Development of a Hydrogenerator Prognosis Approach

摘要

A prognosis does not strictly assess the remaining useful life of repairable systems, but rather predicts the future state of the system and how its degradation will evolve, thus providing the basis for predictive maintenance. In past years, Hydro-Québec has adopted a maintenance strategy based on three types of maintenance—corrective, scheduled and, more recently, condition-based (CBM)—the third directly linked to diagnostic tests. Such a CBM strategy is now widely used for transmission assets. As for generation assets, an integrated hydrogenerator diagnostic application provides information about the overall condition of generators fleet-wide. This system ranks generator condition both for individual units and for all of Hydro-Québec’s power plants as a whole. The health index (HI) ranges from 1 to 5, the highest being the worst condition. This information is used to prioritize generators for maintenance. However, the application does not suggest any particular maintenance action that should be performed to prevent the effects of specific failure mechanisms affecting a generator.To identify the best maintenance actions for any given unit, the specific active failure mechanisms must be known. Currently, this requires that knowledgeable experts study all symptoms and relate them to all possible failure mechanisms. Much of this tedious work could be performed much faster by an automated prognostic tool that reproduces the expert’s judgement. Moreover, such a tool would use a common set of accepted rules so the conclusion would not be subjective.The work presented in this paper systematizes the approach of model-based generator prognostics. It is based on failure mechanism and symptom analysis (FMSA) applied here to hydrogenerators. Each failure mechanism leads to a failure mode after which the equipment can no longer perform its function.All failure mechanisms originate from one or a combination of four sources of stress: thermal, electrical, ambient and mechanical (TEAM) and have identifiable root causes. They are defined by unique logical sequences of physical states. Several mechanisms may sometimes evolve together on a given generator, but only one will eventually lead to failure.Each physical state is uniquely defined by a specific set of symptoms with their respective threshold values (health indices). These symptoms are detectable with diagnostic tools and their associated health indices are logged in the integrated diagnostic database. The prognostic model under development is based on a process identifying active physical states and hence failure mechanisms with the help of logical rules. This paper shows that it is not necessary to instrument every single generator with all possible sensors or run all existing periodic tests to identify the active failure mechanisms. In fact, the most efficient way to proceed is first to identify trees of failure mechanisms originating from a common root cause based on a set of on-line “wide-band” diagnostic tools. Thereafter, a few off-line “narrow-band” tools can be used to pinpoint individual active failure mechanisms with a high level of confidence. The selected on-line wide-band diagnostic tools are the same for all generators and cover all possible failure trees. In contrast, the off-line, narrow-band tools differ for every generator, since they depend on the trees that have been identified, but lead to a high probability of identification with only a small number of tests.