Condition monitoring becomes difficult when the most useful quantities are expensive or impractical to measure continuously. Bismarck Louw’s research asks whether a model can fill that gap without becoming a black box.
The work combines calibrated physical sub-models of a Cummins QSB6.7 engine with data-driven mappings from engine speed, load and fuel consumption. Controlled dynamometer tests provided temperature, pressure, power and ECU data across operating conditions.
The hybrid model predicted groups of thermal, pressure and power indicators with mean absolute percentage errors of 6.21%, 5.08% and 4.12% respectively. That matters because the additional indicators can support performance monitoring and predictive maintenance while relying on measurements already available from an engine control unit.
The research completed as a distinction-level MEng and became an accepted 2025 manuscript. Its next utility step is a bounded interface that preserves the validated operating domain and communicates uncertainty with every result.