Researcher: Kay Eiloof
Project Duration: September 2021 – August 2022
Project Partner: Spie
Research DESCRIPTION:
Predictive maintenance (PdM) is a maintenance policy that is gaining the interest of researchers and companies. In this policy, the maintenance decision-making is based on diagnostic- and prognostic information, achieved by monitoring specific parameters that indicate the asset's condition in real-time. PdM has some common barriers, such as the inability to process a massive amount of monitored data and its poor quality. These barriers also apply to the lock Eefde project. In this project, a centrifugal pump is monitored in real time, which is the focus of this case study. SPIE Nederland B.V. (SPIE) maintains the pumps, and in collaboration with Rijkswaterstaat (RWS), they aim to implement fault diagnosis and prognosis for them.
SPIE has successfully achieved fault detection for the pumps by implementing a Condition Monitoring System (CMS) of the pumps according to ISO-NEN 10816-7. Even though the CMS was set up according to a standard, the data for developing a fault diagnostic and prognostic program is limited. This problem has been addressed by gaining insight into the limitations of a CMS set up according to ISO standards. A data-driven framework is recommended for this study to ensure the development process of the fault diagnostic and prognostic program.
To identify the limitations of the CMS, a case study was conducted where the CMS of the pumps was studied and compared to the CMS's requirements that enable diagnosis and prognosis. This GAP analysis highlights the gaps between the case study and the requirements found in the literature review. Several frameworks were studied, and a synthesised PdM implementation framework was designed. During the framework's design, a crucial step was identified that is mainly overlooked by researchers, i.e. the asset-component selection procedure. The PdM ambition (detection, diagnosis, and prognosis) should be defined here. The findings of this research show that if this is not properly defined at the inception of the project, then it can lead to selecting the wrong CMS. The root causes of the identified gaps are the lack of a properly defined PdM ambition and the pumps' data acquisition system (sensors). The solutions that are provided based on the identified gaps are given for the asset-component selection procedure, the data acquisition system, and the ML model development process.
For the asset-component selection procedure, it is recommended that the stakeholders apply the Funnel approach instead of the conventional methods such as Failure Mode Effect Analysis and Fault Tree Analysis. The Funnel approach considers the definition of PdM ambition and the economic and technical feasibility study of the asset. For the pumps' data acquisition system, velocity transducers should be replaced with accelerometers. Accelerometers have a higher frequency range and are more accurate in detecting failures.
An experimental setup has been recommended to develop an ML model for fault diagnosis and prognosis. This experiment includes setting up a pump with a data acquisition system installed. The experimental setup is needed to collect quality and sufficient data for the ML model development process.
In conclusion, this research achieved its objectives by answering the research questions. The limitations were identified, and based on these, solutions were provided so that a fault diagnostic and prognostic program could be developed.