Condition monitoring of tool performance using a machine learning-based on-machine vision system during face milling of Inconel 718

Banda, Tiyamike (2024) Condition monitoring of tool performance using a machine learning-based on-machine vision system during face milling of Inconel 718. PhD thesis, University of Nottingham.

PDF (Thesis - as examined) - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
Available under Licence Creative Commons Attribution.
Download (40MB) | Preview


The superior properties of Inconel 718 necessitate its use in manufacturing more than 50% of aircraft engine structural components, including high-pressure compressor blades, casings, and discs. However, literature attributed the synergistic impact of these properties and process parameters as the primary cause of wear complexity, notably affecting the performance of PVD-coated carbide inserts during CNC milling of Inconel 718. Features stemming from the wear complexity include uncontrolled wear mechanisms, failure modes, and a rapid flank wear rate, serving as significant indicators of sub-optimal cutting conditions. In trying to diagnose tool wear, previous Tool Condition Monitoring (TCM) techniques could not decipher, explore, and synthesise the diverse features essential for the predictive control of tool performance in challenging CNC machining conditions. Therefore, the successful implementation of advanced feature engineering and Machine Learning (ML) models in Machine Vision-based TCM (MVTCM) offers a proactive approach in predicting and controlling the performance of PVD-coated carbide tools in challenging CNC machining domains.

The hypothesis of this study encompassed three aspects. The first aspect focused on the study of tool wear complexity by characterizing the dominant wear mechanisms, failure modes, and flank wear depth (VB) during face milling of Inconel 718. These features were correlated with the process parameters to establish a coherent tool wear dataset for training the feature engineering and ML models. The second aspect involved the development of feature engineering and ML models, including the multi-sectional singular value decomposition (SVD), a YOLOv3 Tool Wear Detection Model (YOLOv3-TWDM), a multi-layer perceptron neural network (MLPNN), and an inductive-reasoning algorithm. The final aspect pertained to the development of a volatile MV-TCM system’s design, which was integrated with feature engineering and ML techniques to create an enhanced ML-based MV-TCM system. The system was vigorously validated by conducting an online experiment, where the predicted were compared with the actual wear measurements. Furthermore, the inductive reasoning algorithm was devised to regulate process parameters for in-process control of flank wear evolution.

The findings demonstrate that the Diverse Feature Synthesis Vector devised in this research was superior in representing the complex flank wear morphology as compared to some data reported by relevant literature, where geometric and fractal features were used to predict VB progression online. In addition, the ML-based MV-TCM system successfully utilized the DFSV to predict and control VB rate during face milling of Inconel 718. The system achieved higher predictive efficiency than image processing-based MV-TCM systems applied in the previous studies, with an offline validation RMSE of 45.5µm, R2 of 96.52%, and MAPE of 2.36%, as well as an online validation RMSE of 29.09µm, R2 of 97%, and MAPE of 3.52%. Additionally, the system employed a multi-stage optimization strategy that regulated process parameters at different VB levels to minimize the magnitudes of flank wear and chipping. This strategy extended tool life by 63.63% (relative to the conventional method) and 56.52% (relative to the GKRR soft-computing technique). Therefore, this research demonstrates the significance of applying ML-based MV-TCM system for predictive control of tool wear evolution during CNC milling of Inconel 718.

Item Type: Thesis (University of Nottingham only) (PhD)
Supervisors: Lim, Chin Seong
Farid, Ali Akhavan
Keywords: Inconel 718, tool wear, machine learning (ML), tool condition monitoring (TCM), CNC machining
Subjects: T Technology > TJ Mechanical engineering and machinery
Faculties/Schools: University of Nottingham, Malaysia > Faculty of Science and Engineering — Engineering > Department of Mechanical, Materials and Manufacturing Engineering
Item ID: 77371
Depositing User: Banda, Tiyamike
Date Deposited: 11 Mar 2024 03:54
Last Modified: 11 Mar 2024 03:54

Actions (Archive Staff Only)

Edit View Edit View