Online Tip Damage Diagnosis of Atomic Force Microscope Based on Statistical Pattern Recognition

Purpose: Atomic force microscope (AFM) is a key component of vibration-assisted tip-based nanomachining equipment. Its tip radius determines the machining accuracy and surface smoothness of nanostructures. During the process, the AFM tip directly contacts and relatively slides to the workpiece surface, and tip wear will be caused by the action between the tip and the cutting chip. Meanwhile, the cutting force, vibration amplitude, and cutting temperature will rise, thus affecting the accuracy and surface quality of the machining pattern. Therefore, the research on the online diagnosis of AFM tip damage is of great significance. Methods: An online diagnosis algorithm framework of the AFM tip damage based on statistical pattern recognition (SPR) technology can be used for real-time monitoring of nanomachining tool status. The voltage signal of lateral force generated in the cutting process was collected, and its feature variables were extracted and mapped into the feature space. Further, two models including Incremental Adaptive Support Vector Machine (IASVM) and Gaussian Mixture Model (GMM) were established, and they were used to process the time series feature data of processing force, to achieve real-time diagnosis of tip states. Results: According to the results obtained, IASVM has a strong robustness and generalization ability. For moving window data with 2000 sample points, its average recognition time in ten experiments is 0.180 s, and the average accuracy is 97.63%. On the other hand, as an unsupervised learning method, GMM’s significant feature is its short recognition time, only 0.06 s for 2000 samples recognition. Conclusions: The time series data of nano-processing forces represented by vibration signals provide important clues for AFM tip health monitoring. The SPR method has significant advantages in computational time and accuracy, as well as physical consistency, making it a diagnostic tool for real-time detection of AFM tip states.