This paper deals with the development of an online monitoring system based on feature-level sensor fusion and its application to OD plunge grinding. Different sensors are used to measure acoustic emission, spindle power, and workpiece vibration signals, which are used to monitor three of the most common faults in grinding—workpiece burn, chatter, and wheel wear. Although a number of methods have been reported in recent literature for monitoring these faults, they have not found widespread application in industry as no single method or feature has been shown to be successful for all setups and for all wheel-workpiece combinations. This paper proposes a systematic approach, which allows the development and deployment of process-monitoring systems via automated sensor and feature selection combined with parameter-free model training, both of which are especially crucial for implementation in industry. The proposed algorithm makes use of “sparsity-promoting” penalty terms to encourage sensor and feature selection while the “hyperparameters” of the algorithm are tuned using an approximation of the leave-one-out error. Experimental results obtained for monitoring burn, chatter, and wheel wear from a plunge grinding test bed show the effectiveness of the proposed method.