# Transfer stiffness of resilient elements (ISO 10846)

> One page of the phonometry documentation, as Markdown. Index: https://docs.jmrp.io/phonometry/llms.txt

Canonical: https://docs.jmrp.io/phonometry/vibration/structural/transfer-stiffness/
Language: en
Alternate: https://docs.jmrp.io/phonometry/es/vibration/structural/transfer-stiffness/index.md
Version: phonometry 4.0.0
Install: pip install "phonometry[full]"
Author: José Manuel Requena Plens (https://orcid.org/0000-0003-1250-6212)
Published: 2026-07-11
Updated: 2026-11-16T21:36:14Z
Standards: ISO 10846-1:2008; ISO 10846-2:2008; ISO 10846-3:2002; ISO 10846-4:2003; ISO 10846-5:2008
Verified: 20 conformance checks (ISO 10846-1:2008; ISO 10846-2:2008; ISO 10846-3:2002; ISO 10846-4:2003; ISO 10846-5:2008) https://docs.jmrp.io/phonometry/reference/conformance/#room-building-acoustics
License: MIT (https://github.com/jmrplens/phonometry/blob/main/LICENSE)
Summary: ISO 10846 dynamic transfer stiffness of resilient elements: its level and loss factor, and the direct, indirect and driving-point methods (Parts 2 to 5).

---

The vibro-acoustic transfer property of a resilient element (a vibration
isolator, mount, bellows or hose) is its **dynamic transfer stiffness**
$k_{2,1}$, the frequency-dependent ratio of the *blocking force* on the output
(receiver) side to the displacement on the input (source) side
(ISO 10846-1, 3.7):

$$
k_{2,1} = \frac{F_{2,\mathrm{b}}}{u_1} \quad [\mathrm{N/m}].
$$

Because a vibration isolator is only effective between structures of large
driving-point stiffness, the force it delivers to the receiver approximates this
blocking force (ISO 10846-1, Eq. 7), so $k_{2,1}$ is the quantity that
characterises the isolator's transmission. $k_{2,1}$ feeds the structure-borne
source and building prediction standards: ISO 9611, EN 15657 and EN 12354-5.

The whole series is written around one bench, and it is the bench that says
what the quantity means: the element is sandwiched between a driven input mass
and an output that is either rigidly blocked or loaded with a known mass, under
a stated static preload, and the two directions it is loaded in during service
are measured separately.

![ISO 10846 transfer-stiffness rigs. Top, the direct method with the isolator between a driven excitation mass and a blocked output on a force transducer, and the indirect method with the same input mass over a blocking mass on soft supports, both carrying an accelerometer at the edge of the excitation mass for the fifteen-decibel unidirectionality check. Middle, the two ways of applying the static preload: gravity from the output-side mass, and a frame with an actuator loading the element to the permissible static load while auxiliary springs decouple the blocking mass from the frame. Bottom, the transverse arrangement, with roller-bearing guiding, a force-distribution plate as the excitation mass and the output shear force summed from two transducers.](https://docs.jmrp.io/phonometry/media/diagram_transfer_stiffness_rig.svg)

## 1. The transfer-stiffness level and loss factor

Results are reported as a **level** re the reference stiffness
$k_0 = 1\ \text{N/m}$ (ISO 10846-2 and -3, 3.17), and in the low-frequency range
where inertial forces in the element are negligible the **loss factor** is the
tangent of the phase angle of $k_{2,1}$ (ISO 10846-1, 3.8):

$$
L_k = 10\log_{10}\frac{|k_{2,1}|^2}{k_0^2} = 20\log_{10}\frac{|k_{2,1}|}{k_0}, \qquad
\eta = \frac{\mathrm{Im}(k_{2,1})}{\mathrm{Re}(k_{2,1})}.
$$

```python
from phonometry import vibration

# A resilient mount with |k2,1| = 1 MN/m and a 5 % loss factor:
k = 1e6 * (1.0 + 0.05j)
print(round(float(vibration.transfer_stiffness_level(k)), 2))   # 120.01  dB re 1 N/m
print(round(float(vibration.loss_factor(k)), 3))                # 0.05
```

## 2. Direct and indirect determination

Why a *blocked* force rather than, say, the isolator's transmissibility?
Because a transmissibility is a property of a whole assembly: it changes with
whatever masses and stiffnesses the isolator happens to connect, so data
measured on one rig would not transfer to another installation. The blocked
force per unit input displacement is a property of the element alone, and it
predicts the force the element delivers to any receiver that is much stiffer
than the element itself, which is exactly the situation a vibration isolator
is designed for. ISO 10846 therefore sandwiches the isolator between a driven
input mass and an output that is either rigidly blocked or loaded with a
known mass (the rig drawn at the top of this page).

### The test conditions the number belongs to

A transfer stiffness without its preload and its temperature is not a result.
ISO 10846-1 clause 6.1 opens by saying so: $k_{2,1}$ depends on frequency, and
"in addition, it is also dependent on static preload and, in many cases, on
temperature", and may depend on relative humidity as well. The laboratory
conditions of every part of the series include the application of a static
preload, because a rubber or spring isolator characterised unloaded is working
at a different point on its load-deflection curve from the same isolator
carrying its service load.

- **Preload, applied and specified.** ISO 10846-1 clause 6.3.3.1 gives the two
  rig topologies the diagram above draws. Gravity loading, where the
  output-side mass *is* the preload, is simple but unstable for large isolators
  at high preloads; a frame with an actuator (typically hydraulic) applies the
  load instead, with auxiliary springs decoupling the blocking mass from the
  frame, which is also what makes the indirect method less vulnerable to
  flanking transmission through the frame. Elements that are not resilient
  supports are loaded in the way their service demands: a static torque for a
  flexible coupling, a representative internal pressure for a bellows or hose
  (6.3.3.2).
- **Creep.** ISO 10846-2 clause 7.1: elements containing rubber-type components
  change load or deflection through creep, so the preload shall be applied to
  **100 % of the permissible static load**, and the change of load or
  deflection due to creep should be **less than 10 % per day** before
  measurements begin.
- **Temperature.** ISO 10846-2 clause 7.6.1: measurements are made at one or
  more specified load conditions and temperatures covering the range met in
  practice; the temperature is monitored during the measurements; and the
  elements are exposed for at least **24 h** to that temperature, within
  **3 °C**, before testing.
- **Contact and configuration.** Clause 7.1 again: the element contacts the
  flanges over their whole surface (grease or double-sided tape helps, and
  large flanges may need flattening), and any device that is not part of the
  element in service is removed or de-activated.

Those conditions belong on the fiche of section 9 with the result:
`ReportMetadata` carries them in its `test_room` and `notes` fields.

### Three methods, not two

The **direct method** (ISO 10846-2) measures the blocked output force and the
input displacement, $k_{2,1} = F_{2,\mathrm{b}}/u_1$. The **indirect method**
(ISO 10846-3) loads the output with a compact blocking mass $m_2$ and measures
the vibration transmissibility $T = u_2/u_1$; the blocking force is then the
inertia force of the mass (ISO 10846-3, Eq. 1):

$$
k_{2,1} = -(2\pi f)^2\,(m_2 + m_\mathrm{f})\,T \qquad (T \ll 1),
$$

with $m_\mathrm{f}$ the mass of the output flange, which the `flange_mass=` keyword
carries (it defaults to zero). It is not a refinement: ISO 10846-3 measures the
vibration of the mass centre of the *compact body* formed by the blocking mass
together with the output flange of the test element, so $m_\mathrm{f}$ is whatever is
weighed with it: the flange and any auxiliary fixture that moves as part of the
test element. Left at zero it reports a stiffness low by $m_2/(m_2 + m_\mathrm{f})$. The
approximation is valid well above the mass/spring resonance, where $T$ is
small.

```python
import numpy as np
from phonometry import vibration

# Indirect method: a 10 kg blocking mass, transmissibility 0.01 at 500 Hz,
# and the 1.2 kg output flange of the test element bolted on top of it.
k = vibration.transfer_stiffness_indirect(0.01, frequency=500.0, blocking_mass=10.0,
                                          flange_mass=1.2)
print(f"{abs(complex(k)):.3e}")            # 1.105e+06  N/m
# flange_mass defaults to 0, which reports 9.870e+05 N/m: 12 % low here.

# Bundle a swept measurement into a result carrying its level and loss factor:
f = np.logspace(1.5, 3.3, 200)
t = vibration.base_transmissibility(f, mass=8.0, stiffness=1e6, damping=120.0)
res = vibration.indirect_transfer_stiffness_result(t, frequencies=f, blocking_mass=8.0)
print(round(float(res.levels[-1]), 1))      # 125.1  dB re 1 N/m (high-f)

res.plot()   # the Lk(f) level spectrum, as in the figure above (needs matplotlib)
```

The `TransferStiffnessResult` carries the complex $k_{2,1}$ and exposes `.levels`,
`.loss_factor`, `.magnitude`, `.to("impedance"/"apparent_mass")`,
`.band_average()` (section 4) and `.plot()`.

ISO 10846-1 clause 6.1 counts **three** methods, not two, "because they are
complementary with respect to their strong and weak points". The third is the
**driving-point method**, which measures the input displacement and the input
force and so yields the driving-point stiffness $k_{1,1}$ rather than
$k_{2,1}$. Only at low frequencies, below the element's first internal
resonance, where the driving-point and transfer stiffnesses are **equal**, can
it be used to determine the transfer stiffness, which is why the series covers
it at all: owners of expensive driving-point rigs can use them for the
low-frequency transfer stiffness. That same identity is why the fiche of
section 9 boxes the low-frequency *plateau* rather than a value at one
frequency: below the internal resonances there is only one stiffness to report.
Part 5 of the series specifies that method, and section 6 below computes it,
with the frequency up to which it may stand for the transfer stiffness.

The methods split the frequency axis between them, and the two limits of the
indirect method have names. It is valid between $f_2$ and $f_3$, "typically
20 Hz $\le f_2 \le$ 50 Hz and 2 kHz $\le f_3 \le$ 5 kHz" (ISO 10846-3, clause
1), and the two are moved by opposite things:

- **$f_2$** is where Inequality 2 starts to hold. Below it the rig's own
  resonances (the element, the load-distribution plate, the blocking mass and
  the auxiliary springs together) break the impedance mismatch. It sits at
  roughly **three times the highest natural frequency** of that assembly, and a
  *heavier* blocking mass lowers it.
- **$f_3$** is where the blocking mass stops moving as a rigid body. It is
  raised by making the block *compact* and of a dense, high-wave-speed material,
  steel being the usual choice. It follows from the block's dimension or mass
  through the standard's own curves, or is found experimentally from the
  effective mass (clause 6.2.3): support the block alone on soft springs with a
  mass-spring resonance below 10 Hz, drive it through its mass centre and read
  two accelerometers placed symmetrically inside the contact area at a spacing
  $D = \sqrt{S}$, then take $f_3$ as the lowest frequency at which
  $m_{2,\text{eff}} = 2F_2/(a'_1 + a''_1)$ departs from $m_2$ by more than 1 dB.

The two requirements pull against each other (heavy for a low $f_2$, compact
for a high $f_3$), which is why a wide range sometimes needs more than one
blocking mass, and why a full isolator dataset is usually the direct and
indirect methods spliced together. The direct method covers the bottom, from
1 Hz (the lower bound of the ISO 10846-2 scope; in practice the floor is set by
the rig and its instrumentation) up to where the rig's own resonances intrude,
typically a few hundred hertz for large elements.

**Excitation and averaging** (ISO 10846-2, 7.5). The source may be a discretely
stepped sine, a swept sine, a periodically swept sine or band-limited noise. It
shall be applied long enough that doubling the averaging time changes the
result by no more than **0.1 dB**; and for stepped or periodically swept sine
the source frequencies shall be spaced so that every one-third-octave band for
which stiffness is reported contains at least **five** of them.

**The other directions.** A mount under a machine is loaded in shear as well as
in compression, and its transverse stiffness is usually the smaller of the two.
ISO 10846-2 clause 5.2 and ISO 10846-3 clause 5.2 standardise separate
arrangements for the transverse translations, drawn in the lower panel of the
rig figure: the difficulty there is suppressing unwanted input motion, done
with low-friction roller-bearing guiding on the input or with two symmetrically
placed nominally equal elements, an input force-distribution plate acting as
the excitation mass, and an output shear force summed from two transducers,
$F_2 = F_2' + F_2''$. `transfer_stiffness_direct` and
`transfer_stiffness_indirect` are scalar FRF relations and therefore
direction-agnostic: the library does not stop you reporting $k_{2,1}$ per
direction, and the standard expects you to.

## 3. Validity of the indirect method

ISO 10846-3 (clause 6) requires the $T \ll 1$ approximation to be accurate
within **1 dB** (12 % of the stiffness magnitude), which bounds the usable
frequency range on both sides:

* **Impedance mismatch (Inequality 2).** Valid only where
  $\Delta L_{1,2} = L_{a1} - L_{a2} \ge 20\ \text{dB}$, i.e. $|T| \le 0.1$, the
  constant `TRANSMISSIBILITY_LIMIT`. `transfer_stiffness_indirect` computes
  $|T|$ at every frequency line and emits a `TransferStiffnessWarning` when any
  line exceeds it (routine near or below the mass/spring resonance, as in the figure
  above); the result it builds marks those lines as not valid, and
  its band average leaves them out.
* **Rigid blocking mass (Inequality 3).** Above an upper frequency $f_3$ the
  blocking mass no longer moves as a rigid body; results are valid only while
  its measured effective mass $m_{2,\text{eff}} = 2F_2/(a'_1 + a''_1)$ (Eq. 4)
  stays within 1 dB of the rigid mass:
  $|10\log_{10}(m_{2,\text{eff}}^2/m_2^2)| \le 1\ \text{dB}$.
  `effective_blocking_mass` finds $f_3$ from that measurement (section 5).
* **Linearity (clause 7.6).** Two input spectra 10 dB apart must give
  transfer-stiffness levels within 1.5 dB. The number is not arbitrary and it
  is the one criterion here that is really about the *material*: ISO 10846-1
  Annex D collects the physics. Dynamic properties depend on preload, vibration
  amplitude, frequency and temperature; for filled rubber the in-phase modulus
  and the phase angle are essentially amplitude-independent below shear strain
  amplitudes of order $10^{-3}$ and fall significantly above about
  $2 \times 10^{-3}$, the more so the higher the carbon-black content. So the
  elements that fail are the highly filled rubbers and anything with a designed
  amplitude dependence: a hydraulic mount is strongly amplitude-dependent *by
  design*, and cannot be described by a single $L_k$ spectrum at all. That
  makes the criterion actionable rather than pass/fail: the strain amplitudes
  occurring **in service** dictate the test amplitudes, and clause 7.6 says
  what the result then means, namely that the data may only be claimed valid
  for input amplitudes equal to or lower than the higher of the two tested. The
  upper bound of validated input level goes in the test report.

### What the rig and the instrumentation must do

Every criterion above is a property of the element and the model. ISO 10846-2
clauses 6 and 7 add a set of pre-runs and hardware requirements that decide
whether a result exists at all. One of the two level differences, the
unidirectionality pre-run, can be judged from the measured spectra
(`check_unwanted_input`, section 5); the background pre-run and the rest are
left to the operator.

* **A background pre-run** (7.6.1). With the source switched off, measure
  $L_{F_2}$ and $L_{a_1}$; the source output shall be adjusted so that they
  exceed the background by at least **15 dB** in every band of interest.
* **A unidirectionality pre-run** (6.4, Inequality 3). The acceleration in the
  excitation direction shall exceed the unwanted perpendicular components by at
  least **15 dB**, measured at the edge of the excitation mass in the plane of
  the input flange: the accelerometer drawn on the rig figure. Bands that fail
  are excluded from the result; `check_unwanted_input` judges them. (Clause
  7.6.1 sends this exclusion to "6.1, Inequality (1)", the 20 dB of the
  blocked output; the condition it means is 6.4, Inequality (3), as the
  companion parts show. See the [errata](https://docs.jmrp.io/phonometry/reference/errata/).)
* **Transducers** (6.5, 6.6). Accelerometers and force transducers shall have a
  sensitivity level that is frequency-independent within **0.5 dB**, be
  calibrated at the laboratory temperature, and have a cross-axis sensitivity
  below 5 %. Where several transducer signals are summed, the resultant
  sensitivity function shall meet the same 0.5 dB requirement.
* **Analyser** (6.8). The spectral resolution shall give at least **five
  distinct frequencies** in every one-third-octave band of interest, and the
  frequency-response difference between the input-acceleration and
  output-force channels shall be under 0.5 dB or else applied as a correction.
* **Coherence** between the input and output signals is the recommended running
  indicator (7.6.2): it flags poor signal-to-noise and non-linearity, and it is
  read exactly as the [mobility guide's Annex A.1 list](https://docs.jmrp.io/phonometry/vibration/structural/mechanical-mobility/)
  reads it.

Annex B of Part 2 says what each of these is worth. Its uncertainty budget
assigns 0.3 dB to signal processing and background noise, **0.5 dB** to
instrumentation when only the minimum requirements above are met, **0.5 dB** to
the test rig (from the residual output motion the two inequalities bound), and
**0.5 dB** to linearity. The last two are rectangular terms over a 1.5 dB
range, $1.5/(2\sqrt{3}) = 0.433$ dB each, which the annex rounds up to
0.5 dB. The three are comparable, which is the argument for not skipping any
of them: no one of them dominates, so no one of them is optional.

![Three stacked panels sharing a logarithmic frequency axis for a Kelvin-Voigt isolator on an 8 kilogram blocking mass. The upper panel gives the true transfer-stiffness level and the indirect-method estimate: the estimate runs far above the truth below and around the mass-spring resonance and lies on it above. The middle panel gives the transmissibility magnitude, which peaks above ten at the resonance and crosses the 0.1 limit at 187.5 hertz, with the level error of the estimate on a second axis staying inside a shaded plus-or-minus one decibel band from that crossing onwards. The lower panel gives the loss factor, rising in proportion to frequency as the Kelvin-Voigt model demands. Every panel is shaded up to the 0.1 crossing, the frequencies at which no result exists.](https://docs.jmrp.io/phonometry/media/transfer_stiffness.svg)

**Show the code for this figure**

```python
import matplotlib.pyplot as plt
import numpy as np
from phonometry import vibration

# Kelvin-Voigt isolator k + jwc loaded by an 8 kg blocking mass.
k, c, m2 = 1.0e6, 120.0, 8.0
f0 = np.sqrt(k / m2) / (2.0 * np.pi)
f = np.logspace(np.log10(f0 / 5.0), np.log10(f0 * 40.0), 600)

k_true = k + 1j * 2.0 * np.pi * f * c
t = vibration.base_transmissibility(f, m2, k, c)
k_indirect = vibration.transfer_stiffness_indirect(t, m2, frequency=f)  # warns where T is not small

# One line: the indirect determination bundled as a result draws its own
# Lk(f) level spectrum:
res = vibration.indirect_transfer_stiffness_result(t, frequencies=f, blocking_mass=m2)
res.plot()
plt.show()

# By hand, the true element stiffness against the indirect-method estimate:
fig, ax = plt.subplots()
ax.semilogx(f, vibration.transfer_stiffness_level(k_true),
            label="true $L_k$ of $k+j\\omega c$")
ax.semilogx(f, vibration.transfer_stiffness_level(k_indirect), "--",
            label="indirect method $-(2\\pi f)^2 m_2 T$")   # no flange here
ax.axvline(f0, color="0.6", linestyle=":", label="resonance $f_0$")
ax.set(xlabel="Frequency [Hz]", ylabel="Transfer stiffness level $L_k$ [dB re 1 N/m]")
ax.grid(True, which="both", alpha=0.3)
ax.legend()
plt.show()
```

*The criterion, and the error it buys. The indirect method is valid from where
$|T|$ falls below 0.1 (187.5 Hz for this element, not the mass/spring resonance
at 56 Hz and not the $3f_0$ rule of thumb) and that limit is chosen because it
is what holds the stiffness error inside the 1 dB the clause requires, as the
middle panel's second axis shows. The lower panel is the loss factor: for a
Kelvin-Voigt element $\eta = \omega c/k$ rises in proportion to frequency,
which is the model's main deficiency as a description of real rubber, whose
loss factor is far flatter. The definition holds only in the low-frequency
range where inertial forces inside the element are negligible.*

The blocking-force idealisation itself is quantified by ISO 10846-1, Eq. (6):
for an isolator of output driving-point stiffness $k_{2,2}$ on a termination of
stiffness $k_\mathrm{t}$, the delivered force is $F_2/F_{2,\mathrm{b}} = 1/(1 + k_{2,2}/k_\mathrm{t})$,
within 10 % of the blocking force for $|k_{2,2}| < 0.1\,|k_\mathrm{t}|$ (Eq. 7):

```python
import warnings
from phonometry import vibration

# |T| = 0.5 violates Inequality (2): the indirect result is flagged.
with warnings.catch_warnings(record=True) as caught:
    warnings.simplefilter("always")
    vibration.transfer_stiffness_indirect(0.5, frequency=50.0, blocking_mass=10.0)
print(caught[0].category.__name__)                     # TransferStiffnessWarning

# Blocking-force approximation at the 10 % limit (ISO 10846-1, Eq. 6):
print(round(abs(complex(vibration.blocking_force_ratio(1e5, 1e6))), 4))   # 0.9091
```

## 4. One value per one-third-octave band

A swept or stepped measurement gives $k_{2,1}$ at hundreds of lines, and every
part of the series reports it as one value per one-third-octave band: the
squared magnitude averaged over the $n$ lines the band holds (ISO 10846-2
Formula (6), -3 Formula (7), -4 Formula (11), -5 Formula (6)),

$$
k_\mathrm{av} = \left\{ \frac{1}{n} \sum_{i=1}^{n} \lvert k_{2,1}(f_i) \rvert^2 \right\}^{1/2},
\qquad L_{k,\mathrm{av}} = 10 \lg \frac{k_\mathrm{av}^2}{k_0^2},
$$

"where the summation is performed over a minimum of $n = 5$ frequencies".
Averaging the square rather than the level weights the peaks, which the parts
call "usually the most important ones", and the phase is lost. The five lines
are also a requirement on the measurement: the analyser shall resolve at least
five distinct frequencies per band, and a stepped or swept sine shall put five
in every band for which a value is reported. So `band_averaged_stiffness`
gives a band of one to four lines no value (NaN) and a
`TransferStiffnessWarning`, rather than an average of too few lines. It assigns
each line to the base-ten band that encloses it, names the bands by their
ISO 266 centres, and leaves out first the lines that failed an adequacy
condition, since every part excludes those from the evaluation: the indirect
result marks the lines with $|T| > 0.1$, so its `.band_average()` averages only
the valid part of the sweep.

```python
import warnings
import numpy as np
from phonometry import vibration

# Direct method, a Kelvin-Voigt mount k + jwc, lines every 2 Hz:
f = np.arange(90.0, 1120.0, 2.0)
k21 = 1.0e6 + 1j * 2.0 * np.pi * f * 80.0
direct = vibration.TransferStiffnessResult(
    frequencies=f, transfer_stiffness=vibration.transfer_stiffness_direct(k21 * 1e-6, 1e-6)
)
bands = direct.band_average()
print(bands.nominal_frequencies[[0, -1]], bands.line_counts[[0, -1]])  # [ 100. 1000.] [ 12 114]
print(round(float(bands.levels[-1]), 2))                               # 120.99  dB re 1 N/m

# Indirect method on an 8 kg blocking mass: |T| falls to 0.1 at 187.5 Hz, so the
# lines up to 186 Hz fail Inequality (2) and never reach the average; the 200 Hz
# band averages its 18 lines from 188 Hz up.
t = vibration.base_transmissibility(f, mass=8.0, stiffness=1.0e6, damping=120.0)
with warnings.catch_warnings():
    warnings.simplefilter("ignore", vibration.TransferStiffnessWarning)
    indirect = vibration.indirect_transfer_stiffness_result(t, frequencies=f, blocking_mass=8.0)
bands = indirect.band_average()
print(bands.nominal_frequencies[0], bands.line_counts[0])  # 200.0 18
bands.plot()   # the band levels, the undetermined bands marked (needs matplotlib)
```

## 5. Checking the rig from the measured spectra

Three of the adequacy conditions of the series are level differences or a
mass, and can be judged from what the rig measured. Each check returns the
condition frequency by frequency with an overall `.passes`, draws itself with
`.plot()`, and warns (`TransferStiffnessWarning`) where the condition fails.

* **The output is blocked**, $\Delta L_{1,2} = L_{a1} - L_{a2} \ge 20$ dB
  (ISO 10846-5 Inequality (1); Inequality (1) of Part 2, (2) of Parts 3 and
  4): `check_blocked_output`. A smaller difference means too little stiffness
  mismatch between the element and the foundation, or flanking transmission.
* **The input moves in one direction**, the excitation direction at least
  15 dB above each perpendicular one (Part 5 Inequality (2); (3) of Part 2,
  (5) of Part 3, (7) of Part 4): `check_unwanted_input`, which takes one row
  of levels per unwanted direction and lets the loudest decide.
* **The mass in front of the output force transducers is light enough** for
  the direct method (ISO 10846-4 Inequality (3), ISO 10846-2 Inequality (2)):
  the output flange, the force-distribution plate and half the transducers,
  $m_0$, carry an inertia force $m_0 a_2$ the transducers read as part of the
  blocking force, and it shall not exceed 6 % of the force,

  $$
  m_0 \le 0.06 \times \frac{10^{L_{F2}/20}}{10^{L_{a2}/20}}\ \text{kg},
  $$

  with $L_{F2}$ re 1 µN and $L_{a2}$ re 1 µm/s², so the bound is
  $0.06\,|F_2|/|a_2|$. At the bound the force levels differ by
  $20 \lg 1.06 = 0.51$ dB when the inertia force is in phase with the
  measured one and by $-20 \lg 0.94 = 0.54$ dB when it opposes it: the
  0.5 dB of NOTE 1 (which ISO 10846-4 prints as "05 dB"; see the
  [errata](https://docs.jmrp.io/phonometry/reference/errata/)). `check_output_mass` returns the
  limit and that worst-case bias, `bias_bound_db`.

For the indirect method the question is how high the blocking mass stays rigid.
ISO 10846-4 Formula (6) (ISO 10846-3 Formula (4)) measures its effective mass
with the block on soft springs, driven by a force $F_2$ through its centre of
mass and read by two accelerometers $\sqrt{S}$ apart inside the contact area,
$m_{2,\mathrm{eff}} = |2F_2/(a'_1 + a''_1)|$; the upper limit $f_3$ is the
lowest frequency at which it departs from $m_2$ by more than 1 dB (12 %),
Inequality (5), with a departure below 40 Hz ignored as the mass-spring
behaviour of the block on its supports. `effective_blocking_mass` interpolates
the crossing between the last line inside the 1 dB and the first line outside
it.

```python
import numpy as np
from phonometry import vibration

f = [63.0, 125.0, 250.0]
check = vibration.check_blocked_output([110.0, 110.0, 110.0], [85.0, 88.0, 86.0], frequencies=f)
print(check.holds, check.passes)                        # [ True  True  True] True

# m0 = 0.4 kg (output flange, force distribution plate and half the transducers)
# under an output force of 1 N (120 dB re 1 µN) and an output acceleration of
# 0.1 m/s² (100 dB re 1 µm/s²):
mass = vibration.check_output_mass(0.4, [120.0], [100.0], frequencies=[125.0])
print(mass.mass_limit_kg, mass.passes, mass.bias_bound_db.round(3))   # [0.6] True [0.355]

# A 20 kg block whose effective mass grows as (f/3 kHz)^2 above rigid:
fe = np.geomspace(20.0, 5000.0, 400)
m_eff = 20.0 * (1.0 + (fe / 3000.0) ** 2)
ones = np.ones(fe.size, dtype=complex)
block = vibration.effective_blocking_mass(m_eff * ones, ones, ones, frequencies=fe, blocking_mass=20.0)
print(round(block.upper_frequency_limit_hz, 1))         # 1047.9  Hz
```

## 6. The driving-point method (ISO 10846-5)

With the output of the element blocked, the input force and the input
acceleration give the **driving-point stiffness** (Formula (3)),

$$
k_{1,1}(f) = \frac{F_1}{u_1} = -(2\pi f)^2 \frac{F_1}{a_1},
$$

which equals the transfer stiffness only at low frequencies. Two things pull it
away: the mass between the element and the input force transducers, whose
inertia the transducers read with the element, and the element's own internal
resonances. Clause 6.2 fixes where the approximation ends. The low-frequency
value is "the average for 1 Hz to 20 Hz", and the **upper limiting frequency**
$f_\mathrm{UL}$ is "the lowest frequency, at which the driving point stiffness
level becomes 2 dB smaller than the low-frequency stiffness". Up to it the band
averages of $k_{1,1}$ stand for those of $k_{2,1}$ within 2 dB (Formula (7),
$k_\mathrm{av} = k_{\mathrm{av}(2,1)} \approx k_{\mathrm{av}(1,1)}$), and above
it the method has nothing to say. Typically $f_\mathrm{UL}$ lies between 50 Hz
and 200 Hz, so this is a low-frequency method, meant for the owners of
driving-point rigs.

`driving_point_stiffness` computes $k_{1,1}$, takes the low-frequency value as
the Formula (6) average of the lines from 1 Hz to 20 Hz (for the flat stiffness
the clause presumes, the same as the mean of the levels), and interpolates the
crossing of the 2 dB threshold. Given the output acceleration or the unwanted
input accelerations, it checks Inequalities (1) and (2) line by line and leaves
the lines that fail out of everything, as 7.6.1 requires. Its `.band_average()`
averages only the valid lines, those at or below $f_\mathrm{UL}$ (8.3 states
the 2 dB "if $f \le f_\mathrm{UL}$"). For those bands the 2 dB of Formula (7)
is the standard's own statement, which Annex B assumes again for its budget
(B.3.5): the criterion of 6.2 only watches $k_{1,1}$ fall below its own
low-frequency value, so nothing in the average proves the bound line by line.
Below 20 Hz, 7.5 asks for a 0.2 Hz
line spacing rather than five lines per band, which leaves the lowest bands
short of five; those bands have no value, without a warning, and the note to
clause 9 m) accepts narrow-band data there instead.

![Two stacked panels sharing a logarithmic frequency axis from 1 to 200 hertz for a 1 meganewton per metre element with a 5 per cent loss factor driven through a 2 kilogram force distribution plate. The upper panel gives the driving-point stiffness level, flat at 120 decibels at low frequency and falling away as the plate's inertia grows, the transfer stiffness as a dashed horizontal line at 120 decibels, a dotted threshold 2 decibels below the 1 to 20 hertz value, a vertical line at the upper limiting frequency of 52.2 hertz, and red dots for the band averages from 4 to 50 hertz; everything above 52.2 hertz is shaded as not evaluated. The lower panel gives the difference between the driving-point and transfer stiffness levels, narrow band and band by band, inside a shaded band of plus or minus 2 decibels that every band average respects.](https://docs.jmrp.io/phonometry/media/driving_point_stiffness.svg)

**Show the code for this figure**

```python
import matplotlib.pyplot as plt
import numpy as np
from phonometry import vibration

# A 1 MN/m element with a 5 % loss factor, driven through a 2 kg force
# distribution plate: the input force transducers read the plate's inertia with
# the element.
f = np.arange(1.0, 200.0, 0.2)          # the 0.2 Hz lines of 7.5
w = 2.0 * np.pi * f
k21 = 1.0e6 * (1.0 + 0.05j)
res = vibration.driving_point_stiffness((k21 - w**2 * 2.0) * 1e-6, -(w**2) * 1e-6, frequencies=f)
print(round(res.upper_limiting_frequency_hz, 1))    # 52.2  Hz

# One line each: the level with the 2 dB threshold and f_UL, and the bands.
res.plot()
res.band_average().plot()
plt.show()

# By hand, the two panels above: k1,1 against k2,1, and their difference.
bands = res.band_average()
centres = bands.center_frequencies[bands.determined]
band_levels = bands.levels[bands.determined]
l21 = 20.0 * np.log10(abs(k21))
f_ul = res.upper_limiting_frequency_hz
fig, (top, bottom) = plt.subplots(2, 1, sharex=True, figsize=(10, 8.8))
for panel in (top, bottom):
    panel.axvspan(f_ul, f[-1], color="0.9")            # above f_UL: not evaluated
    panel.axvline(f_ul, color="C1")
    panel.grid(True, which="both", alpha=0.3)
top.semilogx(f, res.levels, label="driving-point stiffness $L_{k_{1,1}}$")
top.axhline(l21, color="C2", linestyle="--", label="transfer stiffness $L_{k_{2,1}}$")
top.axhline(res.threshold_level, color="k", linestyle=":",
            label="2 dB below the 1 Hz to 20 Hz value")
top.plot(centres, band_levels, "o", color="C1",
         label="band average $L_{k,\\mathrm{av}}$, Formula (6)")
top.set(ylabel="$L_k$ [dB re 1 N/m]", ylim=(l21 - 7.0, l21 + 2.5))
top.legend(loc="lower left")
bottom.axhspan(-2.0, 2.0, color="C2", alpha=0.15, label="$\\pm$2 dB of Formula (7)")
bottom.semilogx(f, res.levels - l21, label="$L_{k_{1,1}} - L_{k_{2,1}}$, narrow band")
bottom.plot(centres, band_levels - l21, "o", color="C1",
            label="band averages below $f_\\mathrm{UL}$")
bottom.set(xlabel="Frequency [Hz]", ylabel="Difference [dB]",
           xlim=(f[0], f[-1]), ylim=(-7.0, 3.0))
bottom.legend(loc="lower left")
plt.show()
```

*Where the method stops. The force distribution plate between the input force
transducers and the element is 2 kg, $m_\mathrm{p}$, so the measured stiffness
is $k_{2,1} - \omega^2 m_\mathrm{p}$ and falls away from the transfer
stiffness as the frequency rises; at 52.2 Hz it is 2 dB below its 1 Hz to 20 Hz value, and
nothing above that is evaluated. Every band average below it lies inside the
±2 dB of Formula (7); the last one, at 50 Hz, averages only the 37 lines below
$f_\mathrm{UL}$ and sits 1.8 dB down.*

**The uncertainty budget** (Annex B) writes the band level as the measured one
plus five corrections of zero estimate, each with sensitivity coefficient 1
(Formula (B.1)), combines them by root sum of squares (B.2) and takes $U = 2u$
for 95 % coverage (B.3). `driving_point_uncertainty` builds it on
`metrology.uncertainty_propagation`: signal processing 0.3 dB and instrumentation
0.5 dB (normal), installation repeatability $p/\sqrt{3}$ for a spread $2p$ of
repeated installations, and three rectangular terms, the test rig
$1/(2\sqrt{3}) = 0.289$ dB, the driving-point discrepancy
$2/\sqrt{3} = 1.155$ dB for its $\pm 2$ dB, and linearity
$1.5/(2\sqrt{3}) = 0.433$ dB. Those three are the expressions the annex
prints. Table B.1 carries them rounded up to one decimal, 0.3, 1.2 and 0.5 dB,
the conservative rounding an uncertainty may take (ISO/IEC Guide 98-3, 7.2.6),
which puts $u$ at 1.456 dB instead of 1.394 dB with no repeatability spread.
The library keeps the expressions, and every term can be passed in to
reproduce the table or to state a reasoned estimate. The discrepancy term
dominates: it is the price of measuring $k_{1,1}$ for $k_{2,1}$.

```python
from phonometry import vibration

budget = vibration.driving_point_uncertainty(119.3, repeatability_range_db=0.6)
print(round(budget.combined_uncertainty_db, 3))    # 1.405  dB
print(round(budget.expanded_uncertainty_db, 2))    # 2.81  dB (U = 2u)
budget.plot()   # one bar per input quantity, with u and U
```

## 7. Reading $L_k$, and what it buys

A level in dB re 1 N/m needs a scale before it means anything. Each decade of
stiffness is 20 dB, so the working range spans roughly:

| Element | $\lvert k_{2,1}\rvert$ | $L_k$ |
|---|---|---|
| a soft rubber mount or a steel coil spring | $10^5$ N/m | 100 dB |
| a general-purpose machine mount | $10^6$ N/m | 120 dB |
| a stiff bonded pad or a hard elastomer | $10^7$ N/m | 140 dB |
| a rigid connection (for comparison) | $10^8$ N/m and up | 160 dB and up |

Lower is better, and the shape matters more than the value: in the figure above
the level **rises** with frequency, which is the isolator getting stiffer, and
therefore less effective, exactly in the range where structure-borne noise
matters. That rise is the single most useful thing an $L_k(f)$ spectrum tells
an engineer about a mount, and it is what a datasheet quoting one static
stiffness cannot say.

On its own $k_{2,1}$ still predicts nothing, because the blocked force is only
what the element delivers to a receiver much stiffer than itself. `blocking_force_ratio`
prices that assumption against a real termination:

```python
# The same 1 MN/m mount under a concrete floor and under a light timber one.
for k_t in (3e9, 2e6, 5e5):
    ratio = abs(complex(vibration.blocking_force_ratio(1e6, k_t)))
    print(f"{k_t:.0e}", round(ratio, 3))
# 3e+09 1.0
# 2e+06 0.667
# 5e+05 0.333
```

On the concrete floor ($k_\mathrm{t} = 3$ GN/m) the delivered force *is* the blocked
force, Equation (7) is satisfied by three orders of magnitude, and the
datasheet number is directly usable. On a lightweight floor of 2 MN/m it is
not: only two thirds of the blocked force arrives (−3.5 dB), Equation (7) fails,
and the mount's characterisation no longer predicts what the installation does.
That is not the isolator failing (a mobile receiver takes *less* force, so the
real installation is quieter than the blocked-force estimate); it is the
*single-number description* failing, and the route out is a prediction that
carries both mobilities:
[installed structure-borne sound (EN 12354-5)](https://docs.jmrp.io/phonometry/buildings/design/installed-structure-borne/),
fed by a source characterised through
[EN 15657](https://docs.jmrp.io/phonometry/buildings/design/structure-borne-power/). The rule behind
both is the source/receiver mobility rule of the
[mobility guide](https://docs.jmrp.io/phonometry/vibration/structural/mechanical-mobility/#why-two-mobilities-decide-the-power).

## 8. Relation to the FRF family

The dynamic stiffness is a member of the frequency-response-function family
(ISO 10846-1, Annex A / Table A.2): it is the reciprocal of the receptance and
relates to the mechanical impedance $Z$ and effective mass $m_\text{eff}$ by
$k = j\omega Z = -\omega^2 m_\text{eff}$. These conversions are the same as the
[mechanical mobility](https://docs.jmrp.io/phonometry/vibration/structural/mechanical-mobility/) `convert_frf`
pivot:

```python
from phonometry import vibration

k = 1e6 + 5e4j                                  # N/m, at 250 Hz
Z = vibration.convert_frf(k, "dynamic_stiffness", "impedance", frequency=250.0)
print(round(abs(complex(vibration.convert_frf(
    Z, "impedance", "dynamic_stiffness", frequency=250.0))), 1))    # 1001249.2
```

## 9. Test-report fiche

`TransferStiffnessResult.report(path)` renders a one-page
dynamic-transfer-stiffness characterisation report for a resilient element
(ISO 10846-1:2008 definition; determined by the direct method, ISO 10846-2:2008,
or the indirect blocking-mass method, ISO 10846-3:2002). The sheet shows the
$L_k(f)$ level spectrum beside a compact table of characteristic points (the
determination method, the blocking mass for the indirect method, the frequency
range, and the low-frequency stiffness plateau $|k_{2,1}|$, its level $L_k$
and the loss factor there), then the one-third-octave band levels
$L_{k,\mathrm{av}}$ of section 4, which the test report of both parts presents
(ISO 10846-2 9 m), ISO 10846-3 10 j)), and a boxed low-frequency $L_k$ (the
plateau that characterises the element below its internal resonances). The
characteristic points are read at the lowest *valid* line and the spectrum
draws the excluded lines apart, so an indirect result never reports the
resonance region that Inequality (2) rules out; a band of fewer than five valid
lines prints its line count instead of a level. It is a characterisation, so
there is no pass/fail verdict; `language="es"` renders the Spanish fiche. The
fiche always embeds the $L_k(f)$ spectrum, so it needs both the report and plot
extras (`pip install "phonometry[report,plot]"`).

```python
import numpy as np
from phonometry import ReportMetadata, vibration

# Ten lines in every one-third-octave band from 20 Hz to 2 kHz:
freqs = 1000.0 * 10.0 ** ((np.arange(-175, 35) + 0.5) / 100.0)
k21 = 1e6 + 1j * (2 * np.pi * freqs) * 80.0     # Kelvin-Voigt element k + jwc
u1 = 1e-6 + 0j
k = vibration.transfer_stiffness_direct(k21 * u1, u1)     # direct method: k2,1 = F2,b/u1
res = vibration.TransferStiffnessResult(frequencies=freqs, transfer_stiffness=k)
res.report(
    "transfer_stiffness.pdf",
    metadata=ReportMetadata(
        specimen="Rubber vibration isolator",
        measurement_standard="ISO 10846-2",
    ),
)   # one-page fiche (needs phonometry[report,plot])
```

The example fiche is regenerated with `make reports` and kept rendered in the
repository; click the preview to open the PDF.

**Report:** [ISO 10846 dynamic-transfer-stiffness example report (PDF)](https://docs.jmrp.io/phonometry/media/iso10846_transfer_stiffness_example.pdf): One-page dynamic-transfer-stiffness fiche: a metadata header, a table of the FRF characteristic points (the determination method, the frequency range and the low-frequency stiffness, level and loss factor) beside the transfer-stiffness level spectrum, the 21 one-third-octave band levels from 20 Hz to 2 kHz, and the boxed low-frequency level.

Dynamic-transfer-stiffness fiche (TransferStiffnessResult.report): the FRF characteristic points, the transfer-stiffness level spectrum and its one-third-octave band levels.

## What this guide covers

- **Covered.** ISO 10846-1:2008's $k_{2,1} = F_{2,\mathrm{b}}/u_1$ definition, the level (ISO 10846-2
  and -3, 3.17) and loss factor (ISO 10846-1, 3.8) (`transfer_stiffness_level`,
  `loss_factor`), the ISO 10846-2:2008 direct method
  (`transfer_stiffness_direct`), and the ISO 10846-3:2002 indirect method of
  Formula 1 (`transfer_stiffness_indirect`, including its `flange_mass=` term).
  The blocking-force approximation of Equations 6-7 (`blocking_force_ratio`) and
  the Annex A / Table A.2 FRF relations (`convert_frf`) are implemented too.
  `TransferStiffnessResult` bundles a swept determination, and `.report()`
  renders the characterisation fiche with its one-third-octave band levels.

- **Covered.** The one-third-octave band average of every part, ISO 10846-2 Formula (6), -3
  Formula (7), -4:2003 Formula (11) and -5:2008 Formula (6), with no value for a
  band of fewer than five lines (`band_averaged_stiffness`,
  `TransferStiffnessResult.band_average`). The adequacy conditions ISO 10846-4:2003
  and -5:2008 state as inequalities: the 20 dB of the blocked output and the
  15 dB of the unwanted input directions (`check_blocked_output`,
  `check_unwanted_input`) and the output-mass limit of ISO 10846-4 Inequality (3)
  (`check_output_mass`), and the effective blocking mass of ISO 10846-4 Formula (6)
  with its $f_3$ (`effective_blocking_mass`). The ISO 10846-5:2008 driving-point
  stiffness of Formula (3), its upper limiting frequency (6.2) and the band
  averages of Formula (7) (`driving_point_stiffness`), and the Annex B
  uncertainty budget (`driving_point_uncertainty`). Neither part prints a
  worked example, so conformance is anchored on closed forms.

- **Not covered.** The test arrangements of ISO 10846-4 for couplings, bellows, hoses, pipe
  hangers and cables are described by the standard, not modelled: the same
  functions serve them once the forces and accelerations are measured. The
  Figure 11 nomograms of $f_3$ for steel cubes and cylinders are not tabulated;
  $f_3$ is found from a measured effective mass. The linearity test (two input
  spectra 10 dB apart within 1.5 dB, clause 7.7 of ISO 10846-2, -4 and -5 and
  clause 7.6 of ISO 10846-3) is not checked in code, and neither are the preload,
  creep and temperature conditioning, the background pre-run, the 0.5 dB
  transducer flatness and inter-channel matching or the 0.2 Hz line spacing
  ISO 10846-5 asks for below 20 Hz: the page describes them and leaves them to
  the operator. Hydraulic mounts are outside the single-spectrum description
  altogether: their stiffness is amplitude-dependent by design, so one $L_k(f)$
  does not characterise them.

## See also

- [Mechanical mobility and the FRF family (ISO 7626-1)](https://docs.jmrp.io/phonometry/vibration/structural/mechanical-mobility/):
  the source/receiver mobility rule that says when a blocked force is the right
  description at all, and the FRF vocabulary this page's conversions belong to.
- [Structure-borne sound power of equipment (EN 15657)](https://docs.jmrp.io/phonometry/buildings/design/structure-borne-power/):
  the source characterisation that sits on the other side of the isolator.
- [Installed structure-borne sound (EN 12354-5)](https://docs.jmrp.io/phonometry/buildings/design/installed-structure-borne/):
  the prediction that consumes $k_{2,1}$ together with both mobilities and
  turns it into a level in a receiving room.
- API reference: [`vibration.structural.transfer_stiffness`](https://docs.jmrp.io/phonometry/reference/api/vibration/transfer-stiffness/) and [`vibration.structural.mechanical_mobility`](https://docs.jmrp.io/phonometry/reference/api/vibration/mechanical-mobility/).
- Theory: [Point mobilities and radiation efficiency](https://docs.jmrp.io/phonometry/reference/theory/vibration/#point-mobilities-and-radiation-efficiency-cremer-5-hopkins-29): the mobility and impedance definitions the transfer stiffness is the reciprocal of.

## References

- **Cremer, L., Heckl, M., & Petersson, B. A. T. (2005).** *Structure-borne sound: Structural vibrations and sound radiation at audio frequencies* (3rd ed.). Springer. https://doi.org/10.1007/b137728
  Vibration isolation theory: why an isolator's performance depends on the source and receiver mobilities, the physics behind the blocked-force characterisation. ISBN 978-3-540-22696-3.
- **International Organization for Standardization. (2002).** [*Acoustics and vibration — Laboratory measurement of vibro-acoustic transfer properties of resilient elements — Part 3: Indirect method for determination of the dynamic stiffness of resilient supports for translatory motion*](https://www.iso.org/standard/34560.html) (ISO 10846-3:2002).
  The indirect method k₂₁ = −(2πf)²(m₂+m_f)T (Formula 1) with its validity conditions (clause 6: Inequalities 2 and 3; clause 7.6 linearity). Conformance is anchored on the indirect inertia relation, the \|T\| = 0.1 ↔ ΔL₁,₂ = 20 dB validity limit and its 1 dB (12 %) accuracy bound, and the 7.6 linearity criterion.
- **International Organization for Standardization. (2003).** *Acoustics and vibration — Laboratory measurement of vibro-acoustic transfer properties of resilient elements — Part 4: Dynamic stiffness of elements other than resilient supports for translatory motion* (ISO 10846-4:2003).
  Couplings, bellows, hoses, pipe hangers and cables by the direct and the indirect method: the output-mass limit of Inequality (3), the effective blocking mass of Formula (6) with its f₃ (Inequality 5), and the band average of Formula (11). It prints no worked example, so conformance is anchored on closed forms; the 0.5 dB of its 6.2 NOTE 1 is printed as 05 dB (see the errata).
- **International Organization for Standardization. (2008).** [*Acoustics and vibration — Laboratory measurement of vibro-acoustic transfer properties of resilient elements — Part 1: Principles and guidelines*](https://www.iso.org/standard/38936.html) (ISO 10846-1:2008).
  The principles part of the series: the blocking-force idealisation and the FRF relations this page implements: the dynamic transfer stiffness k₂₁ = F₂,b/u₁ and its FRF relations (clause 5 and Annex A / Table A.2) and the blocking-force approximation (Eqs. 6/7). Parts 4 and 5 of the series extend the same quantities to elements other than supports and to the driving-point low-frequency method. Conformance is anchored on the closed-form Table-A.2 identity k = jω·Z and the Eq. (6) force ratio 1/1.1.
- **International Organization for Standardization. (2008).** [*Acoustics and vibration — Laboratory measurement of vibro-acoustic transfer properties of resilient elements — Part 2: Direct method for determination of the dynamic stiffness of resilient supports for translatory motion*](https://www.iso.org/standard/38937.html) (ISO 10846-2:2008).
  The direct method, and (with Part 3) the level L_k re 1 N/m and the loss factor (clauses 3.8/3.17). Conformance is anchored on the closed-form level of a decade of stiffness.
- **International Organization for Standardization. (2008).** *Acoustics and vibration — Laboratory measurement of vibro-acoustic transfer properties of resilient elements — Part 5: Driving point method for determination of the low-frequency transfer stiffness of resilient supports for translatory motion* (ISO 10846-5:2008).
  The driving-point stiffness k₁,₁ of Formula (3), the upper limiting frequency of 6.2, the band averages of Formulas (6) and (7), Inequalities (1) and (2), and the Annex B budget. No worked example either; Table B.1 rounds its three rectangular terms up to one decimal (the linearity term 1.5/(2√3) = 0.433 dB becomes 0.5 dB), and the library keeps the expressions.
