CASE STUDY: AeroMech
Overview: Aviation manufacturer, 5000 employees, $2B revenue. 100 engines, 10k sensors/engine, 1GB data/flight. On-prem Hadoop.
Business Req: Predictive maintenance, secure data sharing with airlines, monetize data.
Execs: CEO wants new revenue; CFO demands ML ROI; CTO says on-prem storage unfeasible.
Tech Req: High-throughput ingestion, PB-scale storage, train ML on historical data, deploy ML to edge (aircraft).
Constraints: Intermittent low-bandwidth flight connectivity, aviation data compliance, data scientists use Python/Jupyter.
QUESTION:
How should you design the ingestion pipeline to handle the intermittent connectivity and high data volume from the aircraft engines?
GCP PCA · Question 15 · Compute Systems
CASE STUDY: AeroMech
Overview: Aviation manufacturer, 5000 employees, $2B revenue. 100 engines, 10k sensors/engine, 1GB data/flight. On-prem Hadoop.
Business Req: Predictive maintenance, secure data sharing with airlines, monetize data.
Execs: CEO wants new revenue; CFO demands ML ROI; CTO says on-prem storage unfeasible.
Tech Req: High-throughput ingestion, PB-scale storage, train ML on historical data, deploy ML to edge (aircraft).
Constraints: Intermittent low-bandwidth flight connectivity, aviation data compliance, data scientists use Python/Jupyter.
QUESTION:
To meet the requirement of deploying ML models to the aircraft for real-time anomaly detection, which approach should you use?
Answer options:
Host the model on Vertex AI Endpoints and have the aircraft query it via REST API.
Export the trained model from Vertex AI to a TensorFlow Lite format and deploy it to an edge computing device on the aircraft.
Deploy a full Kubernetes cluster on each aircraft using Anthos to run the models.
Use Cloud Functions to process the sensor data as it streams in.
50 questions · hints · full answers · grading
Expert