Milling Tool Wear Prediction Method Based on Deep Learning Under Variable Working Conditions

Mingwei Wang,Jing Gao,Jingtao Zhou,Ziqiu Li,Enming Li

doi:10.1109/access.2020.3010378

Abstract

Tool wear prediction is essential to ensure part quality and machining efficiency. Tool wear is affected by factors such as the material, structure, process, and processing time of the parts. Tool wear under the variable working conditions and the above factors show a complex coupling and timing correlation, which makes it challenging to predict tool wear under variable working conditions. This article aims to resolve this issue. First, we establish a unified representation of working condition factors. The stacked autoencoder (SAE) model adaptively extracts tool wear features from the machining signal. The extracted wear features and respective working conditions then combine into a working condition feature sequence for predicting tool wear. Finally, the advantages of the long short-term memory (LSTM) model to solve memory accumulation effects learn the regular wear pattern of the working condition feature sequence to realize the prediction of the tool wear. An experiment illustrates the effectiveness of the proposed method.

Highlights

The tool is a direct executor of machining operation, and its wear prediction is of great significance for ensuring the quality of parts, improving efficiency, and reducing costs
MAIN PROCESS The milling tool wears prediction method under variable working conditions we propose is based on the stacked autoencoder (SAE) network and long short-term memory (LSTM) network
Our innovative method based on deep learning uses engineering requirements of the tool wear prediction to deal with the complicated relationship between working condition factors and tool wear; this method extracts working condition characteristics

Summary

INTRODUCTION

The tool is a direct executor of machining operation, and its wear prediction is of great significance for ensuring the quality of parts, improving efficiency, and reducing costs. M. Wang et al.: Milling Tool Wear Prediction Method Based on Deep Learning Under Variable Working Conditions (parts materials, parts structure, processing technology, processing time), and machining signals (force signal, vibration signal). The research group proposed to use the unique advantages of the LSTM model to solve the problem of complex correlation and memory accumulation effect, and established the prediction model of remaining tool life under variable working conditions [24]. This paper attempts to establish the LSTM model for tool wear prediction, to solve the problem of tool wear related to the historical data of tool use and time sequence correlation. MAIN PROCESS The milling tool wears prediction method under variable working conditions we propose is based on the SAE network and LSTM network.

TOOL WEAR FEATURE EXTRACTION

ESTABLISHMENT OF TOOL WEAR PREDICTION MODEL

Findings

CONCLUSION