Smart asset management system with health degradation identification and remaining useful lifetime estimation in electrical distribution system

The electric infrastructure reliability depends not only on the manufacturing design quality but mainly on the operating conditions and the different types of stresses on the grid apparatus causing repairable and non-repairable failures. Statistics indicate that 87% of stakeholders claim that the ageing infrastructure has impacted the equipment to be prone to degradation and escalated their maintenance cost. Therefore, continuous health monitoring, ageing inspection, and remaining lifetime and operational expenditure prediction are enabling tools for smarter and adequate assets management. To date, smart grid deployment seeks the most appropriate strategy to maximize the profitability of the asset documentation and alleviates data-quality and data-depth issues in terms of information integrity, lacks in detail and high-data-rate. Moreover, electrical power assets are usually overstrained and suffer from overaged conditions including atmospheric instability. The situation is worse in Qatar due to the harsh environment with soared temperature, humidity, and dust. The electric equipment in Qatar strives for operations excellence, sustainability and optimality, for extending their useful life cycle. Observing and maintaining the health of widely dispersed and heterogeneous devices and rotating machines is indispensable in early warning of defects, prevention from major outages and tragic damages. In order to implement such effective maintenance, it is necessary to monitor various parameters of the condition of those electric assets, identify the primary degradation mechanisms, estimate the ageing, and predict the remaining useful lifetime, which allows optimal performance, reduce unplanned shutdowns, and identify the causes of possible failures at early stages. The predictive maintenance technique will play a key role in predicting the future failure point of a machine component before it occurs. The lifetime of the equipment is then maximized and the downtime is minimized. Equipment reliability could be also predicted by assessing the operation history and failures of equipment over time. Such reliability assessment could be generalized to the whole working plant, industry or utility. With the increased machinery complexity, the reliability assessment gained remarkable relevance. Although the huge number of datasets existing, some unusual failures do not have any records for predictive models and cause catastrophic problems with estimations uncertainty. It is for these reasons a single reinforcement based machine learning model might not be enough. Instead, in this project it will be shown that reinforcement learning combined with advanced control system tools such as MPC and Kalman filters, could be a crucial step to properly manage and analyze each individual asset and the set of assets and to estimate their health, degradation process, and remaining useful lifetime. The proposed project could be the first step to mitigate the time-to-failure in electric assets, which allows utilities and industry to reflect the physical state of different equipment, estimate remaining useful lifetime, and identities future maintenance requirements. The project consists of three main layers, each of which is with specific activities, goals, and outcomes, as described below: Health monitoring platform. The purpose of this platform is to monitor vital electric equipment characteristics in the utility or industrial plant, identify their health condition and predict any failures before existence. Condition monitoring and automated fault detection will be achieved under this platform. Degradation and useful lifetime estimation platform The purpose of this platform is to develop a multi-dimensional model that discovers hidden patterns and relationships in order to estimate the degradation process of the assets and the remaining useful life span. The challenges in this platform is identifying and estimating the parameters contributing to the health degradation of various assets at various operating conditions. This layer will include the identification of the assets damaging mechanisms and how those mechanisms degrade the equipment lifetime. Multidimensional remaining lifetime modeling will be conducted. Big Data platform. This platform involves the collection, preparation, parsing, validation, and management of the various electric equipment characteristics parameters, operating electrical measurement data, asset population data, asset failure data, asset diagnosis data, equipment historical data, and distribution system data. The platform involves processing and analyses of the data in real-time.
Research Area Keywords: Machine learing applications; Asset management; Predictive control; Predictive maintenance; Smart grid