MSc Projects (finished)

Robbin Gajadin - Dry Gas Seal Reliability and the Development of Prognostic Models

(June – December 2017)

Link to report in repository:


In the Oil & Gas industry, service logistics support and maintenance of systems constitute a significant fraction of the annual operating costs. Since these assets are often operated at remote locations around the world, unplanned maintenance requires significant logistic effort and thus becomes very costly. Consequently, in an ultimate attempt to decrease said unplanned maintenance, a comprehensive approach to Maintenance- and Reliability Engineering has been explored in this study. Tools and methodologies of what constitutes the two fields have been widely applied towards deriving better insights into the failure modes and mechanisms of centrifugal compressor dry gas seals (DGS), one of the more critical components used in this industry. Three distinct analyses were performed on the DGS, to further expand the understanding of its failures to different degrees and ultimately determine whether its maintenance can be made more predictive instead of reactive. The first being of the qualitative type, whereby a FMECA and FTA were performed to analyze the seal’s failure modes on a high level. The results hereof showed that most seal failures stem from some form of contamination ingress into the DGS, causing issues with its secondary sealing element (SSE), including other failure mechanisms such as wear and extrusion. Hereafter, a statistical reliability study was performed in order to extract the seal’s failure patterns. The conclusion of which was that the seals show a predominantly wear-out pattern and high correlation with the number of compressor starts. Though conclusive remaining useful life (RUL) calculations could not be made due to a lack of sufficient data, the probability of failure of the seals as a group was reasonably accurately determined for running hours and starts to failure. Lastly, a physics of failure (PoF) model of the SSE was developed, which uses operation and material specific parameters as an input, returning estimated RUL values pertaining to wear as an output. The results of which showed good agreement with expectations. Said model consisted of a combined Finite Element and analytical approach, and allowed room for expansion to additional failure mechanisms. Moreover, in addition to wear and extrusion, the model was able to deduce under what circumstances seal hanging can take place, which is one of the primary DGS failure modes. The conclusion of the study was that DGS maintenance can theoretically be made more predictive, given the right inputs and condition monitoring system. The latter requiring appropriately high vibration sampling rates and frequency response spectrum analysis capabilities. Ultimately, the recommendations state that the statistical as well as PoF model can significantly be improved, given that more failure data is logged in a structured manner and sufficient experimental (material) data is available to quantitatively validate the model.