AI FORSchung

Fiber-optic sensing (FOS) is a well-established set of technologies that use coherent laser pulses to record linear or non-linear backscattering signals which in turn carry information about strain and temperature changes along a glass or polymer fiber. FOS can sense strains or temperature variations at a huge number of locations along the fiber almost continuously and moreover can operate in harsh environments over very long periods of time. Though well-established, the FOS market sees rapid development related to both technology (extreme amounts of data that are continuously generated along the fibre, the lack of application-agnostic AI models and the ensuing incomprehensiveness of anomaly-detection models) and to expanding needs in its use areas (water scarcity forcing utility companies to monitor their pipelines more closely, demand for less x-ray exposure in minimally-invasive medical treatment, aging civil infrastructure asking for continuous structural health monitoring). Without a solution, the rising costs related to storage and processing of extreme data streams might stop the FOS market in its tracks and cripple the growth ambitions of many promising start-ups and SMEs in the field. The overall objective of AI FORSchung is therefore to accelerate innovation and growth of fiber-optic sensing by augmenting innovations in large-scale signal and data analysis with cross-domain validated AI methods. These industrial-grade embedded AI technologies are necessary enablers for advanced, robust and widely accessible fiber-optic sensing applications in the biomedical, construction, and utilities sectors. Resulting implementations will advance adoption of fiber-optic sensing across the spectrum of applications through cost-effective and easy-to-use products that extract rich important information from fiber-optic data. We bring together an industrial partner (Philips), 3 SMEs (Fluves, ACI Monitoring,Thunderbyte.ai) and 2 academic partners (TU/e and TUG) with expertise on deep learning and the necessary deep physics and application knowledge base necessary to develop AI solutions which are robust against domain shift whilst delivering superior performance in each specific domain. Our focus is on three innovative distributed fiber-optic sensing applications: leak detection, shape sensing for minimally-invasive interventions, and structural health monitoring of civil infrastructure. More concretely, we will develop and leverage neural networks to build a solution that centers around three AI-enabled innovations: i. intelligent data compression, ii. precise parameter estimation and iii. robust anomaly detection & characterization from the resulting complex FOS data sets.