Assistance to dynamic maintenance tasks by ann-based models

Abstract

Reliability requirements are increasingly demanded in all economic sectors, practical applications of the reliability theory are pursued often as a tailor-made suit for each case, in order to manage the assets effectively according to specific operating and environmental conditions. In sectors like renewable energy, these conditions can be changing importantly over time and reliability analysis is periodically required. At the same time, adapting a unique model to similar systems placed in different plants has proven to be troublesome. This paper adapts reliability models in order to incorporate monitored assets operating and environmental conditions. This paper introduces a logic decision tool which is based on two artificial neural networks models; allowing updating assets reliability analysis in relation to changes in operating and/or environmental conditions, and even more, this model could be easily automated within a SCADA system. Thus, by using the model, reference values and the corresponding warnings and alarms can be dynamically generated, serving as an online diagnostic or prediction of a potential degradation of the asset. The models are developed according to the available amount of failure data and they are used to detect early degradation in the energy production due to power inverter and solar tracker failures, and to evaluate the associated economic risk to the system under existing conditions. This information can then trigger preventive maintenance activities.