The main objective of prognostic is to provide the efficient information to predicted time to failure (TTF), or the remaining useful life (RUL). In addition, prognostic can be used to indicate the degree of certitude of the future predicted failure time. The predicted time to failure (TTF) is the relative probability to the occurrence of the failure event. This measure reveals the expected time to perform preventive actions. The required information to perform actions includes: asset decomposed structure and tag data, failure history, past operating conditions, current condition, identified failure patterns, maintenance history, degradation patterns.
Similar to diagnosis, prognostic methods can be classified as model-based or data-driven. Each one of these approaches has its own advantages and disadvantages.
Model-based approaches require specific failure mechanism knowledge relevant to the monitored asset. The model-based methods assume that an accurate mathematical model can be constructed. Moreover, if the understanding of the system improves, the model can be improved to increase its accuracy to address problems. The weakness of this approach is that it can be difficult, even impossible to catch the system’s behavior. The prognostic tools must evolve in the same pace as the system degradate.
Data-driven approaches use real data (sensors) to reveal degradation and to forecast the behavior of an asset. Data is the major source for a deeper understanding of the system degradation. Data-driven approaches can be divided into two categories: artificial intelligence (AI), neural networks, fuzzy logic, decision trees and statistical techniques. Case-based Reasoning (CBR) have been considered as potential candidates for prognostic algorithms too.
Due to the algorithms involved, quantitative data are processed with statistical techniques, the result is a stochastic estimation. With data-driven techniques noisy data is transformed into information. In theory, AI techniques have the ability to out-perform conventional approaches. In practice, however it isn’t easy to apply AI techniques due to the lack of training data and specific knowledge. So far, data-driven approaches are highly-dependent on the quantity and quality of data.
Nowadays ‘prognostic’ is recognized as a hype in maintenance. Considering the benefits that ‘prognostic’ may bring, it is still a novel axis of development (a few decades). The variety of published articles in maintenance literature is of good omen. However, prognostics is suited to solve the prediction problem it is not sufficient to make a choice: one must have a closer look on implementation requirements and constraints.