ppak10's picture
Merge branch 'main' of hf.co:datasets/ppak10/Agentic-SLS-ASTM
564d78e
Raw
History Blame Contribute Delete
24 kB
"""Shared helpers for the per-session specimen extract scripts.
Each `scripts/specimens/{NN}_*.py` declares one TestWorks session as a SESSION
dict, then calls `process_session(SESSION)`. Output: one JSONL file per
specimen under `data/{standard}/`.
A session dict looks like:
SESSION = {
"session_folder": "2026_05_26", # relative to source/
"test_folder": "TST1.Test", # almost always this
"xlsx_name": "tensile_testing_5.26.xlsx",
"astm": {"standard": "D638", "type": "Type I", "year": "2022"},
"batch_label": "A",
"test_runs": [(tsr_idx, material_class, db_print_date_or_None), ...],
}
"""
import json
import logging
import re
from pathlib import Path
import h5py
import openpyxl
ROOT = Path(__file__).parent.parent.parent
SOURCE_DIR = ROOT / "source"
DATA_DIR = ROOT / "data"
DATABASE_ROOT = ROOT.parent / "Agentic-SLS-Database"
DATABASE_JOBS = DATABASE_ROOT / "data" / "jobs.jsonl"
DATABASE_PROFILES_DIR = DATABASE_ROOT / "source" / "PrintProfiles"
# Substring that identifies the matching STL object in a Database job, by standard.
STANDARD_TO_STL_NEEDLE = {"D638": "d638", "D790": "d790"}
logging.basicConfig(level=logging.INFO, format="%(message)s")
log = logging.getLogger("specimens")
def load_database_jobs() -> dict[str, list[dict]]:
"""Index Database jobs.jsonl by print_date. A single date can have multiple
jobs (e.g. 2026-06-27 has both `D790 and D638 and other objects` and
`Hex Coasters and Nameplates and Benchies`), so values are lists —
resolve_database_fk() picks the one containing the standard's STL."""
jobs: dict[str, list[dict]] = {}
with DATABASE_JOBS.open() as f:
for line in f:
row = json.loads(line)
jobs.setdefault(row["print_date"], []).append(row)
return jobs
def find_print_profile(profile_id: str) -> dict | None:
for p in DATABASE_PROFILES_DIR.glob(f"*{profile_id}*.json"):
with p.open() as f:
return json.load(f)
return None
def pick_object(job_row: dict, standard: str) -> dict | None:
needle = STANDARD_TO_STL_NEEDLE[standard]
for obj in job_row["objects"]:
if needle in obj["name"].lower():
return obj
return None
def safe_float(x):
"""Coerce to float and turn NaN/inf into None for JSON safety."""
if x is None:
return None
try:
v = float(x)
except (TypeError, ValueError):
return None
if v != v or v in (float("inf"), float("-inf")):
return None
return v
def get_either(d: dict, *keys):
"""Return the first key from `keys` present in `d`, or None.
Tensile and flex persistent.h5 spell the same concept differently
(e.g. StrnAtPeak vs StrainAtPeak)."""
for k in keys:
if k in d:
return d[k]
return None
def read_xlsx_scalars(xlsx_path: Path, sheet_index: int) -> dict:
"""Pull {DisplayName: Value} pairs (plus their units) from cols D-F of a
TestWorks export sheet."""
wb = openpyxl.load_workbook(xlsx_path, data_only=True, read_only=True)
sheet_name = wb.sheetnames[sheet_index]
ws = wb[sheet_name]
scalars, units = {}, {}
for i, row in enumerate(ws.iter_rows(values_only=True)):
if i < 2:
continue
name = row[3] if len(row) > 3 else None
value = row[4] if len(row) > 4 else None
unit = row[5] if len(row) > 5 else None
if name is None:
continue
scalars[str(name)] = value
units[str(name)] = unit
wb.close()
return {"scalars": scalars, "units": units, "sheet_name": sheet_name}
def read_xlsx_curve(xlsx_path: Path, sheet_index: int) -> dict:
"""Pull the raw (col A, col B) curve embedded in a TestWorks export sheet,
for sessions where no persistent.h5/DAQ h5 survived (see Dataset H flex).
Row 2 (0-indexed) holds the units for these two columns."""
wb = openpyxl.load_workbook(xlsx_path, data_only=True, read_only=True)
ws = wb[wb.sheetnames[sheet_index]]
a_unit = b_unit = None
a_vals, b_vals = [], []
for i, row in enumerate(ws.iter_rows(values_only=True)):
if i == 1:
a_unit, b_unit = row[0], row[1]
continue
if i < 2:
continue
a, b = row[0], row[1]
if a is None or b is None:
continue
a_vals.append(a)
b_vals.append(b)
wb.close()
return {"col_a": a_vals, "col_a_unit": a_unit, "col_b": b_vals, "col_b_unit": b_unit}
_LENGTH_TO_M = {"mm": 1e-3, "m": 1.0, "cm": 1e-2, "in": 0.0254}
_FORCE_TO_N = {"n": 1.0, "kn": 1e3, "lbf": 4.4482216153}
_STRESS_TO_PA = {
"pa": 1.0, "kpa": 1e3, "mpa": 1e6, "gpa": 1e9,
"n/mm2": 1e6, "kn/mm2": 1e9, "psi": 6894.757,
}
def _normalize_unit(unit) -> str | None:
if unit is None:
return None
return str(unit).strip().lower().replace("²", "2").replace(" ", "")
def _is_stress_unit(unit) -> bool:
return _normalize_unit(unit) in _STRESS_TO_PA
def convert_unit(value, unit, table: dict) -> float | None:
"""Convert `value` from its xlsx-reported `unit` to the SI unit `table` maps to."""
v = safe_float(value)
if v is None:
return None
factor = table.get(_normalize_unit(unit))
if factor is None:
log.warning(f" unrecognized unit {unit!r} for value {value}")
return None
return v * factor
def flexural_modulus_pa(width_mm, thickness_mm, extension_m, load_n):
"""ASTM D790 chord flexural modulus from the load-deflection curve.
TestWorks' D790 method exported a Modulus scalar computed from a
mis-zeroed strain channel (20-150 GPa on SLS specimens — physically
impossible; PA12-GF TDS flexural modulus is 2.4 GPa), so D790 modulus
is recomputed here: E = L^3 m / (4 b d^3) with support span L = 16 d
(span-to-depth 16:1 — confirmed against the export's own stress
channel, which matches sigma = 3FL/(2 b d^2) with this L exactly),
where m is the least-squares slope of load vs crosshead deflection
over the 10-40%-of-peak-load window (chord through the linear region,
above the seating toe). Validated: batch H recomputes to 2.0-2.6 GPa,
at the TDS value. Crosshead-based deflection understates E for very
compliant specimens, so treat low-density batches' values as lower
bounds.
"""
if not width_mm or not thickness_mm or not extension_m or not load_n:
return None
b, d = width_mm / 1000.0, thickness_mm / 1000.0
span = 16.0 * d
peak = max(load_n)
if peak <= 0:
return None
peak_i = load_n.index(peak)
pts = [(e, l) for e, l in zip(extension_m[:peak_i], load_n[:peak_i])
if e is not None and l is not None and 0.10 * peak <= l <= 0.40 * peak]
if len(pts) < 5:
return None
n = len(pts)
sx = sum(p[0] for p in pts)
sy = sum(p[1] for p in pts)
sxx = sum(p[0] * p[0] for p in pts)
sxy = sum(p[0] * p[1] for p in pts)
denom = n * sxx - sx * sx
if denom == 0:
return None
slope = (n * sxy - sx * sy) / denom
if slope <= 0:
return None
return span ** 3 * slope / (4.0 * b * d ** 3)
def flexural_stress_strain_pa(width_mm, thickness_mm, extension_m, load_n):
"""ASTM D790 chord stress/strain arrays derived from a raw load-deflection
curve, for xlsx_only D790 sessions whose export has no analyzed
StressArray/StrainArray to fall back on (e.g. Batch H, K, L flex).
Same span assumption as flexural_modulus_pa (L = 16*d, validated against
TestWorks' own stress channel): sigma = 3*P*L / (2*b*d^2),
epsilon = 6*D*d / L^2, where D is crosshead deflection (extension_m).
"""
if not width_mm or not thickness_mm:
return [], []
b, d = width_mm / 1000.0, thickness_mm / 1000.0
span = 16.0 * d
strain, stress_pa = [], []
for e, l in zip(extension_m, load_n):
if e is None or l is None:
continue
strain.append(6.0 * e * d / span ** 2)
stress_pa.append(3.0 * l * span / (2.0 * b * d ** 2))
return strain, stress_pa
def resolve_database_fk(material_class: str, db_print_date: str | None, standard: str,
jobs_by_date: dict, log_ctx: str):
"""Look up the Database job/print-profile FKs for an SLS specimen, if the
print job has been backfilled into Agentic-SLS-Database."""
job_id = print_profile_id = object_hash = None
print_profile_snapshot = None
if material_class == "SLS" and db_print_date:
candidates = jobs_by_date.get(db_print_date, [])
# Prefer a job that actually contains the standard's STL — this
# disambiguates dates with multiple prints (e.g. 2026-06-27).
job = next((c for c in candidates if pick_object(c, standard)), None)
if job is None and candidates:
log.warning(f" {log_ctx}: {len(candidates)} Database job(s) on "
f"{db_print_date} but none contain a {standard} STL; "
f"falling back to first")
job = candidates[0]
if job is None:
log.warning(f" {log_ctx}: no Database job for {db_print_date}")
else:
job_id = job["metadata"]["AutomaticJob"]["Id"]
print_profile_id = job["print_profile_id"]
obj = pick_object(job, standard)
object_hash = obj["hash"] if obj else None
print_profile_snapshot = find_print_profile(print_profile_id)
return job_id, print_profile_id, object_hash, print_profile_snapshot
def _h5_array_to_list(arr) -> list:
return [safe_float(v) for v in arr]
def read_persistent_h5(path: Path) -> dict:
"""Return analyzed scalars + curve arrays from an AnalysisRun."""
with h5py.File(path, "r") as f:
v = f["Values"][0]
scalars, arrays = {}, {}
for name in v.dtype.names:
val = v[name]
if isinstance(val, (bytes, str)):
continue
if hasattr(val, "__len__"):
arrays[name] = _h5_array_to_list(val)
else:
scalars[name] = safe_float(val) if isinstance(val, float) else val
return {"scalars": scalars, "arrays": arrays}
def read_daq_h5(path: Path) -> dict:
"""Return raw DAQ scans (extension_m, load_N, time_s).
Both tensile and flex DAQs store SI units per the Signals dataset
(Crosshead=m, Load=N, Time=s)."""
with h5py.File(path, "r") as f:
g = f["Session0000000000000000"]
scans = g["Scans"][...]
return {
"extension_m": [float(v) for v in scans[:, 0]],
"load_n": [float(v) for v in scans[:, 1]],
"time_s": [float(v) for v in scans[:, 2]],
}
def build_row(session: dict, tsr_idx: int, material_class: str,
db_print_date: str | None, sample_id: str | None,
jobs_by_date: dict) -> dict:
session_folder = session["session_folder"]
test_folder = session["test_folder"]
base = SOURCE_DIR / session_folder / test_folder
tsr_dir = base / "TestRuns" / f"TSR{tsr_idx}.TestRun"
persistent_path = tsr_dir / "AnalysisRuns" / "ANR1.AnalysisRun" / "persistent.h5"
daq_path = tsr_dir / "Data" / "DaqTaskActivity1.h5"
xlsx_path = SOURCE_DIR / session_folder / session["xlsx_name"]
# Disambiguate sessions where more than one Test project (TST1.Test,
# TST2.Test, ...) shares a session_folder — e.g. Batch H/I tensile, both
# under source/2026_06_30/Tensile/.
specimen_path = session_folder if test_folder == "TST1.Test" else f"{session_folder}/{test_folder}"
persistent = read_persistent_h5(persistent_path)
daq = read_daq_h5(daq_path)
xlsx = read_xlsx_scalars(xlsx_path, sheet_index=tsr_idx - 1)
x_scalars = xlsx["scalars"]
ps = persistent["scalars"]
width_mm = safe_float(x_scalars.get("Width"))
thickness_mm = safe_float(x_scalars.get("Thickness"))
area_mm2 = (width_mm * thickness_mm) if (width_mm and thickness_mm) else None
# Gauge length only exists for tensile (D638). Flex (D790) uses support span,
# which is not surfaced by TestWorks here.
adj_gage = ps.get("AdjGage")
gauge_length_mm = round(adj_gage * 1000, 4) if adj_gage else None
job_id, print_profile_id, object_hash, print_profile_snapshot = resolve_database_fk(
material_class, db_print_date, session["astm"]["standard"], jobs_by_date,
f"{session_folder}/TSR{tsr_idx}")
metrics = {
"modulus_pa": safe_float(ps.get("Modulus")),
"peak_load_n": safe_float(ps.get("PeakLoad")),
"peak_stress_pa": safe_float(ps.get("PeakStress")),
"strain_at_peak": safe_float(get_either(ps, "StrnAtPeak", "StrainAtPeak")),
"load_at_break_n": safe_float(ps.get("LoadAtBreak")),
"stress_at_break_pa": safe_float(ps.get("StressAtBreak")),
"strain_at_break": safe_float(get_either(ps, "StrnAtBreak", "BreakStrain")),
"energy_to_break_j": safe_float(ps.get("EnergyToBreak")),
"yield_stress_pa": safe_float(ps.get("StressAtYield")),
"strain_at_yield": safe_float(get_either(ps, "StrnAtYield", "StrainAtYield")),
}
if session["astm"]["standard"] == "D790":
# TestWorks' D790 Modulus scalar is unusable — see flexural_modulus_pa.
metrics["modulus_pa"] = flexural_modulus_pa(
width_mm, thickness_mm, daq["extension_m"], daq["load_n"])
return {
"sample_id": sample_id,
"batch_label": session["batch_label"] if material_class == "SLS" else None,
"specimen_id": f"{specimen_path}/TSR{tsr_idx}",
"test_date": session_folder.split("/")[0].replace("_", "-"),
"session_folder": session_folder,
"test_run_name": f"TSR{tsr_idx}",
"specimen_index": tsr_idx,
"material_class": material_class,
"astm": session["astm"],
"test_end_reason": x_scalars.get("Test Run End Reason"),
"geometry": {
"width_mm": width_mm,
"thickness_mm": thickness_mm,
"area_mm2": area_mm2,
"gauge_length_mm": gauge_length_mm,
},
"job_id": job_id,
"print_date": db_print_date,
"print_profile_id": print_profile_id,
"object_hash": object_hash,
"session_id": None,
"print_profile_snapshot": print_profile_snapshot,
"metrics": metrics,
"curves": {
"time_s": daq["time_s"],
"extension_m": daq["extension_m"],
"load_n": daq["load_n"],
"strain": persistent["arrays"].get("StrainArray", []),
"stress_pa": persistent["arrays"].get("StressArray", []),
},
"notes": session.get("notes", ""),
"source_paths": {
"persistent_h5": str(persistent_path.relative_to(ROOT)),
"daq_h5": str(daq_path.relative_to(ROOT)),
"xlsx": str(xlsx_path.relative_to(ROOT)),
"xlsx_sheet": xlsx["sheet_name"],
},
}
def build_row_from_xlsx(session: dict, sheet_idx: int, material_class: str,
db_print_date: str | None, sample_id: str | None,
jobs_by_date: dict) -> dict:
"""Build a specimen row from a TestWorks xlsx export only, for sessions
whose persistent.h5/DAQ h5 files did not survive (Dataset H flex: its
TestRuns folder was overwritten when a later TestWorks project reused the
same default TST1.Test name) or were never captured to begin with (Batch
K/L, 2026-07-15: only the xlsx exports were handed off). Cols A-B carry
a curve whose columns vary by export: some sessions (e.g. Batch H flex)
export the raw load/extension trace (col A in mm, col B in N); others
(e.g. Batch K/L) export the already-analyzed strain/stress trace (col A
dimensionless mm/mm, col B in a stress unit like kN/mm²). Which one we
got is detected from col B's unit rather than assumed, and only the
matching curve field gets populated — the other stays empty rather than
being derived. Cols D-F carry whatever scalar metrics that particular
export included. Metrics/curves not present in the export (yield, full
raw DAQ trace, etc.) stay null/empty rather than being approximated."""
session_folder = session["session_folder"]
xlsx_path = SOURCE_DIR / session_folder / session["xlsx_name"]
xlsx = read_xlsx_scalars(xlsx_path, sheet_index=sheet_idx - 1)
curve = read_xlsx_curve(xlsx_path, sheet_index=sheet_idx - 1)
x_scalars, x_units = xlsx["scalars"], xlsx["units"]
width_mm = safe_float(x_scalars.get("Width"))
thickness_mm = safe_float(x_scalars.get("Thickness"))
area_mm2 = (width_mm * thickness_mm) if (width_mm and thickness_mm) else None
job_id, print_profile_id, object_hash, print_profile_snapshot = resolve_database_fk(
material_class, db_print_date, session["astm"]["standard"], jobs_by_date,
f"{session_folder}/Sheet{sheet_idx}")
curve_is_analyzed = _is_stress_unit(curve["col_b_unit"])
if curve_is_analyzed:
curve_ext_m, curve_load_n = [], []
curve_strain = [safe_float(v) for v in curve["col_a"]]
curve_stress_pa = [convert_unit(v, curve["col_b_unit"], _STRESS_TO_PA) for v in curve["col_b"]]
else:
curve_ext_m = [convert_unit(v, curve["col_a_unit"], _LENGTH_TO_M) for v in curve["col_a"]]
curve_load_n = [convert_unit(v, curve["col_b_unit"], _FORCE_TO_N) for v in curve["col_b"]]
curve_strain, curve_stress_pa = [], []
if session["astm"]["standard"] == "D790":
# Derive the analyzed curve via the same chord formulas/span
# assumption as flexural_modulus_pa, so raw-curve D790 exports
# (no StressArray/StrainArray to fall back on) still plot.
curve_strain, curve_stress_pa = flexural_stress_strain_pa(
width_mm, thickness_mm, curve_ext_m, curve_load_n)
if session["astm"]["standard"] == "D790":
# flexural_modulus_pa needs the raw load-deflection trace; an
# already-analyzed strain/stress export doesn't carry deflection, so
# there's nothing to recompute from — leave modulus null.
modulus_pa = (flexural_modulus_pa(width_mm, thickness_mm, curve_ext_m, curve_load_n)
if not curve_is_analyzed else None)
else:
# Tensile Modulus, when the export includes it, is a direct TestWorks
# scalar (not the mis-zeroed-channel D790 case) — use it as-is.
modulus_pa = convert_unit(x_scalars.get("Modulus"), x_units.get("Modulus"), _STRESS_TO_PA)
metrics = {
"modulus_pa": modulus_pa,
"peak_load_n": convert_unit(x_scalars.get("Peak Load"), x_units.get("Peak Load"), _FORCE_TO_N),
"peak_stress_pa": convert_unit(x_scalars.get("Peak Stress"), x_units.get("Peak Stress"), _STRESS_TO_PA),
"strain_at_peak": None,
"load_at_break_n": None,
"stress_at_break_pa": convert_unit(x_scalars.get("Stress at Break"), x_units.get("Stress at Break"), _STRESS_TO_PA),
"strain_at_break": safe_float(x_scalars.get("Strain at Break")),
"energy_to_break_j": None,
"yield_stress_pa": None,
"strain_at_yield": None,
}
return {
"sample_id": sample_id,
"batch_label": session["batch_label"] if material_class == "SLS" else None,
"specimen_id": f"{session_folder}/Sheet{sheet_idx}",
"test_date": session_folder.split("/")[0].replace("_", "-"),
"session_folder": session_folder,
"test_run_name": f"Sheet{sheet_idx}",
"specimen_index": sheet_idx,
"material_class": material_class,
"astm": session["astm"],
"test_end_reason": x_scalars.get("Test Run End Reason"),
"geometry": {
"width_mm": width_mm,
"thickness_mm": thickness_mm,
"area_mm2": area_mm2,
"gauge_length_mm": None,
},
"job_id": job_id,
"print_date": db_print_date,
"print_profile_id": print_profile_id,
"object_hash": object_hash,
"session_id": None,
"print_profile_snapshot": print_profile_snapshot,
"metrics": metrics,
"curves": {
"time_s": [],
"extension_m": curve_ext_m,
"load_n": curve_load_n,
"strain": curve_strain,
"stress_pa": curve_stress_pa,
},
"notes": session.get("notes", "") or (
(
"Raw persistent.h5/DAQ h5 unavailable for this specimen; strain/"
"stress curve and metrics extracted from the TestWorks xlsx "
"export only (already analyzed — no raw load/extension trace to "
"derive it from). Yield metrics are null; D790 modulus is null "
"since it needs the raw load-deflection trace to recompute (see "
"flexural_modulus_pa)."
) if curve_is_analyzed else (
(
"Raw persistent.h5/DAQ h5 unavailable for this specimen; curve "
"and metrics extracted from the TestWorks xlsx export only. "
"Strain/stress curve and modulus are both derived from the raw "
"load-deflection trace via the D790 chord formulas with span = "
"16*depth (see flexural_stress_strain_pa / flexural_modulus_pa) "
"rather than TestWorks' own analysis. Yield metrics are null."
) if session["astm"]["standard"] == "D790" else (
"Raw persistent.h5/DAQ h5 unavailable for this specimen; curve "
"and metrics extracted from the TestWorks xlsx export only. "
"Strain/stress arrays and yield metrics are null."
)
)
),
"source_paths": {
"persistent_h5": None,
"daq_h5": None,
"xlsx": str(xlsx_path.relative_to(ROOT)),
"xlsx_sheet": xlsx["sheet_name"],
},
}
_FILENAME_SAFE = re.compile(r"[^A-Za-z0-9_-]")
def _row_filename(row: dict) -> str:
"""Pick a per-specimen filename: sample_id for SLS rows,
{material_class}_TSR{n} for non-SLS controls."""
if row["sample_id"]:
stem = row["sample_id"]
else:
stem = f"{row['material_class']}_{row['test_run_name']}"
return _FILENAME_SAFE.sub("_", stem) + ".jsonl"
def process_session(session: dict) -> None:
"""Build per-specimen JSONL files for one TestWorks session."""
standard = session["astm"]["standard"]
out_dir = DATA_DIR / standard
out_dir.mkdir(parents=True, exist_ok=True)
jobs_by_date = load_database_jobs()
log.info(f"== {session['session_folder']} ({standard}, batch {session['batch_label']})")
builder = build_row_from_xlsx if session.get("xlsx_only") else build_row
seq = 0
written = 0
for tsr_idx, material_class, db_print_date in session["test_runs"]:
if material_class == "SLS":
seq += 1
sample_id = f"{session['batch_label']}{seq}"
else:
sample_id = None
row = builder(session, tsr_idx, material_class, db_print_date,
sample_id, jobs_by_date)
out_path = out_dir / _row_filename(row)
with out_path.open("w", encoding="utf-8") as out:
out.write(json.dumps(row, ensure_ascii=False) + "\n")
written += 1
log.info(f" TSR{tsr_idx} [{sample_id or material_class}] -> {out_path.relative_to(ROOT)}")
log.info(f" wrote {written} specimens to {out_dir.relative_to(ROOT)}/")