Early detection of fatigue damage is critical for the safety of aerospace structures, yet current structural health monitoring (SHM) methods struggle to extract robust, real-time insights from high-dimensional sensor data. Fiber Bragg grating (FBG) sensors provide rich broadband spectra but are challenging to process for damage identification during operation. Digital deep learning offers strong predictive capability but is computationally intensive and introduces latency. This study therefore investigates optical neural networks (ONNs) as a fast, hardware-efficient approach for interpreting FBG spectra to estimate fatigue damage. Fatigue experiments were conducted on compact tension specimens instrumented with FBG sensors. Artificial neural networks were developed on dimension-reduced FBG spectra to estimate image-based crack measurements, achieving high performance. While ONNs matched this in simulation but exposed a key limitation: photonic implementations using Mach–Zehnder interferometers suffer optical losses, limiting practical applicability. These findings highlight both promises and limitations of ONNs for aerospace SHM.