Abriliam Consulting — Industrial Energy Management
This notebook investigates the condenser water loop hydraulics to determine whether the plant operated in distinct hydraulic regimes during the monitoring period. A "regime change" — such as a valve position change, pump staging difference, or bypass condition — would alter the relationship between pressure drop (delta-P) and flow, and could explain the efficiency degradation observed in earlier notebooks.
We use a combination of scatter analysis, hydraulic coefficient calculation, and Gaussian Mixture Model (GMM) clustering to identify and classify operational regimes.
Index(['oat_C', 'wb_C', 'occ', 'tons', 'chw_sup_C', 'chw_ret_C', 'chw_dT_C',
'chw_flow_m3h', 'cw_sup_C', 'cw_ret_C', 'cw_dT_C', 'cw_flow_m3h',
'approach_C', 'dp_kpa', 'chiller_kw', 'tower_fan_kw', 'chw_pump_kw',
'cw_pump_kw', 'plant_kw', 'kw_per_ton', 'plant_kw_per_ton',
'tower_fan_kw_per_ton', 'pumping_kw_per_ton', 'kw_per_ton_15_sma',
'kw_per_ton_5_sma', 'kw_per_ton_24_sma', 'kw_per_ton_240_sma'],
dtype='object')This dual-axis plot shows the condenser water loop pressure drop alongside tower fan power per ton of heat rejected. Two distinct operating bands are visible — the system appears to operate at different pressure setpoints during the monitoring period. The fan efficiency metric (kW per ton rejected) helps normalize for load variation.
The histogram of CW loop delta-P reveals a bimodal distribution — two distinct clusters of operating pressure. This is strong evidence of a hydraulic regime change. A healthy, stable system would show a single peak that shifts with load. Two peaks suggest the system switched between configurations at some point during the monitoring period.
The scatter plot confirms two operating clusters. At higher differential pressures, the tower fans don't necessarily work harder — the excess pressure is being absorbed elsewhere in the system (possibly through a throttled valve or bypass). This is wasted pump energy.
The hydraulic coefficient k = delta-P / Q-squared characterizes the system's flow resistance. In a fixed piping configuration, this coefficient should be roughly constant regardless of flow rate. A shift in k indicates a physical change — a valve opened or closed, a pump was staged differently, or a bypass condition changed.
Plotting pressure drop against flow-squared should yield a single line for a fixed hydraulic configuration (since delta-P is proportional to Q-squared by the Darcy-Weisbach equation). The presence of two distinct linear relationships confirms that the condenser water system operated in two different hydraulic configurations during the monitoring period.
The time-series view of delta-P clearly shows when the regime change occurred — there's a visible step-change in the operating pressure around early July. This aligns with the efficiency degradation detected by the CUSUM analysis in Notebook 03.
Counts by regime: regime 0 422 1 922 Name: count, dtype: int64 Median confidence by regime: regime 0 0.931134 1 0.999971 Name: confidence, dtype: float64
A Gaussian Mixture Model with two components cleanly separates the two hydraulic regimes. The point opacity represents classification confidence — darker points are confidently assigned to one regime, while lighter "bridge" points represent transitional hours.
The two regimes correspond to:
This unsupervised classification can be used to automatically segment operating data for regime-specific analysis and alerting.