A Generic Model-based Prognosis Architecture – ICSL

The prognosis task estimates how quickly the damage of a system, which is taken to be already at fault, will progress. We take into consideration that progression of the damage depends on how the system is used (damage progression rates may be affected by changes in environmental conditions, amount of load in the system, changes in the system’s process, etc.), and we may or may not know how the system will be used in the future. We also consider that, although we know the system is already faulty, we may not be certain about the exact amount of damage present. The diagnosis task will only give us an inexact estimate of it, bounded by uncertainty measures. Even so, we would like to make an “educated guess” of how much confidence we can place on our system to complete a given task before either it is stopped for maintenance or it breaks down. Said in other terms, we want to know the remaining useful life of our yet-operational system while the detected fault conditions linger and damage progresses.

This aspect of the research intends to provide a generic method of designing model-based prognostic systems. Focus is placed on prognostic systems that require anteceding diagnostic activities, although it is reasonable to expect that a development of this kind is applicable to systems that do not require diagnostic updates (e.g., systems where a prognosis deals with wear and tear and is thus exclusively based on the time the system has been operating).

The present research effort intends to offer a methodical approach to designing a model-based prognostic system that provides functionality in the distinct blocks represented in the image below. The approach depicted in the figure may be applicable to many kinds of systems.


The prognostics aspect of the research attempts to predict how a fault in a dynamical system will evolve given a specific expected usage profile. With this information, it is possible to make estimates of the Remaining Useful Life (RUL) or the Time to Failure (TTF) of the system. This task is very complex and typically requires support from different engineering disciplines, as is the case of the three sub-blocks within the model-based prognosis block, each of which uses specialized techniques from varied areas of engineering that can include Engineering Mechanics, Reliability Engineering, Electrical Engineering, Computer Science, Information Science, Material Science, and Statistics and Mathematics, among others.