To address the significant application value of ultrasonic rolling surface strengthening technology in the aerospace field, a multi-sensor fusion-based processing monitoring system was independently constructed using existing robotic ultrasonic rolling equipment. This system collected and analyzed physical signals in real-time during processing. The relationship between physical signals and processing states was explored, leading to the development of a multi-sensor fusion signal processing and feature extraction method, which resolved issues with signal synchronization. Time-frequency domain analysis methods were employed to extract signal features, followed by feature selection and dimensionality reduction. Machine learning models, including support vector machine (SVM), random forest (RF), and multi-layer perceptron (MLP), were utilized for intelligent prediction of surface integrity in ultrasonic rolling machining. Hyperparameters of these models were optimized using particle swarm optimization (PSO) and grid search to enhance prediction accuracy. Results show that the interval prediction accuracy for surface roughness is 91.4%, while the average relative errors for surface hardness and residual stress are 10.82% and 11.99%, respectively.