The reliability engineering community has been battling with the difficulty of modelling the complex behaviour of asset deterioration. One of the key research areas receiving increasing attention is to use mathematical models to capture the phenomenon of dependences among components within multi-component assets.

Amongst these, ‘stochastic dependence’, which represents the failure and deterioration dependence among aging components, is the most challenging and underdeveloped. Existing research in stochastic dependence predominantly examines either a correlated deterioration that is inherently embodied among the aging components, or shocked damage induced by a malfunctioned component on other components. However, in some cases, these two dependences may also interact. In a multi-component asset, a fault induced by a malfunctioned component may further propagate among other components, resulting in the accelerated deterioration of the asset. This will refer to this meta-dependent feature among components as fault propagation.

A two-layered approach was design to mimic the deterioration of a multi-component asset with fault propagation. In the lower layer, the phenomenon of fault propagation is modelled using a vector-valued continuous-time Markov chain. In order to overcome the state space explosion problem that might arise when modelling this behaviour for complex multi-component assets, a novel partitioning algorithm for Markov aggregation was implemented to reduce the state-space. This results in a compact upper layer deterioration model. This is referred to as the multi-dependent deteriorating paths model. This model is then used to examine the impact of fault propagation on the failure time distribution and expected lifetime of the asset. The results show that the meta-dependent characteristic of fault propagation has a noteworthy impact on those two indicators.

To alleviate the impact of fault propagation, a conditional fault mitigating maintenance model is designed to use fault preventive maintenance to mitigate the risk of fault propagation. It is seen that such a policy increases asset availability and lowers the operational cost of the asset. Simulated Annealing is employed to search for the optimised maintenance policy aimed at maximising availability and minimising operational cost.

The models developed are validated using a real case example of high-voltage power transformers. The case study clearly shows the potential practical value of deterioration and maintenance models presented. Experimental findings, historical data, theoretical propositions and practical knowledge of power transformer deterioration and maintenance are parameterized to provide inputs to the designed deterioration and maintenance model. As a result, it is shown that the risk of fault propagation can be strategically mitigated in high-voltage power transformers.

The model and the case study has the potential to stimulate the realization of other interesting applications and evoke awareness of the phenomenon of fault propagation, and provide asset managers with a more sophisticated maintenance strategy for multi-component assets.

Contact Zhenglin Liang for further information.