Glosarium AI
Kamus lengkap Kecerdasan Buatan
RUL (Remaining Useful Life)
RUL, or Remaining Useful Life, is a key metric in predictive maintenance that estimates the remaining operational time before equipment fails, based on current state and historical data.
Survival Analysis
Survival analysis is a branch of statistics used to model the time until an event occurs, such as a failure, taking into account censored data where the equipment has not yet failed by the end of the observation period.
Prognostics and Health Management (PHM)
PHM is an engineering discipline that combines health state assessment (diagnostics) and prediction of future evolution (prognostics) of a system to optimize its maintenance and reliability.
Time-to-Failure (TTF)
Time-to-Failure (TTF) is a random variable representing the operating duration of a component or system from its commissioning or last repair until its next failure.
Censored Data
Censored data are observations where the event of interest (failure) did not occur before the end of the study, which is common in predictive maintenance and requires specific analysis techniques like survival analysis.
Weibull Analysis
Weibull analysis is a powerful statistical method for modeling equipment lifespan, allowing characterization of failure phases (infant mortality, useful life, wear-out) through its shape and scale parameters.
Health Index (HI)
Health Index (HI) is a composite score, often normalized between 0 and 1, that aggregates multiple performance indicators to provide a single quantitative measure of an equipment's overall health status.
Recurrent Neural Networks (RNN) for RUL
Recurrent Neural Networks (RNN) are particularly suited for RUL prediction as they can process time series data from sensors, capturing long-term dependencies to model degradation evolution.
Convolutional Neural Networks (CNN) 1D
1D CNNs are used in predictive maintenance to automatically extract relevant features from raw temporal signals (such as vibrations) before feeding them to a regression model for RUL prediction.
Feature Engineering for RUL
Feature engineering for RUL prediction is the process of creating descriptive variables (statistical, frequency, temporal) from raw sensor data to improve the performance of machine learning models.
Run-to-Failure Data
Run-to-failure data are chronological datasets that track equipment from its normal operating state to complete failure, forming the ideal learning basis for RUL models.
Similarity-Based Approach
The similarity-based approach estimates the RUL of equipment by comparing its current degradation trajectory to a database of historical failure trajectories, assuming that similar behaviors lead to similar failure times.
Ensemble Methods for RUL
Ensemble methods, such as Random Forest or Gradient Boosting, combine predictions from multiple base models to produce a more robust and accurate RUL estimate, reducing variance and bias.
Remaining Useful Life Distribution
The RUL distribution provides not only a point estimate of remaining time but also a measure of uncertainty associated with this prediction, often in the form of confidence intervals or probability density.
End-of-Life (EOL) Prediction
End-of-Life (EOL) prediction is a classification task that aims to determine whether equipment will reach the end of its useful life within a specified future time window, complementary to RUL regression.
Attention Mechanisms in RUL Models
Attention mechanisms, integrated into models like Transformers, allow the model to focus on the most relevant parts of the temporal data sequence to refine RUL prediction.
Domain Adaptation for RUL
Domain adaptation is a transfer learning technique that allows applying a RUL model trained on one type of equipment (source domain) to another similar but different equipment (target domain) with little data.