The reliable and economical operation of polymer electrolyte membrane (PEM) fuel cells in practical applications requires effective condition monitoring, particularly with respect to internal phenomena spatially distributed along the channel. Since measurements of internal cell states are difficult and costly in practice, model-based approaches are required to access these along-the-channel quantities. Such models must be both accurate and real-time capable, while capturing the nonlinear, spatially distributed thermodynamic and electrochemical processes responsible for local degradation. To efficiently assess local aging stressors such as oxygen partial pressure and temperature, this work employs a network of locally linearized models to approximate the nonlinear dynamics of computationally expensive fuel cell models. This approach preserves essential nonlinear behavior while retaining favorable properties of linear models, such as computational efficiency. However, conventional local model networks typically rely on expert knowledge for the selection of local operating points and membership functions, which limits automation and applicability to new fuel cell systems. This work presents a systematic data-driven design methodology for local model networks using commonly available fuel cell polarization curve data. Local operating points and membership functions are identified via fuzzy c-means clustering, enabling the model structure to represent the nonlinear operating behavior of the fuel cell without manual tuning. A nonlinear high-fidelity fuel cell model is subsequently used to derive locally linearized models at the operating points obtained from the clustering process. The proposed approach is validated in a simulation study using artificial measurement data from a simulated reality. The resulting clustering-based local model network is tested in a dynamic operating scenario and compared to a knowledge-based local model network, with the nonlinear fuel cell model serving as a benchmark. The results show improved reproduction of spatially distributed aging stressors while maintaining computational efficiency, enhancing automated and system-specific condition monitoring for PEM fuel cells.