Content uploaded by Osvaldo Bascur
Author content
All content in this area was uploaded by Osvaldo Bascur on Jun 18, 2019
Content may be subject to copyright.
PROCESS ANALYTICS: TRANSFORMING MINERAL PROCESS PLANT
DATA INTO ACTIONABLE INSIGHTS
J. Steyn (1), O.A. Bascur (2)* and B. Gorain (3)
Flank Engineering Inc.
2234 Colbeck St, Oakville, ON, Canada
OSIsoft, LLC.
14701 St. Mary’s Lane, Suite600, Houston TX, 77079, USA
Barrick Gold Corporation
161 Bay Street, Suite 3700, ON M5J 2S1, Canada
(*Corresponding author: osvaldo@osisoft.com)
ABSTRACT
The advent of digital revolution has now enabled us with numerous tools that could be leveraged to
transform our operational data into actionable insights.
Key opportunities with digitization include better visualization, transparency, integrated planning and
execution for value-chain optimization, that results in smarter production, intelligent response to
changes in ore, process and equipment conditions, reduce energy and waste along with prevention of
asset breakdown, safety and environmental issues.
It is important to realize that these digital tools have limited value, from a metallurgical operational
context, if we cannot bring-in the appropriate domain expertise along with getting the basics right. The
focus within Barrick is to ensure that there is adequate depth of different disciplines built into our
platforms along with breadth to integrate other disciplines such as geology and mining for an effective
Mine-to-Mill integration. A gold recovery improvement strategy based on optimal Gaudin size
distribution, leaching performance monitoring and guidance via feed grade maps is discussed.
Simultaneous identification of process and equipment constraints enable finding the best overall
conditions for Gold recovery.
This paper discusses the methodology, findings, and challenges in the ongoing journey of
implementing a Metallurgy Analytics platform that evolves from being retrospectively descriptive to
anticipatively prescriptive.
KEYWORDS
Mine to Mill, Digital Plant Template, Business Intelligence, Machine Learning, Predictive Analytics,
Grinding and Leaching Optimization, Operational Intelligence,4 Steps to Maximizing Metal Recovery
INTRODUCTION
Guidelines to authors for paper preparation - IMPC 2018, Moscow
The last few decades have seen some remarkable advancement in information technology, but
recently various digital technologies have resulted in several synergies to deliver innovation in diverse
applications. The mining industry has been relatively slow in adopting these technologies compared to
the other industries, but this scenario is changing fast. Mining companies globally are going through a
series of challenges such as declining ore grades with complex metallurgy, increasing depth of ore
bodies, high capital and operating costs, lower productivity, shortage of technical resources, significant
pressure from stakeholders to deliver, along with growing environmental and corporate social
responsibility issues. Improvement in profitability would need significant improvements in operational
performance.
Consequently, over the last few years, there has been a major focus amongst some mining
companies to imbibe Operational Excellence. This has been only possible because of the replacement of
a silos based approach by an integrated view of the organization. Such an integrated view is now possible
via the evolving digital technologies. Even so, such a transformation calls for a change in culture, which,
for most organization is not trivial. Attempts in the past to integrate functions such as through Mine-to-
Mill initiatives in the mining industry have demonstrated significant benefits to many operations
(McKee, 2013). Still, sustaining these benefits has clearly been a challenge due to labor turn over and
lack of proper systems and structures that integrate information between business functions within an
operation and between operations and the corporate.
Opportunities
The advent of digital technology has opened-up opportunities to integrate inter-departmental
information for a unified understanding of, and across, an entire organization. The advances in
instrumentation, equipment control systems, supervisory control systems (SCADA/Historian) and
industrial networks have helped in the direction of enterprise wide integration. The availability of real
time information across an enterprise has given rise to numerous possibilities to improve organization
performance. As suggested by Seshan & Gorain (2016), current state-of-the-art tools allow production
performance to be linked to planning, design and the supply chain. Integrated software tools have
allowed an organization to share not just raw process data but intelligence that is abstracted from such
data to drive well-informed decisions (called “actionable intelligence’). Therefore, changes in
performance happening in any part of the production/mining value chain (ecosystem) can have an
instantaneous impact on the other parts – making the interactions between the various stakeholders of
the ecosystem dynamic and real time. This provides a better capability to all stakeholders in the
ecosystem to predict outcomes that are relevant to their respective scope. It is now possible for the mining
and metallurgical companies to leverage recent advancements in information and telecommunications
technologies as part of their operational excellence drive (Bascur, 2016, 2019).
JOURNEY WITHIN BARRICK: TOWARDS AND INTEGRATED OPERATIONS
Guidelines to authors for paper preparation - IMPC 2018, Moscow
Optimization Goals
Integrating the various functional elements of an operation from strategic, tactical and
operational points of views, allows for a simultaneous focus on both production and cost control. This
initiative provides the industry with an ability to deal with frequently changing business dynamics.
In the mining industry, the target for tangible improvements through this initiative could be 5-
10% increase in productivity and a 15-20% reduction in operating costs as a first step. These targets are
realistic and are based on improvements that have been demonstrated by some mining companies
pursuing even a simpler strategy such as Mine-to-Mill. It is important to realize that these digital tools
have limited value, if we cannot bring-in the appropriate domain expertise along with getting the basics
rights. The focus within Barrick is to ensure that there is an adequate depth of different disciplines built
into our platforms along with breath to integrate other disciplines such as geology and mining for an
effective Mine-to-Mill integration.
Plant Performance Management Strategies
Operational efficiencies, i.e. improved equipment availability and utilization, increased
tonnages and reduced dilution, are a key and necessary focus for increasing production and lowering
operating costs. In addition to continuous improvement and innovation, understanding where the
opportunities exist is the key to increased profits. This is where measuring, managing and maximizing
metallurgical performance is enables by using an industrial data infrastructure.
Figure 1 shows the type of operating events in a plant. It is shown that every area of the plant has
operational events running on target based on the daily schedule, in troubles, idle, down and
maintenance. By maximizing the running on target each area of the plant, we can find the opportunities
for improvements in the whole value chain. The main objective is to maximize the gold extraction. As
such, finding the right particle size distribution (PSD) based on the ore type is the key. This paper will
show a strategy to find the operating modes and to track the production and operating costs based on
each unit operating conditions. A digital plant template is used to identify these minor losses.
Figure 1 - Operational mode tracking and particle size management for maximizing gold recovery
Guidelines to authors for paper preparation - IMPC 2018, Moscow
In addition, we will show how the particle size analyzer data was augmented by finding the
particle size distribution (PSD) shape estimation that as major effect in the grinding viscosity and in the
gold extraction by leaching. A Gaudin Modulus estimate is used to improve the gold recovery empirical
model.
A digital plant template was used to model the plant data in a simplify way to build the process
model. Real time streaming data is transformed into information by using the online analytics generating
the operational events to aggregate the production and consumable data into improvement workflows
generated by the automation of our current operational knowledge. The classification of the data allows
the aggregation of data at the desired level of detail for determining where the improvement opportunities
are. As such, collaboration between production, finance and planning, maintenance and all the safety
and environmental support become active and not passive as in the past.
Figure 2 - Real time operational intelligence strategy
Figure 2 shows a conceptual diagram of the process of detecting abnormal operating modes for
continuous improvement and innovations. The schematic shows a process workflow that inputs the
targets from the daily plant schedule and the process inputs into a unit template analytics, which classifies
the operating modes of all the process units in an industrial plant. The variance is the difference between
expected and actual results. The expected results are generally specified in the operational budget or the
current production schedule (Bascur and Kennedy, 1999, Bascur, 2016). In essence, an innovative
strategy automates the Theory of Constraints (Goldratt, 2014) in our digital era.
The use of the variance is a management tool for implementing an Operational Intelligence
strategy. As such, we can use this classification of operating time modes to aggregate the information
for all the process units. The schematic presents two questions:
• Are we on target? This step is automated to track when the unit is not on target and an
event frame is set to aggregate all production and consumable variables
• Are we satisfied? Here generated operational insight are evaluated for improvements.
Guidelines to authors for paper preparation - IMPC 2018, Moscow
The first question can be tracked in real time, defining the interval of time that the process units
are running on target, in trouble, idle, down or maintenance. These operational events are then used by
the continuous improvement team to aggregate the data to look for opportunities. The large amount of
data calls for the use of modern tools such as Microsoft Power BI to visualize the information and to
assess the operational losses and gains (Bascur, 2016, 2019).
MATERIALS AND METHODS
PI Asset Framework (AF) is a repository for asset centric models, hierarchy, objects and
equipment in processing plant. It integrates, contextualizes, refines, references, and further analyzes
data form multiple sources. Figure 3 shows the plant block flow diagram. A process unit template was
configured with standard analytics and calculations to assess and analyze metrics of common interest,
e.g. down time. A top down approach was followed whereby the Stockpile and crusher, SAG and Ball
Mill, Grinding Thickener, CIL, CIC and Tailing Sections were modeled as individual units.
Figure 3 - Mineral Gold Processing Block Diagram for digitization
The total feed, water consumption, electricity consumption and reagent consumption are
aggregated for each section based on the real time operating modes detected automatically by the system
as shown in Figure 5. Each section of the plant have designated operating modes (Running OK = on
Target, Trouble = off target by 10%, Idle, Down = unscheduled down time and maintenance = scheduled
down time). Because the base template is not equipment or process specific, the real value of this
approach come with its scalability if applied across the organization to render a high-level overview of
plant throughputs, utilities consumption, and reagent consumption.
Guidelines to authors for paper preparation - IMPC 2018, Moscow
Figure 4 - Fishbone diagram showing the relationship between the yield maximization and the possible
causes based on process variables, operational, people and ore quality events.
Figure 4 shows how the process data can be aggregate using PI Event Frames to assess the KPIS
based on major operational, people or quality events to extract the operational insights using
visualization tools such as PI Vision or Microsoft Power BI.
RESULTS AND DISCUSSION
A major objective of the Process Analytics initiative is the development of an integrated
operations process model to be used for the day-to-day coordination and optimization of operations, and
as a planning tool linked to the geology resource model. The integrated model will consist of a series of
connected process models, derived empirically from operations data or using established process
simulation tools such as JKSimMet or the Integrated Extraction Simulator (IES) developed through
AMIRA P9Q. A Digital Plant Template is used to transform raw data into operational insights. These
insights are then used for process improvements. Then, the operational data is used to develop predictive
models.
Figure 5 - Microsoft Power BI showing the event frame data converted into operational insights
Guidelines to authors for paper preparation - IMPC 2018, Moscow
One of the main constraints in developing production models is the quality of the data. As such,
the strategy to classify the operating modes has proven indispensable for analysis, process model
development and process optimization.
Figure 5 shows the event-framed data by PI Analytics using PowerBI visualization tools. The
left hand side for the time selection, the asset selection for the whole plant and the people shift selection
to present the operating mode duration for unit. The right hand side of the figures shows the percent
duration in each state for the selected period of time, process areas and shift. Many other displays
showing the Total Production Rate for each unit, Total Energy, Water, Cyanide consumption are built
using the same tool. As such, a great visualization of the key process variables is presented for the whole
plant in term of the operating conditions for a selected time interval.
Figure 6 - PI Vision real-time performance monitoring using augmented information
Figure 6 shows in PI Vision the real time calculated variables from the raw data that is validated
and classified by operational modes using the Digital Plant Template. This displays shows augmented
information by preprocessing the data with the master algorithm of the standard template. Operations
transparency was enhanced by producing high level, cloud-based process diagrams, available to the
entire Barrick network, highlighting major process flows and section specific process KPIs in real-time.
This is due to ability to share these displays in your cellular phones or smart devices over the cloud. This
transparency between different sites and between sites and corporate management, allows for
identification of issues and more efficient allocation of resources to problem areas.
Analytics and Process Models
While defining the best grinding size for the optimal gold recovery we have found that the
particle size analyzer has availability problems. This availability of the sensors is due to very harsh
conditions. To have a more reliable particle size value over time a soft sensor was built using machine
learning tools. A benefit of configuring operating mode event frames (Running, Down, etc.) is that the
quality of the data is significantly enhanced by filtering operating data for “Running” mode periods. The
Guidelines to authors for paper preparation - IMPC 2018, Moscow
cleansed data can be exposed to advanced analytics tools, e.g. Python, R and MS Azure Machine
Learning Studio, through PI Web API and other connectors, to build customized, operation-specific
advanced regression and classification models for use as soft sensors, predictors, and process advisory
tools.
Figure 7 Particle size measurement and prediction using a soft sensor
Figure 7 shows the predicted particle size and the measured one, particularly including a period
during which the actual equipment failed, and the soft sensor continued to deliver a reliable estimate of
the actual value. PI Analytics is used to incorporate the predictive model coefficients and the process
variables to estimate the particle size. The user built a template to simplify the configuration and the
system is capable of reading the parameters automatically from the machine-learning tool.
Advanced process analytics – Particle Size Distribution shape on line analysis
It is known that the gold leaching is affected by the particle size distribution. Since the size
analyzer provide several sizes, a PSD shape estimator was build. A least square model was implemented
to estimate the Gaudin Size modulus to have an indication of the variations of the Sag Feed Size
Distribution and the Cyclone Overflow size distribution. PI Analytics can also be used to perform online
linear regression of models as a softsensors to see changes in the slope of a linear model. It is good
practice to model the size distribution to a Gaudin Schumann model, as such; we can estimate the slope
of the distribution (a) and the maximum size (k). The “a” parameter is measure of the amount of fines
and related to the hardness of the ore.
Guidelines to authors for paper preparation - IMPC 2018, Moscow
Figure 8 - Gaudin Schumann particle size distribution used to find the shape.
Figure 9 - Gaudin Schuman PSD model parameters for SAG and Leaching Feed streams
Figure 9 shows the continuous estimate of the Gaudin Modulus of the SAG Feed and
hydrocyclone overflow products, which could be related to ore hardness. These PSD Estimates will be
used in the predictive analytics models.
Process Models – Gold recovery
An important outcome of a data driven integrated process model is the understanding of and subsequent
predicting of gold recovery from various ore feeds. Furthermore, to evolve to a prescriptive control
strategy that maximizes gold recovery, or more precisely minimizes tails gold grade (losses), the effects
of various contributing variables should be understood and qualified, and where possible controlled
proactively, e.g. the adjustment of cyanide concentration across the circuit in preparation for changes in
expected cyanide consumption with different ore types.
Guidelines to authors for paper preparation - IMPC 2018, Moscow
Figure 10 - Particle size distribution shape gold recovery model analysis
While there are many known variables that effect gold recovery, e.g. grind size (liberation), cyanide
concentration, residence time the prediction results are often distorted by the unknown or unmeasured
variables that effect recovery. Figure 12 shows a sample of the results showing the effect of the PSD
shape. This analysis was done using data set extracted using PI Datalink and imported in MS Studio
ML. Gold Recovery preliminary estimates can be obtained from the following operating variables:
Au Recovery = f (Mill Feed, D80, Gaudin Size Shape, % Feed Assay, Cyanide Concentration, etc.).
Figure 11 - Tails grade prediction distortions caused by unmeasured variables not included in model
parameters.
CONCLUSIONS
A digital plant template has proven to facilitate the configuration of a consistent plant model.
Each process area use the same nomenclature and analytics to classify the data into operational modes
for overall process effectiveness. The production and consumables variables are simply aggregated by
PI Event Frames. The operational insights can then be analyzed for all KPis by unit, time interval, shift,
operating conditions and feed quality.
Guidelines to authors for paper preparation - IMPC 2018, Moscow
Having an online estimation of operating conditions enabled the calculation of simple mass
balances to augment the process data into business indicators such as yield, particle size softsensors,
particle size distribution shape estimates, energy and water estimates. These aggregated variables are
then reused in predictive analytics empirical models. The running mode operating mode event framed
sub data is valuable to develop particle size and gold recovery simple model estimators from operating
conditions.
The aggregation of data using the trouble-operating mode enables to have an estimate of the
production and utility losses. Usually, these minor losses are difficult to estimate. This novel strategy
using PI Analytics is being called follow the money strategy.
There is a lot more work that needs to be done to develop improved operational models. The
strategy provided here is a first step in our journey to take more proactive ways of optimizing gold
recovery improvements. The recent advances in information technology have allowed the mining
industry leap forward in digital transformation that has opened-up opportunities to integrate inter-
departmental information for a unified understanding of an entire organization. Integrating the various
functional elements of an operation from strategic, tactical and operational points of views, allows for a
simultaneous focus on both production, cost control, and most importantly in equipment condition
improvements.
Following an integrated operational model initiative, Metallurgy Analytics aims to build an
integrated process model, using established process simulation and modern advanced analytical tools.
To achieve this, an open and common platform for data flow between different operational functions
need to be established to share dashboards and encourage cross-functional cooperation to achieve a
common enterprise goal.
ACKNOWLEDGEMENTS
We thanks both Barrick Gold and OSIsoft for their support and the opportunity to share this
strategy to transform data into operational insights.
REFERENCES
Bascur, O.A. 1990a. Profit based grinding controls. Mineral Metallurgy Processing. 7:9–10.
Bascur, O.A. 1991b. Integrated grinding/flotation controls and management, in: Dobby, G.S.,
Argyropoulos, and Rao, S.R., Mineral Processing and Process Control, Proceedings of the Copper
91/Cobre 91 International Symposium – Volume II, August 18-21, Ottawa, Ontario, Canada. pp.411-
427
Bascur, O.A., and Kennedy, J.P. 1999. Real time business and process analysis to increase
productivity in the process industries. 1999 ISA Conference Proceedings, Houston, TX.
Guidelines to authors for paper preparation - IMPC 2018, Moscow
Bascur, O.A. 2016. A journey toward a digital transformation in the process industries.
https://pisquare.osisoft.com/docs/DOC-2935-journeydigitaltransformation-draft-c1-4-binder1pdf.
Bascur, O.A., 2019, Process Control and Operational Intelligence in Mineral and Metallurgical
Processing, Chapter 10.3., SME Handbook, to be published, 2018.
BHP-Billiton, 2013 http://www.stepchangeglobal.com/wp-content/uploads/2013/09/IROC-
Brochure-201305031.pdf.
“Enterprise Gateway, 2010 – A MOM solution for Integrating Production and Maintenance in
Real Time”, 5G Automatica Internal Report 240310.
Goldratt, E.M., 2014, The Goal – A process of OnGoing Improvement – Theory of Constraints,
The North River Press Publishing Company, Great Barrington, MA.
Kelleher, J.D., Namee, B.M., and D’Arcy, A., 2015. Fundamentals of Machine Learning For
Predictive Analytics, The MIT Press, Cambridge, Massachusetts.
McKee, D., 2013. Understanding Mine to Mill, A monograph published by the Co-operative
Research Centre for Optimizing Resource Extraction (CRC ORE), Brisbane, Australia, 2013.
Seshan, A. and Gorain, B.K., 2016. Innovative Process Development in Metallurgical Industry,
V.I. Lakshmanan et al. (eds.), Springer International Publishing Switzerland 2016 393
