A standby set that only ever runs its weekly no-load exercise is an untested machine. We at Enpower see this on factory sites across Punjab and Sindh: the set starts every Friday, idles for ten minutes, and nobody knows whether it can carry its rating until the grid drops in the middle of a shift. A load bank test answers that question on your schedule instead of the feeder's. For an industrial generator that protects a production line, it's one of the most useful days in the maintenance year.
This post is for the engineers and maintenance people who run that test. Choosing the set in the first place is a separate job, and our guide to industrial brands and capacities covers it. Here we'll plan the load steps, decide what to record, and use a short pandas script to check the log for the things people miss when they read numbers off a controller screen.
Resistive or reactive bank?
Most hire load banks are purely resistive, so they load an industrial generator at unity power factor. That's a proper test of the engine, because every kilowatt is real. It's a weaker test of the alternator, which is rated in kVA at a lagging power factor, usually 0.8.
Take a set rated 500 kVA at 0.8 PF, which makes 400 kW. A resistive bank at 400 kW loads the engine fully, but the alternator only carries 400 kVA worth of current, four fifths of its rating. If alternator heating or voltage regulation under reactive load is what worries you, ask for a resistive-reactive bank. They're less common, so book one early.
Planning the load steps
Build load in steps and hold each one long enough for temperatures to settle. A common pattern is 25, 50, 75 and 100 percent of the prime kW rating, with a longer hold at full load. Agree the actual steps and hold times with your supplier or the engine manufacturer's guidance, because some engines come with their own recommendations.
Use the prime rating from the data plate as your 100 percent, not the standby figure. Standby is the emergency ceiling under ISO 8528, and it's not a sensible target for a planned test.
What to record when you test an industrial generator
Log at short intervals, ideally every few seconds, through the whole test. Most modern controllers can record many of these channels for you, and a power analyser on the output fills in the rest.
- kW, kVA and power factor, plus voltage and current on each phase.
- Frequency.
- Coolant temperature and oil pressure.
- Ambient air temperature at the radiator inlet, not in the office.
- Exhaust temperature, if the set has a sensor or you can fit a probe safely.
- Fuel used at each step, from a flow meter or careful tank dips.
Add a "step" column to the export, either by hand or from the bank's own log, so every row knows which load step it belongs to.
Checking the log with pandas
The script below takes the last ten minutes of each step as the "settled" window and summarises it. The limits are deliberately empty. Fill them in from the engine and alternator data sheets for your set, not from a blog post, ours included.
import numpy as np import pandas as pd # Fill every limit from the engine and alternator data sheets. LIMITS = { "freq_band_hz": None, # allowed max-minus-min swing in a settled window "volt_imbalance_pct": None, # allowed phase-to-phase spread "coolant_alarm_c": None, # high coolant temperature alarm point "settle_c_per_10min": None, # above this, coolant is still climbing } missing = [k for k, v in LIMITS.items() if v is None] if missing: raise SystemExit(f"Set these limits first: {missing}") log = pd.read_csv("loadbank_log.csv", parse_dates=["timestamp"]) log = log.set_index("timestamp").sort_index() # columns: step, kw, freq, v_l1, v_l2, v_l3, coolant_c, oil_kpa, ambient_c rows = [] for step, g in log.groupby("step"): settled = g[g.index >= g.index.max() - pd.Timedelta("10min")] mins = (settled.index - settled.index[0]).total_seconds() / 60 slope = np.polyfit(mins, settled["coolant_c"], 1)[0] * 10 volts = settled[["v_l1", "v_l2", "v_l3"]] spread = (volts.max(axis=1) - volts.min(axis=1)) / volts.mean(axis=1) rows.append({ "step": step, "kw_mean": settled["kw"].mean(), "freq_mean": settled["freq"].mean(), "freq_swing": settled["freq"].max() - settled["freq"].min(), "volt_imbalance_pct": 100 * spread.max(), "coolant_end_c": settled["coolant_c"].iloc[-1], "rise_over_ambient_c": (settled["coolant_c"] - settled["ambient_c"]).iloc[-1], "coolant_c_per_10min": slope, "oil_min_kpa": settled["oil_kpa"].min(), }) report = pd.DataFrame(rows).set_index("step") report["freq_ok"] = report["freq_swing"] <= LIMITS["freq_band_hz"] report["volts_ok"] = report["volt_imbalance_pct"] <= LIMITS["volt_imbalance_pct"] report["settled"] = report["coolant_c_per_10min"] <= LIMITS["settle_c_per_10min"] report["coolant_ok"] = report["coolant_end_c"] < LIMITS["coolant_alarm_c"] print(report.round(2)) Reading the results
The column people skip is the coolant slope. If the coolant is still climbing at the end of a hold, that step hasn't told you anything about cooling yet, and you should extend the hold before moving up. A radiator half blocked with dust often looks fine at half load and only gives trouble an hour into full load.
Judge frequency per step, not against a flat 50 Hz. If the governor runs in droop, frequency falls a little as load rises by design. What you're looking for is a steady value inside each settled window, and a swing that stays within the band your supplier commits to.
Watch the oil pressure column alongside coolant temperature. Some drop as oil warms up is normal, but a pressure that keeps falling through a hold is worth a call to your service engineer.
Finally, keep the ambient figure next to the coolant rise. A comfortable result on a mild November morning doesn't prove the set will cope with a 45°C afternoon in Sukkur or Multan. Ask your supplier what ambient the cooling package is rated for, and compare that with how much margin your test showed.
Running the test on a Pakistani site
During the test, the industrial generator is off the plant. If the grid fails halfway through, you have no backup at all. Pick a window when your feeder is normally on, and agree in advance how quickly the team can drop the bank and put the set back on the changeover.
Place the bank so its hot exhaust air can't blow back into the radiator intake, which is easy to get wrong in a cramped generator yard. Size the temporary cables properly, keep them out of walkways, and have fuel on site for the whole test plus a margin. Sets that spend their lives lightly loaded may smoke for the first few steps as deposits burn off. Note it, and see whether it clears.
Keep the summary, not just the certificate
File the report table with the date, the ambient and the bank used. Next year's test is only useful if you can compare it with this one, and a slowly rising coolant figure at the same load and ambient is exactly the early warning you're paying for. That record is worth more than any certificate in a drawer, and it's how we'd want every industrial generator on a production site to be looked after.
If you'd like help planning a load bank test or reading the results, talk to the engineering team at Enpower.

