AI-Based Tool-Condition Classification and Tool-Replacement Warning Using Internal MCT Control Signals in Cylinder-Block Machining

Determining an appropriate cutting-tool replacement point in a mass-production machining environment is difficult when direct tool-wear measurement is unavailable and replacement decisions depend largely on operator experience or fixed preventive replacement intervals. This study develops an AI-based tool-condition classification and tool-replacement warning system using internal control signals acquired from a machining center (MCT) without additional external sensors. Rather than predicting a continuous remaining useful life (RUL) value, the proposed approach classifies the current tool condition as normal (Class 0) or requiring replacement (Class 1) at the end of each machining cycle. Raw machine data were acquired at 1 Hz through MT-LINKi/PoC from MCT #19, filtered to retain actual cutting intervals, transformed into 60 s features, and used to train and evaluate MLP, Random Forest, RNN, and LSTM models. Data partitioning was performed at the tool-life-cycle level using 48% training, 12% validation, and 40% test data to reduce information leakage between correlated samples from the same tool-life cycle. Because the dataset was highly imbalanced, model selection emphasized Precision, Recall, F1-score, and other imbalance-sensitive metrics rather than Accuracy alone. Among the four models, the MLP achieved the highest F1-score (0.98477), followed by the LSTM (0.64516), Random Forest (0.61134), and RNN (0.48077). For practical decision support, a replacement warning was generated only when the predicted replacement probability exceeded a validation-derived threshold for three consecutive machining cycles. In the evaluated MLP case, the first sustained trigger occurred at Cycle 251 and, after seven additional quality-verified cycles, the final usable replacement point was determined to be 258 machining cycles. Compared with the conventional preventive replacement interval of approximately 200 cycles, this corresponds to a 29.0% increase in usable cycles per tool and an estimated reduction of approximately 22.5% in tool consumption for an equivalent production volume. The proposed system functions as an operator-centered decision-support layer and does not directly modify CNC/MCT machining control logic.

Authors

Institutions

Publication Details

Journal
Lubricants
Published
2026-11-19
DOI
https://doi.org/10.3390/lubricants14110442
Primary Topic
Advanced machining processes and optimization
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

AI-Based Tool-Condition Classification and Tool-Replacement Warning Using Internal MCT Control Signals in Cylinder-Block Machining

Dae-Kyu Kwon, Sang‐Woo Kang, Jae-Hak Lee
Lubricants
Advanced machining processes and optimization
article

AI-Based Tool-Condition Classification and Tool-Replacement Warning Using Internal MCT Control Signals in Cylinder-Block Machining

Dae-Kyu Kwon, Sang‐Woo Kang, Jae-Hak Lee
article en

Abstract

Determining an appropriate cutting-tool replacement point in a mass-production machining environment is difficult when direct tool-wear measurement is unavailable and replacement decisions depend largely on operator experience or fixed preventive replacement intervals. This study develops an AI-based tool-condition classification and tool-replacement warning system using internal control signals acquired from a machining center (MCT) without additional external sensors. Rather than predicting a continuous remaining useful life (RUL) value, the proposed approach classifies the current tool condition as normal (Class 0) or requiring replacement (Class 1) at the end of each machining cycle. Raw machine data were acquired at 1 Hz through MT-LINKi/PoC from MCT #19, filtered to retain actual cutting intervals, transformed into 60 s features, and used to train and evaluate MLP, Random Forest, RNN, and LSTM models. Data partitioning was performed at the tool-life-cycle level using 48% training, 12% validation, and 40% test data to reduce information leakage between correlated samples from the same tool-life cycle. Because the dataset was highly imbalanced, model selection emphasized Precision, Recall, F1-score, and other imbalance-sensitive metrics rather than Accuracy alone. Among the four models, the MLP achieved the highest F1-score (0.98477), followed by the LSTM (0.64516), Random Forest (0.61134), and RNN (0.48077). For practical decision support, a replacement warning was generated only when the predicted replacement probability exceeded a validation-derived threshold for three consecutive machining cycles. In the evaluated MLP case, the first sustained trigger occurred at Cycle 251 and, after seven additional quality-verified cycles, the final usable replacement point was determined to be 258 machining cycles. Compared with the conventional preventive replacement interval of approximately 200 cycles, this corresponds to a 29.0% increase in usable cycles per tool and an estimated reduction of approximately 22.5% in tool consumption for an equivalent production volume. The proposed system functions as an operator-centered decision-support layer and does not directly modify CNC/MCT machining control logic.

LubricantsVol. 14(11)
Changshin University (KR)
Advanced machining processes and optimization
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon