You are on page 1of 3

Machine Learning for the Mining

Industry
David Willingham, MathWorks Australia

Copyright 2014 The MathWorks, Inc.

This seminar has 2 examples for Machine Learning in the Mining Industry:

Example 1 - Steel Plate Manufacturing Defects

Fault detection (classification

Example 2 - Iron Ore Processing Plant Impurities

Predict Impurities (regression)

Example 1: Steel Plate Manufacturing Defects Fault detection


(classification)

Goal is to build a model that can detect defects in steel plate

manufacturing.

This demo shows:

a) Machine Learning techniques (Neural Networks, Naive Bayesian, Tree Bagger),


Feature Selection,

b) Parallel Computing

c) MATLAB Compiler & Builder Ex

d) Spreadsheet Link
Example 2: Iron Ore Processing Plant Predict Impurities (regression)

This demo shows how machine learning can be used to improve the accuracy of
modelling and predicting the impurities output of an iron ore processing plant. A
number of variables in the plant were measured over time including the silica (SiO2)
and magnesia (MgO) concentration at the output of the plant. The goal of the
modelling is to determine what parameters in the plant need to be adjusted to keep
the silica and magnesia concentration at the desired level. The demo uses 1 year of
real data captured in an iron-ore processing plant. The mining company has given
MathWorks permission to use this data.

This demo shows:

a) Preparing time series data for analysis

b) Interpolating missing data

c) Time aligning and joining multiple data sets using tables

d) Using decision trees and neural networks to improve the model calculated using
multiple linear regression

e) Using sequential feature selection to identify the important parameters in the


plant

The utility is provided as a MATLAB-deployed standalone executable. It requires


Windows 7 64-bit with Internet Explorer 8, MS Excel 2010 and the MATLAB Compiler
Runtime R2013b for Windows 64-bit.
The fastest way to install the MCR is to copy the installer from here to your local
machine and then run it.

Steps to run the Seminar:


To run the seminar, please follow the steps below. The seminar was updated as of
R2014a.

Required Products:

MATLAB, Statistic Toolbox, Neural Network Toolbox, Spreasheet Link Ex, MATLAB
Compiler, MATLAB Builder EX

Example1: Steel Plate Manufacturing Defects (classifi cation)

1. Running from MATLAB: Run steel_plate_fault_detection.m from \steelplates


2. Running from Excel:
a. Spreadsheelink:
i. Open fault_detection.xlsm in tab Spreadsheet Link
ii. Launch Spreadsheet link addin from Excel
iii. In columns AC call =matlabfcn("steelfault",B2:AB2)
b. Builder Ex
i. Open project fault_detect.prj and rebuild the project on your
machine
ii. Open fault_detection.xlsm in tab Builder Ex
iii. In columns AC call =steelfault(B2:AB292)

Example2: Iron Ore Processing Plant Predict Impurities (regression)

1. There are 2 versions you can run:


a. Based on Excel Spreadsheet input. proc_con.m
b. Based on Text file input proc_conc_txt.m
2. I recommend opening up the spreadsheet concentrator_data.xlsx to show
the attendees that data:
a. Output data is Chemistry Data May-02 to May-03 Columns P SiO2 &
V MgO
b. Input data is tabs Chemistry Data May-02 to May-03 column A->N
and 'Process Data Values'
3. Then running the through the text file version proc_conc_txt.m

You might also like