Knowledge Transfer in Engineering Fleets: Hierarchical Bayesian Modelling for Multi-Task Learning

We propose a population-level analysis to address issues of data sparsity when building predictive models of engineering infrastructure. By sharing information between similar assets, hierarchical Bayesian modelling is used to improve the survival analysis of a truck fleet (hazard curves) and power prediction in a wind farm (power curves). In each example, a set of correlated functions are learnt over the asset fleet, in a combined inference, to learn a population model. Parameter estimation is improved when sub-fleets of assets are allowed to share correlated information at different levels in the hierarchy. In turn, groups with incomplete data automatically borrow statistical strength from those that are data-rich. The correlations can be inspected to inform which assets share information for which effect (i.e. parameter).