A Tool Wear Prediction Method Based on Convolutional Temporal Networks with Attention Mechanism

By using an attention-based convolutional temporal network and audio feature-based method, a sound barometer is used to collect sound signals. Combined with an attention mechanism and a long short-term memory module, the problem of high cost and insufficient accuracy in existing tool wear prediction is solved, and high-precision tool wear monitoring is achieved. This method is suitable for high-end equipment manufacturing in the aerospace and nuclear power fields.

CN116690312BActive Publication Date: 2026-07-17HANGZHOU DIANZI UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
HANGZHOU DIANZI UNIV
Filing Date
2023-07-27
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Existing methods for predicting tool wear suffer from high instrument costs, complex installation, and insufficient prediction accuracy. In particular, tool wear has a significant impact on surface quality and production efficiency in the manufacturing of high-end equipment in the aerospace and nuclear power fields.

Method used

A tool wear prediction method based on attention convolutional temporal networks and audio features is adopted. The method uses a low-cost sound barometer to collect the sound signal during the cutting process, extracts the sound signal features by frame segmentation, and combines the attention mechanism and long short-term memory module to build a wear amount identification model, so as to achieve adaptive feature weight allocation and efficient prediction.

Benefits of technology

It achieves an average absolute error of 0.3% in tool wear prediction, reducing instrument costs and improving prediction accuracy, and is suitable for tool wear monitoring in high-end equipment manufacturing.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116690312B_ABST
    Figure CN116690312B_ABST
Patent Text Reader

Abstract

This invention discloses a tool wear prediction method based on an attention-based convolutional temporal network. The method comprises: 1. Placing three sound barometers around the tool; establishing a dataset based on the acoustic signals. 2. Constructing and training a wear amount identification model. The wear amount identification model includes a shallow convolutional network module, an attention mechanism module, a long short-term memory (LSTM) network module, and a deep fully connected module. 3. Placing three sound barometers around the tool under test; the obtained three-dimensional audio signals, after feature extraction, are input into the wear amount identification model to obtain the wear amount of the tool under test. This invention reduces data redundancy by framing the original signal before extracting the acoustic signal features of the cutting process; it uses an attention mechanism module to perform attention allocation on the features of the shallow network to achieve adaptive feature weight allocation; considering the monotonicity of tool wear in the time dimension, a bidirectional LSM module is introduced, resulting in an average absolute error of 0.3% for tool wear prediction.
Need to check novelty before this filing date? Find Prior Art