Skip to content
Draft
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Original file line number Diff line number Diff line change
@@ -0,0 +1,5 @@
"""NASA C-MAPSS FD001 predictive-maintenance connector."""
from timenet_connectors.datasets.musitu.cmapss_fd001.connector import (
CONNECTOR as CONNECTOR,
CmapssFd001Connector as CmapssFd001Connector,
)
Original file line number Diff line number Diff line change
@@ -0,0 +1,151 @@
"""NASA C-MAPSS FD001 TimeNet connector (NASA PCoE simulated engine data).

The original archive is obtained from NASA's Zenodo-hosted record at build
time and verified against its published MD5. It is not committed to TimeNet.

Source: https://zenodo.org/records/15346912
Reference: A. Saxena, K. Goebel, D. Simon, N. Eklund (PHM 2008),
"Damage Propagation Modeling for Aircraft Engine Run-to-Failure Simulation".

The dataset record does not specify a license. We label it "other" rather than
assert rights we cannot verify; review NASA's general data policy and source
terms before distributing any derived data artifact.
"""
from __future__ import annotations

import hashlib
import os
from pathlib import Path
from urllib.request import Request, urlopen
from zipfile import ZipFile

import numpy as np
import pandas as pd

from timenet.connectors import BaseConnector
from timenet.dataset import OrdinalAxis, Record, Signal, Source, TimeFDataset
from timenet.types import Annotation, ScalarPredictionTask, Split, TimeSeriesSpec, ureg

ARCHIVE_URL = "https://zenodo.org/records/15346912/files/CMAPSSData.zip?download=1"
ARCHIVE_MD5 = "79a22f36e80606c69d0e9e4da5bb2b7a"
COLS = ["unit", "cycle"] + [f"setting_{i}" for i in range(1, 4)] + [
f"sensor_{i}" for i in range(1, 22)
]
SPEC = TimeSeriesSpec(
spec_type="engine_telemetry", name="Engine sensor observation",
unit_value=ureg.dimensionless,
)
WINDOW = 30
STRIDE = 15


def _dataset_files(root: Path) -> list[Path]:
return [root / n for n in ("train_FD001.txt", "test_FD001.txt", "RUL_FD001.txt")]


def _read(path: Path) -> pd.DataFrame:
frame = pd.read_csv(path, sep=r"\s+", header=None, engine="python")
frame = frame.dropna(axis=1, how="all")
if frame.shape[1] != len(COLS):
raise ValueError(f"Expected 26 columns at {path}; got {frame.shape[1]}")
frame.columns = COLS
if not np.isfinite(frame.to_numpy(dtype=float)).all():
raise ValueError(f"Non-finite value in {path}")
frame[["unit", "cycle"]] = frame[["unit", "cycle"]].astype(int)
if frame.duplicated(["unit", "cycle"]).any():
raise ValueError(f"Duplicate engine-cycle rows in {path}")
return frame


class CmapssFd001Connector(BaseConnector[Path]):
"""Build a leak-aware RUL prediction dataset from NASA FD001 trajectories."""

def download(self, cache_dir: Path) -> Path:
"""Use local source files when provided; else fetch the checksummed NASA archive."""
root = Path(os.environ.get("CMAPSS_DATA_DIR", str(cache_dir))).resolve()
root.mkdir(parents=True, exist_ok=True)
if all(path.is_file() for path in _dataset_files(root)):
return root

archive = root / "CMAPSSData.zip"
if not archive.exists() or hashlib.md5(archive.read_bytes()).hexdigest() != ARCHIVE_MD5:
request = Request(ARCHIVE_URL, headers={"User-Agent": "TimeNet-cmapss-fd001/0.1"})
parts, received = [], 0
with urlopen(request, timeout=90) as stream:
while part := stream.read(1024 * 1024):
received += len(part)
if received > 100 * 1024 * 1024:
raise ValueError("Unexpectedly large NASA dataset download")
parts.append(part)
payload = b"".join(parts)
if hashlib.md5(payload).hexdigest() != ARCHIVE_MD5:
raise ValueError("NASA data archive MD5 mismatch")
archive.write_bytes(payload)
with ZipFile(archive) as z:
members = {Path(name).name: name for name in z.namelist()}
for path in _dataset_files(root):
if path.name not in members:
raise ValueError(f"NASA source archive missing {path.name}")
path.write_bytes(z.read(members[path.name]))
return root

def convert(self, root: Path) -> TimeFDataset:
"""Create one 30-cycle multivariate record per sampled training window."""
train, test = _read(root / "train_FD001.txt"), _read(root / "test_FD001.txt")
offset_df = pd.read_csv(root / "RUL_FD001.txt", sep=r"\s+", header=None)
offset_df = offset_df.dropna(axis=1, how="all")
test_ids = sorted(test.unit.unique())
if len(test_ids) != len(offset_df) or offset_df.shape[1] != 1:
raise ValueError("NASA official test RUL labels do not match engine count")
test_offsets = dict(zip(test_ids, offset_df[0].astype(int).tolist()))
dataset = TimeFDataset(metadata=self.metadata())
train_ids = sorted(train.unit.unique())
# Deterministic by ENGINE, not by window; no shared engine in train and validation.
val_ids = set(train_ids[::5])

for cohort, source in (("development", train), ("official_test", test)):
for unit, group in source.groupby("unit", sort=True):
group = group.sort_values("cycle").reset_index(drop=True)
if len(group) < WINDOW:
continue
if cohort == "official_test":
split = Split.TEST
ends = [len(group)-1]
else:
split = Split.VALIDATION if unit in val_ids else Split.TRAIN
ends = range(WINDOW-1, len(group), STRIDE)
for end in ends:
observed = group.iloc[end-WINDOW+1:end+1]
current_cycle = int(group.cycle.iloc[end])
target = (
int(group.cycle.iloc[-1]) - current_cycle
if cohort == "development" else test_offsets[unit]
)
signals = [
Signal.from_values(
observed[col].to_numpy(dtype=np.float32),
spec=SPEC, name=col, time_axis=OrdinalAxis(),
) for col in COLS[2:]
]
record = Record(
sources=[Source("NASA simulated engine telemetry", signals=signals)],
annotations=[
Annotation("engine_id", value=int(unit)),
Annotation("cohort", value=cohort),
Annotation("observation_cycle", value=current_cycle),
],
)
dataset.add_record(record)
dataset.add_task(
ScalarPredictionTask(
inputs=[record],
targets=float(target),
unit="cycle",
target_name="remaining_useful_life_cycles",
split=split,
)
)
return dataset


CONNECTOR = CmapssFd001Connector
Original file line number Diff line number Diff line change
@@ -0,0 +1,19 @@
# yaml-language-server: $schema=https://docs.timenet.ai/schemas/dataset-card-v1.schema.json
yaml_schema_version: 1
dataset_id: musitu/cmapss-fd001
dataset_version: 0.1.0
name: NASA C-MAPSS FD001 Predictive Maintenance
description: >-
NASA Ames C-MAPSS simulated turbofan engine histories in 30-cycle
multivariate windows. Each task predicts remaining useful life in cycles.
Engine IDs do not cross the train/validation partition. NASA official test
engines are used for test-only endpoint evaluations.
license: other
source_url: https://www.nasa.gov/intelligent-systems-division/discovery-and-systems-health/pcoe/pcoe-data-set-repository/
domains:
- engineering
tags:
- predictive-maintenance
- time-series
- simulation
- rul