Surface Roughness and Tool Wear Prediction in Machining of Nimonic 90 Through Machine Learning Under Various Lubrication Conditions

Technical Summary: Surface Roughness and Tool Wear Prediction in Machining of Nimonic 90 Through Machine Learning Under Various Lubrication Conditions

Problem Statement
Nimonic 90, a nickel-based superalloy essential for aerospace and gas turbine applications, presents significant machining challenges due to its low thermal conductivity, rapid work hardening, and high chemical affinity for tool materials. These properties classify it as a "difficult-to-cut" material, leading to excessive tool wear, thermal softening, and poor surface integrity. Traditional experimental optimization is cost-prohibitive due to the high expense of superalloys and the limitations of small sample sizes in capturing complex, non-linear machining dynamics. Furthermore, conventional lubrication methods often fail to manage the intense heat generated at the tool-chip interface, while existing data-driven models struggle with the scarcity of experimental data typical in superalloy machining research.

Methodology
The study employed a multi-faceted approach combining experimental machining, advanced lubrication characterization, data augmentation, and machine learning (ML) modeling.

• Experimental Design: Turning experiments were conducted on Nimonic 90 using a Taguchi L19 orthogonal array. Three lubrication environments were compared: traditional oil-based Minimum Quantity Lubrication (MQL), Ionic Liquid (IL)-assisted MQL using [1-Butyl-3-Methyl Imidazolium tetrafluoroborate], and Cryogenic Liquid Nitrogen (LN₂) cooling. Cutting speeds (410, 620, 830 m/min) and feed rates (0.1, 0.5 mm/rev) were varied with a constant depth of cut (0.5 mm).
• Characterization: Lubricants were analyzed for pH, zeta potential, and contact angles. Machining responses included Surface Roughness ($R_a$), Flank Wear ($V_b$), and cutting temperature. Chip morphology, microstructure, and tool wear mechanisms were examined via SEM, optical microscopy, and microhardness testing.
• Data Augmentation: To address the limitation of a small initial dataset (18 samples), the authors utilized a Gaussian Process Regression (GPR) framework to probabilistically augment the data. While the abstract mentions enriching the sample size to 27, Section 3.2.1 explicitly states that the final augmented dataset utilized for statistical representation consisted of the original 18 samples, bridging the gap between discrete experiments and continuous predictive modeling.
• Machine Learning Modeling: Four algorithms—Linear Regression (LR), Support Vector Machines (SVM), Random Forest (RF), and Extreme Gradient Boosting (XGBoost)—were trained and evaluated using 5-fold cross-validation. The models aimed to predict $R_a$ and $V_b$ based on the input parameters.
• Optimization: A Pareto frontier analysis was conducted to identify optimal machining regimes balancing productivity and surface integrity.

Key Results

• Lubrication Performance: Cryogenic LN₂ machining yielded the superior results, achieving the lowest surface roughness ($R_a$ = 0.45–0.82 µm) and the lowest tool wear. Specifically, LN₂ reduced flank wear by 46% compared to oil-based machining and reduced the total wear volume by 73% compared to oil-based machining. Ionic liquid-assisted MQL demonstrated improved wettability (contact angle of 39.6° vs. 51.8° for oil) and formed protective tribo-films, resulting in intermediate performance between oil and cryogenic cooling.
• Microstructural and Morphological Changes: LN₂ cooling resulted in the finest grain structure (8–12 µm) compared to IL (15–20 µm) and oil (25–30 µm), indicating enhanced microstructural homogeneity. Chip morphology analysis revealed that LN₂ minimized saw-tooth formation and abrasive wear, producing smoother, more continuous chips with reduced thickness.
• Machine Learning Performance: The XGBoost model outperformed all other algorithms, achieving an $R^2$ of 0.948, RMSE of 0.122, and MAE of 0.094 for surface roughness prediction. This significantly surpassed Linear Regression ($R^2$ = 0.627), confirming the highly non-linear nature of the machining process. Feature importance analysis identified Feed Rate as the primary driver for surface roughness (0.68 relative importance) and the Cooling Environment as the dominant factor for tool wear (0.71 relative importance).
• Optimization: The Pareto frontier analysis identified a "Golden Regime" at a cutting speed of 620 m/min, feed rate of 0.1 mm/rev, and LN₂ cooling, balancing sub-micron surface quality ($R_a$ = 0.82 µm) with extended tool life.

Significance and Claims
The paper claims to establish a data-driven paradigm for the precision machining of high-performance alloys. Its primary contributions include:

1. Validation of Cryogenic and Ionic Liquid Strategies: Demonstrating that LN₂ cooling and IL-assisted MQL offer superior tribological and thermal management compared to traditional oil-based MQL for Nimonic 90.
2. Data Augmentation Framework: Successfully applying GPR to overcome the "small data" problem inherent in expensive superalloy experiments, allowing for robust model training without introducing physical inconsistencies.
3. Predictive Modeling: Proving that ensemble tree-based models, specifically XGBoost, are highly effective in capturing the complex thermodynamic interactions of superalloy machining, outperforming linear and other non-linear models.
4. Industrial Applicability: Highlighting the computational efficiency of the developed framework (training time < 0.5 seconds, memory footprint < 20MB), making it suitable for edge computing, real-time process optimization, and integration into Digital Twin systems for autonomous manufacturing in the aerospace sector.

The study concludes that integrating advanced lubrication strategies with machine learning-based predictive modeling provides a viable pathway toward sustainable, autonomous, and economically optimized manufacturing processes for aerospace components.