A feature selection method based on an improved fruit fly optimization algorithm in the process of numerical control milling

Automatic control is the key to improved production quality and efficiency of numerical control milling operations. Because the milling cutter is the most important tool in milling operations, the automatic monitoring of the tool wear state is of great significance. This work establishes a set of time domain and time-frequency domain features based on measurements of the cutting force for a computer numerical control milling machine and develops a method incorporating the Fisher criterion in an improved fruit fly optimization algorithm for selecting features most indicative of the tool wear state. A back propagation neural network was employed to test the effectiveness of the proposed feature selection method. Experimental comparisons with three other feature selection methods demonstrate that the proposed improved fruit fly optimization algorithm offers the advantages of the selection of a small number of significant features, easy implementation, precise optimization, rapid training, and good back propagation network performance. The proposed method has great potential for facilitating the practical monitoring of the milling tool wear state.