Framework for a Hybrid Prognostics

• Main Authors: K. Medjaher, N. Zerhouni
• Format: Article
• Language: English
• Published: AIDIC Servizi S.r.l. 2013-07-01
• Series: Chemical Engineering Transactions
• Online Access: https://www.cetjournal.it/index.php/cet/article/view/6223

Fault detection and isolation, or fault diagnostic, of physical systems has been subject of several interesting works. Detecting and isolating faults may be convenient for some applications where the fault does not have severe consequences on humans as well as on the environment. However, in some situations, detecting and diagnosing faults may not be sufficient. In these cases, it is more interesting to anticipate the time of the fault, what is the purpose of prognostics. This latter activity aims at estimating the remaining useful life of systems by using three main approaches: data-driven prognostics, model-based prognostics and hybrid prognostics. Data-driven prognostics concerns the transformation of the raw monitoring data to relevant models or trends which are then used to continuously assess the health state of the system and predict its remaining useful life. This approach is easy to implement, but suffers from precision. In addition, the method is applied in most cases on single physical components (bearings, gears, belts, etc.) and thus the interaction between the components of the whole system is not addressed. Model-based prognostics (also called physics of failure prognostics) deals with analytical modelling of the system including its degradation. This approach gives more precise results, but it is difficult to apply on complex systems for which the construction of the behaviour and degradation models is not a trivial task. Finally, the hybrid approach combines both model-based and data-driven approaches by keeping their advantages. This paper presents a framework allowing the development of hybrid prognostics. The framework relies on two main phases: an offline phase and an online phase. The first phase concerns the construction of the nominal and degradation models, and the definition of the faults and performance thresholds needed to calculate the remaining useful life of the system. The second phase deals with the utilisation of the models and thresholds obtained in the first phase to detect the fault initiation, assess the current health state of the system, predict its future health state and calculate its remaining useful life.