On Conserving Privacy in Structural Health Monitoring

Abstract

The ability to make informed decisions with regards to maintenance and service life is critical across many high-value industries. Making accurate predictions about the condition of assets is a key goal and one supported by structural health monitoring (SHM) strategies. To help support this aim, the development of a population-based approach to structural health monitoring (PBSHM), has been making progress in recent years. This solution for dealing with structures, aims to group them together in “populations” and transfer information between individuals, within the population. PBSHM necessitates the sharing of information between structures, which may not all be controlled by a single stakeholder. Therefore, a concern arises that PBSHM may compromise “private” information between different organisations or individuals if their assets are included in a PBSHM scheme. This work will seek to demonstrate a solution to this problem. Differential Privacy is a probabilistic privacy regime that retain population trends in a dataset whilst keeping individual data private from others by the injection of random noise into the data. In this context, privacy is a one-way mapping which obscures sensitive information in a dataset, in a non-reversible manner; as opposed to encryption which may be “broken”. Being differentially-private gives a bound on the amount of information which is ever recoverable. Being able to protect discriminating features of different structure designs from other parties will promote and accelerate the adoptions of PBSHM strategies in industrial settings, where protecting information is a key concern. This paper will demonstrate the application of differential privacy to a simulated homogeneous population of structures.