# Building Your Own Balancing Machine: Supports, Bed, Drive, and Measurement System > Building a balancing rig in your own shop is realistic, and it's done more often than people tend to think. The expensive part of a project like this isn't the electronics — it's the mechanics: supports, bed, drive, and rotor location. Below we break the construction down by assembly, show where the physical limits on accuracy lie, and give a step-by-step commissioning procedure. The Balanset-1A OEM serves as the measurement core in the examples, because for soft-bearing machines that's its standard application. **In short:** Yes, you can build a rig on your own, and the soft-bearing (above-resonance) scheme is the most accessible way to do it. You need four things: a rigid, heavy bed, supports with a known suspension natural frequency (the frequency at which the support's moving part oscillates on its own) well below the running speed, a drive with stable speed, and a two-channel measurement system with a phase-angle sensor. The electronics get solved by buying a ready-made measurement core; everything else has to be designed and verified by you. The key point that separates a working rig from an expensive piece of hardware: acceptance testing by geometry, by dynamics, and against a reference rotor with a known trial unbalance. Source: https://axiline.pt/en/articles/building-your-own-balancing-machine/ Publisher: AXILINE · Vila Nova de Gaia, Portugal · +351 931 831 229 · axilinegeral@gmail.com ## Do You Need Your Own Rig: An Honest Fork in the Road Your own rig is justified when you have a steady stream of similar parts and you balance them before assembly. Fan wheels, pump impellers, pulleys, screw conveyors, drums, knife rolls, motor rotors after rewinding, tooling. A rig gives you the same location from part to part, removes the hassle of mounting sensors each time, and lets you work from saved influence coefficients — the rig's remembered response to a trial weight, from which the instrument calculates the correction without repeat trial runs. A rig isn't justified when the machine is already sitting on its foundation. Balancing in the machine's own bearing housings accounts for the real bearing housings, the real frame stiffness, the operating temperature, and the running condition. A rig doesn't reproduce any of that. You'd end up with a part balanced on an arbor, and you'd still have to go finish the job on site. A ready-made machine wins wherever a certificate is needed. A factory machine comes with a data sheet, a stated minimum achievable residual unbalance, and a verification procedure. A homemade rig doesn't inherit those figures: you'll have to establish them yourself, and that's a separate job. | Situation | Sensible choice | Why | | --- | --- | --- | | One-off rotors, a new type every time | On-site balancing in the machine's own bearing housings | Designing tooling for every part costs more than balancing it on site | | A run of similar parts before assembly | Your own rig | Location is repeatable, influence coefficients carry over, run time is minimal | | Repairing fans and pumps for customers | On-site balancing | The rotor operates in its own machine — a rig can't reproduce its bearing housings and running condition | | Need a report for external acceptance | A ready-made machine or a certified service | Declaring an accuracy grade requires a verified procedure and a reference rotor | | You have sound old mechanics on hand | Retrofit the measurement side | The bed and supports outlive the electronics — no reason to replace them | > Practical advice: start with on-site balancing of your own part range. Over a month or two you'll build up statistics on masses, diameters, speeds, and the number of correction planes — rotor cross-sections where correction weights are mounted. Designing a rig from this data is far easier than designing it from guesses. ## Soft-Bearing and Hard-Bearing Schemes: Where the Running Speed Sits The entire difference between the two schemes comes down to one question: where the running speed sits relative to the supports' suspension natural frequency. In the soft-bearing (above-resonance) scheme, the support is compliant and oscillates noticeably under the centrifugal force from unbalance. The running speed sits above the suspension's natural frequency — hence "above-resonance" — roughly 2–3 times higher or more. What's measured here is the support's motion: vibration velocity or vibration displacement. In the hard-bearing (below-resonance) scheme, the support is essentially motionless. It's built as a thick plate with a U-shaped slot that effectively splits the plate into a rigid part and a compliant part. Under the unbalance force, these parts shift relative to each other by microns, and this shift is picked up with a force sensor or a highly sensitive displacement sensor. The running speed is kept 2–3 times below the support's natural frequency — hence "below-resonance." For a homemade rig, the soft-bearing scheme is simpler for three reasons. Amplitudes at the supports are several times higher, so ordinary accelerometers are good enough. The suspension's natural frequency is easy to tune with spring geometry and easy to measure. And finally, the support's response doesn't need a calculation model: the system calibrates itself with a trial weight. The Balanset-1A OEM fits into this scheme as a standard application: two accelerometers on the supports, a laser phase sensor on a reflective marker, a two-channel USB module, and software on a laptop. | Feature | Soft-bearing (above-resonance) | Hard-bearing (below-resonance) | | --- | --- | --- | | Running speed | 2–3 times above the suspension's natural frequency, or more | 2–3 times below the support's natural frequency | | What's measured | Support oscillation | Relative deformation between parts of the support, or force | | Sensors | Accelerometers, vibration velocity sensors | Force sensors, non-contact displacement sensors | | Lower speed limit | Limited by the suspension: with a band-suspension at 1–2 Hz, you can balance from 200 rpm | Limited by the measurement channel's sensitivity, realistically 200–500 rpm and below | | Versatility across rotor mass | Lower: the natural frequency depends on the part's mass | Higher: a rigid support barely reacts to mass | | Difficulty of building it yourself | Moderate | High: the slot's transverse stiffness has to be tuned experimentally | | Balanset-1A OEM | Standard application per the manual | Requires separate verification: the accelerometer needs the support to actually move | > A soft-bearing rig's support natural frequency depends on the mass of the mounted rotor. That means your working speed range will differ for a light blower and a heavy drum. Calculate the speed window for the extreme masses in your part range, not for one average part. ## Supports: The Assembly That Determines Everything Else The support is the rig's moving part that carries the rotor and passes its vibration on to the sensor. The requirements on it are contradictory: it needs to be compliant in the measurement direction and rigid in every other direction, while holding its own coaxial alignment and not losing its geometry over years of use. The most common solution for homemade rigs is flat (leaf) springs. Two steel plates connect the lower base plate to the upper one, which carries the support assembly. Always make the springs from spring steel or good-quality alloy steel. Ordinary structural steel with a low elastic limit picks up permanent deformation under static and dynamic load, its coaxial alignment drifts off, and in the worst case the support loses stability. A band suspension for the carriage gives a very low natural frequency, on the order of 1–2 Hz, and therefore a wide working speed range. What matters here is choosing the band's thickness and width: it has to hold the rotor's weight while also ruling out torsional oscillation of the suspension, which shows up as face runout. Cylindrical compression springs under the plate are the simplest to make, but they carry a built-in flaw. With an asymmetric rotor, the front and rear springs compress by different amounts, the rotor axis tilts in the vertical plane, and forces appear that push the rotor in the axial direction. - [x] Stiffness and its direction: compliance only in the measurement plane — the support must be rigid crosswise and vertically. - [x] Suspension natural frequency in the transverse direction: at least 2–3 times below the running speed. - [x] Natural frequency in the axial direction: it caps the range from above — keep the running speed no higher than 0.3–0.5 of it. - [x] Matching frequencies across all the machine's supports: a spread between the spindle supports and the intermediate ones narrows the working speed window down to the worst case. - [x] Friction: any dry friction and play turn into scatter in amplitude and phase from run to run. - [x] Adjustment: in height for setting the support's coaxial alignment, and along the bed's length for different rotor spans. - [x] Axial force: the support has to absorb the axial force from the drive and from misalignment without its geometry drifting. - [x] Lock: a mechanism to fix the support during run-up and coast-down, or the suspension wears itself out passing through resonance. - [x] Rigid-support thickness as a guideline: about 20 mm for rotors up to 50–100 kg, 30–40 mm up to 300–500 kg, 50–60 mm and more for rotors from 1000 kg. | Support type | Natural frequency, guideline | Where it applies | What to watch for | | --- | --- | --- | --- | | Flat (leaf) springs | 10–15 Hz at typical dimensions of 200–300 mm in height and 3 mm thick | Shafts, driveshafts, rotors up to a few hundred kilograms | Steel grade, matching natural frequencies across all the machine's supports, the locking mechanism | | Band suspension for the carriage | 1–2 Hz | Wide speed range, balancing from 200 rpm, small turbines and tooling | Band thickness and width, torsional oscillation of the suspension | | Plate on cylindrical springs | Tuned to 2–3 times below the running speed | Rigs for a fully assembled mechanism: fans, pumps, abrasive wheels | Uneven spring compression, axis tilt, axial rotor shift | | Rigid plate with a U-shaped slot | Above 100 Hz in the transverse direction | Below-resonance machines, a wide range of rotor masses | Plate thickness, slot depth, and web width are tuned by trial and error | | Rocking frame on bearings | Set by the frame's elastic element | Compact rigs for small rotors | Friction and play in the hinges directly ruin repeatability | > Adjusting the natural frequency of a leaf-spring support isn't hard. Mill longitudinal or transverse slots into the spring, lower the stiffness, remeasure the frequency. This is how spindle-support and intermediate-support frequencies get matched after assembly. ## Suspension Natural Frequency: Calculate First, Then Measure An estimating calculation is done using a single-degree-of-freedom model: a mass on a spring. The natural frequency equals the square root of stiffness over mass, divided by 2π. For a symmetric between-bearings rotor, take the mass as the support's moving-part mass plus the rotor's mass divided by the number of supports involved. Stiffness is easier to get experimentally: load the support with a static force using a force gauge, measure the displacement with a dial indicator, and divide the force by the deformation. Two rules follow from this formula, and they're easy to forget. Frequency depends on stiffness and mass through a square root: doubling the stiffness raises the frequency by only a factor of 1.4, and doubling the moving-part mass lowers it by that same factor of 1.4. And second: a calculation from a simplified model carries a large error, so always get the final figure by measurement. Two methods are used to measure it. Impact excitation works for both soft and rigid supports. Determination from coast-down only applies to the soft-bearing scheme: on a hard-bearing machine, the running speed is deliberately below the support's natural frequency, and coast-down won't be able to excite it. 1. **Calculation** — Estimate the support's stiffness by static loading and calculate the expected natural frequency. This is so you know what order of magnitude you're looking for. 2. **Impact** — Mount the vibration sensor so its measurement axis lines up with the direction of the impact. Strike it with a rubber-faced hammer or mallet whose mass is about 10% of the excited assembly's mass. Record the time trace and the spectrum of the decaying oscillation: the peak in the spectrum is the natural frequency. 3. **Two directions** — Repeat the impact in the transverse direction (where you measure unbalance) and in the axial direction, along the rotor's rotation axis. On soft supports, the axial frequency often turns out to be the limiting factor from above. 4. **Coast-down** — Spin the rotor up to maximum, disconnect the drive, and record the amplitude and phase of the running-speed component — vibration at the rotation frequency — through the cycle. Resonance gives itself away with a local spike in amplitude and a sharp roughly 180° phase reversal. 5. **Speed window** — Set the working range: no lower than 2–3 times the transverse natural frequency, and no higher than 0.3–0.5 of the axial natural frequency. Check that the window stays viable for both the lightest and heaviest part in your range. > An example from practice. A leaf-spring support measuring 300×200×3 mm gave 11–12 Hz unloaded. That means balancing on this machine can't go below 22–24 Hz — roughly 1320–1440 rpm. The intermediate supports on the same machine, with 200×200×3 mm springs, showed 13–14 Hz, and their frequencies had to be matched to the spindle supports. ## Bearing Assemblies and the Sensor Location The bearing assembly is what the rotor actually spins on. It determines friction, the repeatability of location, and runout that no amount of weights will ever balance out. Give separate thought to the sensor pad. Mount the vibration sensor on the support's moving part, as close as possible to the line of load action, on a clean, flat, machined surface, using a stud or a magnet. The measurement direction is usually horizontal-radial, perpendicular to the shaft, and it needs to stay the same from one reading to the next. A non-rigid pad, a bracket made from strip stock, a sensor on a cover or a guard — these all introduce their own resonances into the measurement band and phase scatter. ### Prism (V-block) assemblies Simple and cheap, usually for rotors up to 50–100 kg. Often made from phenolic laminate, PTFE, or nylon. One serious drawback: sliding friction on the journal. It makes spin-up harder and hits repeatability directly. A good solution is two cylindrical bushings set at an angle to each other: contact along a line, a minimal friction zone, and when worn, the bushing gets rotated a bit further around its axis. ### Grounding the rotor If the assembly is non-metallic, be sure to provide a ground connection from the rotor to the machine frame. Otherwise static charge builds up on the rotating part. That means both noise in the measurement channel and a risk to the operator. ### Roller assemblies The main choice for rotors 50 kg and up: rolling friction instead of sliding friction, easier spin-up, less journal wear. The rollers are built on standard rolling bearings, whose outer races turn on fixed axles. For heavy rigs, self-aligning rollers with two extra angular degrees of freedom are used. ### Roller runout The tolerance on a roller's radial runout is tight: no more than 3–5 µm. Check it with a dial indicator on every roller before assembly, and periodically during operation. Experience shows that even new, bought-in sets sometimes come with 10–11 µm of runout, and that alone puts a ceiling on the achievable accuracy. ### Spindle assemblies Needed for rotors with a flange mount. Take the distance between the bearing supports as four to five times the spindle diameter at the front support. The shaft diameter accounts for 70–80% of the assembly's stiffness, so trying to raise stiffness with pricier bearings alone gets you almost nothing. For mid-sized machines, aim for a stiffness of at least 50 kg/µm and a spindle natural frequency of at least 500–600 Hz. ### Spindle balance The spindle's own unbalance carries over onto the workpiece as a systematic error. Balance the spindle in two planes to a grade 1–2 steps tighter than the parts you'll be balancing on it. The same goes for the motor with its pulley and any intermediate pulleys. - [x] Roller and bearing fits with no play, with runout checked after assembly, not taken from the bearing's data sheet. - [x] Assembly adjustment in height, for setting support alignment, and along the bed's length, for different rotor spans. - [x] A machined pad for the vibration sensor on the support's moving part, close to the line of load. - [x] The same measurement direction at every support, fixed in the operator's instructions. - [x] The sensor cable is secured and doesn't act as a spring: a dangling cable adds its own oscillation. - [x] Repeatable rotor location: a stop, a mark, a fixed position, so the part seats into the supports the same way every time. - [x] Spindle-bearing outer-race temperature no higher than 70°C, and for high accuracy, no higher than 40–45°C. ## The Bed, Installation, and Vibration Isolation The bed holds the whole rig's geometry. Four requirements apply to it: bending and torsional stiffness, geometric accuracy of the ways, vibration resistance, and wear resistance of the ways. The bed's mass needs to noticeably exceed the mass of the heaviest rotor in your range, or the supports will start rocking the base, and you'll end up measuring the structure's behavior instead of unbalance. The fastest route for a shop is a cast-iron bed from a used metalworking or woodworking machine. You get a rigid load-bearing frame with ways right away, and any type of support stand can be mounted on them. The main work comes down to restoring the ways' geometric accuracy, which in shop conditions is done by fine scraping. Welded and bolted-up beds made from channel steel also work, but demand discipline. After welding, stress-relief heat treatment is needed, or the bed's geometry will drift within six months. A bolted joint lets you skip that step, since it doesn't introduce welding distortion. Either way, the channel's top flanges will need machining: grinding or fine milling for the ways. - Polymer concrete with a vibration-damping coating: a higher damping coefficient, lower thermal conductivity and thermal deformation, no internal stresses, doesn't corrode. The top section is reinforced with steel inserts for the ways. - Floor mounting: either rigidly anchored into the foundation, or on vibration isolators. Rigid mounting holds the geometry better; isolators protect better against external interference from the shop floor. - Bed resonances: check them with the same impact method as the supports. A bed resonance peak inside the working speed range produces unstable amplitude and phase, and no software setting fixes that. - External sources: a press, an overhead crane, a compressor near the rig all produce vibration at frequencies that fall inside the measurement band. Synchronous filtering suppresses them, but not without limit. - The ways: their length and accuracy determine what rotor span you can service and how repeatably the intermediate supports land in the same spot. > Geometry guidelines worth aiming for at acceptance: bed leveling in the longitudinal and transverse directions no worse than 0.02–0.04 mm/m, straightness deviation of the ways around 10 µm at lengths up to 500 mm, 15 µm up to 1000 mm, and 20 µm up to 2000 mm, and way twist no more than 10 µm. These are working guidelines for a homemade rig, not normative requirements. ## The Drive: How to Spin the Rotor Without Ruining the Measurement The drive exists for one purpose: to bring the rotor up to a stable speed. Anything it adds to the vibration goes straight into the error. Hence a simple rule: the less the drive is involved in the measurement, the cleaner the result. The overwhelming majority of homemade rigs are built on an induction motor with a VFD. It's cheap and gives a wide adjustment range. Rough main-drive power ratings: 0.25–0.72 kW for rotors up to 5 kg, 0.72–1.2 kW up to 50 kg, 1.2–1.5 kW up to 100 kg, 1.5–2.2 kW up to 500 kg, 2.2–5 kW up to 1000 kg, 5–7.5 kW up to 3000 kg. The type of drive transmission affects the measurement more than it seems to. For spindle assemblies, a flat-belt drive is preferred. V-belts and toothed belts introduce extra dynamic loads from geometric errors in the belt and pulleys, and those loads arrive at the supports right alongside the useful signal. Mount the drive pulley on the spindle's rear end, as close to the bearing assembly as possible, with minimal overhang. A pulley set out on a long overhang increases the dynamic-load moment on the spindle's supports. - [x] The motor is rigidly mounted to the bed or the foundation, not to the support's moving part. - [x] The motor, together with its mounted pulley, is balanced before installation on the machine, or in place after installation. - [x] Drive-pulley radial runout within 20 µm: the measurement system reads pulley runout as unbalance. - [x] The belt is tensioned so it doesn't slip: slipping produces speed jumps, and the synchronous filter's tuning and phase accuracy both depend on speed. - [x] Line filters sit on the VFD's input and output, either factory-made or homemade on ferrite rings. - [x] The machine is reliably grounded: without grounding, the VFD induces interference directly into the measurement channel. - [x] Run-up and coast-down happen with the supports locked, if a locking mechanism is provided. - [x] A removable drive or coast-down measurement is provided as a way to escape drive interference entirely. > Coast-down measurement is a legitimate technique for a soft-bearing rig. You spin the rotor above the running speed, disconnect the drive, and record amplitude and phase at the target rotation frequency with the motor off. Interference from the belt, pulley, and VFD disappears in the process. The trade-off is that the speed keeps falling continuously, and on low-speed rotors there may not be enough time left for synchronous averaging. ## The Measurement System: Sensors, Marker, Two Channels The measurement system has one job: precisely measure the amplitude and phase of the vibration's running-speed component at the rotor's rotation frequency. Everything else is a means to that end. The useful signal has to be pulled out of a mix of interference, so the signal chain runs, in sequence, through broadband filtering, amplification, integration, a narrowband tracking filter, analog-to-digital conversion, synchronous filtering, and harmonic analysis. Sensible bandpass-filter limits for a balancing instrument are 2–3 Hz on the low end and 50–100 Hz on the high end. Filtering the low end suppresses low-frequency amplifier noise; filtering the high end removes combination frequencies and resonances of individual machine components. Synchronous filtering works as averaging across revolutions: the useful signal's energy grows in proportion to the square of the number of averaging cycles, while the noise energy grows only in proportion to their number. The cost is that at low speed the measurement takes time: at 120 rpm and 64 averaging cycles, a single reading takes more than 30 seconds. Two channels at once aren't a convenience — they're needed because a two-plane problem is solved as a system of equations across both supports at once. Measure the supports one after the other, and you get two readings taken in different states of the system, and the influence coefficients end up calculated from inconsistent data. ### Accelerometers The main choice for a soft-bearing rig. A sensitivity of 10–30 mV/(m/s²) covers most jobs; for especially precise balancing, 100 mV/(m/s²) and up is used. The signal is integrated in the chain into vibration velocity, and on low-speed rigs sometimes twice, into vibration displacement. With capacitive accelerometers, low-frequency noise in the 0.5–3 Hz band creeps in after integration, and that caps the lower speed limit. ### Vibration velocity sensors Induction (velocity) sensors give a high sensitivity — tens to hundreds of mV per mm/s, and versions with mechanical amplification even more. They're labor-intensive to make, so they're rare on homemade rigs. Piezoelectric vibration-velocity sensors with a built-in charge amplifier and integrator are simpler to use. ### Non-contact displacement sensors Capacitive and inductive sensors work from 0 Hz, which matters for low-speed rotors at 120 rpm and below. Sensitivity reaches 1000 mV/mm and up. The limitation is the working gap: it's tied to the coil diameter and is usually a few millimeters. ### Force sensors Used in the rigid, below-resonance scheme, where the support essentially doesn't move. Piezoelectric and strain-gauge sensors are clamped into the support's slot, or bolted against its rigid section through a flat washer for even pressure across the sensor's face. ### Phase-angle sensor Without it there's no extracting the running-speed component and no reading the phase. A practical option is a laser (photoelectric) sensor aimed at a reflective marker. The Balanset-1A OEM has a range of 100–100,000 rpm and a phase measurement error of ±1°. Where optics aren't desirable, an inductive non-contact sensor is used instead. ### Marker and the zero reference The marker is placed on the rotor if it's driven directly by a belt, and on the drive spindle if the rotor is located by its flange. Align the marker's position with the machine's angular-scale zero. A simple check: slowly turn the rotor by hand and confirm the beam lands exactly on the marker and the tooling doesn't block it. - What the measurement core's software gives you: correction calculation in one and two planes, influence coefficients saved for a specific rotor type. - The 1x running-speed component with phase, separate from overall vibration, the FFT spectrum, and a polar diagram for checking that you're actually fighting unbalance. - A target residual vibration value and a within-tolerance flag, plus tolerance calculation against grades G. - Fixed positions: the software gives a position number and a mass instead of an angle, which removes the operator's most common mistake with the reference direction. - Splitting the weight between two adjacent positions, and calculating drilling parameters when metal is removed instead of added. - A measurement archive and report printing: without this a rig gives you no traceability, and you can't pull up the history for a specific part. - Location-eccentricity compensation using the rotor-flip method: it subtracts the arbor's or spindle's contribution from the result. > The Balanset-1A's measurement band per the manual is 5–200 Hz for vibration velocity RMS, with a running-speed-component range of 0.02–80 mm/s. Assessing machine condition per the ISO 20816 series usually requires a broadband measurement in the 10–1000 Hz band. These are different things, and if your acceptance criteria are tied to the second band, check applicability in advance. Confirm the applicable part and edition of the standard separately. Sources: [Balanset-1A operation manual](https://vibromera.eu/balanset-1a-operation-manual/) · [Balanset-1A manufacturer specification](https://vibromera.eu/product/balanset-1/) · [ISO 20816-1:2016](https://www.iso.org/standard/63180.html) ## Arbors, Location, and Eccentricity Compensation Your rig's accuracy comes down to rotor location, not electronics. If the arbor has eccentricity, the rotor sits on it with an axis offset, and the measurement system will honestly show that eccentricity as unbalance. You'll balance the part together with the arbor's error, take it off the rig, put it in the machine, and get vibration. The same goes for the spindle. Its radial and face runout carries straight into the result. A guideline tolerance: no more than 5 µm on both parameters, if you want to skip an extra 180° rotor flip. For mid-sized machines, hold the bearing assemblies' alignment within 10–15 µm. There's a standard technique against eccentricity: balancing with a rotor flip, which measurement systems call eccentricity compensation. The idea is simple. You take a reading, flip the rotor on the arbor 180°, and take a second reading. The rotor's own unbalance rotates along with the flip, while the arbor's eccentricity stays put. The vector difference lets you separate the two contributions and subtract the location error from the result. - [x] The arbor has minimal runout and is itself balanced in two planes, to a grade 1–2 steps tighter than the part. - [x] The part's fit on the arbor is repeatable: a taper, a shoulder, a stop — not "by eye, hand-tight." - [x] The part's clamping is torque-specified: different torque means a different fit and a different result. - [x] The marker on the rotor or spindle is tied to the machine's angular-scale zero. - [x] For parts with a tight residual-unbalance requirement, an eccentricity-compensation mode is provided, not just a direct reading. - [x] The correction planes are accessible with the rotor stopped, without removing the part from the arbor, or every iteration breaks the location. - [x] The weight-mounting radius is marked and recorded: mount the correction at the same radius as the trial weight. > A practical limit. If the arbor's runout accounts for a noticeable share of the required residual unbalance, no measurement system will save you: you'll be capped by the tooling. Work out this figure before building the rig, because it determines what accuracy grade is even achievable with your construction. ## Rig Acceptance: Geometry and Dynamics, Step by Step A homemade rig is a one-off piece of equipment with no production documentation. That's why it needs a thorough check at commissioning, and the order here matters: geometry first, then dynamics, then the measurement channel, and only after that, accuracy on a reference rotor. Checking accuracy before the geometry is set is pointless. For tools you'll need a machinist's level (spirit or electronic), a straightedge, a stand with a dial indicator, a measuring bridge, a reference arbor, and a force gauge. All of this is available to any repair shop. 1. **Installation and leveling** — Anchor the bed into the foundation, or set it on vibration isolators. Level it in the longitudinal and transverse directions, taking readings at points evenly spaced along the bed's length. 2. **Ways** — Check the ways' straightness with a straightedge and indicator, or with a level advanced in steps equal to its base length. The deviation equals the largest algebraic difference between readings across all sections. Separately check for twist using a level set crosswise on the ways. 3. **Spindles and pulleys** — Measure the radial and face runout of the drive and driven spindles with a dial indicator, on the cylindrical and face surfaces of the locating washer. Then measure the drive pulley's radial runout. 4. **Rollers** — Check each roller's radial runout on its cylindrical surface. While you're at it, look for nicks and raceway damage, which can appear both during manufacturing and in service. 5. **Support alignment** — Set the reference arbor into the bearing assemblies of the front and rear supports. Locate the bridge with the indicator stand against the flat and side surfaces of the ways, and roll it back and forth between the supports. Take measurements in two mutually perpendicular planes, vertical and horizontal. 6. **Natural frequencies** — Use the impact method to determine the natural frequencies of every support in the transverse and axial directions, and of the bed too if needed. Check that the supports' frequencies are close to each other. 7. **Coast-down** — Spin the rotor up and record the 1x amplitude and phase on coast-down. Confirm that within the assigned working range there's neither an amplitude spike nor a phase reversal. Only after that, fix the working speed in the instructions. 8. **Measurement channel** — Record the time trace on both vibration channels and the phase-angle channel. The sine waves should be clean, and their period should match the marker pulse rate. 9. **Stability** — Take three readings in a row on the same rotor without stopping and compare them: the speed shouldn't deviate from the average by more than 1–2 rpm, the running-speed-component magnitude by more than 5%, or the phase by more than 1–2°. | What's checked | Working guideline | What to measure with | | --- | --- | --- | | Bed leveling | 0.02–0.04 mm/m in both directions | Machinist's level | | Straightness of the ways | 10 µm up to 500 mm, 15 µm up to 1000 mm, 20 µm up to 2000 mm | Straightedge with indicator, or level | | Way twist | up to 10 µm | Level set crosswise on the ways | | Spindle radial and face runout | up to 5 µm | Dial indicator | | Drive pulley radial runout | up to 20 µm | Dial indicator | | Roller radial runout | 3–5 µm | Dial indicator | | Bearing-assembly alignment | 10–15 µm for mid-sized machines | Reference arbor, bridge, indicator | | Speed spread between readings | 1–2 rpm | Measurement system and tachometer | | 1x magnitude spread between readings | up to 5% | Measurement system | | 1x phase spread between readings | 1–2° | Measurement system | > Unstable readings almost always have one of two causes. Electrical: no grounding, no filters on the VFD, welding running on the same mains nearby. Mechanical: support or bed resonance, worn rollers and raceways, play in the spindle bearings, pulley runout, ovality or nicks on the rotor's own journals. A convenient way to tell them apart: drive interference disappears on coast-down, a mechanical defect stays. ## Verifying Accuracy on a Reference Rotor This is the step that homemade rigs skip most often, and it's exactly what separates a verified machine from an unverified one. The idea is to take a rotor with a known, small residual unbalance, introduce known trial unbalances into it, and see how consistently the machine finds them. Ideally the reference rotor should be a certified standard, but in shop conditions a carefully balanced production part will do. The requirement on it: its residual unbalance shouldn't exceed ten times the minimum residual unbalance achievable on your machine. The test rotor's mass is taken as 0.2–0.3 of the maximum mass you're designing the rig for. Verification runs in two stages. First, iterative convergence: the standard procedure with trial runs, followed by a few more cycles that drive the residual unbalance down to the limit of the machine's capability. In addition, the marker gets shifted by 60° and a verification reading is taken in the new angular coordinate system: this checks that the result doesn't depend on where the zero sits. The second stage is walking a reference weight around the rotor. A weight of known mass is moved through a sequence of angles in 30° steps — 12 runs — or 60° steps for a shortened check. For each correction plane, calculate the average of the measured unbalance and the ratio of each reading to that average. The spread of these relative values is the honest characterization of your machine. As a guideline drawn from standard practice, a band of 0.88 to 1.12 is used. 1. **Preparation** — Balance the reference rotor down to a state where its residual unbalance doesn't exceed ten times the minimum. Apply a marker aligned with the machine's scale zero. 2. **Trial weight** — Calculate the trial weight's mass from the stated relative residual unbalance, the rotor's mass, and the mounting radius. Weigh the weight and enter the actual mass and radius into the software. 3. **Iterations** — Run the standard procedure with setup runs, then several successive-convergence cycles. Log the initial and running residual unbalance values on a verification sheet. 4. **Changing the zero** — Shift the marker by 60° and repeat the reading. The result shouldn't drift: if it does, look for location runout and errors in the angle reference. 5. **Walking the weight around** — Mount the reference weight in sequence at 12 (or 6) angular positions and record the measured unbalance for both correction planes at each position. 6. **Processing** — Calculate the average values and relative deviations, and plot a chart for each plane. A spread that falls outside the band means a defect remains somewhere in the machine — not that the tolerance in the data sheet needs to be loosened. 7. **Eccentricity compensation** — Separately check the eccentricity-compensation chain: simulate the rotor flip by moving the reference weights 180°. Once this mode is working correctly, the residual unbalance should drop to a few percent of the reference unbalance introduced. > What to do if verification fails. Don't rewrite the tolerance in the data sheet until you've worked out the cause. The troubleshooting order runs opposite to the acceptance order: measurement channel and grounding first, then resonances, then roller and spindle runout, then location and the arbor, and only then the support construction itself. Sources: [ISO 21940-11:2016](https://www.iso.org/standard/54074.html) · [Balanset-1A operation manual](https://vibromera.eu/balanset-1a-operation-manual/) ## Long and Multi-Support Rotors Two correction planes are enough for a rigid rotor that doesn't deform at the running speed. The geometric threshold is simple: at a length-to-diameter ratio up to about 0.5, the part behaves like a disc, and one plane is often enough. Above that ratio you need two, or couple unbalance will remain — a pair of forces that twists the rotor — and vibration at the second support won't go away. From there it becomes a different problem. A composite shaft on three or four supports is balanced in three or four planes, and it has to be solved as a system of equations with three or four unknowns, using the influence coefficients of all the supports. Every trial run fills one column of the matrix: you mount a trial weight in one plane and record the response at all the supports at once. For three planes that's a minimum of four runs; for four planes, five. A flexible rotor, one that changes shape while running, stands apart entirely. Here the two- and three-plane rigid-rotor scheme doesn't work at all, and modal methods are used instead. That's no longer a job for a homemade rig, and it's worth honestly labeling it as such. - A rig for a multi-support shaft requires longitudinal travel of the intermediate supports along the ways, and fixtures for adjusting the intermediate bearing assemblies in height. - All the supports on a rig like this need to have close natural frequencies: a spread narrows the working speed window down to the worst support. - A two-channel measurement system solves the problem for two planes. Three and four planes require a matching number of channels and a different calculation algorithm. - A practical workaround for three planes on a two-channel system: break the problem down into stages by shaft section, though repeatability suffers, and the result needs to be checked with a verification run across all the supports. - The classification of rotors into rigid and flexible, along with approaches to balancing flexible rotors, is described in the ISO 21940 series. Check the applicable part and edition separately. Sources: [ISO 21940-11:2016](https://www.iso.org/standard/54074.html) · [ISO 21940-12:2016](https://www.iso.org/standard/50429.html) ## Safety and Common Mistakes in Homemade Rigs A balancing rig is a machine with a spinning mass and removable weights bolted onto it. A correction weight flying off at running speed isn't a hypothetical risk. Build guarding and a work procedure into the design from the very first drawing, not after the first incident. ### A flexible bed A light welded frame under a heavy rotor. The supports start rocking the base, the amplitude depends on where the rig happens to be sitting, and the result doesn't repeat from one setup to the next. The cure is mass and stiffness, not software settings. ### Suspension resonance inside the working range The most common mistake. Nobody measured the support's natural frequency, and the speed was chosen for the drive's convenience. In resonance the phase drifts by up to 180° from a small speed change, and the correction calculation turns into a lottery. ### Friction in the supports Prism assemblies with no attention paid to the contact zone, dry or worn rollers, play in the hinges of a rocking frame. The rotor lands in a slightly different position every time, and you get scatter that gets blamed on "the instrument is lying." ### Inconsistent location The part gets mounted on the arbor differently each time, torque isn't specified, there's no stop. Repeatability from one setup to the next disappears, and with it the whole point of saved influence coefficients. ### Sensor on a non-rigid pad A bracket made from strip stock, a magnet on paint, a sensor on a guard. The mount ends up with a natural frequency inside the measurement band, and you're measuring the bracket, not the support. ### Unstable speed A slipping belt, an undersized motor, a VFD with no filters. Speed wanders, the synchronous filter tunes imprecisely, the phase gets noisy. Check it by comparing the software's speed reading against an independent tachometer. ### No verification on a reference rotor The rig is assembled, it shows something, parts get balanced somehow. There's nothing to back up an accuracy claim, nothing to catch degradation with, and the first contested acceptance ends in a dispute. Verification on a reference rotor is done once at commissioning, then periodically after that. ### Mismatched support frequencies Spindle supports on one set of springs, intermediate ones on another. The working speed window shrinks down to the worst support, and readings become unstable across part of the range for no apparent reason. - [x] Guarding around the entire rotating zone, with an interlock: the drive won't start with the guard open. - [x] An emergency stop within the operator's arm's reach, separate from the VFD's buttons. - [x] Software and mechanical limits on maximum speed, set for the heaviest and largest-diameter part. - [x] Mount trial and correction weights as securely as a permanent one: welding, a bolt, a pin — not putty and not tape. - [x] Sanity-check the trial weight's mass: it shouldn't create dangerous vibration, overload the supports, or touch any stationary parts. - [x] Securing the rotor on the arbor with torque control and an axial stop. - [x] Locking the supports during run-up and coast-down, if the running speed is above the suspension's resonance. - [x] Grounding the rotor and the machine, especially with non-metallic bearing assemblies. - [x] A separate procedure for coast-down: the operator doesn't approach the rotor until it's fully stopped, not "once it's nearly there." ## What You Can Honestly Claim, and How AXILINE Can Help Keep two claims separate. The first: "our rig delivers a repeatable result and is fit for internal control." You can confirm this yourself by passing acceptance testing on geometry, dynamics, and the reference-weight walk-around. The second: "our rig achieves accuracy grade X and we issue reports for external acceptance." This requires a verified procedure, a reference rotor, and agreement with whichever party accepts the report. Don't blur the two together in correspondence with a customer. A separate note on what "within tolerance" means in the software. It means exactly one thing: the residual running-speed component has dropped below the target value you entered yourself. It's not confirmation of grade G per the ISO 21940 series, and it's not an assessment of the machine per the ISO 20816 series. Three criteria exist in parallel: residual 1x against your own setpoint, residual unbalance in g·mm/kg against grades G, and overall vibration velocity against the zones. Check the applicable part and edition of the standard separately in each case. Balanset instruments are designed and manufactured by engineers who do their own on-site balancing with them. In practice that means questions like "where will the sensor land on a support like this," "will one plane be enough for this impeller," or "what speed window is left with a mass spread from 3 to 40 kg" get answered directly, not out of a catalog. We provide consulting support on choosing the measurement configuration, the sensor and marker mounting scheme, software setup, and training your operator. Where to start the conversation. Send us the mass and diameter range for your rotors, their length-to-diameter ratio, the required accuracy grade, the planned speed, the drive type, and a sketch or photos of the intended support construction. That's enough for us to tell you whether the Balanset-1A OEM will fit your scheme as a standard application, where you'll be capped by the tooling, and what to check before you build. If your data shows a rig isn't needed, we'll say so: for machines already sitting on their foundation, on-site balancing in their own bearing housings remains the more honest choice. > A middle path people forget about. Start with on-site balancing of your own part range using a full Balanset-1A kit. You'll get figures on initial unbalance, on the number of planes needed, and on achievable residual values, you'll see the instrument working on your own parts, and only then decide whether to build a rig and what it should look like. Sources: [Balanset-1A operation manual](https://vibromera.eu/balanset-1a-operation-manual/) · [Balanset-1A manufacturer specification](https://vibromera.eu/product/balanset-1/) · [ISO 21940-11:2016](https://www.iso.org/standard/54074.html) · [ISO 20816-1:2016](https://www.iso.org/standard/63180.html) ## Frequently asked questions **Can you build a proper balancing rig without an engineering department?** Yes, if you go with the soft-bearing scheme and don't try to cover the whole rotor range from the start. A support on leaf or band springs forgives calculation inaccuracy, because its natural frequency is easy to measure with the impact method and easy to correct by milling slots into the spring. The rigid, below-resonance scheme requires tuning the support slot's transverse stiffness experimentally — there's no ready-made calculation method for it, and without experience it's an expensive iteration. A realistic plan looks like this: a narrow part range, one support size, a ready-made measurement core, and full acceptance testing against a reference rotor. Extend the mass range with a second rig rather than reworking the first one. **How do you choose the rig's working speed?** Not by the drive's convenience — by the supports' natural frequencies. For the soft-bearing scheme, the running speed needs to be at least 2–3 times above the support's transverse natural frequency, and no higher than 0.3–0.5 of its axial natural frequency. Both frequencies are measured with the impact method, separately for each direction and each support. If the supports' frequencies differ, base the window on the worst one. And remember that the natural frequency depends on the mass of the mounted rotor: calculate the window for both the lightest and the heaviest part, or you'll find there's simply no working range for part of your range. **What vibration sensor should go on a homemade rig?** For the soft-bearing scheme, an accelerometer. A sensitivity of 10–30 mV/(m/s²) covers most jobs; for high accuracy, 100 mV/(m/s²) and up is used. The Balanset-1A OEM ships with two single-axis accelerometers at around 20–30 mV/(mm/s), and the entire chain (preamplifiers, integrators, ADC) is matched to exactly these, so we recommend the standard sensors rather than arbitrary ones. For the rigid, below-resonance scheme, an accelerometer doesn't fit: the support barely moves, and you need force sensors or non-contact displacement sensors there instead. For very low speeds, 120 rpm and below, non-contact displacement sensors win, because they work from 0 Hz. **Why do you need two channels at once, instead of one sensor you move around?** Because a two-plane problem is solved as a system of equations across both supports at once. The software needs each support's response to the trial weight in each plane, taken in one and the same state of the system. If you measure the supports one after another, speed, temperature, belt tension, and part fit all drift between readings, and the influence coefficients end up calculated from inconsistent data. This error doesn't show up as an obvious glitch: the result simply stops converging in one iteration, and you start adding weights by guesswork. **What does eccentricity compensation give you, and is it mandatory?** It subtracts the arbor's or spindle's contribution from the result. You take a reading, flip the rotor on the arbor 180°, and take a second one. The rotor's own unbalance rotates along with the flip, while the location's eccentricity stays put, and the vector difference lets you separate the two. It's mandatory whenever your arbor's runout accounts for a noticeable share of the required residual unbalance. For coarse accuracy grades you can skip it; for precision parts, without compensation you'll simply balance the part together with the tooling's error and end up with vibration once it's in the machine. **Can you issue a customer a report with results from your own rig?** You can issue a report as an internal-control document with the actual measured residual-unbalance values, and that's normal practice. Declaring the machine's accuracy grade and issuing a report for external acceptance is a different matter: that requires a verified test procedure, a reference rotor, and agreement with the accepting party. The minimum that makes your report defensible: passed acceptance testing on geometry and dynamics, a recorded result from the reference-weight walk-around, and the report stating speed, weight-mounting radius, number of planes, support type, and the date of the machine's last verification. The metrological status required for mandatory calibration depends on your industry's requirements and national legislation, and that needs sorting out before you promise the document. **A rig from scratch or a retrofit of an old machine: which is cheaper?** If you have sound mechanics already, a retrofit is almost always the better deal. A cast bed and supports outlive several generations of electronics, and there's no reason to pay twice for iron that's already working for you. The typical scope of retrofit work is small: machine sensor pads, mount a bracket for the laser phase sensor, apply a marker, get the drive and belt into shape, check the support fastener torque. A rig from scratch makes sense when there's no old machine to work with, or when its scheme is fundamentally wrong for the job — say, the supports are rigid and essentially motionless, but you're planning to measure their oscillation with accelerometers. **How long does commissioning a rig take?** The honest answer: the mechanics get done faster than the acceptance testing. Assembling and installing the measurement side takes days, while setting the geometry, measuring every assembly's natural frequency, assigning the working speed window, checking reading stability, and walking a reference weight through 12 angular positions in two planes is separate work with its own iterations. Plan it as a project stage, not as "flip it on and see." This is exactly the stage where a flexible bed, suspension resonance, roller runout, and inconsistent location get caught — everything that later can't be fixed with software settings.