Maintaining Privacy for Structural Health Monitoring in Population Based Classification

Population based structural health monitoring (PBSHM) is a recent approach to the problem of asset management which seeks to leverage information from multiple structures (the population) to improve inferences about an individuals condition. This viewpoint has been shown to improve outcomes across a range of tasks relevant to SHM. In most of these cases the model requires the construction of a classifier leveraging data from each available structure. One practical drawback of building such a classifier is that parties with access to the classifier may be able to uncover insight about the training data used by other parties. While this recovery may not be a problem when all assets are owned by a single company or there is a high degree of trust between the asset owners, it is worth considering: can PBSHM be deployed between untrusted collaborators? This work will introduce a classification tool, built on the foundations of differential privacy, which will enable application of PBSHM while bounding the privacy of the classifier itself.