Longitudinal and time to event models


To minimise the number of unplanned pump failures at a pulp mill in Canada (Irving Pulp and Paper), an observational study [1, 2, 3] was conducted whereby vibrations and failure times of pumps were recorded during operation. In the study were 18 identical pumps (Goulds 3175L) with vibrations recorded using hand-held accelerometers and readings taken from 8 locations on each pump. The dataset comprises a survival table of pump failure times and a longitudinal table of vibration measurements during operation.

The idea is to use the vibration readings together with the theory of Vibrational Analysis to probabilistically model bearing failure in pumps, and use this to perform preventative maintenance prior to a costly failure during operation. A model and decision rule based on this dataset was published by C-MORE researchers, Sundin et al [3]. This thesis takes another look at this dataset and carefully retraces the steps from Sundin et al. to assess the method, examine the assumptions and construct and fit alternative models.


[1] D Banjevic and AKS Jardine. Calculation of reliability function and remaining useful life for a markov failure time process. IMA Journal of Management Mathematics, 17(2):115–130, 2006.

[2] Aiwina Heng, Andy CC Tan, Joseph Mathew, Neil Montgomery, Dragan Banjevic, and Andrew KS Jardine. Intelligent condition-based prediction of machinery reliability. Mechanical Systems and Signal Processing, 23(5):1600–1614, 2009.

[3] PO Sundin, N Montgomery, and AKS Jardine. Pulp mill on-site implementation of cbm decision support software. In Proceedings of International Conference of Maintenance Societies, Melbourne, Australia, 2007

Leave a Reply