Quantitative characterization of crack formation and growth allows for efficient and timely management of structural health. By using real-time strain data and a probabilistic diagnosis process to inform a persistent digital twin of a structure, failure modes can be monitored and addressed with greater confidence by decision makers. It has been shown that a probabilistic approach can be used to diagnose cracks with full-field strain data using digital image correlation [1]; this project aims to perform a comparable diagnosis process by using embedded fiber-optic strain sensors that are more practical in real-world applications. A geometrically complex part was manufactured with embedded fiber-optics and subjected to fatigue loading to initiate and grow a crack. A finite element (FE) model of the part was created and crack growth was simulated under identical loading conditions. Strain data recorded from the FE model were used to train a Gaussian process surrogate model to reduce the computational cost of the diagnosis process. Surrogate model output and measured strain data were compared using uncertainty quantification techniques to estimate cracks lengths at different times throughout the test. Diagnoses resulted in good predictions for crack length up until the crack grew through all the fibers in the test specimen, at which point modeled and measured strain fields showed significant differences. While it was shown that accurate crack length predictions can be made using this process, it would be necessary to create more complex algorithms to monitor and select only strain data relevant to a diagnosis for use beyond the laboratory environment.