Abstract: Developing early-warning sensor-based maintenance systems for ageing railway infrastructure, such as masonry arch bridges, can be a challenging task due to the difficulty of identifying degradation/damage as the source of small, gradual changes in sensor data, as opposed to other environmental and loading effects. This paper offers a new method of applying statistical modelling and machine learning to enhance the interpretation of fibre optic sensing data, and, therefore, improve deterioration monitoring of railway infrastructure. Dynamic strain and temperature monitoring data between 2016 and 2019 from a fibre Bragg grating (FBG) network installed in a Victorian railway viaduct are first presented. The statistical shape analysis adopted in this study is modified to track changes in the shape of FBG signals directly linked to train speed and dynamic strain amplitudes. The method is complemented by a support vector machine, which is trained to identify different classes of trains. After distinguishing train types, dynamic strain was found to be clearly correlated to temperature, verifying previous findings. No correlation with train speed was observed. The integrated system is then able to compensate for changes in the structural performance due to variations in train loading and ambient temperature, and identify changes in dynamic deformation caused by degradation, in an order comparable to the signal noise (± 2 micro-strain). As a result, the new procedure is shown to be capable of detecting small magnitudes of local degradation well before this degradation manifests itself in typical global measures of response.