TL;DR
This paper introduces an improved Northern Goshawk Optimization (INGO) algorithm to optimize Support Vector Machines (SVMs) for accurate tool wear state identification in machining. By combining advanced feature selection with a novel optimization technique, the INGO-SVM model achieves a remarkable 97.9% accuracy, outperforming existing methods and paving the way for more reliable real-time tool condition monitoring.
What Problem Does This Paper Address?
In manufacturing, accurately identifying the state of tool wear is crucial for maintaining machining precision, maximizing equipment uptime, and reducing production costs. Traditional methods often struggle with noisy sensor data, the complexity of tool wear patterns, and the challenge of optimizing machine learning models, especially with limited datasets. This paper tackles the problem of developing a robust and highly accurate tool wear identification system that can overcome these limitations, enabling more efficient and reliable manufacturing processes.
Methods
The core of this research lies in a multi-stage approach:
- Signal Preprocessing: Wavelet packet threshold denoising (WPTD) is employed to clean multi-source machining signals (force, vibration, acoustic emission), removing noise while preserving important information.
- Feature Extraction: A comprehensive set of 245 features from time, frequency, and time-frequency domains are extracted from the denoised signals.
- Feature Selection: Support Vector Machine Recursive Feature Elimination (SVM-RFE) is used to identify the most relevant features, reducing dimensionality and preventing overfitting.
- Optimized Identification Model: A novel Improved Northern Goshawk Optimization (INGO) algorithm is proposed to optimize the key parameters (penalty factor C and kernel parameter γ) of a Support Vector Machine (SVM). The INGO algorithm itself is an enhancement of the original Northern Goshawk Optimization (NGO) and incorporates:
- Tent Chaos Mapping: For a more diverse and uniform initialization of the population, improving global search capabilities.
- Adaptive Weight Factor: Dynamically adjusts the search strategy based on iteration progress, balancing global and local search.
- Levy Flight Strategy: Enhances exploration and prevents premature convergence to local optima.
Datasets
The study utilizes the Prognostic and Health Management Society 2010 (PHM 2010) real-world dataset for tool wear state identification. This dataset is derived from milling experiments and contains data from multiple milling cutters under various conditions.
Key Insights and Ideas
- INGO's Superiority: The INGO algorithm demonstrates significant improvements in convergence speed and stability compared to its predecessor (NGO) and other popular optimization algorithms (PSO, GWO, WOA) on benchmark test functions. This suggests its effectiveness in finding optimal solutions.
- Feature Sensitivity: The research highlights that cutting force signals, particularly features like Peak-to-Peak (PP) and Standard Deviation (Std), are highly sensitive to tool wear. This underscores the importance of selecting the right features for accurate identification.
- SVM-RFE Effectiveness: SVM-RFE proves to be a powerful tool for selecting a concise yet highly informative feature set, crucial for building efficient and accurate identification models.
- Real-World Validation: The successful application of the INGO-SVM model to the PHM 2010 dataset validates its practical utility and effectiveness in a real-world machining scenario.
Key Results
- The proposed INGO-SVM model achieved an overall identification accuracy of 97.9% on the test set.
- This represents an approximate improvement of 15.6% over standard SVM and 3.2% over NGO-SVM.
- The model demonstrated high performance across different wear states, with accuracies of approximately 83.3% (initial wear), 99.5% (normal wear), and 100% (severe wear).
- The INGO algorithm outperformed other tested optimization algorithms (PSO, GWO, WOA, NGO) in terms of mean and standard deviation on most benchmark test functions.
Limitations and Future Work
- Remaining Useful Life Prediction: The authors acknowledge that while this paper focuses on tool wear state identification, predicting the remaining useful life (RUL) of a tool is another critical aspect of tool management.
- Future Research Direction: They propose integrating the INGO algorithm with other advanced predictive technologies to develop a more comprehensive solution for tool management, including RUL prediction.
Summary updated