Wilcoxon Vibration Sensors for Predictive Maintenance

A rolling element bearing announces itself in the kilohertz band long before it shows up at shaft speed. An accelerometer reads that band directly, so the choice comes down to three numbers: how many millivolts per g, how the sensor is fixed to the housing, and how fast the data is sampled.

This guide works through those three numbers against published Wilcoxon specifications, for route-based or permanent monitoring of motors, pumps, fans and gearboxes below about 100 kW, running at fixed speed.

Why bearings need acceleration, not velocity or displacement

A bearing with a spalled race does not vibrate at a tidy frequency. Each time a rolling element passes the defect it produces an impact, and the impact excites the housing at its own structural resonance, somewhere in the low kilohertz. The result is a burst of high-frequency energy repeating at the defect rate. At an early stage the repetition rate is often invisible as a discrete line in the spectrum; only the high-frequency burst is measurable.

Velocity and displacement are the wrong currency for that burst. For a given acceleration amplitude, velocity falls in proportion to frequency and displacement falls with the square of it. A defect producing 1 g at 3 kHz corresponds to a velocity of about 0.5 mm/s and a displacement of roughly 0.03 µm. The same 1 g at 50 Hz is 31 mm/s and 100 µm. So the kilohertz content arrives at the displacement channel some 3,000 times smaller than the running-speed content, and a displacement of 0.03 µm sits below the resolution of the eddy-current probes used to measure it.

This is scaling, not deafness. A piezovelocity sensor such as the 793V is specified from 2.5 to 7,000 Hz, so it does respond at 3 kHz. But its output is integrated, and at 3 kHz the integrated signal is around sixty times smaller than at 50 Hz for the same acceleration, so it disappears into the noise floor of the channel. Measure acceleration, and integrate in software when a velocity reading is required for the ISO 20816 assessment.

Sensitivity and range are one number, spent two ways

An IEPE accelerometer carries its own charge amplifier, so it needs no charge amplifier in the cabinet. It needs only a constant-current supply: 18 to 30 VDC fed through a 2 to 10 mA current-regulating diode on the same two wires that carry the signal back. The amplifier sits at a bias of about 12 VDC and the vibration signal swings either side of that bias until it runs out of supply rail.

That swing is a fixed budget. A 500 mV/g sensor spends 500 mV for every g, so a 5 V swing buys 10 g. A 10 mV/g sensor spends 10 mV per g and the same swing buys 500 g. The published ranges show the trade directly.

Three conclusions follow from that table.

1. 100 mV/g is the general-purpose choice, and it is not a compromise. The 786A covers 80 g with a noise floor of 5 µg/√Hz, which resolves a bearing defect months before it becomes audible. Most motors, pumps and fans never exceed 20 g in service, so the headroom is there for start-up transients and run-up resonance.
2. Higher sensitivity buys resolution, and pays for it in range. On a slow-turning machine the 1× component sits at a few hertz, and because acceleration scales with the square of frequency, the acceleration produced by a given mechanical looseness is tiny. The 786LF-500 spends its swing on 10 g and extends down to 0.1 Hz, which is what a cooling tower fan or a slow gearbox needs. Fit it to a 1,500 rpm motor and you will clip it on the first start.
3. Low sensitivity is for impacts, not for general monitoring. The 793-10 reaches 500 g with a 25 kHz mounted resonance, which suits diesel engines, hammer mills, crushers and spindles above 60,000 CPM. Its noise floor is 40 µg/√Hz, eight times that of the 786A, so it is the wrong instrument for a point where the defect signal is small.

One datasheet condition is easy to miss. Wilcoxon quotes the 80 g range of the 786A for a supply above 25 V, because the swing is bounded by the rail. Run the same sensor from a 24 V loop and the usable range is slightly smaller than the label suggests. If your points run hot, the HT780A keeps the same 100 mV/g and 80 g to 165 °C, though its noise floor rises from 5 to 7 µg/√Hz at that temperature.

Mounting decides the band you actually get

Every datasheet band assumes a stud mount into a flat, tapped, spot-faced surface. Any other method inserts compliance between sensor and machine, and compliance lowers the mounted resonance. Wilcoxon’s mounting guide gives the practical ceilings.

The table is a ceiling on the mount, not on the sensor. With a 786A, the sensor’s own 14 kHz response ends the argument well before the stud does. Put the same sensor on a two-pole magnet and the mount closes the band at 2–5 kHz, which cuts off exactly the region where an early bearing defect lives. The usable band is always the lower of the two limits.

Five installation details decide whether you get the band you paid for.

1. Prepare the surface. Spot-face the pad so the sensor base sits flat, tap perpendicular to it, and keep the stud short. A burr or a chip under the base costs more high-frequency response than a cheaper sensor does.
2. Torque it. A 1/4-28 stud wants about 30 in-lb, roughly 3.4 N·m. Under-torque leaves the joint soft; over-torque damages the sensor base.
3. Use a coupling film. A thin layer of silicone grease raises the mounting stiffness and improves the high-frequency response. It is standard practice for permanent points.
4. Break the ground path when you need to. The sensor case is isolated from its circuit internally, but a steel stud still ties the case to the machine frame. Isolated studs such as the SF4 and SF4M have a non-conductive layer for exactly that case.
5. Stay off guards and thin panels. Mounting on a guard, a cover or a thin bracket measures the resonance of that bracket. Wilcoxon’s own guidance is to avoid thin sections and vibration-free areas, and to place the sensor as close to the load path as the geometry allows.

Where the points go

The bearing housing is the reference location. It sits on the load path between the shaft and the structure, so defect energy arrives there before it reaches the frame.

1. Motor. Drive end and non-drive end, each in two radial directions at 90°, horizontal and vertical. Add an axial point at the non-drive end where the thrust bearing sits.
2. Pump. Drive end and non-drive end bearing housings, radial in two directions, plus axial on the thrust bearing. On a close-coupled pump the motor non-drive end and the pump drive end are separate points even though they are 300 mm apart.
3. Gearbox. Input and output shaft bearings, in all three directions. Mesh forces are axial as well as radial, and an axial point is where a misaligned or cocked gear shows first.
4. Fan or blower. Both bearings, radial. The axial point on the drive end catches unbalance and belt tension problems that the radial points under-report.
5. Anything belt-driven. Both the motor and the driven shaft, because belt tension faults appear on both sides of the drive.

Count bearings, not kilowatts. A motor-pump set with four bearings usually carries eight to ten points; each additional gearbox shaft adds two or three. Log the load, speed and bearing temperature with every reading, because a trend compared across different operating conditions is not a trend, and take a baseline at commissioning while the machine is known good.

Matching the acquisition side

The sensor is half of the measurement. The other half is the input it is connected to.

• Supply. Confirm 18 to 30 VDC and 2 to 10 mA constant current per channel. Many cards fix the current at 4 mA, which is fine; what matters is that the compliance voltage is high enough to carry 12 V of bias plus the signal swing.
• Cabling. For the MIL-C-5015 two-pin types, Wilcoxon lists the J9T2 and J10 cable assemblies. Use low-noise cable, keep the run clear of drive output cables, and strain-relieve at the connector.
• Ground loops. The symptom is a mains-frequency spike with a raised noise floor, and it appears when the shield is earthed at both ends. Earth the shield at the acquisition end, rely on the isolated case, and use an isolated stud if the machine frame is bonded elsewhere.
• Sampling and anti-aliasing. Set the band you need first: 1 kHz for general trending, 5 to 10 kHz for bearings and gear mesh. Then sample at 2.56 times that upper frequency with the anti-alias filter set at the same value. Sampling at twice the frequency of interest without a filter folds high-frequency energy back into the band and invents peaks that are not in the machine.
• Analysis depth. 16-bit resolution is enough for trending and for an FFT. Envelope or shock-pulse work needs the high-frequency band, which means a higher sample rate and more storage.
• Continuous versus route-based. For a trend into a PLC or DCS, a loop-powered transmitter such as the PC420AP-10 gives 4–20 mA proportional to 10 g full scale over 1 to 2,000 Hz on two wires. It will not give you a spectrum, so keep IEPE points where diagnosis has to happen.

Sourcing accelerometers and the parts around them

We supply the Wilcoxon range with matching cables, studs and mounting pads, and match the sensor to the machine, the mounting and the acquisition system rather than to a part number alone. Every unit is original and new, bought through the manufacturer and established distribution channels, with procurement documentation for verification on arrival.

• Browse the Wilcoxon range we carry.
• Compare the wider sensor range across all brands.
• Read more selection guides on measurement and drive components.
• Start from the complete brand directory.
• Send the machine type, the point list and the acquisition system to the inquiry form.

Sensitivities, ranges, bands and temperature limits quoted here are taken from current Wilcoxon datasheets and mounting documentation. Specifications change without notice, so confirm the figures for the exact type against the official datasheet before ordering.