Time-Series Forecasting

Time-Series Forecasting

animation
iiot
maintenance
Try Change ML Model Type: and Training Examples: to compare performance. Then press Calculate expected performance.
Observe Watch accuracy, precision, recall, and F1 score as Training Examples: changes.
Explain Describe how Historical Failure Rate: changes expected failures and missed failures after you press Calculate expected performance.

Technical boundaries. Performance uses fixed base metrics with training data adjustments capped at 200 examples. Expected failures and alerts use 100 assets.

Time-Series Forecasting

Approach: Predicts remaining useful life (RUL) based on degradation trends.

Algorithms:

  • LSTM neural networks
  • Prophet (trend + seasonality)
  • Gaussian Process Regression

Output: “How many hours/days until failure?” (RUL estimate with confidence interval)

Key metrics:

  • Mean Absolute Error (MAE)
  • Root Mean Square Error (RMSE)
  • Percentage within 10%/20% tolerance

Back to the chapter