This project addresses technology gaps in improving bitumen production performance, by establishing the proof of concept of using hyperspectral imaging for ore classification, and assessing its potential impact on the observability of batch, laboratory-scale, water-based extraction processes.
This project aims to determine hyperspectral feature sets, classification schemes and predictive models to discriminate samples with processibility issues. It examines the trade off of implementing these tools with instrumentation collecting broadband data. This would build on previous work which has established that this approach can predict bitumen content and < 2micrometre fines content.
Principal Investigator: Benoit Rivard
Research Team: Michael Lipsett, Jilu Feng
Project Number: COSI 2009-08
Projected Completion Date: August 31, 2011
Figure 1: Reflectance spectra of oilsand converted to wavelet representation for two samples (1L07 and SUN6). A number of oil (O), clay (C), and water (W) features controlling predictive models (e.g Figure 3) are identified.
Figure 2: Broadband spectral predictive model of % total bitumen content (TBC). Circles: modeling data suite, Stars: Validation data suite. The models were derived from the analysis of hyperspectral data to optimized the selection of feature sets.