For an AI-based predictive maintenance system, PMI-style AI lifecycle guidance emphasizes that the first critical step is defining a comprehensive data collection strategy aligned with the business objective and risk profile. Predictive maintenance models require a blend of historical failure records, maintenance logs, operational sensor readings (e.g., temperature, vibration, pressure), usage patterns, and contextual data such as environment and flight profile. The project manager is expected to ensure clarity on what data is needed, from which sources, at what frequency, and under what quality standards, before investing in pipelines, cleaning routines, or pilots.

Option A (setting up real-time streaming) and B (data cleaning and preprocessing) are important implementation tasks, but they come after the fundamental question of “which data and why?” has been answered. Option D (pilot with a small dataset) is a useful validation step, but it still depends on having the right data identified and collected in the first place. PMI-oriented AI governance stresses making data requirements explicit and traceable to model objectives, performance metrics, and regulatory constraints.

Thus, the project manager should develop a comprehensive data collection strategy (option C) to define and structure all required data for training the predictive maintenance model.