⚡ electrical engineering

Predicting the process parameters for torque and thrust force in wood plastic composite drilling using machine learning algorithms

This study utilizes Taguchi-based drilling experiments and machine learning algorithms to demonstrate that feed rate is the dominant factor affecting thrust force and torque in Wood Plastic Composite drilling, with the Random Forest model emerging as the most accurate predictor for optimizing process parameters to minimize these forces.

Original authors: Palanikumar Kayaroganam, Sundarapandiyan Murugesan, Sreedevi B, Latha Balasubramanian

Published 2026-09-04

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Palanikumar Kayaroganam, Sundarapandiyan Murugesan, Sreedevi B, Latha Balasubramanian

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Wood has served humanity for millennia, valued for its strength and versatility in building everything from simple shelters to complex machinery. However, as forests face pressure from deforestation and the demand for sustainable materials grows, engineers have turned to a modern alternative: wood plastic composite. This material is a blend of wood fibers, such as sawdust or bamboo flour, mixed with melted plastic and various additives to create a durable, weather-resistant product. It is now a common sight in decking, fencing, and furniture, offering the look of wood without the same susceptibility to rot or insects. Yet, working with this composite presents a unique challenge. Because it is a mixture of rigid fibers and flexible plastic, it behaves differently than solid wood or metal when cut. Drilling holes into it—a necessary step for assembly—can easily cause damage. If the drilling process is not carefully controlled, the material can crack, the fibers can tear out, or the layers can separate, ruining the final product's strength and appearance. The key to avoiding this damage lies in understanding the forces at play: the downward pressure the drill exerts and the twisting force required to turn it.

In a recent study, researchers set out to master these forces by teaching a computer to predict exactly how much pressure and twisting will occur during the drilling process. The team, based at engineering colleges in India, focused on a specific type of wood plastic composite made from wood fibers and thermoplastic. They wanted to find the perfect settings for a drill to minimize the stress on the material. To do this, they treated the drilling process like a series of controlled experiments. They used a computer-controlled machine to drill holes into panels of the composite material, systematically changing three main variables: how fast the drill spun, how quickly it moved forward into the material, and the size of the drill bit itself. They tested these variables across a wide range of values, from slow speeds to fast ones, and from small drill bits to larger ones. For every combination of settings, they measured the exact downward force and the twisting torque generated, repeating each test three times to ensure the results were reliable.

The researchers then analyzed these measurements to see which variable mattered most. They discovered that the speed at which the drill bit moved forward was the single most important factor, accounting for more than sixty percent of the changes in both downward force and twisting torque. The size of the drill bit was the next most significant factor, while the speed of the spinning drill had a smaller, though still noticeable, effect. Generally, they found that moving the drill faster into the material increased the forces, while spinning the drill faster actually helped reduce them. This is because a faster spin generates heat that softens the plastic component of the composite, making it easier to cut through. By combining these findings with a specific type of computer algorithm known as a random forest, the team built a model that could predict the forces with remarkable accuracy. This model learned from the experimental data to recognize patterns, allowing it to estimate the forces for any new set of drilling conditions without needing to run a physical test first.

The results showed that this computer-based approach was highly effective. The model successfully predicted the downward force and twisting torque with an accuracy that exceeded ninety-nine percent when compared to the actual physical measurements. This level of precision means that manufacturers could use such a system to determine the best drilling settings before they even touch the material, ensuring that the holes are clean and the composite panels remain undamaged. The study confirmed that by carefully selecting the drill speed, feed rate, and bit size, it is possible to drill wood plastic composites without causing the delamination or surface defects that often plague the industry. While the researchers noted that their work was limited to specific drill bits and dry drilling conditions, the success of their machine learning approach suggests a clear path forward. It demonstrates that by letting algorithms learn from data, engineers can optimize manufacturing processes for these complex, eco-friendly materials, ensuring they perform as well in the real world as they do in the laboratory.

Technical Summary: Predicting Process Parameters for Torque and Thrust Force in Wood Plastic Composite Drilling Using Machine Learning Algorithms

Problem Statement
Wood Plastic Composites (WPC) have become increasingly prevalent in domestic and industrial applications, including automotive interiors, flooring, and furniture, due to their durability, eco-friendliness, and cost-effectiveness. However, machining these materials, particularly drilling, presents significant challenges. The anisotropic and layered nature of WPC leads to defects such as delamination, fiber pull-out, micro-cracking, and poor surface finish when excessive thrust force ( $F_z$ ) and torque ( $F_t$ ) are generated. These defects compromise the structural integrity and aesthetic quality of the final product. While traditional statistical methods like Response Surface Methodology (RSM) exist, they often lack the flexibility to generalize across different materials or process conditions. There is a recognized need for accurate, data-driven models to predict $F_z$ and $F_t$ to optimize process parameters, minimize defects, and enhance production efficiency in WPC manufacturing.

Methodology
The study employed a systematic approach combining experimental design, statistical analysis, and machine learning (ML) modeling:

Experimental Design: Drilling experiments were conducted on WPC panels (composed of 40% wood fibers, 54% thermoplastic, and 6% additives) using a CNC vertical machining center. The study utilized Taguchi's $L_{27}$ orthogonal array to investigate the effects of three input parameters: Spindle Speed ( $N$ ), Feed Rate ( $f$ ), and Drill Diameter ( $d$ ).
- Ranges: $N$ (1000–3000 rpm), $f$ (75–225 mm/min), $d$ (6, 8, and 10 mm).
- Tooling: Brad center point drill bits (carbon steel with aluminum finish) were used.
- Measurement: A Kistler multi-component dynamometer (Model 9277B) measured $F_z$ and $F_t$ . Each of the 27 experimental runs was repeated three times, resulting in an 81-point dataset.
Statistical Analysis: Analysis of Variance (ANOVA) was performed to determine the significance of input parameters and their interactions on the output responses.
Machine Learning Modeling: Three ensemble learning algorithms were developed and compared to predict $F_z$ and $F_t$ :
- Extra Trees Regressor: Utilized for its ability to handle noisy data through extreme randomization.
- Bagging with Decision Trees: Employed to reduce variance by averaging predictions from bootstrapped subsets.
- Random Forest Regressor: Used for its robustness in capturing non-linear correlations through random feature selection at node splits.
Evaluation: The dataset was split into 70% training and 30% testing. Model performance was evaluated using six metrics: Mean Absolute Percentage Error (MAPE), Mean Absolute Difference (MAD), Root Mean Square Error (RMSE), Standard Error (SE), Mean Squared Error (MSE), and the Coefficient of Determination ( $R^2$ ).

Key Results

Process Parameter Influence: ANOVA results identified Feed Rate ( $f$ ) as the most dominant factor influencing both thrust force and torque.
- For Thrust Force ( $F_z$ ): Feed Rate contributed 65.16%, Drill Diameter 27.33%, and Spindle Speed 4.90%.
- For Torque ( $F_t$ ): Feed Rate contributed 64.41%, Drill Diameter 20.39%, and Spindle Speed 12.84%.
- Optimal conditions for minimizing forces were found to be 3000 rpm, 75 mm/min, and a 6 mm drill diameter. Conversely, the highest forces occurred at 1000 rpm, 225 mm/min, and 10 mm diameter.
Model Performance: All three ML models demonstrated high predictive capability, with $R^2$ values exceeding 0.988. However, the Random Forest model consistently outperformed the others:
- Thrust Force ( $F_z$ ): Random Forest achieved an $R^2$ of 0.998, MAPE of 1.601, and RMSE of 0.275.
- Torque ( $F_t$ ): Random Forest achieved an $R^2$ of 0.997, MAPE of 0.718, and RMSE of 0.1406.
- The Random Forest model also exhibited the most normally distributed prediction errors, as validated by Anderson-Darling normality plots, indicating higher reliability compared to Bagging and Extra Trees.

Significance and Claims
The authors assert that this study provides a novel approach for evaluating and identifying suitable process parameters to minimize thrust force and torque during the drilling of WPC composites. By demonstrating that machine learning models, specifically Random Forest, can accurately predict machining forces based on input parameters, the research offers a tool to improve production efficiency and optimize resources. The study claims that developing such accurate forecasting models is crucial for enhancing industrial productivity and ensuring the quality of mechanical joints in WPC applications. The work establishes a foundation for using data-driven methods to replace or supplement traditional empirical modeling in the machining of composite materials.

Future Directions
The paper suggests that future work could expand the scope by estimating additional quality indicators (such as delamination factor, surface roughness, circularity, and cylindricity errors), testing different tool materials and geometries, exploring Minimum Quantity Lubrication (MQL) to reduce temperature and wear, and comparing traditional drilling with Alternative Water Jet Machining (AWJM).

More like this