MFAFENet: A Multi-Sensor Collaborative and Multi-Scale Feature Information Adaptive Fusion Network for Spindle...

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Accurate classification of spindle rotational errors is crucial for CNC machine tool precision and reliability.
• Existing methods struggle with subtle feature differences and electromechanical interference in vibration signals.

Purpose of the Study:

• To propose a novel deep learning model, MFAFENet, for enhanced spindle rotational error classification.
• To leverage multi-sensor collaboration and adaptive multi-scale feature fusion for improved accuracy.

Main Methods:

• Vibration signals from three positions were converted to time-frequency representations using Short-time Fourier Transform.
• MFAFENet adaptively fuses multi-scale features via a branch attention mechanism and recalibrates features using channel attention.
• Network parameters were optimized using the cross-entropy loss function.

Main Results:

• MFAFENet achieved an average test accuracy of 93.37% on a spindle reliability test bench.
• The model outperformed comparative methods, demonstrating the effectiveness of adaptive fusion and multi-sensor collaboration.
• Multi-sensor fusion improved accuracy by 7.23% compared to single-sensor setups.

Conclusions:

• MFAFENet provides an effective end-to-end solution for mapping vibration signals to rotational errors.
• The proposed method offers a promising approach for high-precision spindle condition monitoring and enhancing machining reliability.