Modeling method of in-situ identification model for hob wear and in-situ identification method for hob wear

Abstract

The invention discloses a hob wear on-site identification model modeling method and a hob wear on-site identification method, which improves and expands the acquisition method and sample type of the original signal sample database, so that the sample database can accurately reflect thermal deformation and tooth deformation. The influence of the inhomogeneity of the blank machining allowance on the signal-to-noise ratio of the current signal sample, combined with the advantages of artificial feature extraction in excluding low-value information and the advantages of deep learning algorithms in high-quality feature extraction and classification, cooperate with each other to establish the influence on error The weight change has a strong expressive ability and can reflect the classification feature group of different working mode modes to realize the high-quality overall representation of the tool wear state. Based on this, the in-situ recognition model of the hob wear in the parallel recognition framework is established. , to achieve high-precision and robust in-situ recognition of hob wear.

Description

technical field [0001] The invention relates to the field of process detection, in particular to a hob wear on-site recognition model modeling method and a hob wear on-site recognition method. Background technique [0002] The currently developed in-situ recognition methods for tool wear based on current signal characterization can be divided into three categories: (1) Time domain / frequency domain / small packet analysis is performed on the signal samples, and feature quantities related to tool wear are manually extracted. Establish a mathematical recognition model to judge the wear state of the tool; (2) Graphically process the sample signal, input it into the deep learning mathematical model, and use semi-supervised / unsupervised image feature learning and hierarchical feature extraction efficient algorithms to realize Recognition of tool wear state (3) Extract deep features from the original time domain signal, and add a classifier on the top layer to realize the recognition...

AI Technical Summary

Problems solved by technology

If the data samples are not full, the deep learning algorithm will not be able to make an unbiased estimate of the law of the data, and the recognition effect may not be as good as some existing simple algorithms

In addition, if the input samples are too large, the time complexity of the algorithm will increase sharply. In order to ensure the real-time performance of the algorithm, users need to have good parallel programming skills and better and more hardware support.

Images