<p>During the dressing of grinding wheels with a diamond roller, the roller’s surface profile, cylindricity, and topography gradually deteriorate as service time increases. Such degradation in surface quality directly compromises the accuracy of wheel dressing and, consequently, indirectly affects the grinding quality of the workpiece. Conventional quality assessment relies on frequent machine downtime for wheel disassembly and offline inspection, which not only lowers production efficiency but also introduces errors during reinstallation. To overcome these limitations, this study proposes an online monitoring method for evaluating the performance degradation of diamond rollers. Acoustic emission (AE) signals generated during the interaction between the roller and the grinding wheel were collected using an AE sensor. These signals were processed through the empirical mode decomposition (EMD) algorithm to extract intrinsic mode functions (IMFs), from which frequency-domain features were analyzed to construct a representative dataset. Three machine learning approaches—support vector machine (SVM), particle swarm optimization-enhanced SVM (PSO-SVM), and long short-term memory (LSTM) networks—were then applied for data analysis. Their performance was systematically evaluated and compared, demonstrating the feasibility of online monitoring of diamond roller performance degradation.</p>