From equipment documentation to a validated physics-based model of an industrial pump unit: Modelica modelling, calibration and monitoring functions evaluated on open plant data

Physics-based models are the core of digital twins of rotating equipment. Building and validating such a model for a specific installation remains slow and expert-intensive. We report an end-to-end case study on a boiler feed pump unit: a seven-stage centrifugal pump driven by a 110 kW induction motor. The model was obtained from manufacturer documentation with our model-synthesis method. It was compiled and simulated in OpenModelica using the Modelica Standard Library only, calibrated on open plant telemetry (three CIRA P507 units, eight campaigns, 17.5 h of operation), and validated under a protocol registered before computation. The model has 9 state variables and 320 equations. It passed all 143 behavioural acceptance checks on its first build. Documentation alone over-predicted the motor frame temperature by up to 11 K. Calibrating four site properties reduced the errors on held-out campaigns to 0.7–3.0 K. The calibrated parameters transferred from one unit to another unit with the same motor type (RMSE 0.74 and 1.05 K). Outside their training range, the model was an order of magnitude more accurate than ARX and output-error models fitted to the same data. With the full Modelica model, we quantify functions that run on the existing sensors: • a residual alarm that detects a 20 % loss of motor cooling at any ambient temperature within 1.3–2.2 h of running, where the best fixed threshold misses it below 40 °C ambient; • frame-temperature forecasts with 0.8–1.0 K error one hour ahead; • motor-current signatures of pump wear: efficiency loss and head loss move the current in opposite directions. We also show what the present data cannot identify — the stator winding temperature — and which single measurements would remove this limit. The work is an intermediate step towards site-specific digital twins built from documentation at low cost.