.K., Baker, E.W., Susarapu, S.R., and Kasper, G.M. "A Review of IS Research Activities and Outputs Using Pro forma Abstracts," Information Resources Management Journal (20:4) 2007, pp 65-79. Baskerville, R., and Myers, M.D. "Information Systems as a Reference Discipline," MIS Quarterly (26:1) 2002, pp 1-14. Inmon, W.H. Building the data warehouse John Wiley, New York, 1992. Karr, C.L., and Weck, B. "Computer Modeling of Mineral Processing Equipment Using Fuzzy Mathematics," Minerals Engineering (9:2) 1996, pp 183-194. Kasper, G.M., and Andoh-Baidoo, F.K. "Advancing the Theory of DSS Design for User Calibration," in: Human-Computer eraction in Management Information Systems: Applications, D.F. Galletta and P. Zhang (eds.), M.E. Sharpe, Inc, Armonk, NY, 2006. Larson, B. Delivering Business elligence with SQL Server 2008 McGraw Hill, New York, 2008. Lindström, E.B., Gunneriusson, E., and Tuovinen, O.H. "Bacterial Oxidation of Refractory Sulfide Ores for Gold Recovery," Critical Reviews in Biotechnology (12:1-2) 1992, pp 133-155. March, S.T., and Smith, G.F. "Design and Natural Sciene Research on Information Technology," Decision Support Systems (15:4) 1995, pp 251-266. Mundy, J., Thornthwaite, W., and Kimball, R. The Microsoft Data Warehouse Toolkit: with SQL Server 2005 and the Microsoft Business elligence Wiley, Indianapolis, 2006. Shim, J.P., Warkentim, M., F., C.J., Power, D.P., Sharda, R., and Carlson, C. "Past, present, and future of decision support technology," Decision Support Systems (33) 2002, pp 111-126. The Gartner Group "Gartner EXP Worldwide Survey of More than 1,500 CIOs Shows IT Spending to Be Flat in 2009," 2009. Watson, H.H., and Wixom, B.H. "The Current State of Information Systems," Computer (40:09) 2007, pp 96-99. Watson, H.J. "Tutorial: Business elligence - Past, Present and Future," Communications of the AIS (25:Article 39) 2009, pp 487-510. Wright, G.P., Chaturvedi, A.R., Mookerjee, R.V., and Garrod, S. "egrated modeling environments in organizations: an empirical study," Information Systems Research (9:1) 1998, pp 64-84.