Resourcing

Tomorrow
Creating Value
for Society

Proceedings

HOSTED BY

WMC2023.ORG
The WMC 2023 program looks to the future of mining and resources
in a global context. Themes will be addressed in plenary and
concurrent sessions, special interest group meetings, workshops
and discussion panels. The focus is on active participation, giving
attendees opportunities to present and participate in important
discussions on the major current and future issues and challenges
facing mining and resources across the globe.

This structure will allow delegates to take “deep dives” into

the latest developments and new research in these specific
areas. These parallel streams will be interspersed with
joint-stream and plenary sessions where delegates come together to
explore nexus issues.

The Congress will explain and explore how technology is transforming

the sustainable production of minerals and fuels creating value
that continues to lift significant segments of the world’s population
out of poverty and contributes in an essential manner to an improved
way of life.

Published by the 26th World Mining Congress (WMC 2023)

© 2023

ISBN: 978-0-646-87565-1

Copyright © 2023 by the 26th World Mining Congress 2023 (WMC 2023), Inc. All rights reserved.

Copyright and Reprint Permission

Material in this publication is protected by copyright but may be used providing both the authors and publisher
are acknowledged.

Disclaimer
Material presented in this document is the responsibility of the authors. The opinions expressed do not
necessarily represent the views of The World Mining Congress.

The World Mining Congress accepts no liability (including liability in negligence) and takes no responsibility for
any loss or damage which a user or any third party may suffer or incur as a result of reliance on the document.

We acknowledge the Traditional Owners of the lands and waters throughout Australia, and pay
respect to the Elders past, present and emerging. We recognise the importance of connection to
culture, land, kinship and community to the health and wellbeing of Aboriginal & Torres Strait
Islander families. We acknowledge the cultural practices and traditions still carried out today and
being passed down to future generations.
Applying Machine Learning to Drill and Blast
Optimisation

*A.A. Labriola1
1
Digital Solutions Division, Orica Ltd, Melbourne, Victoria, Australia
(*Presenting author: [email protected])

ABSTRACT
Blast Domaining is the classification of rock for blasting purposes. Conventional blast domaining involves
manual processes and typically fails to use increasingly available blast hole measurement data such as
Measure while Drill (MWD) data. A machine learning approach has been applied to Blast Domaining that uses
multiple data sets to automate rock classification for blasting. The solution also establishes the data
infrastructure and data pipeline automation that facilitates usage in a mine production environment. Named
‘Design for Outcome’ (DfO), two modular, web-based user applications have been created. The first is a Post-
Drill classification application, that automates the extraction and processing of MWD data into blasting
domains based on Mechanical Specific Energy. This is used to refine blast loading designs to more closely
match rockmass variability. The other is a Pre-Drill classification application. This applies machine learning to
MWD and block model data to classify into blasting domains ahead of drilling, and therefore inform drill
pattern selection. Drill and blast users benefit from the integration of these applications with a digitalized
blast design and on-bench management system. This completes the workflow through to tailored loading on
an inter and intra hole basis. The complete workflow has been demonstrated through to smart Mobile
Manufacturing Units (MMUs) that automate the preparation and loading profile of bulk explosives. The Roy
Hill iron ore operation in Western Australia has been a foundational partner with Orica in implementing the
Post-Drill end-to-end solution. At time of writing, the operation is approaching its 500th production blast using
this system. Significant geological variability led Roy Hill to historically implement variable loading across
blasts. By using a data driven and automated approach, the solution has enabled Roy Hill to streamline the
process and distribute explosives energy at a higher resolution. In doing so, the solution has been able to be
applied to reduce excessive energy usage while maintaining excavator productivity. This has reduced drill
and blast costs by more than A$7.5m in just under two years and yielded a reduction of 1,867 tonnes Scope
1 and 8,618 tonnes Scope 3 CO2 equivalent emissions respectively.

KEYWORDS
Drill and Blast, Machine Learning, Automation, Blasting Domains, Measure While Drill, Design for Outcome,
Roy Hill, Orica.

392
26th World Mining Congress (WMC 2023) 26 – 29 JUNE BRISBANE AUSTRALIA ISBN: 978-0-646-87565-1
1. INTRODUCTION
Despite an increasing proliferation of data in mining, the blast design process has remained largely
unchanged. The engineer’s interpretation of rock classifications for blasting purposes, herein referred to as
Blasting Domains, is typically determined by judgement and use of selected block model parameters.

This raises questions, including:

• Are the best parameters being selected from the geological block model to define Blasting Domains?
• Is the method of determining Blasting Domains consistent?
• Is the rich data set from sensor-equipped drills or blast hole logging tools used in a workflow to refine
blast designs?

The limitations of traditional design methods and workflows mean that most mines cannot answer yes to all
these questions. This is especially the case when there is significant rock mass variability and where Measure
While Drilling (MWD) data exists. This paper describes a machine learning-based approach for automatically
defining Blasting Domains, and a software interface created by Orica, supported by enabling data
infrastructure and pipeline automation.

2. BLAST DOMAINING – COMMON INDUSTRY PRACTICE

Conventionally, Blast Domaining is a manual and subjective process based on a few observable geological
parameters. Hence it is usually done at a very coarse scale, and usually doesn't use high resolution data even
when it is available. Therefore, the more heterogeneous the orebody is, the less likely it is that Blast Domains
reflect the variability within a blast. As drill and blast designs are based on Blast Domains, this impacts the
consistency of blast outcomes.

Production blast hole MWD data produces a rich data set that can be used to refine Blast Domains. However,
challenges with the data itself and with the ability to make use of it in time for blast loading means that MWD
is typically not used for this purpose.

3. BENEFITS OF AUTOMATED BLAST DOMAINING

Better allocation of Blast Domains means better allocation of blast energy, which is important for maximising
value from the orebody.

Automation of Blast Domain decisions should improve consistency and accuracy by removing subjectivity and
making better use of large, rich data sets.

Economic benefits are realized through time savings through automation and either reduction of drill and
blast costs or an increase in blasting energy where appropriate to improve downstream productivity. ESG
benefits are also realized through reducing carbon emissions generated through wasted energy.

4. DESIGN FOR OUTCOME – A MACHINE LEARNING APPROACH TO BLAST

DOMAINING
Design for Outcome (DfO) describes a series of applications built by Orica Digital Solutions that:

393
26th World Mining Congress (WMC 2023) 26 – 29 JUNE BRISBANE AUSTRALIA ISBN: 978-0-646-87565-1
• use machine learning to domain rock for blasting purposes;
• automate data ingestion and processing for use in a production mining workflow;
• interface with a digital blasting system to efficiently implement tailored blast designs

Two applications are currently available. The first is the Post-Drill classification application. This uses high
resolution blasthole MWD data to refine Blasting Domains and adjust blast charging designs per hole.

The second is the Pre-Drill classification application. This combines block model data and historical MWD
data to create Blasting Domains. These are used to generate drill patterns and initial loading designs.

4.1 Post-Drill Classification

The Post-Drill classification application uses high spatial density blast hole measurement data to classify the
rock. The most commonly available data of this type is MWD data.

4.1.1 Measure While Drill data

MWD data provides high spatial density measurement. In the X, Y plane, measurements are captured at blast
hole burden and spacing resolution. Z elevation measurements are at 0.3m intervals or less.

Teale (1965) published a Mechanical Specific Energy (MSE) calculation that identified a broad correlation
with the crushing strength of rock as follows:

Specific Energy (Teale, 1965)

𝑘𝑘𝑘𝑘
𝐸𝐸! # '𝑚𝑚" ( = 𝐹𝐹⁄𝐴𝐴 + 2𝜋𝜋𝜋𝜋 𝑇𝑇⁄𝐴𝐴 𝑉𝑉

F = thrust on bit (kN)

A = area removed by drill bit (m2)
N = rotation speed (rps)
T = rotation torque (kN . m)
V = drilling speed (m/s)

Numerous other publications have investigated the use of MWD data to determine blastability, including and
not limited to Leighton (1982), Liu & Yin (2001), and Mozaffari (2007). As noted by Khorzoughi (2013) and
Khorzoughi et al (2018), MSE does not fully define rock blastability as it does not consider structural/fracture
properties. Nevertheless, it is a proxy for rock strength, which is a key blastability parameter.

Engineers might make limited or no use of MWD data because they:

• Have low confidence in data quality

• Cannot, or do not have time to, correlate drilling measurements to blasting designs
• Find it difficult to transfer hole-by-hole blast designs to blast crews, or have blast crews willing and able
to implement them

The quality of MWD data fundamentally determines whether it can be used for Blast Domaining. The blast
domaining solution cannot resolve poor sensor data however it does provide a valuable use case for
maintaining MWD sensors.

394
26th World Mining Congress (WMC 2023) 26 – 29 JUNE BRISBANE AUSTRALIA ISBN: 978-0-646-87565-1
4.1.2 Applying Machine Learning to determine Blasting Domains

An example of MWD data plotted along the hole depth is shown in

Figure 1.

Figure 1 - An example of MWD sensor data output along the depth of a hole

The variability with depth presents the question of how to resolve a single domain, which may be required
due to limitations in explosives selection and delivery. Is it the average value along the entire depth, the
maximum value, or the 75th percentile value? These options result in different classifications that may not be
optimally aligned with the user's blasting objectives.

The solution applies machine learning to the problem. This involves data spanning all drills and rock
conditions. Suitable data is cleansed and processed using an unsupervised machine learning model.

Normalisation is required before unsupervised machine learning because the data from two identically
equipped production drills drilling the same rock can show variation. Figure 2 demonstrates this.

395
26th World Mining Congress (WMC 2023) 26 – 29 JUNE BRISBANE AUSTRALIA ISBN: 978-0-646-87565-1
Figure 2 - Classification based on non-normalised (middle image) versus normalised (image on right)

The plan on the left shows colour variation by drill unit. The middle plan shows the results of relative hardness
clustering based on MSE without normalisation. The clustering is influenced by the drill type. The plan on the
right shows the clustering results using normalised data. Figure 3 shows the resulting clusters based on
feature similarity.

Figure 3 - Identification of five clusters representing relative hardness domains

A supervised machine learning model is then applied to new MWD data to determine the relative hardness
domain per blast hole using the Post-Drill classification application.

4.1.3 Practical limitations related to blast hole charging

Generation of Blasting Domains needs to be considered based on achievable loading configurations. Mines
are generally not able to vary explosives energy at 0.3m increments. For mines using conventional triple

396
26th World Mining Congress (WMC 2023) 26 – 29 JUNE BRISBANE AUSTRALIA ISBN: 978-0-646-87565-1
product explosives delivery units (ie: ANFO, Heavy ANFO, and water-resistant emulsion blends), some
flexibility may exist in altering product energies via multiple product decks in some but not all products.

Furthermore, the extent of variability in the rock mass should be considered. A blast bench level can extend
laterally for several hundred metres yet only 10 – 15m in depth. Therefore, at many mines a single domain
per hole may be adequate, whereas others may require multiple domains per hole.

4.1.4 Data pipeline automation

Equally important as Blast Domaining itself is the automation of the data flow, processing and presentation
to the user. The Post-Drill classification application automates the data pipeline from the mine drill systems
through to availability of results on demand via a web-accessed User Interface.

4.1.5 The Post-Drill Classification User Interface

The Post-Drill Classification User Interface (UI) provides a simple and intuitive means for Drill and Blast
Engineers to apply refined Blast Domains to blast designs. The UI is accessible as a module within the Orica
BlastIQ™ cloud-based platform, as shown in Figure 4.

Figure 4 - The Post Drill classification user interface

The user selects a blast they wish to update (1). This uploads the blast showing domains by hole (2), based
on colour coding of relative hardness domains (3). Details by blast and hole are displayed (4) & (5). The user
downloads the processed information (6) to use in the blast design application.

4.1.6 Loading rules and the automation of hole-by-hole loading design

The Drill and Blast Engineer imports the Blast Domain data into the blast design application, where loading
rules are applied to automate loading designs. Figure 5 shows an Orica SHOTPlus™ blast design loading rule
template that uses Post-Drill classification data.

397
26th World Mining Congress (WMC 2023) 26 – 29 JUNE BRISBANE AUSTRALIA ISBN: 978-0-646-87565-1
Figure 5 - Example of a loading rules template detailing loading configuration for multiple scenarios

4.1.7 Field implementation of tailored blast loading designs

While tailored blast designs improve energy matching to the rock, they also increase complexity for blasting
crews.

This is addressed by implementing tailored designs with a digital blast management system. BlastIQ™ Blast
Control system integrates the blast design application with field devices, including BlastIQ™ Mobile tablets
and smart Mobile Manufacturing Units (MMU™s) equipped with the LOADPlus™ control system, as shown in
Figure 6. All units synchronise blasthole depth and condition and update the loading design automatically,
based on the loading rules.

Figure 6 - A smart MMU™ with control system fully integrated to a digital blasting management system,
connecting blast domaining to tailored blast loading

398
26th World Mining Congress (WMC 2023) 26 – 29 JUNE BRISBANE AUSTRALIA ISBN: 978-0-646-87565-1
4.2 Pre-Drill Classification

The Pre-Drill classification application applies machine learning to MWD and mine block model data. The
purpose is to generate refined Blasting Domains ahead of drilling that better informs drill pattern selection.

For mining operations with limited explosives energy options, the Pre-Drill Classification module assists to
better inform drilling, which becomes the main mechanism to alter energy in a blast.

5. DESIGN FOR OUTCOME BLAST DOMAINING – AN END-TO-END

WORKFLOW EXAMPLE
The following example describes the workflow from drill pattern design to final loading using the blast
domaining solution. Refer to Figure 7.

• The bench polygon is imported into the Pre-Drill classification application (1) and selected to reveal the
Pre-Drill Blast Domains (2).
• In this example, two drill patterns in one blast are designed in the blast design application using Pre-
Drill application output. Yellow holes denote tighter pattern (3).
• The Blast ID is selected in the Post-Drill classification module after drilling (4); showing post-drill
domaining by hole (foreground)
• Loading rules update the blast design (5); this example shows some differences to Pre-Drill
classification.

Figure 7 - The Pre and Post Drill domaining user interfaces (top left and right, respectively), the drill pattern
designs (bottom left), and post-drill blast domains (bottom right)

In (5) there are a selection of holes at the boundary between patterns that were expected to have medium
hardness and were subsequently classified as hard, and holes that were expected to be hard that were
classified as medium-soft.

399
26th World Mining Congress (WMC 2023) 26 – 29 JUNE BRISBANE AUSTRALIA ISBN: 978-0-646-87565-1
Figure 8 - Automatically generated loading profiles, adjusted based on Post-Drill classification

Each example in Figure 8 shows how the loading rules automatically apply an appropriate increase (hole D13
vs C12) or decrease (hole AC2 vs E9) in explosives energy. For hole D13, higher energy is applied in the wider
spaced pattern. For hole AC2, lower energy is applied in the more closely spaced pattern.

6. A CASE STUDY – IRON ORE MINE, WESTERN AUSTRALIA

Roy Hill is a world-class surface iron ore mine situated in Western Australia’s Pilbara region. The Roy Hill
operation spans an integrated iron ore mine, processing plant, rail and port. Roy Hill mines 60 Mt of iron ore
and 250Mt of waste per annum. The company has a focus on sustained, low-cost production of consistent
quality and grade and has established a Smart Mine program to create ‘Tomorrow’s Mine, Today’ by working
across the business to identify, develop, and deliver new solutions in four focus areas:

• Smart People
• Smart Assets
• Smart Orebody
• Smart Integrated & Intelligent Systems

6.1 The need for improved Blast Domaining

The geology of the Roy Hill deposit consists of the following major units - Roy Hill Shale Member which is
overlain by unmineralised Nammuldi BIF, altered Nammuldi shale, mineralised Nammuldi (iron ore), hard-
capped by partially mineralised Canga, and overlain by detritals and alluvials.

Within these units are horizontal bedded geological domains (categorised by grade and physical attributes)
that stretch across the tenement, represented in Figure 9.

400
26th World Mining Congress (WMC 2023) 26 – 29 JUNE BRISBANE AUSTRALIA ISBN: 978-0-646-87565-1
Figure 9 - Roy Hill bedded geological units and domains

Not all geological domains are present in every pit, as each domain pinches and swells in thickness and each
pit region can have a differing distribution of the domains. The bedded geological domains also often present
as steeply dipping. Therefore, the presentation of geology can vary significantly across a blast.

Prior to the application of the blast domaining solution, Roy Hill did apply varying blast designs to reflect
changes in the rockmass. However, the process was time-consuming, complex, and did not utilize multi-deck
designs. The Blast Domaining solution streamlined the use of MWD data and produced multi-deck designs as
necessary. The data driven and automated domaining approach provided consistency, which increased
confidence in undertaking blast optimisation projects.

A Design for Excavation project was jointly implemented by Roy Hill and Orica. The objective was to maintain
excavator productivity while reducing the overall blasting energy and cost by accurately domaining ore and
waste for blasting. Substantial cost savings and productivity gains were anticipated.

6.2 The application of Post-Drill Blast Domain Classification

The Design for Excavation project at Roy Hill involved:

• Implementing digital blast management with BlastIQ™ Blast Control and five LOADPlus™-enabled
Bulkmaster™ 7 MMU™s.
• Building a Post-Drill blast domaining model and establishing an automated data pipeline from the
autonomous production blast hole drills.
• Providing a user interface for Post Drill Blast Domaining.
• Integrating and transforming Fleet Management System data to automatically calculate instantaneous
dig rate and compare it to the pre-project baseline.

401
26th World Mining Congress (WMC 2023) 26 – 29 JUNE BRISBANE AUSTRALIA ISBN: 978-0-646-87565-1
• Implementing a project governance system and energy reduction via staged drill pattern expansion.
• Establishing project reporting and success evaluation criteria for each blast.

The end-to-end workflow integration from autonomous drills to smart MMU™s loading tailored hole-by-hole
designs was a critical enabler of the solution. The MMU™s on site can load multiple energy decks per hole.
An example of an implemented blast design is shown in Figure 10. This design shows lateral and vertical
variability across two loading horizons, bottom (left in image) and top (right in image).

Figure 10 - Plan view of two horizons within a blast, showing variation in explosives energy applied between
top and bottom

To monitor blast-related digging performance a Dig Performance Monitoring dashboard was established in
BlastIQ™. This integrates with the mine FMS system and automates calculation and representation of the
blast-related digging performance. Error! Reference source not found. shows digging performance for the
same blast as in Figure 10. This automated blast analysis helps sustain management of the Design for
Excavation project at Roy Hill.

Figure 11 - The Dig Performance Monitoring dashboard that provides automated blast-related digging KPI

402
26th World Mining Congress (WMC 2023) 26 – 29 JUNE BRISBANE AUSTRALIA ISBN: 978-0-646-87565-1
analysis. For the blast shown, the average instantaneous dig rate was 9,006tph versus a baseline of 7,380tph
6.3 Benefits realised

The Post-Drill classification solution was implemented at Roy Hill, emerging from trials into full scale blasting
between November 2020 and February 2021. As of September 2022, project summary details and benefits
are:

• 482 blasts fired

• 394 blasts validated (the difference relates to fired blasts not yet fully excavated)
• $7.6m validated savings (D&B cost reductions while maintaining instantaneous digging productivity
targets)
• 1,867 tonnes Scope 1 CO2 equivalent emissions reduction
• 8,618 tonnes Scope 3 CO2 equivalent emissions reduction
• Significant time saving and consistency in blast domaining methodology

5. CONCLUSION
The Design for Outcome Post-Drill classification technology continues to be in active use for production blasts
at Roy Hill. Since its introduction the solution has been refined, with improvements in the algorithm using
the growing volume of data and the data pipeline automation and user interface resulting in an improved
user experience.

Further initiatives related to drill and blast optimisation continue to be progressed at Roy Hill. Block model
data is being analysed for potential application in Pre-Drill domain classification. Automated blast
fragmentation measurement options are also being investigated. The introduction of in-pit fragmentation
measurement would provide further opportunity to investigate refining blast domaining for fragmentation
outcomes.

ACKNOWLEDGEMENTS
The author wishes to express appreciation to the multi-disciplinary team from Roy Hill and Orica involved in
the implementation of the solution at Roy Hill. Also to Stuart Hawke, Mateus Rocha and Benn Hamer for
taking Design for Outcome from concept to operational product. For assistance with this paper, the author
wishes to acknowledge and thank Juan Fonseca Bracho of Roy Hill, and Eden Paki, Luke Armistead, Joey
Girsang and Martin Adam of Orica for their contribution and review.

REFERENCES
Teale, R. (1965). The Concept of Specific Energy in Rock Drilling. (P. Press, Ed.) International Journal of Rock
Mechanics and Mining Science, Volume 2, pp. 57-73.

Leighton, J. (1982). Development of a Correlation between Rotary Drill Performance and

Controlled Blasting Powder Factors. MSc. Thesis, Department of Mining and Mineral Process
Engineering, University of British Columbia, Canada

Liu, H. and Yin, K. (2001). Using information extracted from drill data to improve
blasting design and fragmentation. Fragblast Volume 5, Issue 3: pp.157 – 179.

403
26th World Mining Congress (WMC 2023) 26 – 29 JUNE BRISBANE AUSTRALIA ISBN: 978-0-646-87565-1
Mozaffari, S. (2007). Measurement While Drilling System in Aitik Mine. MSc.
Thesis, School of Applied Geosciences and Mining, Luleå University of Technology, Sweden

Khorzoughi, M. (2013). Use of Measurement While Drilling Techniques for Improved Rock Mass
Characterization in Open-Pit Mines. MSc. Thesis, The Faculty of Graduate Studies (Mining
Engineering), University of British Colombia, Canada.

Khorzoughi, M., Hall, R. and Apel, D. (2018). Rock fracture density characterization using measurement while
drilling (MWD) techniques. International Journal of Mining Science and Technology, Volume 28, Issue
6, pp. 859 – 864.

404
26th World Mining Congress (WMC 2023) 26 – 29 JUNE BRISBANE AUSTRALIA ISBN: 978-0-646-87565-1