Vibration is low-amplitude, high-cycle loading. A case may see tens of millions of cycles a year, at accelerations far too small to feel, and that loading does three things.
It fatigues joints and old repairs — adhesives and fills behave differently from the parent material, and a repair that survives one handling event may not survive a million small cycles. It abrades contact interfaces — a stone base on a metal pin, a ceramic on a felt pad; a few micrometres of relative motion per cycle is enough to grind a surface, which is why damage on mounted objects clusters at the fixing points. And it causes fasteners to creep — threads, crimps and wire mounts that were firm at installation can be hand-loose five years later.
Worse, every object has natural frequencies. Thin-walled ceramics, detached panels, a cracked bell — these can amplify floor motion by a factor of ten or more at their own resonance. A floor that satisfies a building code can still be hostile to one specific object.
Most people worry about visitors walking past. They are usually the least of it.
Road traffic, including buses and heavy goods vehicles, sits at four to twenty hertz, random, peaking with every vehicle. Rail and metro structure-borne noise reaches ten to sixty hertz, while ground-borne vibration through soil falls at one to ten hertz and travels a long way. Construction and piling run from one to forty hertz — high amplitude, but finite. Air handling units, chillers and pumps occupy ten to sixty hertz, with blade-pass tones on top. Lifts, doors and loading bays add transients at one to fifteen hertz, footfall sits at one to twelve, and handling or opening a case door puts two to fifty hertz straight into the object through its mount.
One problem deserves naming. A piling rig outside makes everyone nervous. A slightly unbalanced supply fan turning quietly in the ceiling void above a case goes unnoticed, and delivers hundreds of millions of cycles over a single exhibition. In our experience, chronic plant-borne vibration is the most under-diagnosed problem in new galleries — and the easiest to fix, because it is usually a commissioning issue rather than a design failure.
Glass panels, cantilevered shelves and tall plinths have their own modes, typically between ten and forty hertz, right where traffic and plant energy is concentrated. A case can be perfectly isolated at its base and still amplify floor motion at the shelf. Isolation at the floor is necessary, not sufficient.
The VC-A to VC-G family was developed for vibration-sensitive laboratory equipment and has since been widely borrowed by museum specifications. Limits are expressed as RMS velocity in one-third-octave bands, usually covering one to eighty hertz, running from fifty micrometres per second at VC-A down to below one at VC-G.
Most museum display specifications land between VC-C and VC-E, with the more sensitive end reserved for unrestrained, friable or already-damaged material.
ISO 2631 addresses human response to vibration; DIN 4150-3 addresses the risk of structural damage to buildings. Both are useful for the building and largely irrelevant to the object — the latter's limits are in millimetres per second, roughly three orders of magnitude above the levels at which collections are considered at risk.
Put plainly: a floor that passes the building code can still be unacceptable for a display case. This is the hardest point to communicate in a project, and the easiest to settle with data.
For high-value or high-fragility loans, the more reliable approach is to establish the object's own resonant frequencies — from a conservator's assessment, or a tap test on a surrogate of the same material — and set the limit at those frequencies rather than across a generic band. It costs more work, and it avoids the two classic errors: over-engineering for a robust object, and under-protecting a fragile one that happens to resonate exactly where the generic curve is most lenient.
Whichever family is used, the criterion must be written down before design begins, in units and frequency bands that can be meas
An isolator is a spring. It does not absorb vibration; it changes the frequency at which the supported mass responds. For a single-degree-of-freedom system, transmissibility depends on the ratio of forcing frequency to natural frequency, T ≈ 1 ÷ [1 − (f / fₙ)²], where f is the forcing frequency and fₙ the natural frequency; take the absolute value when the bracket is negative.
The consequences are unforgiving. At resonance, transmissibility tends to infinity — a mount that is resonant at a forcing frequency makes the problem five to twenty times worse. Isolation only begins above 1.41 times the natural frequency; below that ratio, a "vibration isolator" is an amplifier. The rule of thumb is to keep fₙ at or below one-third of the lowest significant forcing frequency, which yields roughly 89% isolation there and progressively more above it.
Static deflection relates to natural frequency as δ ≈ 248 / fₙ², with δ in millimetres and fₙ in hertz. Ten hertz needs about 2.5 mm; five hertz about 10 mm; 3.3 hertz about 28 mm; two hertz about 62 mm.
In other words, a 2 Hz vertical natural frequency requires 62 mm of static compression under load. That cannot be achieved with a 10 mm rubber pad, and it cannot be hidden inside a 40 mm plinth. Low-frequency isolation is a space problem before it is an engineering problem — which is why the isolation architecture has to be settled at concept design, not chosen after the case is drawn.
Elastomer pads, high-damping polymers and laminated steel-rubber bearings belong to one family: 6–15 Hz loaded, cheap, thin, maintenance-free, laterally stable, easy to integrate into a plinth — but limited in isolation, ineffective below 10 Hz, and with stiffness that shifts with temperature and age. If the criterion is met at 8–10 Hz and the forcing is mostly above 20 Hz, this family is the honest answer.
Wire rope isolators are the most underrated option in case design. They deliver 3–6 Hz within a 60–100 mm envelope, tolerate the load drift of a changing exhibition, have no consumable parts, include damping and cover a wide temperature range. Their weakness is amplitude-dependent stiffness: the rated natural frequency must come from a test at the actual load and excitation level, not from a catalogue curve.
Further down are steel springs (1.5–4 Hz, needing separate damping and lateral restraint, and taller) and air springs (1–3 Hz, the best practical route to low frequency for a heavy case). Air springs need an air supply or periodic topping, a leveling strategy, and a documented response to pressure loss — in a gallery that opens and closes daily, that is a maintenance commitment, not an installation step.
One common and expensive mistake deserves naming: fitting air springs to a case that does not need them. If the criterion can be met without that complexity, do not buy it.
An isolated floor platform puts a steel-framed, mass-loaded platform on isolators, with cases standing on top. It is the highest-performing option and the only one that reliably reaches below 3 Hz for a group of cases, because mass and height can be added freely. It is a building intervention, coordinated with the floor structure, fire strategy and threshold details.
Plinth-integrated isolation is the normal solution: isolators between the case and the floor, or between the case and a fixed plinth. Performance is limited by available height and case mass — a 300 kg case cannot reach 2 Hz on a 30 mm resilient layer. A realistic target is 4–8 Hz vertical, with particular attention to lateral stiffness so the case does not rock.
Object-level isolation seats the object on an isolated sub-plinth inside the case, decoupling it from the case's own resonances as well as the floor. It works well for a single fragile loan, at the cost of changing the visual presentation, complicating installation, and introducing a compliant interface that must satisfy security and seismic requirements too.
One detail decides whether any of this works.
Isolation holds only if the entire load path passes through the isolators; any parallel rigid connection short-circuits the spring. Bypasses we have found on installed cases include a rigid bracket holding the case against a wall "for stability", alarm cabling in rigid conduit running from floor to case, an earth braid bolted to a floor anchor, leveling feet wound down hard "to be safe" after installation, and a silicone skirt sealing the gap under the case. Any one of these can transmit more vibration than the isolators remove.
The response is to specify flexible services — a service loop with a compliant section, flexible conduit, a soft earth braid with slack — write "no rigid connections" on the installation drawing, and check it at commissioning rather than assume it. The same applies inside the case: objects should not lean on glazing, and shelves must sit in the isolated load path.
Before the case is designed, measure the floor it will stand on: three axes, one to eighty hertz, logged for at least 24 hours and preferably a full week to capture weekday traffic, weekend crowds and weekly plant cycles. Report one-third-octave RMS velocity and a peak-hold spectrum, and identify dominant sources by correlating peaks with events. A single afternoon will miss the delivery lorry that only comes on Tuesdays.
Ambient data tells you what is happening; forced tests tell you how the floor responds. A heel drop at one metre, a walk-past and a door slam reveal the floor's resonant frequencies and damping far more clearly than ambient noise. If plant is nearby, switch it off and on during logging — this resolves more disputes than any other single test.
At acceptance, repeat the measurement at the same points with the same settings and report the achieved attenuation band by band. Measure at the object's support surface, not the case top or the plinth, because that is where the object experiences the motion; test with services connected and the case closed, because that is the real configuration. For critical cases, add a factory shake-table test — a sine sweep from one to one hundred hertz, measured simultaneously below and above the isolators with coherence checked at every frequency. It is the only way to know the real natural frequency and damping, because an assembled case never behaves exactly like the datasheet.
Measure at handover, again after the first exhibition changeover, and again at twelve months. Elastomers creep, air springs lose pressure, fixings loosen and building plant gets rebalanced. A one-off test proves the design; only a repeat test proves the installation.
The two are routinely conflated in specifications, and they pull in opposite directions. Isolation wants a soft, low-frequency mount with small displacement; seismic protection wants either stiffness and strength to limit displacement, or a base isolation system with hundreds of millimetres of travel and explicit energy dissipation. A mount soft enough to isolate at 2 Hz will, in a seismic event, allow displacements far beyond what a case can tolerate.
In a moderate seismic zone there are three realistic answers: restraint first, with the case anchored and the object mounted against seismic forces while vibration is handled by a separate, stiffer layer — often the correct answer; combined isolation with snubbers that engage at seismic displacement, which works but must be engineered and verified as a system; or object-level restraint with case-level isolation. Whichever is chosen, the specification must state both requirements and their displacement assumptions. Design for one while specifying both, and the one nobody calculated is the one that fails.
The criterion: units (µm/s RMS velocity or acceleration), frequency bands, and whether the value is a peak or a percentile. The measurement point and condition: object support surface, services connected, case closed. The architecture and frequencies: platform, plinth-integrated or object-level, with loaded vertical and horizontal natural frequencies — tall cases are usually governed by the horizontal.
Transmissibility and static deflection: a maximum T per one-third-octave band rather than a headline number, plus the static deflection and installed height, because they determine the plinth. Load range and adjustment, covering empty, installed and fully loaded conditions, with leveling tolerance and drift allowance. And finally testing and documentation: baseline, factory test where applicable, on-site acceptance, a twelve-month re-test, and a report with spectra, calibration records and a pass/fail judgement per band.
A;For plinth-integrated systems, 4–8 Hz vertical covers most gallery criteria. Below 4 Hz the plinth becomes a significant structural element; below 3 Hz you are usually looking at an isolated platform or air springs.
A:A properly designed system will not. The softness shows up only as millimetre-scale vertical travel; lateral stiffness is designed separately, with limit stops where required. The movement from a visitor leaning on the case or a cleaning bump is far beyond the isolator's travel and is handled by snubbers — not by abandoning isolation. The real risk is winding the leveling feet down hard "for safety", which short-circuits the isolation.
A:They can, and they often become the dominant high-frequency source. Prefer passive conditioning (silica gel); where a fan is required, choose a low-speed unit on rubber mounts or soft connections and decouple the ducting from the glazing. Such equipment should be running at its highest setting during acceptance testing.
A:Three signals: site measurements show high energy in the target band, the criterion demands less than 3 Hz, or a group of cases must meet the same standard. A platform lets mass and height be added freely, at the cost of a building intervention. When a single case can meet its criterion at 4–8 Hz, plinth-integrated isolation is usually the better value.
Project