# Sound Power by Intensity Scanning

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

Canonical: https://docs.jmrp.io/phonometry/devices/emission/sound-power-intensity/
Language: en
Alternate: https://docs.jmrp.io/phonometry/es/devices/emission/sound-power-intensity/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: 2026-07-28
Updated: 2026-11-16T21:36:14Z
Standards: ISO 9614-2:1996; ISO 9614-3:2002
Verified: 2 conformance checks (ISO 9614-2:1996; ISO 9614-3:2002) https://docs.jmrp.io/phonometry/reference/conformance/
License: MIT (https://github.com/jmrplens/phonometry/blob/main/LICENSE)
Summary: Sound power from scanned intensity: the ISO 9614-2 engineering and survey methods with field indicators and achieved grade, and ISO 9614-3 precision.

---

The pressure routes to the sound power level fail exactly where machines
live: on a factory floor where the neighbouring lines cannot be switched
off. Sound **intensity** is the net energy flux, so it distinguishes energy
*leaving* the source from steady energy merely passing through the
measurement surface, and the scanning methods built on it tolerate
extraneous noise that would defeat any pressure method. This guide covers
the two of them: the ISO 9614-2 engineering/survey determination with its
field indicators and per-band achieved grade, the ISO 9614-3 precision
(grade 1) scan with its tighter criteria, and the accredited-style test
fiche. The probe itself, its finite-difference bias and the ISO 9614-1
discrete-point method live in
[Sound Intensity (p-p)](https://docs.jmrp.io/phonometry/devices/emission/intensity/); which route fits
which job is weighed in [Sound Power](https://docs.jmrp.io/phonometry/devices/emission/sound-power/).

## 1. Intensity scanning (ISO 9614-2)

A p-p probe (see the
[Sound Intensity guide](https://docs.jmrp.io/phonometry/devices/emission/intensity/)) is swept
continuously over each of $N$ segments of a surface enclosing the source,
reporting the segment-averaged signed normal intensity
$\langle I_{\mathrm{n},i} \rangle$: one row per segment, one column per band.

![A two-microphone p-p sound intensity probe: two pressure microphones separated by a spacer, from which the pressure gradient and hence the normal intensity are estimated](https://docs.jmrp.io/phonometry/media/diagram_pp_probe.svg)

The partial powers sum to the total:

$$
P_i = \langle I_{\mathrm{n},i} \rangle\ S_i, \qquad P = \sum_i P_i, \qquad
L_W = 10 \log_{10}\frac{P}{P_0},\quad P_0 = 1\ \text{pW} .
$$

**Video:** [Intensity scanning over a measurement box](https://docs.jmrp.io/phonometry/media/anim_intensity_scan_power.webm): A p-p probe traces the serpentine scan over the top face of the measurement box while the normal-intensity arrows appear behind it, and the partial powers of the five faces accumulate into the sound power level L_W.

### Running the scan

Everything above arrives as an array; this is where it comes from. The
measurement surface is divided into **at least four segments** (clause 8.2),
chosen where possible so that a region of net inflow is not averaged together
with a region of net outflow, and so that each segment faces one component of
the machine. Its mean distance from the source surface is at least **200 mm**
for an extended vibrating surface, reducible to 100 mm for a small compact
source. At that point, action a) of Table B.1 ("move the surface out") is no
longer available to you.

The scan itself is a path, not a set of points. Its basic element is a straight
line; the path must cover the segment uniformly at uniform speed, with adjacent
lines equally spaced, and on the initial measurement surface **the mean line
spacing shall not exceed the mean distance from the segment to the source**
(clause 8.1). Sweep speed is 0.1 m/s to 0.5 m/s by hand, 0 m/s to 1 m/s
mechanically; each segment takes **at least 20 s**, with the time-averaging
started and stopped at the segment boundaries so that the reported
$\langle I_{\mathrm{n},i} \rangle$ belongs to that segment and nothing else. The probe
axis stays normal to the surface throughout, and the operator stands to the side
of the segment being scanned, not in front of it.

For the engineering grade each segment is scanned **twice, and the two paths
must be orthogonal**: the second sweep rotated 90° wherever that is possible
(clause 8.3.1). That is what `normal_intensity_2` is: not a repeat of the same
sweep, but the rotated one. Criterion 3 exists precisely to catch a field that
does not repeat, and repeating an identical path passes it on a determination
that a rotated path would fail. Note 10 of the same clause adds the practical
consequence: above $F_{pI} = 10\ \text{dB}$, scan speeds over 0.25 m/s are
likely to fail criterion 3 whatever the stability of the field.

### The field indicators and their criteria

A band whose net power is **non-positive** ($P \le 0$: more energy entering the
surface than leaving it, from a stronger source outside) is **not determinable**
and reported as `NaN` (clause 9.2). Two normative field indicators qualify the
bands that are. The **surface pressure-intensity indicator** $F_{pI}$ measures
how reactive or how contaminated the field is, and the
**negative-partial-power indicator** $F_{+/-}$ measures how much energy
circulates in and out across the surface:

$$
F_{pI} = \overline{L_p} - L_W + 10 \log_{10}\frac{S}{S_0}, \qquad
F_{+/-} = 10 \log_{10}\frac{\sum_i \lvert P_i \rvert}{\lvert \sum_i P_i \rvert} ,
$$

where $\overline{L_p}$ is the area-weighted surface pressure level over the
segments. These are the ISO 9614-2 names for the two ISO 9614-1 Annex A
indicators the [intensity guide](https://docs.jmrp.io/phonometry/devices/emission/intensity/) calls
$F_2$ and $F_3$: $F_{pI} \equiv F_3$ and $F_{+/-} \equiv F_3 - F_2$ for
equal-area segments, so the field indicators there and the qualification
criteria here are the same two numbers under two standards' notations.

The probe's **dynamic capability** $L_\mathrm{d} = \delta_{pI0} - K$ (pressure-residual
intensity index minus the bias factor $K$, 10 dB for grade 2 and 7 dB for
grade 3) must exceed $F_{pI}$ (criterion 1); $F_{+/-} \le 3\ \text{dB}$ is
criterion 2 (mandatory for grade 2); and the two rotated sweeps must agree
within the Table 2 limit $s$ per segment (criterion 3). A band is
**engineering** grade
when criteria 1, 2 and 3 hold, **survey** when 1 and 3 hold, else `none`.

### Reading the indicators

Neither indicator is a boolean, and both carry a scale. $F_{pI}$ sits near zero
when everything the surface encloses radiates outward and nothing else reaches
it; a few decibels is normal in-situ work; and clause 8.3.2 asks that every
effort be made to keep it **below 10 dB** by the choice of surface. A large
$F_{pI}$ has two causes with opposite fixes: a reactive near field, answered by
moving the surface *out*, and a strong external source, answered by moving the
surface *in* towards the machine or by shielding it. $F_{+/-}$ reads
differently: zero when every segment radiates outward, and 3 dB when the inward
and outward partial powers are already comparable to the net, so above 3 dB the
surface is straddling two sources and the fix is to re-cut it. Averaging longer
does not help, because the problem is where the surface is, not how noisy the
estimate is.

When criterion 1 fails, the remedies are cheap before they are expensive:
enlarge the surface away from the near field, shield or remove the extraneous
source, fit a **larger spacer** to raise $\delta_{pI0}$ (see
[Choosing the probe](https://docs.jmrp.io/phonometry/devices/emission/intensity/#choosing-the-probe)),
and only then consider better hardware. Dropping the claim from engineering to
survey grade buys 3 dB of $L_\mathrm{d}$ on its own, because $K$ goes from 10 dB to
7 dB: the trade-off is the loss of grade.

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

# 6 surface segments x 6 octave bands: signed normal intensity (W/m^2) from two
# repeated sweeps, the segment areas, and the per-segment surface SPL (dB).
freqs = np.array([125, 250, 500, 1000, 2000, 4000], dtype=float)
areas = np.full(6, 0.5)                                 # 0.5 m^2 per segment
rng = np.random.default_rng(0)
scan1 = np.abs(rng.normal(1e-4, 2e-5, size=(6, 6)))     # (segments, bands)
scan2 = scan1 * (1.0 + rng.normal(0.0, 0.02, size=(6, 6)))
pressure = np.full((6, 6), 80.0)

res = emission.sound_power_intensity(
    scan1, areas, normal_intensity_2=scan2, pressure_levels=pressure,
    pressure_residual_index=12.0, frequencies=freqs,
    fraction=1, grade="engineering",
)
print(np.round(res.sound_power_level, 1))               # per-band LW
print(round(res.sound_power_level_a, 1))                # LWA over determinable bands
print(round(float(res.dynamic_capability_index[0]), 1)) # Ld = 12 - 10 = 2.0 dB
print(round(float(res.surface_pressure_intensity_index[0]), 2))   # FpI
print(list(res.achieved_grade))                         # per-band grade

res.plot()   # LW spectrum; non-positive (undeterminable) bands hatched (needs matplotlib)
```

![The intensity-scanning sound power level spectrum of the ISO 9614-2 example, one bar per octave band from 125 Hz to 4 kHz all near 85 dB, with the A-weighted total of 90.9 dB(A) in the title](https://docs.jmrp.io/phonometry/media/sound_power_intensity_result.svg)

*The partial powers $\langle I_{\mathrm{n},i} \rangle\,S_i$ of the six segments sum to
each band's $L_W$; every band here nets positive power and passes the
field-indicator criteria at engineering grade, so all six bars stand, and the
A-weighted total of 90.9 dB(A) heads the title.*

**Show the code for this figure**

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

# 6 surface segments x 6 octave bands: signed normal intensity (W/m^2) from two
# repeated sweeps, the segment areas, and the per-segment surface SPL (dB).
freqs = np.array([125, 250, 500, 1000, 2000, 4000], dtype=float)
areas = np.full(6, 0.5)                                 # 0.5 m^2 per segment
rng = np.random.default_rng(0)
scan1 = np.abs(rng.normal(1e-4, 2e-5, size=(6, 6)))     # (segments, bands)
scan2 = scan1 * (1.0 + rng.normal(0.0, 0.02, size=(6, 6)))
pressure = np.full((6, 6), 80.0)
res = emission.sound_power_intensity(
    scan1, areas, normal_intensity_2=scan2, pressure_levels=pressure,
    pressure_residual_index=12.0, frequencies=freqs,
    fraction=1, grade="engineering",
)

# res is the SoundPowerIntensityResult computed above. One line:
res.plot()
plt.show()

# By hand: a bar spectrum of LW with the A-weighted total in the title.
freqs = res.frequencies
positions = np.arange(freqs.size)
fig, ax = plt.subplots()
ax.bar(positions, res.sound_power_level, width=0.7, color="#1f77b4")
ax.set_xticks(positions)
ax.set_xticklabels([f"{f:g}" for f in freqs], rotation=45, ha="right")
ax.set_xlabel("Frequency [Hz]")
ax.set_ylabel("Sound power level LW [dB]")
ax.set_title(
    f"Intensity-scanning sound power (ISO 9614-2)  "
    f"LWA = {res.sound_power_level_a:.1f} dB(A)")
plt.show()
```

Supplying `normal_intensity_2` (the second, rotated sweep) averages the two for
the
partial powers and evaluates criterion 3; `pressure_levels` enables $F_{pI}$;
`pressure_residual_index` ($\delta_{pI0}$) plus a second sweep enables the
per-band
achieved grade. The probe's finite-difference intensity has a
frequency-dependent bias handled in the [intensity guide](https://docs.jmrp.io/phonometry/devices/emission/intensity/).

### Where the A-weighted total loses bands

`sound_power_level_a` is not a sum over every band, and the difference matters
for anything that will be compared against a limit. Two things take a band out
of it, for different reasons:

- a band whose net power is non-positive is **not determinable at all**
  (clause 9.2) and comes back `NaN`; `negative_band` flags it;
- a band that *is* determinable but fails criterion 1, or at engineering grade
  criterion 2, is a perfectly good number that **clause 10.6 b) forbids you to
  include** in an A-weighted determination unless its contribution is negligible
  per clause 4.3; `a_weighting_omitted_bands` records exactly which ones.

Clause 10.6 b) also makes stating that omission part of the determination: an
$L_{W\mathrm{A}}$ reported without saying which bands were dropped does not conform. The
consequence is one-directional: the total is an **underestimate**, and the
reader can bound it by summing the omitted bands at their computed level and
adding them back. Note also the failure mode of the screening itself: without
`pressure_levels` and `pressure_residual_index` the criteria cannot be
evaluated, so every determinable band is summed and a `SoundPowerWarning` says
so. A quiet total is not the same as a screened one.

### Criterion 3 is not quite a hard gate

Criterion 3 is evaluated per segment and per band, which is why `repeatability`
comes back per segment rather than as one verdict. A failing segment does not
automatically void its band: clause 8.3.1 allows the determination to stand when
the summed partial powers of the failing segments lie **more than 10 dB below**
the source power obtained from the passing ones, their contribution then being
negligible. What it does not allow is silence: the bands in which criterion 3
was not satisfied must carry a statement in the report that their uncertainty
exceeds the Table 2 value for the intended grade.

The `repeatability_limit` default deserves the same care. Table 2 tabulates $s$
per band for the **engineering** grade only; for the survey grade it gives a
single A-weighted value of 4 dB, and this implementation reuses that 4 dB as a
per-band survey limit. That is a deliberate extrapolation, not a normative
value, and it is looser than any grade-2 limit in the table: a survey-grade
criterion-3 pass is therefore not evidence of grade-2 repeatability. Table 2
is also bounded: `_table2_s` raises outside the octave bands 63 Hz to 4 kHz and
the one-third-octave bands 50 Hz to 6.3 kHz, so an 8 kHz octave is an
exception, not a wide limit.

### The instrument and the environment

The fiche's `"Class 1 p-p intensity probe (IEC 61043)"` is a claim with clauses
behind it. The instrument **including its probe** shall comply with IEC 61043,
verified in a laboratory at least annually, or at least every two years if an
intensity calibrator is applied before each power determination (clause 6.2).
Before each series, a class 0 or class 1 IEC 60942 calibrator goes on each probe
microphone (clause 6.2.1), and the probe-reversal test of clause 6.2.2 confirms
in situ that the chain agrees on which way is out; both procedures, and what
their failures mean, are in
[Sound Intensity](https://docs.jmrp.io/phonometry/devices/emission/intensity/#measuring-the-residual-index-and-checking-the-probe).
Clause 10.5 makes reporting both of them part of the deliverable.

The environment has limits too. The mean air-flow velocity over the measurement
surface shall be below **4 m/s**, or a windscreen shall be used (clause 6.1);
static pressure and temperature are measured because they set the density and
the speed of sound the intensity is computed with.

When a criterion fails, Annex B's Table B.1 lists what to change rather than
leaving it at a `False`: increase the averaging or the scan time, move the
measurement surface, double the scan-line density, subdivide the surface into
more segments, or shield the extraneous source. The action codes are keyed to
which criterion failed, which is why the result reports the criteria separately
instead of one boolean.

### `sound_power_intensity()` parameters

| Parameter | Type | Units | Range / default | Notes |
| :--- | :--- | :--- | :--- | :--- |
| `normal_intensity` | 2D array | W/m² | `(N_seg, N_bands)` | Signed segment-averaged normal intensity $\langle I_{\mathrm{n},i} \rangle$ (first sweep) |
| `areas` | 1D array | m² | > 0, `(N_seg,)` | Segment areas $S_i$ |
| `normal_intensity_2` | 2D array | W/m² | same shape | Second sweep → criterion 3 and averaging |
| `pressure_levels` | 2D array | dB | same shape | Segment SPL $L_{pi}$ → $F_{pI}$ |
| `pressure_residual_index` | float or 1D array | dB | — | $\delta_{pI0}$ → $L_\mathrm{d}$ / criterion 1 |
| `frequencies` | 1D array | Hz | octave 63 Hz-4 kHz, third 50 Hz-6.3 kHz | $L_{W\mathrm{A}}$ and Table 2 limits; outside the range Table 2 has no $s$ and the call raises |
| `fraction` | int | — | `3` (one-third octaves, default) / `1` (octaves) | Table 2 lookup |
| `grade` | str | — | `'engineering'` (default) / `'survey'` | Selects $K$ |
| `repeatability_limit` | float or 1D array | dB | default Table 2 (survey: the A-weighted 4 dB reused per band) | Override criterion-3 $s$ |

Returns a `SoundPowerIntensityResult`: `partial_power`/`partial_power_level`
per segment and band, `sound_power`/`sound_power_level` (band total, `NaN`
where `negative_band`), `negative_band` (the non-determinable bands),
`surface_pressure_intensity_index` ($F_{pI}$),
`negative_partial_power_index` ($F_{+/-}$), `repeatability`,
`dynamic_capability_index` ($L_\mathrm{d}$), `achieved_grade`,
`a_weighting_omitted_bands` (the bands clause 10.6 b) kept out of $L_{W\mathrm{A}}$),
`surface_area`, `sound_power_level_a` and `grade`.

`partial_power_level` is the field that pays for the whole method. A pressure
determination returns one number per band; an intensity scan returns one per
segment *and* band, signed, so the result says **where** the power leaves the
machine and where it flows back in:

![The five faces of a measurement box laid out flat as a developed net, each tiled into four segments and shaded by its partial power level in decibels re 1 pW. The front face is the loudest at 73 to 74 dB, the sides near 68 dB, and one segment of the back face is hatched and signed negative because its net power flows inward. Below, a bar chart of the five face totals against the band sound power level](https://docs.jmrp.io/phonometry/media/partial_power_map.svg)

*The measurement box unfolded, one tile per segment. The front face carries most
of the emission; the hatched segment on the shadow side nets **inward**, which
is normal beside a louder neighbour and is exactly what $F_{+/-}$ counts. A
band-summed $L_W$ spectrum cannot show any of this, and it is the map, not the
spectrum, that tells you which panel to treat.*

**Show the code for this figure**

```python
import matplotlib.patches as mpatches
import matplotlib.pyplot as plt
import numpy as np

# A 1.2 x 0.8 x 1.0 m machine boxed at d = 0.25 m: five faces, four segments
# each, one of them taking net inflow from a louder neighbour.
face_area = np.array([1.7 * 1.5, 1.3 * 1.5, 1.7 * 1.5, 1.3 * 1.5, 1.7 * 1.3])
seg_area = np.repeat(face_area / 4.0, 4)
share = np.array([2.6, 2.2, 2.0, 1.8, 0.9, 0.8, 0.7, 0.7,
                  0.5, 0.4, -0.35, 0.3, 0.8, 0.7, 0.6, 0.6,
                  1.1, 1.0, 0.9, 0.8])
res = emission.sound_power_intensity(
    (share * 1.0e-5 / seg_area)[:, None], seg_area,
    frequencies=np.array([500.0]))
level, signed = res.partial_power_level[:, 0], res.partial_power[:, 0]

# The developed net: four side faces in a row, the top above the front.
origin = ((0, 0), (1.7, 0), (3.0, 0), (4.7, 0), (0, 1.5))
size = ((1.7, 1.5), (1.3, 1.5), (1.7, 1.5), (1.3, 1.5), (1.7, 1.3))
fig, ax = plt.subplots(figsize=(11, 5))
for f, ((ox, oy), (w, h)) in enumerate(zip(origin, size, strict=True)):
    for k in range(4):
        value = level[4 * f + k]
        ax.add_patch(mpatches.Rectangle(
            (ox + (k % 2) * w / 2, oy + (1 - k // 2) * h / 2), w / 2, h / 2,
            facecolor=plt.get_cmap("viridis")(
                (value - level.min()) / (level.max() - level.min())),
            edgecolor="black",
            hatch="//" if signed[4 * f + k] < 0 else None))
ax.set(xlim=(-0.2, 6.2), ylim=(-0.4, 3.0), aspect="equal")
plt.show()
```

## 2. Precision intensity scanning (ISO 9614-3)

ISO 9614-3 is the grade-1 scanning method: like ISO 9614-2 it integrates the
normal intensity over a surface enclosing the source, but with a continuous
scan, tighter field-indicator criteria and an explicit uncertainty budget.

![ISO 9614-3 precision sound intensity scanning: a source enclosed by a measurement surface divided into segments, a two-microphone intensity probe scanned along a serpentine path over each segment, and the sound power formed by summing the normal intensity times segment area, subject to the field-indicator acceptance criteria](https://docs.jmrp.io/phonometry/media/diagram_intensity_scan.svg)

**Power and level (Clause 7).** The partial power of each segment is
$P_i = I_{\mathrm{n},i}\,S_i$; the total $P = \sum_i P_i$ gives
$L_W = 10\log_{10}(P/P_0)$, $P_0 = 1\ \text{pW}$. A band whose net power is
non-positive ($P \le 0$) is flagged not-applicable rather than
logged. The field indicators (temporal variability $F_\mathrm{T}$, the signed and
unsigned pressure–intensity indicators, and the non-uniformity $F_\mathrm{S}$) drive the
five acceptance criteria.

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

# A fully enclosing surface with a uniform normal intensity In = W/S recovers
# the source power exactly: LW = 10*lg(W/P0). Here W = 100 uW -> 80 dB.
areas = np.array([0.5, 1.0, 0.25, 2.0])
w = 1.0e-4
i_n = np.full(areas.shape, w / float(areas.sum()))
res = emission.sound_power_intensity_precision(i_n, areas)
print(round(float(res.sound_power[0]), 6))          # 0.0001
print(round(float(res.sound_power_level[0]), 2))    # 80.0
```

Across several bands the result carries the per-band $L_W$ (`NaN` where the net
power is non-positive), flags those bands `not_applicable`, and draws them
with the one-line `result.plot()` of the figure below:

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

# Four partial surfaces scanned over five one-third-octave bands. Each cell of
# partial_intensity is the signed normal intensity In_i (W/m^2); areas are the
# partial-surface areas Si. The 250 Hz band has net-negative power (a locally
# reactive field), so ISO 9614-3 flags it not-applicable (clause 9.2) -> NaN.
freqs = np.array([250, 500, 1000, 2000, 4000], float)
areas = np.array([0.5, 1.0, 0.75, 0.5])
base_intensity = np.array([2.0e-6, 8.0e-6, 2.0e-5, 1.0e-5, 3.0e-6])
partial_intensity = base_intensity[None, :] * np.array([1.0, 1.1, 0.9, 1.05])[:, None]
partial_intensity[:, 0] = [2.0e-6, -3.0e-6, -4.0e-6, -1.0e-6]   # net-negative band

result = emission.sound_power_intensity_precision(partial_intensity, areas, frequencies=freqs)
print(result.not_applicable_band.tolist())   # [True, False, False, False, False]
print(round(result.sound_power_level_a, 1))   # 80.6
result.plot()   # LW spectrum; the not-applicable band is hatched (needs matplotlib)
```

![The precision intensity-scanning sound power level spectrum over five one-third-octave bands, four determinate bars and a hatched, greyed 250 Hz band flagged not-applicable because its net intensity is negative, with the A-weighted total of 80.6 dB(A) in the title](https://docs.jmrp.io/phonometry/media/intensity_scan_power.svg)

*The 250 Hz band nets negative (more energy flowing in than out), so ISO 9614-3
declares it not-applicable; the figure hatches and greys it while the four
determinate bands and the A-weighted total stand.*

**Show the code for this figure**

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

# Four partial surfaces scanned over five one-third-octave bands. Each cell of
# partial_intensity is the signed normal intensity In_i (W/m^2); areas are the
# partial-surface areas Si. The 250 Hz band has net-negative power (a locally
# reactive field), so ISO 9614-3 flags it not-applicable (clause 9.2) -> NaN.
freqs = np.array([250, 500, 1000, 2000, 4000], float)
areas = np.array([0.5, 1.0, 0.75, 0.5])
base_intensity = np.array([2.0e-6, 8.0e-6, 2.0e-5, 1.0e-5, 3.0e-6])
partial_intensity = base_intensity[None, :] * np.array([1.0, 1.1, 0.9, 1.05])[:, None]
partial_intensity[:, 0] = [2.0e-6, -3.0e-6, -4.0e-6, -1.0e-6]   # net-negative band
result = emission.sound_power_intensity_precision(partial_intensity, areas, frequencies=freqs)

# result is the PrecisionIntensityResult computed above. One line:
result.plot()
plt.show()

# By hand: determinate bands as LW bars; a not-applicable band (its LW is NaN)
# is flagged by a full-height greyed, hatched span rather than a zero-height bar.
freqs = result.frequencies
positions = np.arange(freqs.size)
neg = result.not_applicable_band
lw = np.nan_to_num(result.sound_power_level)
fig, ax = plt.subplots()
ax.bar(positions[~neg], lw[~neg], width=0.7, color="#1f77b4")
for pos in positions[neg]:
    ax.axvspan(pos - 0.35, pos + 0.35, facecolor="#888888", alpha=0.28,
               hatch="//", edgecolor="#888888")
ax.set_xticks(positions)
ax.set_xticklabels([f"{f:g}" for f in freqs], rotation=45, ha="right")
ax.set_xlabel("Frequency [Hz]")
ax.set_ylabel("Sound power level LW [dB]")
ax.set_title(
    f"Precision intensity scanning (ISO 9614-3)  "
    f"LWA = {result.sound_power_level_a:.1f} dB(A)")
plt.show()
```

### Qualifying the scan (Annex C)

`sound_power_intensity_precision` returns a power, not a grade: nothing in that
result says whether the determination is a grade-1 one. The acceptance gate is a
separate pair of calls, and without it the precision half of this page is a
computation with its criteria removed.

`precision_field_indicators` computes the Annex B indicators from the same
segment arrays: the signed and unsigned pressure-intensity indicators
$F_{pI_\mathrm{n}}$, the field non-uniformity $F_\mathrm{S}$, and the temporal variability $F_\mathrm{T}$
when a per-time-window intensity is supplied. `precision_qualification` then
evaluates the five Annex C criteria per band:

1. **scan repeatability**, $\lvert L_{I_\mathrm{n}}(1) - L_{I_\mathrm{n}}(2) \rvert \le s/2$ with
   $s$ from Table 1 (Eq. C.1): half the tolerance ISO 9614-2 allows, which is
   most of what "precision" costs;
2. **dynamic-capability adequacy**, $L_\mathrm{d} \ge F_{pI_\mathrm{n}}^{\text{signed}}$ with
   $L_\mathrm{d} = \delta_{pI0} - K$ and $K = 10$ dB (Eq. C.2);
3. $F_{pI_\mathrm{n}}^{\text{signed}} - F_{pI_\mathrm{n}}^{\text{unsigned}} \le 3$ dB (Eq. C.3),
   the precision counterpart of ISO 9614-2's $F_{+/-}$ limit;
4. **field non-uniformity**, $F_\mathrm{S} \le 2$ (Eq. C.4);
5. **scan-density convergence**, $0.83 \le F_\mathrm{S}(1)/F_\mathrm{S}(2) \le 1.2$ between the
   initial density and a doubled one (Eq. C.5).

The conjunction is `qualified`, and it carries one subtlety worth knowing before
you re-scan a machine: clause C.1.6.2 accepts a band that satisfies criterion 5
as a final result **even where criterion 4 is not met**. A non-uniform field
that stops changing when the scan density doubles has been sampled adequately;
it is the change with density, not the non-uniformity itself, that invalidates.

```python
# The same four segments and five bands, now with their surface pressure levels
# and a second (90-degree rotated, doubled-density) scan: enough for Annex C.
segment_pressure = np.full((4, 5), 74.0)
ind = emission.precision_field_indicators(partial_intensity, segment_pressure)
print(np.round(ind.f_pi_signed, 1))            # [12.2  4.9  0.9  3.9  9.2]

scan1 = 10.0 * np.log10(np.abs(partial_intensity.mean(axis=0)) / 1e-12)
scan2 = scan1 + np.array([0.9, 0.2, 0.1, 0.3, 0.4])
crit = emission.precision_qualification(
    ind, scan_intensity_level_1=scan1, scan_intensity_level_2=scan2,
    pressure_residual_index=15.0,              # Ld = 15 - 10 = 5 dB
    field_nonuniformity_1=ind.fs, field_nonuniformity_2=0.95 * ind.fs,
    frequencies=freqs,
)
print(crit.criterion_2.tolist())   # [False, True, True, True, False]
print(crit.qualified.tolist())     # [False, True, True, True, False]
```

Two bands fail here, and they fail for the reason the numbers say: at 250 Hz and
at 4 kHz the signed indicator (12.2 dB and 9.2 dB) exceeds the probe's
$L_\mathrm{d} = 5$ dB, so the instrument cannot resolve the net flux against its own
residual index. That is criterion 2, and the answer is a better probe, a larger
spacer or a surface moved away from the reactive near field. A longer scan will
not help.

## 3. The measurement report (`.report()`)

The intensity-scanning result (`SoundPowerIntensityResult`, ISO 9614-2) writes
a one-page PDF fiche laid out like a sound-power test sheet through its own
`.report()`, sharing the layout and the `ReportMetadata` container of the
[pressure-method fiche](https://docs.jmrp.io/phonometry/devices/emission/sound-power-pressure/#3-the-measurement-report-report).
The standard-basis line
names ISO 9614-2:1996 and the measurement grade, the per-band table lists the
intensity-derived band sound-power level $L_W$, and the boxed $L_{W\mathrm{A}}$ carries
the total $L_W$, the measurement surface $S$ and the determination grade
(the intensity result has no expanded uncertainty $U$). `verbose=True` adds the
field indicators $F_{pI}$ (surface pressure-intensity) and $F_{+/-}$
(negative partial power) and the per-band achieved grade; the basis strip
states the partial-power model (the segment partial powers $P_i = I_{\mathrm{n},i} S_i$
summing to $P$) and the Annex B qualification criteria. A band whose net power
is non-positive is not determinable (clause 9.2) and prints an em dash.

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

freqs = np.array([125, 250, 500, 1000, 2000, 4000], float)
# Six equal 0.5 m^2 segments (S = 3.0 m^2); one uniform normal-intensity
# spectrum scanned twice, with the surface SPL and the instrument residual
# index that qualify every band at engineering grade.
intensity = np.array([0.6e-4, 1.0e-4, 1.5e-4, 1.4e-4, 0.9e-4, 0.5e-4])
scan = np.tile(intensity, (6, 1))
res = emission.sound_power_intensity(
    scan, np.full(6, 0.5), normal_intensity_2=scan.copy(),
    pressure_levels=np.full((6, 6), 80.0), pressure_residual_index=15.0,
    frequencies=freqs, fraction=1, grade="engineering",
)

res.report(
    "sound_power_intensity.pdf",
    metadata=ReportMetadata(
        client="Example manufacturing plant",
        specimen="Hydraulic power pack (floor-standing)",
        test_room="Machine hall with steady background noise",
        instrumentation="Class 1 p-p intensity probe (IEC 61043), s/n 0042",
        laboratory="Phonometry reference example",
        report_id="EXAMPLE-9614",
        requirement=93.0,
    ),
)   # LWA = 90.9 dB(A) re 1 pW -> declared limit 93 dB(A): PASS
```

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

**Report:** [ISO 9614-2 sound power by intensity example report (PDF)](https://docs.jmrp.io/phonometry/media/iso9614_sound_power_intensity_example.pdf): One-page ISO 9614-2 sound-power-by-intensity determination fiche: a header with the client, the noise source, the machine-hall test environment and the intensity probe and climate, the octave-band table (125 Hz to 4 kHz) of intensity-derived band sound-power levels LW, the sound-power spectrum LW(f) with a nominal band axis, the boxed A-weighted sound power level LWA = 90.9 dB(A) re 1 pW with the total LW = 92.5 dB, the measurement surface S = 3.00 m2 and the engineering grade, and a PASS verdict against the declared 93 dB(A) limit, closed by a basis strip stating the partial-power model, the field indicators FpI and F+/- and the Annex B qualification criteria.

Sound power by intensity fiche (SoundPowerIntensityResult.report), an ISO 9614-2 engineering-grade scan with the field indicators and the boxed LWA.

### The precision sheet (ISO 9614-3)

`PrecisionIntensityResult` writes its own fiche, and it is not the part 2 sheet
with a different title: part 3 asks a report to state different things
(clause 10). The per-band table carries the normalized level $L_{W0}$ the
standard reports (Eq. 10) beside $L_W$, and the expanded uncertainty $U$ of
clause 4.3, twice the Table 1 standard deviation of reproducibility of the
band; the caption names the frequency range the determination covers, because
clause 4.3 asks for it whenever that range is narrower than 50 Hz to 6.3 kHz. A
one-third-octave set is printed in two column groups side by side, the way an
accredited sheet fits that many bands on a page. `verbose=True` tabulates the
four Annex B indicators $F_\mathrm{T}$, $F_{p|I_\mathrm{n}|}$, $F_{pI_\mathrm{n}}$ and $F_\mathrm{S}$ per band, the
tabulation clause 10 f) 1) asks for, over the measurement surface as Annex B
defines them, and the qualification cell the criteria decide.

Hand it an Annex C qualification and the sheet does what
clause 10 f) 2) makes mandatory: the bands whose Annex C criteria are not
satisfied are dropped from the A-weighted determination and named on the sheet,
next to the bands the method is not applicable to at all (clause 9.2). That is
why the boxed $L_{W\mathrm{A}}$ can differ from `result.sound_power_level_a`, which is
computed before any criterion is evaluated and therefore sums every applicable
band. Beside it the box states the normalized $L_{W\mathrm{A}0}$, since clause 10 f) 2)
reports the normalized quantity while the headline number is the level a
declared limit is written against. Without `criteria` the fiche boxes the result's own value and says that
no qualification was supplied. `residual_index` puts the probe's
pressure-residual intensity index on the sheet (clause 10 d) 5)) with the
dynamic capability $L_\mathrm{d}$ it yields; the clause 10 items that are free
description rather than numbers, the scan geometry and speed, the scanning
time and the probe-reversal checks, go in the metadata `notes`.

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

# A box measurement surface: five partial surfaces, each scanned in four
# segments, over the one-third-octave bands 200 Hz to 800 Hz. Part 3 works in
# one-third octaves, which is the band set Table 1 tabulates its uncertainty
# and its criterion-1 tolerance for.
third_octave = np.array([200, 250, 315, 400, 500, 630, 800], float)
faces = np.array([1.65, 1.575, 1.575, 1.155, 1.155])
scanned = 1.2e-5 * np.array([1.3, 1.1, 0.8, 0.9, 0.9])[:, None] * np.ones(7)
segment_intensity = np.repeat(scanned, 4, axis=0)
# The pressure-intensity margin per band: 8 dB at 200 Hz, where the hall is
# reverberant, 2 dB above it.
margin = np.array([8.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0])
segment_levels = 10 * np.log10(segment_intensity / 1e-12) + margin

precise = emission.sound_power_intensity_precision(
    scanned, faces, frequencies=third_octave,
    temperature_c=28.0, static_pressure_pa=94_000.0,
)
annex_b = emission.precision_field_indicators(segment_intensity, segment_levels)
scan_intensity = np.average(scanned, axis=0, weights=faces)  # unequal faces
scan_level = 10 * np.log10(scan_intensity / 1e-12)
annex_c = emission.precision_qualification(
    annex_b, scan_intensity_level_1=scan_level,
    scan_intensity_level_2=scan_level + 0.1,
    pressure_residual_index=15.0,          # Ld = 15 - 10 = 5 dB
    frequencies=third_octave,
)
print(annex_c.qualified.tolist())   # [False, True, True, True, True, True, True]

precise.report(
    "sound_power_intensity_precision.pdf",
    metadata=ReportMetadata(
        client="Example manufacturing plant",
        specimen="Hydraulic power pack (floor-standing)",
        test_room="Machine hall with steady background noise",
        instrumentation="Class 1 p-p intensity probe (IEC 61043), 12 mm spacer",
        temperature_c=28.0, relative_humidity_percent=40.0, static_pressure_kpa=94.0,
        laboratory="Phonometry reference example",
        report_id="EXAMPLE-9614-3",
        requirement=98.0,
        notes="Box surface at 0.25 m; five partial surfaces scanned twice.",
    ),
    indicators=annex_b, criteria=annex_c, residual_index=15.0,
)   # 200 Hz fails criterion 2 and is named as omitted from LWA
```

The example fiche in the repository is a fuller determination of the same
machine: sixteen one-third-octave bands from 100 Hz to 3150 Hz over a five-face
box surface, with the 100 Hz band failing criterion 2 and named as omitted.

**Report:** [ISO 9614-3 precision sound power by intensity example report (PDF)](https://docs.jmrp.io/phonometry/media/iso9614_3_precision_intensity_example.pdf): One-page ISO 9614-3 precision sound-power-by-intensity determination fiche: a header with the client, the noise source, the machine-hall test environment, the intensity probe and the 28 degrees Celsius and 94 kPa test atmosphere, the one-third-octave table from 100 Hz to 3150 Hz in two column groups giving the band sound-power level LW, the normalized level LW0 and the expanded uncertainty U, the sound-power spectrum LW(f) with a nominal band axis, the boxed A-weighted sound power level LWA = 96.7 dB(A) re 1 pW with the total LW = 97.9 dB, the normalized total LW0 = 98.5 dB, the normalized A-weighted level LWA0 = 97.3 dB(A), the measurement surface S = 7.11 m2 and the expanded uncertainty U = 2.0 dB, and a PASS verdict against the declared 98 dB(A) limit, closed by a basis strip stating the partial-power model, the meteorological normalization, the Annex B field indicators and the five Annex C criteria, and naming the 100 Hz band as omitted from LWA for criterion 2.

Precision sound power by intensity fiche (PrecisionIntensityResult.report), an ISO 9614-3 grade-1 scan with the normalized levels, the per-band uncertainty and the Annex C omission stated.

## What this guide covers

- **Covered.** The ISO 9614-2 scanning determination (`sound_power_intensity`): the partial
  powers, the $F_{pI}$ surface pressure-intensity and $F_{+/-}$
  negative-partial-power field indicators, the dynamic capability $L_\mathrm{d}$, the
  repeatability criterion, the per-band achieved grade and the clause 9.2
  not-applicable flagging for net-inflow bands, with the clause 10.6 b)
  screening of the A-weighted total (`a_weighting_omitted_bands`). The ISO
  9614-3 precision determination (`sound_power_intensity_precision`) with its
  own not-applicable flagging, its Annex B field indicators
  (`precision_field_indicators`) and the five Annex C acceptance criteria
  (`precision_qualification`). Both results render the accredited-style
  sound-power fiche through `.report()`, each stating what its own part asks a
  report to state: the precision sheet adds the normalized levels $L_{W0}$, the
  per-band expanded uncertainty of Table 1, the Annex B indicator tabulation and
  the clause 10 f) 2) omission of the bands the Annex C criteria reject.

- **Not covered.** ISO 9614-1, the discrete fixed-point intensity method, is not one of the
  routes here: it holds the probe still at each position instead of sweeping it,
  and both its power determination and the Annex A field indicators the two
  scanning parts reuse are in the [sound intensity
  guide](https://docs.jmrp.io/phonometry/devices/emission/intensity/). The probe's
  finite-difference bias
  and the pressure-residual intensity index are measured properties of the
  instrument, covered in the same guide. Choosing among the seven routes, and the
  ISO 4871 declaration a result feeds, live in [Sound
  Power](https://docs.jmrp.io/phonometry/devices/emission/sound-power/).

## See also

- [Sound Power](https://docs.jmrp.io/phonometry/devices/emission/sound-power/): choosing among the seven
  determination routes, the accuracy grades and the ISO 4871 noise-emission
  declaration.
- [Sound Intensity (p-p)](https://docs.jmrp.io/phonometry/devices/emission/intensity/): the two-microphone
  probe, its finite-difference bias, the ISO 9614-1 field indicators behind
  the scanning methods and the discrete-point determination those indicators
  qualify.
- [Sound Power by Pressure Methods (ISO 3744 / ISO 3746 / ISO 3745)](https://docs.jmrp.io/phonometry/devices/emission/sound-power-pressure/):
  the enveloping-surface and anechoic routes for quiet sites.
- [Sound Power in the Reverberation Room (ISO 3741)](https://docs.jmrp.io/phonometry/devices/emission/sound-power-reverberation/):
  the precision diffuse-field route in a qualified laboratory room.
- [Theory](https://docs.jmrp.io/phonometry/reference/theory/environment-transport/): the
  field-indicator derivations.
- API reference: [`emission`: sound power by intensity scanning](https://docs.jmrp.io/phonometry/reference/api/power/sound-power-intensity/).
- Theory: [Sound power determination](https://docs.jmrp.io/phonometry/reference/theory/environment-transport/#sound-power-determination-iso-374437453746-iso-3741-iso-9614-123): the ISO 9614 scanning derivation and the field indicators that qualify it.

## References

- **Fahy, F. J. (1995).** *Sound intensity* (2nd ed.). E&FN Spon. https://doi.org/10.4324/9780203475386
  ISBN 978-0-419-19810-9. The monograph on sound energy flux: why intensity separates the energy leaving the source from steady energy passing through, behind both scanning methods.
- **International Organization for Standardization. (1996).** [*Acoustics — Determination of sound power levels of noise sources using sound intensity — Part 2: Measurement by scanning*](https://www.iso.org/standard/21247.html) (ISO 9614-2:1996).
  The scanning method of section 1: the partial powers, the FpI and F+/- field indicators and the grade criteria.
- **International Organization for Standardization. (2002).** [*Acoustics — Determination of sound power levels of noise sources using sound intensity — Part 3: Precision method for measurement by scanning*](https://www.iso.org/standard/24012.html) (ISO 9614-3:2002).
  The grade-1 scanning method of section 2, its field indicators and the clause 9.2 not-applicable flagging.
