Analyst knowledge: reading-rules.md, stations.md, a phenomenon catalogue in families.md; the label map and knowledge/ archived
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
1 parent
1c8479d962
commit
4688d8ab9e
16 files changed
+201
-17
No files matched your search
@@ -0,0 +1,23 @@
|
||||
# The Barcelona incident (19.09.2026)
|
||||
|
||||
The AIMsight documentation makes it its main regression case: roadmap, line 33 and §15; support context, §7-8 (example file, question "Why did D-300-A jump?"), §12.2, §16.1 and the example "Barcelona reference switching". Its data are in `data/CLARIDEN/` ("L9 BCN" export, from 17/09 22:00 to 20/09 21:50 UTC, 10-minute rate, 29 sensors including 5 references; 432 images at 10 min, one per measurement, zip times in UTC) and, coded, in case E-004 (`events/E-004/`).
|
||||
|
||||
## What is observed
|
||||
|
||||
According to the blind report of 2026-10-15 (`agent-diagnostic/rapports/2026-10-15T1346Z_CLARIDEN_19-09-matin/`), rechecked on the data:
|
||||
- around 00:00 on 19/09, the references move relative to each other (`PR-291` and `PR-293` by +0.7 px in `relative_x`, `D-302-C` motionless); `sigma0` is multiplied by ten; `corr_x` of the targets drops without their raw position moving;
|
||||
- at 06:10, the "-B" targets move alone in the image (0.8 to 1.4 px), the references motionless; `PR_BG` jumps by 17 px, its reflector is no longer visible, its `error_value` stays just under the threshold and its values (~11) are kept;
|
||||
- from 06:40 to 07:30, the 29 sensors move together (+1.8 to +4 px in `y`), then partly come back; `corr_y` keeps a step until the end of the measurements: +0.5 to +1.2 on the "-A" targets, growing with the distance to the references, and −1.0 on `D-306-B`;
|
||||
- no external cause on the images.
|
||||
|
||||
## Two readings
|
||||
|
||||
- **AIMsight documentation**: movement of the station and the camera, instability of the references (`PR_284`, `PR_291`), switches between compensation states, targets that remained measurable; *"primarily reference-network / compensation-state instability, not target tracking"*. The follow-up described is on the system side: pattern reset, reconfiguration, exclusion of `PR_284` considered, reference added.
|
||||
- **Expert** (reported by the user on 2026-10-15): the structure moved, and the camera and the references with it; the system probably overestimates the amplitude; other measurement systems on the site saw a major anomaly, and the site was interrupted; the structure is a bridge, whose arch is sagging (user, 2026-10-24). This is the diagnosis of E-004: *monitored-object motion*, real movement, intervention.
|
||||
|
||||
The two do not exclude each other: an unstable compensation distorts the amplitude of a real movement. The measurements alone do not separate the structure from the camera when everything moves together; blind, the diagnostic agent concluded a *weak/unstable reference solution*, consequence undetermined.
|
||||
|
||||
## Consequences and questions
|
||||
|
||||
- For the correction of the curves (D-044): these jumps are, for the expert, a real movement; do not remove them as an artifact. The amplitude, however, is not established.
|
||||
- Open: the documentation speaks of an SFTP export at 1 min and images at 15 min, here 10 min: is it the same set, reprocessed or not? `PR-284` is here only a local reference (`pattern_type` `ref`), outside the compensation.
|
||||
@@ -0,0 +1,19 @@
|
||||
# Camera images: IRON and DIFF
|
||||
|
||||
The platform keeps two types of images per measurement (measurement type), which the pump stores separately (`~/aimsight-data/extraction/pump.py`).
|
||||
|
||||
- **IRON** (`OPTICAL_SHIFT-IRON`, in `images/`): the classic image of the scene, in JPG, ~1 MB (NORDEND). It is on it that the cause of an event is seen (condensation, rain, train, snow). The name probably comes from "IR on" (taken under infrared lighting): deduction, the doc does not say it.
|
||||
- **DIFF** (`OPTICAL_SHIFT-DIFF`, in `images_diff/`, option `--diff`): the "differential" image. According to the station manual (*AIMsight Test Manual*, "Choosing Between Single and Differential Acquisition"), it **combines several captures** to strengthen the contrast of the reflectors ("*contrast-enhanced view of the reflectors*") and it is **the image that the displacement algorithms measure, by default**. It is **not** the difference between two instants. 16-bit PNG, 3088×2064, ~7 MB (NORDEND).
|
||||
|
||||
## Acquisition modes (station manual)
|
||||
|
||||
- "Single with infrared light": one image under infrared per measurement, for dynamic measurements.
|
||||
- "Differential": several images combined, for quasi-static monitoring; `Stacking` from 20 to 40 if the light is difficult or the targets weakly reflective (better range, lower rate).
|
||||
- Option "Work on IRON images": measure on the classic image, to follow a texture when there is no reflector.
|
||||
|
||||
## What is known
|
||||
|
||||
- NORDEND has DIFF images, one per hour at pumping (~160 MB per day). LENZSPITZE has none in the database on 26 and 27/10 (pumping of 2026-10-28), only IRON, one per hour: an acquisition mode without DIFF, probably (not verified). The other stations: unknown.
|
||||
- On 29/07 at NORDEND (E-002, spider web according to the expert), the DIFF image shows long bright streaks over the whole field and the targets as dots, where the IRON image shows only faint dark bands. What is very close to the lens perhaps stands out better there: to confirm (point 28 of the morning-journal plan).
|
||||
- The reading tools (`tools/scripts/data.py`, tools repository) look only at IRON images and leave out `images_diff/`; `inventory.py` counts them, and the analyst can read them with its own scripts (D-097).
|
||||
- Lead, not verified: if `x`, `y` are measured on the DIFF image, a centroid taken on the IRON image can differ from it (constant offset per sensor seen on NORDEND, `displacement-columns.md`).
|
||||
@@ -0,0 +1,40 @@
|
||||
# Displacement columns (next use case)
|
||||
|
||||
According to the FAQ, the columns of the CSV export are prefixed by the target (e.g. `target_06_rch`):
|
||||
- `rch` / `rcv`: displacement corrected for camera movement [mm]. This is the main signal.
|
||||
- `x` / `y`: raw position in the image [px].
|
||||
- `rx` / `ry`: relative displacement before camera correction. **It is not** `x - x0`.
|
||||
- `error`: target recognition error, **not** an uncertainty. Default threshold: 0.7.
|
||||
- `camera_pan|tilt|rotation` [deg], `camera_scale`, `camera_sigma0` (residual of the reference-network fit).
|
||||
- `inlier_{ch}`, `reproj_err_{ch}`: contribution and residual of each reference.
|
||||
- Sharpness, FWHM, Dominance, Peak Intensity, Flux, SNR: in the "advanced" export of recent cameras only (see below); definitions: `quality-metrics.md`.
|
||||
|
||||
## Exports received (PS des Coudres, D-045)
|
||||
|
||||
Two other formats, observed on the data (not documented in the FAQ):
|
||||
- **"advanced"** (`<Station>_advanced_<start>_<end>.csv`), the source of the analyses: one row per sensor and per measurement, at the native rate (15 min, 10 min depending on the station), `time_utc` in explicit UTC.
|
||||
- `sensor_name`, `pattern_type` (`global_ref` for a compensation reference; `ref` for a *local reference*, treated as a target, outside the compensation, according to the station manual and without `reproj_err`, seen on NORDEND), `x` / `y` [px], `error_value`, `error_threshold` (specific to each sensor: 0.3 to 0.7);
|
||||
- `x` / `y` are not the centroid of the spot in the image (NORDEND, 2026-10-16): by day (12:30 UTC, seven days), centroid − (`x`, `y`) is constant per sensor and differs from one sensor to another (PR17 −2.5 / −0.5 px, PR18 −3.1 / +1.6, PR16 −5.8 / −0.1, CTR02 −1.4 / −2.4, CTR04 −1.0 / −1.4). Centroid weighted above the median of a ±20 px frame, on a ±8 px window. At night, the spot is saturated and the offset changes (PR17 −0.7 to −1.3 px in x). Hypothesis: `x`, `y` designate the position of the reference pattern (station manual), not the spot itself.
|
||||
- `relative_x` / `relative_y` (presumably `rx` / `ry`), `corr_x` / `corr_y` (presumably `rch` / `rcv`, but not in mm, see below);
|
||||
- `reproj_err` and `inlier`: references only; `sigma0`: one value per measurement, repeated on each row; no `camera_pan|tilt|rotation|scale`.
|
||||
- `peak`, `flux`, `sharpness`, `fwhm`, `dominance`, `snr`: the image quality of each target, computed by the GPU of recent cameras (`run.md`); "Empty for most devices today" (handoff G, `docs/aimsight/AIMsight_handoff_G_site_health.md`). Filled on LENZSPITZE and NADELHORN, empty elsewhere (pumping of October 2026); `inventory.py` says, on a batch, which columns carry values. For older cameras, `~/optical_emulator` recomputes them from the DIFF images in `quality/` (`run.md`), with `stacking`, `exposure`, `gain`, `rotation` and the image `flag`, to be joined on `time_utc` and `sensor_name`. Definitions and reading: `quality-metrics.md`. Flux, SNR and sharpness relate to the observability of a target in the image, to be separated from the health of the reference network (`AIMsight_agentic_support_project_context.md`, §5.5 and §5.6).
|
||||
- A rejected measurement (`error_value` > `error_threshold`) keeps its row and its `error_value`; `corr_x` / `corr_y` are empty, not interpolated. "Rejected" is a judgment by AIMsight, not a sorting by the camera: the camera delivers all its measurements, and the platform of a client that integrates the data receives them and can display them (the user, 2026-10-28).
|
||||
- **LENZSPITZE, relative `x` / `y`** (2026-11-04): `x` / `y` there are the offset from the pattern's reference position (medians ~0 px), `relative_x` / `relative_y` are equal to `corr_x` / `corr_y`, `pattern_type` is empty. Provisional conversion, for this station only: `~/aimsight-data/extraction/lenzspitze_absolute.py` rewrites `data/LENZSPITZE/displacement/` in image px (pattern center from `lentzpitze_patterns.json` + the measured offset +330.5 / +15.8 px between the JSON frame and the IRON images + `x` / `y`) and fills `pattern_type`. To rerun after a re-pump. Limits: 1023 and 8019 have no pattern in the JSON, their `x` / `y` are emptied (8019 classed `global_ref`, by its `reproj_err`); `relative_x` in px is not reconstructed. To fix in due course: the frame of positions on the pumping side, and `visibility.py`, `thermal_displacement.py`, `categories.py`, which test `pattern_type` directly instead of `data.is_reference` (empty, they see no reference).
|
||||
- **Camera movements** (`camera/<Station>_camera_<YYYYMMDD>.csv`, pumping only, absent from the exports received): the camera pose at each registration, `time_utc` in UTC, `median_|min_|max_` of `pan`, `tilt`, `rotation` [deg], `scale`, `sigma0` at the minute; one row per measurement (min = max), at the instants of the displacements. A measurement instant without a row: the camera was not registered (NORDEND: 101 instants out of 6,071, of which 49 on 29 and 30/07). Orders of magnitude: `pan`, `tilt` of a few hundredths of a degree, `scale` within 10⁻³ of 1; differences of several degrees on NORDEND (`pan` up to 3.1°, `tilt` −2.8 to 4.0°), not explained. Sign: a positive `tilt` means the camera is nose-diving (the expert, 2026-11-03, BREITHORN: the `tilt` rises all October while the camera tilts). Read by `overview.py` and `inventory.py` (D-073).
|
||||
- **Hourly** (`<Station>_<start>_to_<end>_1hour.csv`): one row per hour, targets only. Per target: `horizontal_displacement`, `vertical_displacement`, their `_min` / `_max`, `_error`, `_count_val`.
|
||||
- `timestamp` **without time zone**: Swiss time (UTC+2 in October), at the start of the hour. Verified: `_error` = mean of the `error_value` of the hour, to within 0.0006 (median).
|
||||
- `count_val` counts the measurements of the hour, rejected ones included.
|
||||
- **Unit of `corr_x` / `corr_y`**: `horizontal_displacement` = factor × hourly mean of `corr_x` + constant offset (likewise `vertical` and `corr_y`, correlation ≥ 0.99). The factor depends on the pattern: `target05` 0.61, `target` 1.6, `target02` 4.3, `target03` 6.85, `target04` 7.9. If the hourly is in mm (FAQ), `corr_x` / `corr_y` are not. According to the user (2026-10-16), they are surely in **px**, the factor then being mm per px. This is compatible with NORDEND: at the jump of PR17 (08/10 02:30), `corr_y` drops by about 0.15, and `relative_y` (in px) by 0.26 px against 0.08 px for the other sensors; over the window, `corr` of PR17 ≈ 0.8 × (`relative` − mean of the `relative` of the global references), but a mean ignores the rotation and scale of the compensation. To confirm. The factor depends on the station: on NORDEND, `vertical_displacement` of PR17 = 55.5 × hourly mean of `corr_y` (r = 0.994, 244 hours), i.e. about 9 mm for its jump of 0.15.
|
||||
|
||||
Causes to tell apart for a displacement anomaly:
|
||||
- real movement of the structure;
|
||||
- camera movement;
|
||||
- instability of the compensation (Barcelona case);
|
||||
- local degradation of a target;
|
||||
- parallax or unsuitable model.
|
||||
|
||||
The authoritative reference is `AIMsight_reference_quality_event_explainer_roadmap.md` (AIMsight documentation, stored in Support-Agent, `docs/aimsight/`). The non-negotiable rules that relate to it:
|
||||
- build nothing on a binary `inlier`: RANSAC will be replaced by a weighted estimator; use continuous weights, `N_eff`, residuals, influence;
|
||||
- `sigma0` alone and the number of references are never sufficient criteria;
|
||||
- do not confuse the image quality of a target with the quality of the reference network, nor a small residual with a well-constrained transformation;
|
||||
- do not confuse common-mode movement with structural movement, nor coherent references with physically fixed references.
|
||||
@@ -0,0 +1,48 @@
|
||||
# Cases annotated by the expert (displacements)
|
||||
|
||||
Two analysis reports handed to the client by the AIMsight team (May 2026, revision 1.0, in French), stored as they are in `docs/aimsight/`:
|
||||
- `Mur de soutènement GN-1.472.pdf` (16 p.): written report;
|
||||
- `Mur de soutènement GN-5.054.pdf` (7 p.): draft, partly in notes; the introduction announces three walls, only one is detailed.
|
||||
|
||||
Masonry retaining walls along a railway line, monitored for nearly two years, with prisms and adhesive reflective strips ("ciblettes"). Verdict of both reports: **no notable structural displacement**; each apparent displacement is attributed to an external cause, with evidence (images, `error`, brightness).
|
||||
|
||||
They serve as reference truth for `displacement`, with two limits:
|
||||
- the exports of these walls are not in `data/`: neither the signatures can be verified nor an event replayed. They therefore do not enter the diagnostic agent's event base (`events/`, D-048);
|
||||
- the signatures below are read from the figures of the reports (axes illegible in GN-5.054); no time is given there.
|
||||
|
||||
## Wall GN-1.472 (01/01 to 15/03/2026, detail from 15/02 to 15/03)
|
||||
|
||||
| Finding | Signature (report figures) | Expert diagnosis | Evidence cited |
|
||||
|---|---|---|---|
|
||||
| H22 and H27, at the top of the wall | vertical: 24 h cycle, correlated between the two, ~10 mm peak to peak, whole period | mutual interference of two very close prisms; persistent anomaly, not a movement of the structure | photo of the two prisms; brightness of H27 well above its neighbors |
|
||||
| H18, H26, H31, H21 | isolated peaks, one target at a time, up to ~30 mm | a neighboring prism that is too bright hampers detection; too large for masonry | images without obstruction; `error` > 0.7 (H18, and H26 or H22: the report contradicts itself); unbalanced brightness |
|
||||
| C17 | horizontal: square waves by day, around −65 mm, return at night | dim prism, insufficient contrast against daylight | day and night images; `error` rises by day, without exceeding 0.7 |
|
||||
| Camera movements | pan, roll, tilt: 24 h cycle badly compensated; scale stable | not a camera movement: 2 references out of 4 masked by vegetation (growing since March 2025), insufficient base for the transformation | images of March 2025 and current |
|
||||
| Snow, 11/01/2026 | noise and apparent movements on all sensors and the camera; brief data loss | snowfall; probably a flake in front of the lens for the loss | photo of the flakes; images before and after, signal quickly back |
|
||||
| Interruption (date not given) | gap, drawn in straight lines | `possible`: frost or condensation ("it could be", cold period) | none |
|
||||
|
||||
- A badly configured offset shifts the horizontal curves of a target below zero: parametric, no effect on interpretation.
|
||||
- Remedies proposed: fewer targets, prisms instead of strips, one more reference, brush clearing or moved references, alarms when a target or a reference becomes invisible.
|
||||
|
||||
## Wall GN-5.054-MU-G-AV (nearly two years)
|
||||
|
||||
103 targets, the largest surface; the most exposed system (altitude, snow clearing that pushes snow against the station).
|
||||
|
||||
| Finding | Signature | Expert diagnosis | Evidence cited |
|
||||
|---|---|---|---|
|
||||
| Snow | the traces of the hidden targets disappear, the others stay stable | outright obstruction by snow | winter images |
|
||||
| R2_C20 | jump in the measurements | a plant grows in front of the strip and gradually hides it (partial obstruction) | two images two days apart |
|
||||
| Data lost in winter | gap | remote solar panel covered with snow, unfavorable orientation: accepted compromise (budget) | operator context |
|
||||
| Station moved after the first winter | (not shown) | station buried, freeze-thaw cycles that destabilized the mast; guy wires added | on-site intervention |
|
||||
| Power connector | (not shown) | partly torn off by a snow projection during clearing; replaced by a more robust variant | on-site intervention |
|
||||
|
||||
- Most delicate obstruction according to the expert: vegetation (growth, leaves in the wind), partial, for which the default acceptance threshold was not enough.
|
||||
- Strips up to 40 m; some, near much brighter prisms, badly measured or eliminated.
|
||||
|
||||
## What these events add to the framing
|
||||
|
||||
Families absent from the table §4.1 of `docs/displacement/cadrage.md`, to be confronted with data when we have some:
|
||||
- **interference between neighboring targets**: clear image, but a neighbor that is too bright; *Dominance* (FAQ) measures this risk, in the HDF5 only;
|
||||
- **progressive occlusion** (vegetation): days or weeks, not reversible from one measurement to the next like a passing occlusion;
|
||||
- **masked references**: 24 h cycle on pan, roll, tilt, not to be read as a camera movement (instability of the compensation, `labeling/knowledge/displacement-columns.md`);
|
||||
- **daytime contrast**: the "Lighting" family, here on a single prism, in day / night square waves.
|
||||
@@ -0,0 +1,16 @@
|
||||
# Gaps: why measurements are missing
|
||||
|
||||
What serves to label a gap (`No data`) and to state its cause. Taken from `docs/domaine/pieges-des-donnees.md` (2026-11-04), which refers here; the rules for labeling a gap are in `labeling/labeling-rules.md`.
|
||||
|
||||
- **A gap is not necessarily an outage.** Acquisition, transmission, ingestion and display are distinct layers.
|
||||
- **At the end of a file**, empty values can be just a synchronization delay (`last_sync`): WILDSTRUBEL, and MISSISSIPPI on 09/10.
|
||||
- **At the start of a file**, empty values precede the first measurement: this is not an outage (VERMONT `Power` and temperatures, from 22/02 to 08/04).
|
||||
- **Simultaneous gaps on several stations** rather point to a common cause outside the station:
|
||||
- on 22/07 at 14:00: FLORIDA, MISSISSIPPI, VERMONT and NORDEND;
|
||||
- on 19/08 at 15:00: the same four stations;
|
||||
- on 22/08 at 17:00: FLORIDA, MISSISSIPPI and NORDEND.
|
||||
- **A gap surrounded by a normal voltage** is not an energy problem.
|
||||
- **A power cut leaves a signature**:
|
||||
- the voltage is below the knee (~12.8 V) just before or just after the gap;
|
||||
- the temperature drops during the gap, more than at the same hours on nights without a gap. On VERMONT, the difference ranges from 2.6 to 15 °C over 12 gaps. Alone, it does not settle: the common gap of 19/08 also comes back ~7 °C below the usual, and an outage that came back after sunrise shows no difference (`docs/temperature/domaine/point-de-depart.md`).
|
||||
- **A `NaN` or `inf` value** in an export must be treated as an absence of measurement.
|
||||
@@ -0,0 +1,55 @@
|
||||
# The measurement chain: where each cause enters
|
||||
|
||||
- **Status**: draft of 2026-10-29 (point 67 of the morning-journal plan), to be reread by the user; not yet among what the analyst reads.
|
||||
- **Purpose**: the why, to reason about a case that the event map does not list. The meaning of the columns is in `labeling/knowledge/displacement-columns.md`, the signatures and categories in `events/carte-des-evenements.md`; this note does not repeat them.
|
||||
- **Sources**, noted in brackets: [U] the user (voice messages of 2026-10-29); [A] the AIMsight documentation (`AIMsight_agentic_support_project_context.md`, `AIMsight_reference_quality_event_explainer_roadmap.md`); [D] seen in the data (note cited); [H] Claude's hypothesis, to validate.
|
||||
|
||||
## 1. How a measurement is made
|
||||
|
||||
1. **The scene**: targets on the structure, references around it, supposed fixed.
|
||||
2. **The view**: the air between the camera and the targets, and what can pass or settle in front.
|
||||
3. **The image**: the lens, then the capture (IRON, or DIFF which combines several captures: `camera-images.md`).
|
||||
4. **The pattern measurement**: each sensor is found in the image: `x`, `y` [px], `error_value` (recognition error, not an uncertainty); on recent cameras, the quality (`peak`, `flux`, `sharpness`, `fwhm`, `dominance`, `snr`). Above `error_threshold`, the measurement is rejected: its row stays, `corr` is empty.
|
||||
5. **The compensation**: a 2D similarity (translation, rotation, scale) fitted on the references (`global_ref`) gives the camera pose (`camera/`: `pan`, `tilt`, `rotation`, `scale`, `sigma0`) and, per reference, `reproj_err` and `inlier` [A, reference §2.5].
|
||||
6. **The corrected curve**: `corr_x`, `corr_y`, the position of the target once the estimated camera movement is removed (in px, probably: `displacement-columns.md`).
|
||||
|
||||
A displayed value is the end of this chain: find the stage that gave way [A, context §4].
|
||||
|
||||
## 2. Where each cause enters, and what it touches
|
||||
|
||||
| Stage | Cause | What it touches | Time | Witnesses |
|
||||
|---|---|---|---|---|
|
||||
| View (air) | fog, rain, condensation: **reduced visibility** | a drop in flux, homogeneous [U] (stronger on distant targets? [H]); `error_value` rises, rejections; raw positions without a step | hours | quality (`flux`), images |
|
||||
| View (air) | **heat haze** (turbulence) | a fast noise; only when the structure really heats up; not under a bridge [U] | the hot hours | temperature, `error_value` by day versus night |
|
||||
| View (obstacle) | object, leaves, snow on a target: **occultation**; passage | one or a few targets: something in front [U]; flux cut at once, neighbors intact [H]; a passage: one measurement | from one measurement to weeks | quality, images |
|
||||
| Target | problem target (badly oriented, dirty, glued) | one target, lastingly | days and more | `error_value` of the target alone |
|
||||
| Lens | **thermal deformation of the optics** (sun on the station) [U] | the whole image deforms [U]; a similarity does not represent a deformation: residuals organized by position in the image, `reproj_err` of the references, `corr` that depends on the target's position [H, from A, reference §2.5] | short: follows the sun on the station [U] | station temperature |
|
||||
| Camera and mount | **expansion of the mount** [U]; wind, jolt, shock | raw positions of all sensors, references included, together; the compensation absorbs it (`corr` flat) or badly: `corr` follows the inflections of the camera movement (JUNGFRAU: mount or lens, the user hesitates [U]) [A, context §12.4] | specific to the mount [U] | `camera/`, station temperature |
|
||||
| References | **reference hidden or less well seen** [U] | the transformation is redone without it: points shifted at once, often grouped in one part of the image; back to the recalibration, when it returns [U]; reversible [A, context §12.2]; shift that grows with the distance to the references (Barcelona [D]) | the time of the nuisance | `error_value`, `reproj_err` of the references, `sigma0`, camera pose |
|
||||
| References | a reference moved | the `corr` of the targets shifted lastingly | lasting | `reproj_err` of the reference |
|
||||
| Structure | **thermal expansion of the structure** [U] | `corr`, amplitude specific to each target, references stable | the longest: slow heating, late cooling [U] | air temperature (absent, see 3) |
|
||||
| Structure | real movement | `corr` of one or several targets, which lasts; references and pose stable, good quality [A, context §5.8] | lasting | all the previous, at calm |
|
||||
|
||||
**A jump that returns** to normal is not serious: it is this return that makes it benign. Saying "jump" is not enough; the question is its cause, quite often a reference hidden or less well seen (row "References") [U].
|
||||
|
||||
## 3. The thermal effects
|
||||
|
||||
- **Three movement effects** [U]: the lens, the mount, the structure. They are told apart by their heating and cooling time, each its own; the structure is the slowest. Heat haze is apart: a noise, not a movement.
|
||||
- **The cases cross** [U]: a camera placed on a bridge has two structures, the one that carries it (the mount) and the one it monitors.
|
||||
- **In waves** [U]: expansion, then contraction; often stronger in Y than in X, but it depends on the orientation.
|
||||
- **The temperatures available** [D, `docs/temperature/domaine/point-de-depart.md`]: those of the compute board (`sensor_*_deg_c`), inside the housing, heated by the sun and by the computing, as hourly means: a witness for the lens and the mount, not for the structure nor the air. `temperature_deg_c`, on a few stations: unknown sensor.
|
||||
|
||||
## 4. The reading rules
|
||||
|
||||
Copied from `CLAUDE.md` ("Non-negotiable rules"):
|
||||
- build nothing on a binary `inlier`: RANSAC will be replaced by a weighted estimator; use continuous weights, `N_eff`, residuals, influence;
|
||||
- `sigma0` alone and the number of references are never sufficient criteria;
|
||||
- the image quality of a target is not the quality of the reference network; a small residual is not a well-constrained transformation [A, reference §2.1 to §2.3];
|
||||
- a common-mode movement is not a structural movement; coherent references are not physically fixed references: the camera cannot prove that a network that moves as a block is fixed [A, reference §2.4] (Barcelona: `barcelona-incident.md`).
|
||||
|
||||
## Leads to revisit
|
||||
|
||||
- The order of the lens and mount times, and what separates them on the curves (point 58, question 2).
|
||||
- The lens signature: deformation organized by position in the image [H].
|
||||
- Reduced visibility, stronger on distant targets; occultation, flux cut at once [H]: to confirm on the stations that have the quality (point 63).
|
||||
- The sensor of `temperature_deg_c` (point 59).
|
||||
@@ -0,0 +1,18 @@
|
||||
# The image-quality metrics of a target
|
||||
|
||||
Six columns of the "advanced" export, per sensor and per measurement: `peak`, `flux`, `sharpness`, `fwhm`, `dominance`, `snr`. Computed by the GPU of recent cameras (`aimsight_optical` 6.7.11) on the sensor window in the DIFF image, in units of that image. Filled on LENZSPITZE and NADELHORN (pumping of October 2026), empty elsewhere; for an older camera, `~/optical_emulator` recomputes them in `quality/` (`run.md`).
|
||||
|
||||
Source: `~/optical_emulator/docs/cuda_phase_correlation.md`, §14 (definitions of the reference code). "Disc": the neighborhood of the peak, of radius `min(H, W)/6` (at least 3 px); "background": all the rest of the window (`bg_mean`, `bg_sigma`).
|
||||
|
||||
| Metric | Definition | What it measures | Reading |
|
||||
|---|---|---|---|
|
||||
| `peak` | `max(I)` | the brightness of the return | high, saturation; low, underexposure or missed flash. All targets dropping: the lighting (LED, flash); a single one: that target (fouling, misalignment) |
|
||||
| `flux` | flux of the disc minus the background | the light of the target, background removed | reflectivity of the target (fouling, aging, flash), not ambient light; drops well before `error_value` exceeds its threshold |
|
||||
| `sharpness` | `Σ(I²) / (ΣI)²` | concentration of the energy (higher = sharper) | drops: drifting focus, thermal turbulence, degrading target, i.e. more measurement noise. Independent of the gain, but not of the background (not removed): to follow over time on one target, not to compare between cameras |
|
||||
| `fwhm` | width at half height around the peak, background removed [px] | width of the spot | rises: defocus, turbulence, motion blur. Truncated to the disc: under strong defocus, it flattens instead of rising |
|
||||
| `dominance` | `(second peak − background) / (peak − background)`, from 0 (clean) toward 1 (ambiguous) | does a single target dominate the window? | high: a comparable second reflection, a stray bright point, or the target has disappeared and something else is the brightest: the correlation can latch onto the wrong pattern, which `error_value` alone does not see. A single hot pixel outside the disc is enough to make it rise |
|
||||
| `snr` | `(peak − background) / bg_sigma` | the signal above the background noise | low: marginal return, even when `peak` looks normal. Independent of exposure and gain. The emulator marks `empty` below 8 |
|
||||
|
||||
In short: `peak` and `flux` give the brightness, `sharpness` and `fwhm` the sharpness, `dominance` and `snr` the validity. Together, they say **why** a measurement degrades, where `error_value` only says that it degrades.
|
||||
|
||||
These are qualities of the image of **one target**: they say nothing about the health of the reference network (the non-negotiable rules, at the end of `labeling/knowledge/displacement-columns.md`; `AIMsight_agentic_support_project_context.md`, §5.5 and §5.6). The readings in the table are those of the source; what they are worth on our phenomena (rain, condensation, obstruction, sun) is confirmed in the data (`labeling/knowledge/measurement-chain.md`).
|
||||
@@ -0,0 +1,37 @@
|
||||
# The thermal cycle: what is raw, what is compensated, what remains
|
||||
|
||||
- **Date**: 2026-10-24 (point 13 of the morning-journal plan). **Sources**: detector `detectors/thermique.py` (2026-10-23; today `detectors/scripts/thermal_displacement.py`), draft of a 24 h sinusoid (2026-10-24, point 41), the user.
|
||||
|
||||
## What the user (the expert) says about it
|
||||
|
||||
- Some structures move a lot during the day through **thermal expansion**, and come back down at night through **contraction**: one then sees **big waves** on the curves. These waves are **entirely normal**.
|
||||
- He no longer remembers the project that showed it (perhaps one of those sent, perhaps turbulence). None of the six sets received shows big waves in the corrected curves.
|
||||
- The wave is an event in its own right, of expansion (rise) or contraction (descent); a "periodic effect" that explains it is an idea he wants to keep (point 41 of the plan).
|
||||
|
||||
## What the data show
|
||||
|
||||
| | Raw positions (`x`, `y`) | Corrected curves (`corr_x`, `corr_y`) |
|
||||
|---|---|---|
|
||||
| 24 h wave | almost everywhere, all targets together: 0.8 px at NORDEND (October), 1.8 px (July), 3.2 px at WILDSTRUBEL, 4.6 px at NIEDERHORN | only at PS des Coudres: 11 targets out of 11, 0.17 px, 08:00 → 18:00 UTC; about 1 mm, up to 2 mm on PM_C and PM_D in the hourly export |
|
||||
| Cause | the camera or its mast bending (common mode) | the targets really move relative to the references: expansion of the structure or incomplete compensation, not settled |
|
||||
|
||||
- **The compensation removes the camera's wave.** The client does not see it; we do not tell him about it (decision of 2026-10-24).
|
||||
- **NORDEND of October** keeps a small cycle (9 targets out of 19, 0.02 px, about 1 mm): not flagged (decision of 2026-10-24); the sinusoid does not find it either.
|
||||
- **CLARIDEN** has no cycle over 3 days, even raw (not explained).
|
||||
|
||||
## A 24 h sinusoid to recognize it (draft of 2026-10-24)
|
||||
|
||||
Per target and per axis, `a + b·sin(2πt/24 h) + c·cos(2πt/24 h)` by least squares; R² = share of the curve explained.
|
||||
|
||||
| Set | Corrected: axes with R² ≥ 0.3 | Raw: axes with R² ≥ 0.3 |
|
||||
|---|---|---|
|
||||
| PS des Coudres | 14 / 22 (0.16 px, peak around 08 h UTC) | 24 / 24 |
|
||||
| NORDEND October | 0 / 38 | 18 / 38 |
|
||||
| NORDEND July | 1 / 38 | 33 / 38 (1 px, peak around 11 h) |
|
||||
| WILDSTRUBEL | 1 / 14 | 4 / 14 |
|
||||
| CLARIDEN | 0 / 48 | 0 / 48 |
|
||||
| NIEDERHORN | 14 / 40 (to check: 3 days, two rainy nights at the same hours) | 2 / 60 |
|
||||
|
||||
The sinusoid separates the station with a wave (PS des Coudres) from the others, and a real movement (CLARIDEN) does not pass for a wave.
|
||||
|
||||
**In the journal (point 41, 2026-10-24)**: `thermique.py` requires its two methods (repeated profile and sinusoid); at PS des Coudres, every day from 09 to 13/10, "Thermal expansion" 08:00 → 15:00 then "Thermal contraction" 15:00 → 18:00 (15:00: largest gap to the night level); nothing elsewhere.
|
||||
Reference in new issue
Block a user