A constrained Langevin-adapted Particle Filter for Aircraft Engines’ Health Monitoring

We examine the application of particle filter in estimating performance indicators of an aircraft engine; these indicators are a crucial aspect in health monitoring and condition-based maintenance for aeronautics. This approach is flexible and not restricted by rigid assumptions often found in other methods; however, it poses three challenges in our context: (i) high computation cost: classical particle filters require a large number of particles, each of them calling a heavy model; (ii) non-observability: in our system, different system states might provide the same measurements; (iii) constraints: constraints on the estimation are required to be integrated dynamically. We propose a version of particle filter, based on Langevin dynamics, to resolve these challenges.