# Plate junctions: wave approach and experimental SEA

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

Canonical: https://docs.jmrp.io/phonometry/vibration/structural/junction-transmission/
Language: en
Alternate: https://docs.jmrp.io/phonometry/es/vibration/structural/junction-transmission/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: 2026-07-20
Updated: 2026-11-16T21:36:10Z
License: MIT (https://github.com/jmrplens/phonometry/blob/main/LICENSE)
Summary: Two routes to the coupling loss factor: bending waves and Kij at rigid X, T, L and in-line plate junctions, and experimental SEA.

---

When a bending wave travelling on a wall or floor reaches a rigid junction with
another plate, part of its energy is reflected and part is transmitted into the
connected plates. The **wave approach** of Cremer et al. (1973), tabulated by
Craik (1981, 1996) and collected in Hopkins (2007, Section 5.2.1.3), gives the
transmission coefficient in closed form for the four most common junctions of
thin, homogeneous, isotropic plates: the **X**, **T**, **L** and **in-line**
junctions. Modelling the junction as a simply supported (pinned) massless beam
forces an incident bending wave to generate only reflected and transmitted
*bending* waves, with no conversion to in-plane waves, and the resulting
coefficients are **independent of frequency**. That is what makes them
convenient closed-form inputs for statistical energy analysis (SEA) and for the
EN 12354 flanking-transmission model, where they feed the coupling loss factor
$\eta_{ij}$ and the vibration reduction index $K_{ij}$.

That convenience has a price, and it is worth knowing before the number is used
as though it were measured. A *real* junction's $K_{ij}$ **is** frequency
dependent, which is why ISO 10848 reports it per band and why EN 12354-1
Annex E tabulates band-dependent values. The wave approach needs plates thin
relative to the bending wavelength, and enough modes in the band for the
diffuse-field angular average of section 3 to mean anything, so it degrades at
low frequency and on small elements. The in-plane conversion it discards grows
with frequency and with junction complexity, so the closed form flatters heavy,
stiff, complicated joints; the clip in section 2 shows exactly the precursor
being neglected. And a massless pinned beam is a poor model of a junction that
in reality carries mass and rotational stiffness. Section 5 hands one
frequency-independent number straight to the EN 12354 model in place of a
band-dependent one; when that matters, the measured counterpart is the
[Laboratory Flanking Transmission guide](https://docs.jmrp.io/phonometry/buildings/insulation/flanking-lab/).

The object itself first: a T-junction where a 140 mm concrete floor runs into
a continuous 200 mm wall. A `junction_transmission` result retains its plates,
so `res.plot_geometry()` draws the junction to scale with the incident bending
wave marked.

![To-scale cross-section of a T2 plate junction: a 140 mm concrete floor arrives horizontally from the left with the incident-sound arrow on it and ends against the continuous 200 mm wall drawn vertically, both plate thicknesses dimensioned](https://docs.jmrp.io/phonometry/media/junction_plate_geometry.svg)

*Drawn to scale: the incident bending wave arrives along the 140 mm floor, and
everything the closed-form coefficients describe happens where it meets the
continuous 200 mm wall.*

**Show the code for this figure**

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

# A 140 mm concrete floor meeting a 200 mm wall at a T-junction.
res = vibration.junction_transmission("T2", 0.14, 3500.0, 320.0, 0.2, 3500.0, 460.0)
res.plot_geometry()
plt.show()
```

![Transmission coefficient against incidence angle for a rigid X-junction of a 100 mm and a 200 mm concrete plate. The corner coefficient tau12 falls from 0.064 at normal incidence to exactly zero at the 45 degree cut-off angle and stays there, and the region beyond the cut-off is shaded; the straight-section coefficient tau13 falls from 0.011 and carries on smoothly to grazing incidence. Both diffuse-field angular averages are marked as horizontal lines.](https://docs.jmrp.io/phonometry/media/junction_transmission.svg)

*The sharpest feature of this figure is the cut-off. Beyond
$\theta_\text{co} = \arcsin\chi = 45°$ the heavier receiving plate has no
propagating bending wave for the corner path to feed, so $\tau_{12}$ is exactly
zero and stays there, while the straight path carries on to grazing incidence.
That is why $\chi$ (a wavenumber ratio) and not $\psi$ sets the shape of the
curve.*

**Show the code for this figure**

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

# X-junction between a 100 mm and a 200 mm concrete plate (cL = 3200 m/s).
res = vibration.junction_transmission("X", 0.1, 3200.0, 240.0, 0.2, 3200.0, 480.0)
res.plot()  # tau(theta) for the corner and straight paths, with averages
plt.show()
```

## 1. Wave parameters chi and psi

With plate $i$ of thickness $h_i$, quasi-longitudinal wave speed $c_{\mathrm{L},i}$ and
surface density $\rho_{\mathrm{s},i}$, the whole family of coefficients depends on just
two dimensionless ratios (Cremer et al. 1973):

$$
\chi = \frac{k_{\mathrm{B}2}}{k_{\mathrm{B}1}}
     = \left(\frac{\rho_{\mathrm{s}2} B_1}{\rho_{\mathrm{s}1} B_2}\right)^{1/4}
     = \sqrt{\frac{h_1 c_{\mathrm{L}1}}{h_2 c_{\mathrm{L}2}}}
     = \sqrt{\frac{f_{\mathrm{c}2}}{f_{\mathrm{c}1}}}
$$

$$
\psi = \frac{B_2 k_{\mathrm{B}2}^2}{B_1 k_{\mathrm{B}1}^2}
     = \frac{h_2 c_{\mathrm{L}2} \rho_{\mathrm{s}2}}{h_1 c_{\mathrm{L}1} \rho_{\mathrm{s}1}}
     = \frac{\rho_{\mathrm{s}2} f_{\mathrm{c}1}}{\rho_{\mathrm{s}1} f_{\mathrm{c}2}}
$$

$\chi$ is the ratio of the plates' bending wavenumbers (equivalently the square
root of their critical-frequency ratio) and sets the total-internal-reflection
cut-off angle $\theta_\text{co} = \arcsin\chi$; $\psi$ is the ratio of their
bending-moment mobilities. For **identical plates** both are 1.

```python
from phonometry import vibration

params = vibration.junction_wave_parameters(0.1, 3200.0, 240.0, 0.2, 3200.0, 480.0)
chi, psi = params.chi, params.psi
#  -> (sqrt(0.5), 4.0)
```

## 2. Corner and straight-section coefficients

For an incident wave on plate 1, transmission **around the corner** (into the
perpendicular plate 2) is $\tau_{12}(\theta)$ (Eq. 5.12), and transmission
**across the straight section** (into the collinear plate 3, X- and
T-junction (1) only) is $\tau_{13}(\theta)$ (Eq. 5.13). Around the corner,

$$
\tau_{12}(\theta) =
\frac{0.5\,J_1 J_2 \psi \cos\theta \sqrt{\chi^2 - \sin^2\theta}}
     {(J_2\psi)^2 + \chi^2
      + J_2\psi\left(\sqrt{(1+\sin^2\theta)(\chi^2+\sin^2\theta)}
      + \sqrt{(1-\sin^2\theta)(\chi^2-\sin^2\theta)}\right)},
\qquad \chi \ge \sin\theta
$$

and beyond the cut-off the coefficient is zero, $\tau_{12}(\theta) = 0$ for
$\chi < \sin\theta$: no propagating wave is transmitted there.
The junction constants $J_1$, $J_2$, $J_3$ select the geometry:

A T-junction has three plates, and **which of them carries the incident wave**
is the whole distinction between its two cases. In **T-junction (1)** the wave
arrives on one of the two collinear members, so it can go straight on into the
collinear plate *and* round the corner into the branch. In **T-junction (2)**
it arrives on the branch, the stem of the T, so both available paths are
corners and there is no straight section at all. The opening figure of this
page is a T2 for exactly that reason: the 140 mm floor is the branch, and the
200 mm wall runs through.

| Junction | API string | $J_1$ | $J_2$ | $J_3$ |
|---|---|---|---|---|
| X | `"X"` | 1 | 1 | 1 |
| T-junction (1) | `"T1"` | 2 | 0.5 | 0.5 |
| T-junction (2) | `"T2"` | 2 | 2 | — |
| L | `"L"` | 4 | 1 | — |
| in-line | — | \* | \* | \* |

\* The in-line junction has no $J$ constants: it is a change of section, not a
corner, so the geometry is one-dimensional. See below.

The straight section is undefined for the T-junction (2) and the L-junction.

Two plates parametrise all of this because the closed forms describe a
**symmetric** junction, in which opposite plates are identical: for the X,
plate 3 is a copy of plate 1 and plate 4 a copy of plate 2; for T-junction (1),
plate 3 is a copy of plate 1; for T-junction (2), plate 4 is a copy of plate 2.
The page's own X-junction example is therefore a cross of two 100 mm plates and
two 200 mm plates, not four different ones. That is a real restriction on a real
building: a floor that differs above and below a junction, or a wall that
continues in a different thickness, has no closed form here. Bound it with the
two symmetric cases, or go to the EN 12354-1 Annex E empirical $K_{ij}$, or
measure it. The L-junction needs no such caveat because it genuinely has two
plates, and so, for a different reason, does the in-line junction.

### The in-line junction: a change of section

The in-line junction is the one case on this page where the two plates are
collinear, so there is no corner and the $J_1/J_2/J_3$ apparatus does not
apply. A 100 mm slab continuing as a 200 mm slab is a common building
situation, and it has its own closed form at normal incidence (Hopkins
Eq. 5.14, after Cremer et al. 1973):

$$
\tau_{12}(0°) = \left[\frac{2(1+\chi)(1+\psi)\sqrt{\chi\psi}}
{\chi(1+\psi)^2 + 2\psi(1+\chi^2)}\right]^2
$$

Only normal incidence is defined for it, for the geometric reason above; the
module notes that the value is within 1 dB of the angular average whenever
$\chi \ge 1$.

`inline_transmission_coefficient(chi, psi)` evaluates it. For the same
100 mm / 200 mm concrete pair as the figure above: $\chi = 0.707$,
$\psi = 4.0$ from `junction_wave_parameters`. It returns **0.936**, against
exactly 1 for identical plates. Doubling the thickness of a continuing slab
still passes 94 % of the incident power. That is the quantitative form of "a
continuous plate transmits fully", and it is why a change of section is a poor
place to look for isolation.

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

theta = np.radians(np.linspace(0.0, 90.0, 91))
tau12 = vibration.corner_transmission_coefficient(theta, chi, psi, "X")
tau13 = vibration.straight_transmission_coefficient(theta, chi, psi, "X")
```

The clip below runs this experiment in the time domain with the library's 2D
[elastic FDTD solver](https://docs.jmrp.io/phonometry/simulation/elastic-waves/): a bending packet on a 10 mm steel plate reaches an
L-junction with an identical plate, and the corner splits it into the
reflected and transmitted waves this section prices at $\tau_{12}(0°) = 0.5$.
The fast in-plane precursor racing down the receiving plate is the mode
conversion the pinned-junction model deliberately leaves out.

**Video:** [Bending waves at an L-junction in a 2D elastic FDTD simulation](https://docs.jmrp.io/phonometry/media/anim_elastic_plate_junction.webm): A 4 kHz tone-burst force launches a bending-wave packet along a 10 mm steel plate. On the straight control plate the packet just runs on and nothing returns. At an L-junction with an identical perpendicular plate it splits: part reflects, part turns the corner as a bending wave descending the vertical plate, and a faster in-plane precursor races ahead, with the closed-form transmission coefficient of 0.50 at normal incidence and the diffuse vibration reduction index of 5.2 dB annotated.

## 3. Diffuse-field angular average

In a diffuse vibration field every angle of incidence is equally probable and
the incident intensity carries a $\cos\theta$ obliquity factor, so the average
transmission coefficient is
$\bar{\tau}_{ij} = \int_0^{\pi/2} \tau_{ij}(\theta)\cos\theta\,\mathrm{d}\theta$
(Eq. 5.6). For **identical plates** the algebra collapses
to exact fractions that serve as the library's first-principles oracle:

- X-junction corner and straight: $\tau_{ij}(\theta) = \cos^2\theta / 8$, so
  $\bar{\tau}_{ij} = 1/12$;
- L-junction corner: $\tau_{ij}(\theta) = \cos^2\theta / 2$, so
  $\bar{\tau}_{ij} = 1/3$;
- in-line junction: $\tau_{12}(0°) = 1$ (a continuous plate transmits fully; for
  a change of section, `inline_transmission_coefficient` of section 2 applies).

```python
from phonometry import vibration

vibration.angular_average_transmission_coefficient(1.0, 1.0, "X", section="corner")  # 1/12
vibration.angular_average_transmission_coefficient(1.0, 1.0, "L", section="corner")  # 1/3
```

The two directions obey the SEA consistency relationship (Eq. 5.7),
$\bar{\tau}_{12} = \chi\,\bar{\tau}_{21}$, so only one direction needs to be
computed.

## 4. Coupling loss factor and vibration reduction index

The angular average is the bridge to the two junction descriptors used in
SEA-based building models: the **coupling loss factor** (Eq. 2.154)
$\eta_{ij} = c_{\mathrm{g},i} L_{ij} \tau_{ij} / (2\pi^2 f S_i)$ and the wave-approach
**vibration reduction index** (Eq. 5.116)
$K_{ij} = 10\log_{10}(1/\tau_{ij}) + 5\log_{10}(f_{\mathrm{c},j}/f_\text{ref})$ with
$f_\text{ref} = 1000\ \text{Hz}$ and
$f_{\mathrm{c},j}$ the critical frequency of the *receiving* plate. Combined with the
Eq. 5.7 reciprocity this form is symmetric, $K_{ij} = K_{ji}$, as EN 12354
requires of the junction descriptor. For the identical 100 mm concrete
X-junction ($f_\mathrm{c} \approx 203\ \text{Hz}$),
$K_{ij} = 10\log_{10} 12 + 5\log_{10}(203/1000) \approx 7.3\ \text{dB}$.

```python
from phonometry import vibration

eta = vibration.coupling_loss_factor(1.0 / 12.0, group_velocity=200.0,
                                     junction_length=4.0, frequency=500.0, plate_area=10.0)
res = vibration.junction_transmission("X", 0.1, 3200.0, 240.0, 0.1, 3200.0, 240.0)
kij = vibration.wave_vibration_reduction_index(res.corner_average,
                                               res.critical_frequency2)  # 7.33 dB
kij = res.corner_reduction_index  # the same, precomputed on the result

res.plot()   # tau(theta) for this junction's corner and straight paths (needs matplotlib)
```

$K_{ij}$ is also what ISO 10848 measures on a built junction: excite one
element, read the velocity level difference on both, and normalize by the
junction length, so the closed-form value above has a direct experimental
counterpart.

![ISO 10848 junction measurement rigs for an L-junction and a T-junction of 140 to 200 mm concrete plates: a shaker or hammer excites element i, accelerometers on elements i and j read the velocity level difference Dv,ij, and the junction length of at least 2.3 m runs along the highlighted corner line](https://docs.jmrp.io/phonometry/media/diagram_junction_rig.svg)

The junction descriptor is a *design* quantity: sweeping the receiving
plate's thickness shows how much a mass change at the junction buys. The
corner paths stiffen quickly with a heavier receiving plate, while the
straight (in-line) path of the X-junction rises fastest of all, since the
perpendicular plates increasingly pin the junction line:

![Wave-approach vibration reduction index Kij versus the thickness ratio of two concrete plates for the X-junction corner and straight paths, the T-junction corner and the L-junction corner, with the identical-plates X-junction value of about 7.3 dB marked](https://docs.jmrp.io/phonometry/media/junction_kij_thickness.svg)

**Show the code for this figure**

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

# Concrete plates (cL = 3200 m/s, rho = 2400 kg/m3): plate 1 fixed at
# 100 mm, plate 2 swept from 50 mm to 400 mm.
h1, cl, rho = 0.1, 3200.0, 2400.0
ratios = np.linspace(0.5, 4.0, 36)
curves = {"X corner": [], "X straight": [], "T-junction (1) corner": [],
          "L corner": []}
for ratio in ratios:
    h2 = h1 * float(ratio)
    res_x = vibration.junction_transmission("X", h1, cl, rho * h1, h2, cl, rho * h2)
    curves["X corner"].append(res_x.corner_reduction_index)
    # Only the corner index is precomputed on the result; the straight path
    # is the angular average tau13 converted the same way.
    curves["X straight"].append(float(vibration.wave_vibration_reduction_index(
        res_x.straight_average, res_x.critical_frequency2)))
    res_t = vibration.junction_transmission("T1", h1, cl, rho * h1, h2, cl, rho * h2)
    curves["T-junction (1) corner"].append(res_t.corner_reduction_index)
    res_l = vibration.junction_transmission("L", h1, cl, rho * h1, h2, cl, rho * h2)
    curves["L corner"].append(res_l.corner_reduction_index)

fig, ax = plt.subplots()
for label, values in curves.items():
    ax.plot(ratios, values, "--" if label == "X straight" else "-", label=label)
# The identical-plate X-junction: Kij = 10 lg 12 + 5 lg(fc2/1000).
res_eq = vibration.junction_transmission("X", h1, cl, rho * h1, h1, cl, rho * h1)
ax.scatter([1.0], [res_eq.corner_reduction_index], zorder=6,
           label="identical plates (tau = 1/12)")
ax.set_xlabel("Thickness ratio h2/h1")
ax.set_ylabel("Vibration reduction index Kij [dB]")
ax.set_title("Wave-approach junction Kij (Hopkins Eq. 5.116)")
ax.legend()
plt.show()
```

## 5. Worked example: feeding Kij into EN 12354

The number this page predicts is exactly what the EN 12354-1 flanking model
consumes. Take the 100 mm / 200 mm concrete X-junction of section 2's
$\tau(\theta)$ figure: its corner path gives $K_{12} = 9.8\ \text{dB}$. Handing it to
`flanking_element` in place of a tabulated Annex E value prices the
junction's three flanking paths and their effect on the apparent rating:

```python
from phonometry import building, vibration

# The 100 mm / 200 mm concrete X-junction of the tau(theta) figure:
res = vibration.junction_transmission("X", 0.1, 3200.0, 240.0, 0.2, 3200.0, 480.0)
k12 = res.corner_reduction_index                     # 9.8 dB (corner path)

# Feed it to the EN 12354-1 simplified model as this junction's Kij:
ff, df, fd = building.flanking_element(
    label="floor", r_flanking=49.0, r_separating=57.0,
    k_ff=k12, k_fd=k12, k_df=k12, separating_area=11.5, coupling_length=4.5)
pred = building.predicted_airborne_insulation(r_direct=57.0,
                                              flanking_paths=[ff, df, fd])
print(round(pred.r_prime_w, 1))                      # 55.4  (Rw 57 direct)
print(pred.dominant.label, round(pred.dominant.energy_fraction, 2))   # Dd 0.68
```

One junction with a moderate $K_{ij}$ already trims 1.6 dB off the direct
$R_\mathrm{w} = 57\ \text{dB}$; a full building repeats this for every junction, which
is the [EN 12354 prediction guide](https://docs.jmrp.io/phonometry/buildings/design/insulation-prediction/).

The measured, EN 12354 counterpart of $K_{ij}$ (from the direction-averaged
velocity level difference) is the separate flanking-transmission
`vibration_reduction_index`; this page is the closed-form *predicted* value from
the wave approach.

## 6. Experimental SEA: coupling loss factors from measured energies

Everything above is the **predictive** route: a coupling loss factor derived
from a wave model of the junction. Real joints (welds, bolt rows, spot welds,
adhesives) are not tractable that way, and the **experimental** route inverts
the steady-state SEA power balance from *measured* subsystem energies instead
(Norton & Karczub 2003, Sections 6.3.3 and 6.3.4). For two subsystems,

$$
\Pi_1 = \omega\left[(\eta_1 + \eta_{12})\,E_1 - \eta_{21}\,E_2\right]
\quad (6.10)
$$

$$
0 = \omega\left[(\eta_2 + \eta_{21})\,E_2 - \eta_{12}\,E_1\right]
\quad (6.11)
$$

with $E_i = M_i \langle v_i^2 \rangle$ the band energy of subsystem $i$ (its
mass times the space- and time-averaged mean-square velocity). Drive
subsystem 1 only, add the SEA consistency (reciprocity) relationship
$n_1\eta_{12} = n_2\eta_{21}$ (Eq. 6.8), and both coupling loss factors follow
from the two measured energies (Eq. 6.15):

$$
\eta_{12} = \frac{\eta_2 E_2}{E_1 - E_2\,n_1/n_2}, \qquad
\eta_{21} = \eta_{12}\,\frac{n_1}{n_2}, \qquad
\Pi_\text{in} = \omega\,(\eta_1 E_1 + \eta_2 E_2)
$$

The input power collapses to the total dissipated power, as it must in the
steady state: substituting Eq. (6.11) into Eq. (6.10) cancels the two coupling
terms exactly, which is a free check on any measurement, and one to run on
real data, not only on paper.

### The measurement those symbols stand for

$E_i$ and $\Pi_\text{in}$ are measured quantities, and the inversion is only as
good as they are. Norton's section 6.6.4 describes the experiment; the
requirements below are the ones that decide whether a result exists.

![Two states of the same experiment, side by side. In each, two plates are joined along an edge, one horizontal and one upright, with the junction line highlighted and several accelerometers distributed across both plates away from the edges and away from the drive point. In the first, a shaker drives the horizontal plate through an impedance head, so the injected power is measured on subsystem one and is zero on subsystem two; in the second, the shaker has been moved to the upright plate. Both states carry the band energies as the mass times the space-averaged mean-square velocity. Underneath, the injected power is given as one half the real part of force times conjugate velocity at the drive point, the space-averaging rule for the velocities, the statement that one run inverts only if the internal loss factors are known from a decay measurement while two runs solve all four, and the requirement that the bands hold several modes of each subsystem.](https://docs.jmrp.io/phonometry/media/diagram_power_injection_rig.svg)

- **Measure the injected power, do not infer it.** Drive through a force
  transducer or an impedance head so that
  $\Pi_\text{in} = \tfrac{1}{2}\operatorname{Re}\{F v^*\}$ is formed at the
  drive point (Norton Eq. 6.50). Taken from the cross-spectrum of the force and
  velocity signals it is a time average and behaves well under broadband random
  excitation; taken from $\operatorname{Re}\{Z\}$ it does not, because the
  impedance is small exactly at a resonance and the loss factor estimate then
  inherits a large error.
- **Space-average the velocities.** $\langle v_i^2 \rangle$ is a space *and*
  time average, so it takes several accelerometer positions per subsystem, away
  from the edges and away from the drive point, and the excitation is repeated
  at several randomly chosen drive points so that the single-point excitation
  satisfies the statistical-independence assumption the model rests on.
- **Choose bands that hold several modes.** The SEA balance is an average over
  the modes in a band, so the band must contain enough of them, which is what
  the modal densities computed in this section are for, and why the same
  section computes them.
- **Get $\eta_1$ and $\eta_2$ independently, or drive twice.** The single-drive
  inversion of Eq. (6.15) takes the internal loss factors as *inputs*. They come
  from a separate experiment on the decoupled subsystems: a reverberation-decay
  measurement, or the steady-state energy-flow technique of Norton 6.5.3
  ($\eta = \Pi_\text{in}/(\omega E)$, which needs the same input-power
  measurement again). If neither is available, drive each subsystem in turn
  instead: that is the two-drive scheme below, which needs no prior value at
  all.
- **Keep the boundary losses out.** The two-subsystem model assumes the energy
  flowing out of the pair at its far boundaries is negligible: fine wire point
  supports or foam pads approximating free-free conditions are what the
  literature uses. Mass loading, contact damping and exciter-structure feedback
  at the drive point are the same errors the mobility guide's
  [transducer section](https://docs.jmrp.io/phonometry/vibration/structural/mechanical-mobility/)
  quantifies, and they apply here unchanged.
- **Read the coupling against the damping.** The figure's own acceptance test:
  the coupling loss factors should sit at least an order of magnitude below the
  internal loss factors, or the subsystems are not weakly coupled and the model
  does not hold. `PowerInjectionResult.coupling_strength` reports that ratio.

`power_injection_clf` performs that inversion, with the modal densities of
Norton Section 6.4.1 alongside it: `flat_plate_modal_density` (Eq. 6.25),
`bar_modal_density` (6.23), `beam_modal_density` (6.24) and
`cylindrical_shell_modal_density` (6.27 to 6.29, the Szechenyi approximations
in three regimes about the `ring_frequency` of Eq. 6.26). The flat-plate
expression $n(f) = S\sqrt{12}/(2 c_\mathrm{L} t)$ is the same quantity as EN 12354-4's
$n = \pi S f_\mathrm{c} / c_0^2$ used by the
[flanking model](https://docs.jmrp.io/phonometry/buildings/insulation/flanking-lab/), only
parametrised by the plate itself rather than by its critical frequency.

![Two-panel figure. Left panel: coupling loss factor on a logarithmic axis against the octave bands from 125 hertz to 2 kilohertz for two aluminium plates at right angles, with the welded line junction falling from about 3 times 10 to the minus 3 to 8 times 10 to the minus 4 as one over the square root of frequency, the twelve-bolt point connection falling twice as steeply from about 1.4 times 10 to the minus 2 to 9 times 10 to the minus 4 as one over frequency, and a dashed horizontal line marking an internal loss factor of 10 to the minus 2. Right panel: a bar chart of the four loss factors of a satellite platform and cylinder in the 500 hertz octave on a logarithmic axis, the internal loss factors 4.40 times 10 to the minus 3 and 2.40 times 10 to the minus 3 clearly above the coupling loss factors 4.26 times 10 to the minus 4 and 3.91 times 10 to the minus 4, with the injected power of 1.31 watts annotated.](https://docs.jmrp.io/phonometry/media/experimental_sea_clf.svg)

*Left: two ways of joining the same two plates. Right: the loss-factor budget
inverted from a pair of measured velocities; the coupling stays an order of
magnitude below the damping, which is the condition for a two-subsystem SEA
model to be trustworthy.*

**Show the code for this figure**

```python
import math
import matplotlib.pyplot as plt
import numpy as np
from phonometry import vibration
from phonometry.vibration.structural.point_mobility import plate_bending_wave_speed

rho, nu, young = 2700.0, 0.33, 7.1e10                 # aluminium
cl = math.sqrt(young / (rho * (1.0 - nu**2)))         # 5432 m/s (Eq. 6.25)
bands = np.array([125.0, 250.0, 500.0, 1000.0, 2000.0])

# Predicted: a 3 mm x 2.5 m x 1.2 m plate meeting a 5.5 mm x 2.0 m x 1.2 m
# plate at right angles along the 1.2 m edge, welded and then bolted.
h1, h2, area1, length = 0.003, 0.0055, 2.5 * 1.2, 1.2
tau = vibration.right_angle_transmission_coefficient(h1, h2, density1=rho, density2=rho,
                                                     wave_speed1=cl, wave_speed2=cl)
cb = plate_bending_wave_speed(bands, vibration.plate_bending_stiffness(young, h1, nu),
                              rho * h1)
welded = [float(vibration.coupling_loss_factor(tau, 2.0 * c, length, area1, frequency=f))
          for c, f in zip(cb, bands, strict=True)]
bolted = vibration.point_connection_coupling_loss_factor(
    bands, 12, thickness1_m=h1, thickness2_m=h2, surface_density1=rho * h1,
    surface_density2=rho * h2, wave_speed1=cl, wave_speed2=cl,
    plate_area1=area1)

# Measured: a 5 mm platform driven directly, a 3 mm cylinder driven only
# through the joints; 27.2 and 13.2 mm/s in the 500 Hz octave.
t_p, t_c, radius = 0.005, 0.003, 0.75
area_c = 2.0 * math.pi * radius * 2.0
area_p = 3.5 * 3.0 - math.pi * radius**2
sea = vibration.power_injection_clf(
    rho * t_p * area_p * 0.0272**2, rho * t_c * area_c * 0.0132**2, 4.4e-3,
    2.4e-3, vibration.flat_plate_modal_density(area_p, t_p, cl),
    vibration.cylindrical_shell_modal_density(500.0, area_c, t_c, radius, cl),
    frequencies=500.0)

print(f"{float(sea.coupling_loss_factor12[0]):.2e}")   # 4.26e-04
print(f"{float(sea.coupling_loss_factor21[0]):.2e}")   # 3.91e-04
print(round(float(sea.input_power[0]), 2))             # 1.31 W
sea.plot()
plt.show()
```

Two closed forms sit between the two routes and are what a measured value is
normally compared against.
`right_angle_transmission_coefficient` gives the wave transmission coefficient
of a right-angle plate junction without any angular integration (Eqs. 6.53 to
6.55, after Bies & Hamid and Cremer et al.); feeding it to
`coupling_loss_factor` reproduces Norton Eq. (6.52) identically, because that
equation and Hopkins Eq. (2.154) are the same expression once $c_\mathrm{g} = 2 c_\mathrm{B}$.
And `point_connection_coupling_loss_factor` covers plates joined at $N$
discrete points instead of along a line (Eq. 6.56). Which one applies is
decided by the bending wavelength: use the point form when it is shorter than
the joint length, and the line form when it is longer. The two differ in slope
as well as in level, $1/f$ against $1/\sqrt{f}$, so bolting and welding are
not interchangeable across the spectrum.

When no independent value of the internal loss factors is available, the
single-drive inversion is underdetermined, and the classical **power-injection
method** drives each subsystem in turn while measuring both energies each time
(the two states the rig diagram above draws). That gives four equations for
$\eta_1$, $\eta_2$, $\eta_{12}$ and $\eta_{21}$
with no prior assumption at all, which `power_injection_matrix` solves band by
band, taking the two energy pairs and the two injected powers and needing no
modal density. It costs one extra run and it removes the measurement that is
usually hardest to get (the decoupled decay), which is why it is the classical
form. Run on the platform-and-cylinder example of the figure above it returns a
`modal_density_ratio` of 0.919, against a geometric
`flat_plate_modal_density / cylindrical_shell_modal_density` of 0.919 for the
same structures; the two agreeing is the check.

Reciprocity then becomes a *check* on the measurement rather than an
input: `PowerInjectionResult.modal_density_ratio` is a property, not a
function, and it returns only the $n_1/n_2$ *implied* by the measured coupling
loss factors, $\eta_{21}/\eta_{12}$. Making it a check is the reader's step:
compare it against the ratio built from the geometry with
`flat_plate_modal_density` and its siblings, and a large disagreement means the
subsystem boundaries were drawn in the wrong place. `coupling_strength` is the
other diagnostic on the same result, and says whether the weak-coupling
assumption the inversion rests on holds at all.

## What this guide covers

- **Covered.** The frequency-independent, rigid-junction transmission coefficients of Cremer,
  Heckl & Ungar (1973), tabulated by Craik (1981/1996) and collected in Hopkins
  (2007, Section 5.2.1.3), for the X, T, L and in-line junctions of thin,
  homogeneous, isotropic plates: the wave parameters $\chi$/$\psi$, the corner
  and straight-section coefficients $\tau_{12}(\theta)$/$\tau_{13}(\theta)$,
  their diffuse-field angular average, the SEA coupling loss factor and the
  wave-approach vibration reduction index $K_{ij}$, via
  `junction_wave_parameters`, `corner_transmission_coefficient`,
  `straight_transmission_coefficient`,
  `angular_average_transmission_coefficient`, `inline_transmission_coefficient`,
  `coupling_loss_factor`, `wave_vibration_reduction_index` and
  `junction_transmission`.

  Section 6 covers the **experimental** route of Norton & Karczub Chapter 6: the
  two-subsystem power balance and its in-situ inversion (`power_injection_clf`,
  `power_injection_matrix`), the modal densities of bars, beams, flat plates and
  thin-walled cylindrical shells with the `ring_frequency` they are split about
  (`bar_modal_density`, `beam_modal_density`, `flat_plate_modal_density`,
  `cylindrical_shell_modal_density`), and the two closed forms that sit between
  the routes (`right_angle_transmission_coefficient`,
  `point_connection_coupling_loss_factor`).

- **Not covered.** This predicted $K_{ij}$ is a closed-form idealisation for a rigid, simply
  supported junction, not a measurement: the measured, empirical $K_{ij}$ from a
  direction-averaged velocity level difference (ISO 10848) is the separate
  `vibration_reduction_index` of the [Laboratory Flanking Transmission
  guide](https://docs.jmrp.io/phonometry/buildings/insulation/flanking-lab/). The straight-section
  coefficient $\tau_{13}$ is undefined for the T-junction (2) and L-junction
  geometries, which have no collinear third plate, so only the corner path
  applies there. The tabulated coefficients also assume a *symmetric* junction
  (opposite plates identical), and there is no closed form here for one that is
  not. On the experimental side, the inversion of section 6 is written for
  **two** subsystems only, and `PowerInjectionResult.modal_density_ratio` is a
  consistency check the reader performs rather than one the library automates.

## See also

- [Predicting Sound Insulation (EN 12354)](https://docs.jmrp.io/phonometry/buildings/design/insulation-prediction/):
  the flanking model that consumes $K_{ij}$ junction by junction.
- [Laboratory Flanking Transmission (ISO 10848)](https://docs.jmrp.io/phonometry/buildings/insulation/flanking-lab/):
  the measurement of $K_{ij}$ from velocity level differences, the empirical
  counterpart of this page's closed forms.
- [Structure-borne sound power of equipment (EN 15657)](https://docs.jmrp.io/phonometry/buildings/design/structure-borne-power/):
  the source power that these junction transmissions carry through a building.
- [Mechanical mobility and the FRF family (ISO 7626-1)](https://docs.jmrp.io/phonometry/vibration/structural/mechanical-mobility/):
  the plate mobilities behind the wave parameters.
- API reference: [`vibration`: junction transmission](https://docs.jmrp.io/phonometry/reference/api/vibration/junction-transmission/).
- 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 picture behind a junction, and the radiation efficiency each element carries.

## References

- **Craik, R. J. M. (1996).** [*Sound transmission through buildings using statistical energy analysis*](https://openlibrary.org/isbn/9780566075728). Gower.
  The SEA treatment of building sound transmission with the tabulated bending-wave transmission coefficients for X, T, L and in-line junctions used here (Eqs 5.12/5.13). ISBN 978-0-566-07572-8.
- **Cremer, L., Heckl, M., & Ungar, E. E. (1973).** *Structure-borne sound: Structural vibrations and sound radiation at audio frequencies* (1st ed.). Springer. https://doi.org/10.1007/978-3-662-10118-6
  The original derivation of the wave parameters chi and psi (Eqs 5.10/5.11 in Hopkins) and the in-line normal-incidence transmission coefficient. ISBN 978-3-540-06002-4.
- **Hopkins, C. (2007).** *Sound insulation*. Butterworth-Heinemann. https://doi.org/10.4324/9780080550473
  Section 5.2.1.3 collects the rigid X, T, L and in-line junction coefficients, the angular average (Eq. 5.6), the SEA consistency relationship (Eq. 5.7), the coupling loss factor (Eq. 2.154) and the wave-approach Kij (Eq. 5.116) implemented on this page. ISBN 978-0-7506-6526-1.
- **Norton, M. P., & Karczub, D. G. (2003).** *Fundamentals of noise and vibration analysis for engineers* (2nd ed.). Cambridge University Press. https://doi.org/10.1017/CBO9781139163927
  Chapter 6 gives the two-subsystem SEA power balance (Eqs. 6.10, 6.11), the consistency relationship (Eq. 6.8), the in-situ inversion of the coupling loss factors from measured energies (Eq. 6.15), the modal densities of bars, beams, flat plates and thin-walled cylindrical shells (Eqs. 6.23-6.29) and the closed-form right-angle and point-connection coupling loss factors (Eqs. 6.52-6.56) implemented on this page. ISBN 978-0-521-49913-2.
