When there’s no instrument to measure the quantity of interest, sometimes you need to develop your own. Coanda’s instrumentation group can search for the best solution – whether that’s commercially available, designed from scratch, or somewhere in between. In this case, in collaboration with Canadian Natural Resources, we used machine-learning algorithms to combine conventional measurements with micron-resolution infrared images from process borescopes. The resulting “instrument” provided a measurement of the flocculation quality of oil sands tailings treatment, enabling real-time feedback for process control.
One of the algorithms developed by our data-science team resulted in a publication in AIChE Journal. In the images suspended solids and flocs reflect IR and appear bright contrasted against dark channels of release water. The algorithm was sensitive to the length scales and shapes present in the images via the Fourier transform and a basis set built from training images. It quantified whether the tailings were under-mixed (no release water), well-mixed (good release water channeling and floc shape), or over-mixed (narrow release water channels and broken flocs).
See this related Coanda paper published by AIChE.