Register or Log In

International Society for Industrial Process Tomography

11th World Congress on Industrial Process Tomography

Application of Unsupervised Machine Learning to Identify Slurry Deposition Velocities

A. Kumar1, S. A. Hashemi2*, R. B. Spelay2, J. Valdes3, G. Durand3, T. D. Machin4, H-Y. Wei5, R. S. Sanders1

1Dept. of Chemical & Materials Engineering, University of Alberta, Edmonton, CA

2Pipe Flow Technology Centreā„¢, Saskatchewan Research Council, Saskatoon, CA

3Digital Technologies Research Centre, National Research Council, Ottawa, CA

4Stream Sensing Ltd, Manchester, UK

5Industrial Tomography Systems Ltd, Manchester, UK

*Email: reza.hashemi@src.sk.ca


ABSTRACT

Numerous applications (e.g., mining and mineral processing, dredging, tunnelling, food processing, pharmaceutical production) involve pipeline transport of coarse-particle (settling) slurries. Below a limiting deposition velocity (Vc), particles form a stationary bed, leading to operational challenges:   increased energy consumption, reduced pipeline capacity, blockages and missed production targets. Dynamic changes in particle size distribution and rheology are typically not tracked, hence Vc cannot be predicted. Real-time non-invasive electrical resistance tomography (ERT) is commonly used to provide 2D spatial distribution of solids in mixing, separation, and slurry pipeline systems, potentially enabling identification of localized events such as solids deposition, through analysis of the 2D solids distribution. The paper describes machine learning to automate analysis of ERT tomograms in real-time to detect solids (coarse) stratification and deposition, using a 100mm pilot-scale pipe loop. Test slurries with mixed coarse (settling) particles in carrier fluids with different rheology represent industrial slurries. Extracted 2D solids distribution data were segregated into 3 groups using K-means clustering: (1) uniform solids (coarse) distribution, (2) stratified coarse concentration, and (3) solids (coarse) deposited at the bottom of the pipe. Results were aligned with experimental measurements of Vc obtained during the pipe loop tests. Use of unsupervised machine learning is shown to extract features from ERT data that can potentially be used to identify specific operating conditions, such as the onset of deposition.


Keywords: Slurry flow, deposition velocity, unsupervised machine learning

Industrial Application: Mining and mineral processing, dredging, food and pharmaceutical production

Sign-in to access the full text