During shield tunneling, cutterhead monitoring data are often contaminated by outliers and noise owing to frequent operation mode switching, complex geological conditions, and interference from multiple coupled sensors. These factors considerably compromise the accuracy of cutterhead condition assessment and service life prediction. To improve data quality, a cutterhead monitoring data-cleaning method integrating dynamic segmentation modeling with multidimensional correlation analysis is proposed. First, monitoring data are coarsely segmented by combining thrust and torque conditions, categorizing the original time-series data into three operation modes—shutdown, segment assembly, and normal tunneling—thereby constructing a labeled baseline dataset. Subsequently, multiscale adaptive segmentation is performed using the comprehensive geological index and tunneling energy index as criteria. The segmentation is optimized using the Fisher discriminant method to extract key features and characterize temporal variation patterns across various operating conditions, with segmentation results and correlation information stored in a correlation cube. Outlier detection and correction are then conducted based on segmentation and correlation information. Specifically, a linear regression model is established, and the prediction interval is derived to accurately identify anomalous data. The positions of the detected anomalies are indexed and stored in the correlation cube. Subsequently, a graph-theoretic approach is introduced to perform weighted correction of outliers, ensuring repaired data maintain consistency and rationality in multidimensional correlations and temporal continuity. Comparative experiments across various data scales, dimensionalities, and anomaly ratios demonstrate that the proposed method significantly outperforms traditional statistical methods, clustering approaches, and classical interpolation strategies in precision, recall, root mean square error, and coefficient of determination. The results confirm high outlier detection accuracy and robust data repair performance for complex, high-dimensional, and time-varying monitoring data, demonstrating practical value in shield tunneling applications.