2026 JOHN OGILVIE RESEARCH INNOVATION AWARD
“ProteinProcessIO: A GPU-Accelerated, Experimentally Validated Digital Twin for the Complete Dry Fractionation of Canadian Pulse Crops”
ProteinProcessIO is the first integrated, physics-based digital twin that predicts the end-to-end performance of the complete dry fractionation chain in a single computational environment. Developed at McGill University in collaboration with NRC Canada under the federal Sustainable Protein Production Program, the platform covers all three interdependent stages — RF dielectric pretreatment, hammer milling, and air classification — and couples them in an automated pipeline where the outlet conditions of each stage feed directly into the inlet of the next. This end-to-end coupling is scientifically critical: a change in pretreatment temperature modifies seed brittleness, which reshapes the milling particle-size distribution, which in turn shifts the air classifier cut-point and protein recovery yield. Stage-isolated models cannot capture this cross-stage sensitivity.
The RF pretreatment module is a full digital twin of the QMTI GP-15 15 kW RF heating system at NRC Canada. It couples ten physics and control sub-models per timestep — including RF field propagation, volumetric dielectric heating, thermal conduction, moisture diffusion, evaporation kinetics, belt kinematics, oscillator circuit dynamics, and full PLC controller logic. The model was calibrated on actual PLC telemetry from a 61 kg NRC Canada production run, simultaneously matching outfeed temperature, anode current trajectory, and electrode gap. It was then blind-validated on an independent 90 kg run, predicting electrode gap within 1.2% (93.0 mm predicted vs. 94.1 mm measured). The milling module reproduces measured D50 to within 0.1 μm at 6,000 RPM using a calibrated Rosin-Rammler breakage function. The air classification module implements Lagrangian particle tracking through 36 fully parameterized equipment components — venturi eductor, zigzag preclassifier, high-speed wheel classifier (to 5,000g), three-stage cyclone cascade, and bag filter — accelerated via NVIDIA Warp GPU kernels.
The platform is released free of charge as a self-contained desktop application for Windows, macOS, and Linux, placing the simulation capacity of a fully equipped pilot plant in the hands of any Canadian research group or processor with a standard laptop.
