Evaluating 1D CNN-Based Sensor Informatics for Run-Level Surface Roughness Prediction from Dynamometer Signals in Ti-6Al-4V Turning under Varying Lubrication Conditions

Surface roughness (Ra) is an important quality indicator for Ti-6Al-4V components in aerospace and biomedical applications. Although stylus profilometry provides reference measurements of the finished surface, it does not provide continuous feedback during cutting. Dynamometer signals capture the mechanical response at the tool-workpiece interface, but the predictive value of their learned temporal representations remains insufficiently investigated. The present study evaluates run-level Ra prediction across 54 Ti-6Al-4V turning runs under dry, minimum-quantity lubrication (MQL) and nano-MQL (NMQL) conditions. Four neural configurations, namely machining parameters alone, parameters with dynamic root-mean-square (RMS) features, parameters with a one-dimensional convolutional neural network (1D CNN), and parameters with both RMS and CNN features, are compared with parameter-only tree models using run-disjoint evaluation. Across five repeated five-fold partitions, the hybrid model achieves mean MAE and RMSE values of 0.292 and 0.388 µm, respectively, without improving pooled RMSE over parameter-only gradient boosting or the RMS-based neural comparator. Its strongest temporal-branch benefit occurs in the nine-run MQL subset, with mean R 2 = 0.829 and RMSE of 0.178 µm, representing RMSE reductions of 44.6% and 26.1% relative to gradient boosting and parameters plus RMS. Gradient boosting performs best under dry conditions, whereas parameters plus RMS performs best under NMQL. These results support the feasibility of run-level prediction while showing that learned temporal features do not provide a uniform advantage. By integrating machining parameters with learned representations of densely sampled dynamometer signals, the framework provides a methodological foundation for sensor informatics and future machining big-data analytics that can be extended to other machining processes.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-18
DOI
https://doi.org/10.21203/rs.3.rs-11599759/v1
Primary Topic
Advanced machining processes and optimization
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Evaluating 1D CNN-Based Sensor Informatics for Run-Level Surface Roughness Prediction from Dynamometer Signals in Ti-6Al-4V Turning under Varying Lubrication Conditions

Rajat Sen, Abhishek Mandal, Lidiya Priyadarsini K
2 links
Research Square
Advanced machining processes and optimization
preprint

Evaluating 1D CNN-Based Sensor Informatics for Run-Level Surface Roughness Prediction from Dynamometer Signals in Ti-6Al-4V Turning under Varying Lubrication Conditions

Rajat Sen, Abhishek Mandal, Lidiya Priyadarsini K
preprint en

Abstract

Surface roughness (Ra) is an important quality indicator for Ti-6Al-4V components in aerospace and biomedical applications. Although stylus profilometry provides reference measurements of the finished surface, it does not provide continuous feedback during cutting. Dynamometer signals capture the mechanical response at the tool-workpiece interface, but the predictive value of their learned temporal representations remains insufficiently investigated. The present study evaluates run-level Ra prediction across 54 Ti-6Al-4V turning runs under dry, minimum-quantity lubrication (MQL) and nano-MQL (NMQL) conditions. Four neural configurations, namely machining parameters alone, parameters with dynamic root-mean-square (RMS) features, parameters with a one-dimensional convolutional neural network (1D CNN), and parameters with both RMS and CNN features, are compared with parameter-only tree models using run-disjoint evaluation. Across five repeated five-fold partitions, the hybrid model achieves mean MAE and RMSE values of 0.292 and 0.388 µm, respectively, without improving pooled RMSE over parameter-only gradient boosting or the RMS-based neural comparator. Its strongest temporal-branch benefit occurs in the nine-run MQL subset, with mean R 2 = 0.829 and RMSE of 0.178 µm, representing RMSE reductions of 44.6% and 26.1% relative to gradient boosting and parameters plus RMS. Gradient boosting performs best under dry conditions, whereas parameters plus RMS performs best under NMQL. These results support the feasibility of run-level prediction while showing that learned temporal features do not provide a uniform advantage. By integrating machining parameters with learned representations of densely sampled dynamometer signals, the framework provides a methodological foundation for sensor informatics and future machining big-data analytics that can be extended to other machining processes.

Research Square
Jadavpur University (IN)
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