Ultimate access to all questions.
You are tasked with designing an architecture for a predictive maintenance system. The goal is to serve asynchronous predictions to determine whether a mission-critical machine part will fail. Your system collects data from multiple sensors on the machine. The model should predict a failure that could occur in the next N minutes, based on the average of each sensor's data over the past 12 hours. How should you design the architecture to meet these requirements?
Explanation:
The correct answer is B. This architecture leverages the strengths of Pub/Sub, Dataflow, and Vertex AI to handle asynchronous predictions in real-time. It is important to process sensor data streams in real-time for predictive maintenance to ensure timely predictions of potential failures. Pub/Sub handles the high volume of incoming sensor data and buffers messages to prevent data loss. Dataflow processes the events in real-time, performs feature engineering and data preprocessing, and invokes the trained ML model for prediction. The predictions are then published to another Pub/Sub topic for downstream consumption. This design ensures scalability, resilience, and real-time processing, addressing the mission-critical requirements effectively.