Orchestrator

polars_ts.iiot_agents.orchestrator

MaintenanceOrchestrator: trains and evaluates maintenance agents.

MaintenanceResult dataclass

Output of a :class:MaintenanceOrchestrator run.

Attributes

actions Per-step maintenance action id from the final greedy policy. health_index Per-step fused health index. rul Per-step Remaining Useful Life estimate (steps). first_maintenance_step Index of the first MAINTAIN action, or -1 if never. total_reward Total environment reward of the final greedy evaluation pass. history Per-step diagnostic records.

MaintenanceOrchestrator

Coordinate predictive-maintenance agents over a machine trajectory.

The scheduler is trained over n_episodes replays of the trajectory, then evaluated once greedily to produce the reported schedule.

Parameters

n_episodes Number of Q-learning training episodes over the trajectory. window Sliding-window length for spectral/health feature extraction. failure_threshold Health level defining failure (shared by env, health, and RUL). seed Seed for the scheduler's RNG.

_health_series(sensors, health)

Per-step fused health index (uses ground truth when provided).

run(sensors, health=None, failure_step=None)

Train the scheduler then return its greedy maintenance schedule.

Parameters

sensors 2D array (n_steps, n_sensors) of sensor readings. health Optional ground-truth health trajectory. failure_step Optional explicit failure index.

Returns

MaintenanceResult