Power Factor Correction for Air Compressor Motor Installations

Put a recording power analyzer on a 160 kW screw compressor supply. Leave it three days. The kVAR channel is flat. 43 kVAR loaded, 41 unloaded, on a WEG W22 at a bottling plant. A Siemens 1LE1 of the same rating next to it read 39 loaded, 37 unloaded. Ten percent difference between identically rated motors. The kW channel is a square wave: 152 kW loaded, 19 kW unloaded. Power factor swings from 0.84 to 0.18 every cycle.

That flat kVAR trace is the magnetizing current. It holds the air-gap flux whether the compressor is doing work or not.

Most engineers size from the nameplate. 0.85 power factor at rated load, calculate correction to 0.95, get ~50 kVAR. Install it. At full load, 0.95 achieved. At no load, the bank is oversized by about 6 kVAR because the leakage reactance component vanished with the torque. And 6 kVAR of surplus capacitance at 400 V becomes a larger surplus at 418 V, because capacitor output follows voltage squared while the motor's magnetizing demand follows the B-H saturation curve and flattens above about 105 percent nominal. Self-excitation territory. ABB's capacitor application guide (Clause 3.4) says 90 percent of no-load magnetizing kVAR as the limit. For compressors, use 80 percent, measured at the highest bus voltage logged over a week. The extra margin absorbs IEC 60831-1 Clause 17 capacitance tolerance of +10/-5 percent and the reality that compressors get tripped off supply by high-pressure cutouts several times per shift.

The star-delta problem is going to take more space than it deserves relative to its engineering complexity, because the engineering is trivial and the failure is common. During the 50 to 100 ms open transition between star and delta contactors, the motor spins free. Capacitors connected at the motor terminal box (which is where they get connected, because the terminal box is right there and the MCC is across the room) remain energized on the motor side. Self-excitation. Motor generates voltage at a drifting frequency. Delta contactor closes into it. Phase angle at reclosure is random. At 180 degrees, the voltage across the contactor tips approaches twice the supply voltage, current transient hits 15 to 20 times rated, and the contactor welds.

Schneider's PFC Application Guide, Section 4.2, documents this. Panel drawings show capacitors upstream of the star-delta group with a dedicated contactor and 3 to 5 second delay after delta confirmation. What happens on site is that the electrician runs cable to the motor terminal box because it is two meters away instead of fifteen back to the MCC. The drawing gets filed. Months later, first cold morning, contactor welds.

Or the fuses blow. And then someone replaces them with a larger size, which shifts the failure from the fuses to the capacitor dielectric.

One line in the installation specification prevents this. The line says capacitors shall be connected upstream of all motor starting equipment. Most specs do not include it. Some specs include it and it gets ignored during installation. Commissioning tests do not check for it. The failure recurs.

This is not an engineering problem. The engineering has been solved for decades. It is a coordination problem between the person who writes the specification, the person who draws the panel, the person who routes the cable, and the person who writes the commissioning checklist, who are typically four different people working for three different companies on a timeline that does not allow them to talk to each other.

Enough about star-delta.

APFC panels. Contactor-switched versus thyristor-switched. IEC 60947-4-1 Table 1 rates AC-6b contactors at 100,000 operations minimum. A compressor cycling every 30 seconds generates 864,000 per month. Thyristor-switched panels per IEC 61921 cost 40 to 60 percent more and have no mechanical wear. Contactor panels keep getting installed on compressor loads because the APFC panel is a line item on an electrical subcontract bid on price.

There is not much more to say about this. The numbers speak.

The tariff question is more interesting than most articles on compressor PFC acknowledge, and it gets compressed into a paragraph when it deserves more room. A kVARh tariff and a kVA demand tariff reward correction of different operating states. Under kVARh, the unloaded condition is where the money goes, because kVARh consumption is nearly identical loaded and unloaded. Under kVA demand, the loaded condition sets the peak. Under average-PF penalty tariffs, unloaded hours drag the average down nonlinearly. Which tariff applies shifts the optimal capacitor sizing by 20 to 30 percent.

Here is the part that does not get discussed: the interaction between the tariff structure and the 15-minute demand interval averaging. Within a single 15-minute window, the compressor cycles many times. The demand meter averages everything. The APFC controller's brief overshoots during transitions are invisible to the meter. A persistent sizing error in the fixed base capacitor is visible in every interval. This means the fixed capacitor is doing the financial heavy lifting and the automatic controller is trimming. Most project budgets allocate disproportionately to the controller.

Consider a specific scenario. A plant in KwaZulu-Natal on Eskom's Megaflex tariff, which bills kVA demand. Three 110 kW compressors. Duty cycle averages 55 percent. The loaded kVA per compressor is about 140 kVA. Unloaded, about 45 kVA. The demand peak is set by loaded operation. Correcting from 0.83 to 0.96 at full load reduces the loaded kVA from 140 to about 120. That is 20 kVA per compressor, 60 kVA total, off the demand peak. At Megaflex demand rates of roughly R45 per kVA (varies by season and time of use), that is R2,700 per month. The capacitor bank costs maybe R35,000 installed. Payback: 13 months. Under a kVARh tariff at a different utility, the same three compressors with 45 percent unloaded running time generate a completely different payback calculation because the unloaded kVARh dominates.

The point is that the same physical installation, same motors, same capacitors, pays back in 13 months under one tariff and 8 months under another, depending on which operating condition the tariff penalizes more.

VFDs on compressors eliminate the displacement PF problem. The input rectifier runs at 0.96+ displacement PF. Do not install PFC capacitors on the motor output side of a VFD. They interact with the PWM switching and trip the drive's overcurrent protection.

The harmonic resonance issue in compressor rooms develops slowly. A PFC bank installed on a bus with three direct-on-line compressors works fine. One compressor gets a VFD two years later. Another gets one a year after that. Fifth harmonic current on the bus rises. The PFC bank on the remaining fixed-speed machine forms an LC circuit with the transformer impedance. If the resonant frequency drifts near 250 Hz, harmonic amplification starts. Capacitors overheat. Fuses go. Replacements fail.

Detuning reactors at 7 percent shift the resonance to ~189 Hz on a 50 Hz system. Add 30 to 40 percent to the bank cost at installation. A detuned bank delivers less fundamental kVAR than its nameplate: 93 percent with 7 percent detuning per IEC 61921 Annex A. "120 kVAR 7% detuned" on a purchase order delivers 112 kVAR effective. If the design target was 120 effective, the spec must say so.

On the distinction between displacement PF and true PF per IEEE 1459: VFDs improve displacement PF. True PF on a six-pulse front end without line reactors is 0.78 to 0.83 due to harmonic current content. Whether this matters financially depends on the utility's metering technology. Most utilities still use displacement-only reactive meters.

Capacitor mounting location.

IEC 60831-1 Clause 13 covers temperature derating. Dielectric life halves per 10°C above rated core temperature. Compressor rooms with air-cooled machines hold 45 to 50°C ambient. Add 12 to 15°C self-heating. Core temperature 60°C or higher. Rated life of 100,000 hours at 45°C becomes roughly 30,000 hours at 62°C.

Oil mist from oil-injected screw compressors condenses on terminal blocks. Traps dust. Conductive film. Tracking faults over months. Micro-arcs. Carbonization. Short circuit.

Adjacent electrical room with filtered air. Extra cable loses a few hundred watts of I²R. Against the cost of replacing the bank every four years instead of every twelve, the cable losses are irrelevant. This calculation takes two minutes. It rarely gets done during design because the electrical room layout is finalized before the PFC requirement is identified.

Anti-cycling timers and capacitor discharge. Compressor OEMs set anti-cycling at 60 to 120 seconds. IEC 60831-1 Clause 22 requires internal discharge resistors to bring voltage to safe levels. Most LV capacitors reach 50 V within 60 seconds. When maintenance technicians shorten the timer below 60 seconds for production reasons, the motor restarts before discharge completes. Residual voltage adds to supply voltage. 160 to 170 percent dielectric stress. Metallized polypropylene film self-heals: micro-patches of metallization vaporize, capacitance drops fractionally. Over a year of this, measurable capacitance loss and elevated internal temperature. In a room already running the capacitor at its thermal limit, the degradation compounds.

The timer change and the capacitor degradation are separated by twelve to twenty-four months and by the boundary between mechanical and electrical maintenance records.

Synchronous motors with brushless excitation on base-load compressors above 300 kW generate reactive power instead of consuming it. Motor cost premium 25 to 35 percent, offset over 15 years by eliminating the PFC bank, detuning, switching gear, and maintenance. Compressor OEMs do not stock them. The procurement coordination required to specify one across mechanical and electrical streams is the barrier, not the cost.

APFC controllers have Modbus or Ethernet ports as standard. Power factor trending, step status, fault flags. Connection to plant SCADA takes a few hours. Most installations never connect the port.

— end —