Files

T

Antigravity Agent f8b21af4be Wiki cleanup: error-doc removal, dedup merge, link normalization

10_Wiki/Topics 대규모 정리:
- 오류 캡처/미완성 stub 문서 227개 제거
- 교차폴더 중복 43클러스터 병합 (63파일 → redirect)
- 링크명 정규화: 깨진 링크 수정·redirect 직결·개념 매핑 ~2,400건
- 카테고리 MOC 6개 신규 생성
- Graph 섹션 미해결 related-keyword 링크 10,058건 제거

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

2026-05-20 23:52:15 +09:00

4.8 KiB

Raw Blame History

id, title, category, status, canonical_id, aliases, duplicate_of, source_trust_level, confidence_score, verification_status, tags, raw_sources, last_reinforced, github_commit, tech_stack

title

Digital Twins

매 한 줄

"매 physical asset/system 의 매 live, bidirectional digital replica". Grieves 2002 의 PLM concept 가 매 IoT, sensor cost 폭락, real-time sim, generative AI 의 만남으로 매 Industry 4.0 의 핵심. 매 NVIDIA Omniverse, Azure Digital Twins, Siemens Xcelerator 의 2026 era — physical + digital 의 매 closed loop.

매 핵심

매 3 layer

Physical: 실제 자산 + sensor (vibration, temp, pressure, camera).
Communication: MQTT/OPC-UA/AMQP, time-series store, edge gateway.
Digital: 3D model + physics sim + ML predictor + control logic.

매 spectrum

Digital model: static 3D, no live data.
Digital shadow: one-way (physical → digital).
Digital twin: bidirectional — twin can command physical.

매 응용

Predictive maintenance (jet engine, wind turbine).
Smart city traffic / energy optimization.
Manufacturing line virtual commissioning.
Healthcare (patient-specific organ twin).
Robot fleet sim (NVIDIA Isaac, Omniverse).

💻 패턴

Sensor → twin (MQTT + Python)

import paho.mqtt.client as mqtt, json, time

def on_message(c, u, msg):
    data = json.loads(msg.payload)
    twin.update_state(asset_id=data['id'], temp=data['temp'], ts=data['ts'])
    if twin.predict_failure(data['id']) > 0.8:
        c.publish(f"cmd/{data['id']}/throttle", "0.5")  # bidirectional!

cli = mqtt.Client(); cli.on_message = on_message
cli.connect("mqtt.factory.local"); cli.subscribe("sensor/+"); cli.loop_forever()

Azure Digital Twins (DTDL)

{
  "@id": "dtmi:com:factory:Pump;1",
  "@type": "Interface",
  "displayName": "Pump",
  "contents": [
    { "@type": "Property", "name": "rpm", "schema": "double" },
    { "@type": "Telemetry", "name": "vibration", "schema": "double" },
    { "@type": "Command", "name": "shutdown" },
    { "@type": "Relationship", "name": "feeds", "target": "dtmi:com:factory:Tank;1" }
  ]
}

Physics-based twin (Modelica via FMU)

from fmpy import simulate_fmu
result = simulate_fmu(
    'pump.fmu',
    start_values={'inlet_pressure': 2.5, 'rpm': 1800},
    output=['outlet_pressure', 'efficiency'],
    stop_time=10.0
)

Omniverse USD scene (live update)

from pxr import Usd, UsdGeom, Gf
stage = Usd.Stage.Open('factory.usd')
pump = UsdGeom.Xform.Get(stage, '/World/Pump_01')
# Stream live sensor pose via OmniGraph / Live Sync
def on_telemetry(rpm, vibration):
    pump.GetPrim().GetAttribute('rpm:live').Set(rpm)
    pump.GetPrim().GetAttribute('vib:live').Set(vibration)

Predictive maintenance (LSTM)

import torch.nn as nn
class FailurePredictor(nn.Module):
    def __init__(self, n_sensors=8):
        super().__init__()
        self.lstm = nn.LSTM(n_sensors, 64, batch_first=True)
        self.head = nn.Linear(64, 1)
    def forward(self, x):  # (B, T, n_sensors)
        h, _ = self.lstm(x)
        return torch.sigmoid(self.head(h[:, -1]))
# Train on (sensor window, RUL label) pairs from CMAPSS / NASA dataset.

매 결정 기준

상황	Approach
Single asset, simple monitoring	Digital shadow (cheap)
Fleet w/ predictive maint.	Twin + ML failure model
Process plant commissioning	Physics twin (FMU/Modelica)
Robotics / AV training	Sim-to-real (Isaac, CARLA)
Smart city / building	Hierarchical twins (DTDL)

기본값: 매 start digital shadow → ML 추가 후 twin → bidirectional 마지막.

🔗 Graph

🤖 LLM 활용

언제: IoT/manufacturing/CPS context, sim2real planning, asset lifecycle. 언제 X: Pure web app, no physical asset. 매 marketing buzzword 화 주의.

❌ 안티패턴

3D model = twin 오해: 매 3D 만으론 shadow도 아님.
Sensor 무인 데이터: garbage in → garbage twin.
No model versioning: physical 변경 시 twin drift.
Closed-loop without safety: bidirectional 시 매 fail-safe + human-in-loop 필수.
Vendor lock-in: proprietary schema → DTDL/USD 표준 사용.

🧪 검증 / 중복

Verified (Grieves 2014, NIST CPS framework, Azure DT docs, NVIDIA Omniverse docs).
신뢰도 A.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — DT layers + DTDL/Omniverse/MQTT patterns

4.8 KiB Raw Blame History