Towards a predictive maintenance methodology of hydraulic pumps

• Autor(a) principal: Santos, Inês Vilela dos
• Data de Publicação: 2021
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10773/31362
• topic: Hydraulic pumps; Maintenance; Fault detection and classification; Remaining useful life estimation

Hydraulic pumps, essential elements in water supply systems, are mainly responsible for the high energy consumption associated with these systems. It is, therefore, relevant to keep the pumps running in their best possible conditions in order to avoid further consumption and costs, and also to anticipate possible pump failures. The best strategy to anticipate the occurrence of failures is to implement preventive and predictive maintenance plans, instead of corrective maintenance that is still widely applied. Thus, with the goal of developing a predictive maintenance methodology applied to hydraulic pumps, this dissertation aims to explore and investigate the applicability of two techniques that can be integrated into a maintenance plan: the detection and classification of failures and the estimation of the remaining useful life (RUL) of the pump. To implement the proposed tasks, simulated data and measured data from real systems were used, taken from online data repositories, with values recorded by sensors and with the identified condition of the system. The first technique allowed, through sensor data with the respectively identified faults, to train classification algorithms able to identify failures. In the first of the evaluated case studies, the best of the implemented algorithms identified the failures associated with the pump data with an accuracy of 82.9%, whereas, in the second of the evaluated case studies, the algorithm that presented the best performance obtained an accuracy of 94.6% in identifying the failure mode associated with the pump. The decision tree and ensemble trees algorithms proved to be the most suitable for the studied purpose. The second technique allowed to estimate RUL values from sensor data recorded from normal operation to system failure. Although the first RUL implemented case study was an engine, the second case study was a water pump. The methodology of the RUL model proved to be relevant because it managed, even with some deviations from the true values, to estimate acceptable values of RUL. An economic analysis was also carried out, highlighting the relevance of applying RUL estimation models in predictive maintenance methodologies for hydraulic pumps