Identification of cement manufacturing raw materials using machine vision

Notley, Lindsay (2016) Identification of cement manufacturing raw materials using machine vision. [USQ Project]

Text (Main Project)
Notley_L_ Maxwell.pdf

Download (9MB) | Preview


In the mining and manufacturing industry, there is a need for a non-extractive system to identify raw materials on conveying systems. Such a system would allow identification of raw materials on conveying systems preventing cross-contamination when the materials arrive at the final storage location.

This project used machine vision techniques to identify cement manufacturing raw materials (clinker, gypsum and, limestone). Firstly, a representative sample (25 x 10kg samples of each material) was collected using a stratified random sampling procedure. This stratified random sampling procedure ensured the sample accurately represented the raw material in the stockpile.

A dual purpose test bed and controlled lighting camera enclosure (for static model development and future dynamic system implementation) were constructed to minimise the effect of varying ambient light. This test bed and camera enclosure allowed the CMOS global shutter industrial camera to take twenty, 24bit colour images (8bit for each colour) of each sample. These images were catalogued and stored in a database for further model training and verification purposes.

These images were pre-processed by a median filter which allowed any over saturated pixels (due to raw material surface moisture reflection) to have their intensity level reduced by replacing its value by the median value of its local neighbours. From the filtered image the individual red, green and blue (RGB) components were passed to a Histogram function which binned (255 bins for 8-bit colour) the various pixel intensities. The statistical features (weighted mean, skewness and kurtosis) of each colour's histogram were then stored in an array which then passed to the image feature database.

A varying amount of feature arrays were used to train and verify the success of a probabilistic neural network (PNN) model. Initial optimisation of the PNN model was conducted using a local search algorithm which changed the smoothing parameter which achieved 94.83% accuracy. This model was then improved by implementing a Supervised Learning Probabilistic Neural Network (SLPNN). This model added data weight which changed the height of the Gaussian distribution function and input variable vector weight which changes the width of Gaussian distribution function. The implementation of the Supervised Learning Probabilistic Neural Network improved the models accuracy to 99.57%.

Further model field testing will be required to verify the system in an operational environment where the camera enclosure will be subjected to dust, noise, varying temperatures and moisture. The Supervised Learning Probabilistic Neural Network outperforms the standard Probabilistic Neural Network which has been proven by this work. This work supports the claim that Machine Vision can be successfully be used to identify cement manufacturing raw materials with a high success rate. It also contributes to the literature by classifying clinker, gypsum and limestone in one body of work.

Statistics for USQ ePrint 31454
Statistics for this ePrint Item
Item Type: USQ Project
Item Status: Live Archive
Additional Information: Bachelor of Engineering (Honours) Major Electrical & Electronic Engineering project
Faculty/School / Institute/Centre: Historic - Faculty of Health, Engineering and Sciences - School of Mechanical and Electrical Engineering (1 Jul 2013 - 31 Dec 2021)
Supervisors: Maxwell, Andrew
Date Deposited: 21 Jul 2017 02:35
Last Modified: 21 Jul 2017 02:35
Uncontrolled Keywords: machine vision techniques; cement manufacturing; conveying systems; raw materials
Fields of Research (2008): 09 Engineering > 0906 Electrical and Electronic Engineering > 090602 Control Systems, Robotics and Automation
Fields of Research (2020): 40 ENGINEERING > 4007 Control engineering, mechatronics and robotics > 400799 Control engineering, mechatronics and robotics not elsewhere classified

Actions (login required)

View Item Archive Repository Staff Only