Data-driven prediction of tool wear using Bayesian-regularized artificial neural networks

Tam T. Truong,Jay Airao,Panagiotis Karras,Faramarz Hojati,Bahman Azarhoushang,Ramin Aghababaei

DOI: https://doi.org/10.48550/arXiv.2311.18620

2023-11-30

Abstract:The prediction of tool wear helps minimize costs and enhance product quality in manufacturing. While existing data-driven models using machine learning and deep learning have contributed to the accurate prediction of tool wear, they often lack generality and require substantial training data for high accuracy. In this paper, we propose a new data-driven model that uses Bayesian Regularized Artificial Neural Networks (BRANNs) to precisely predict milling tool wear. BRANNs combine the strengths and leverage the benefits of artificial neural networks (ANNs) and Bayesian regularization, whereby ANNs learn complex patterns and Bayesian regularization handles uncertainty and prevents overfitting, resulting in a more generalized model. We treat both process parameters and monitoring sensor signals as BRANN input parameters. We conducted an extensive experimental study featuring four different experimental data sets, including the NASA Ames milling dataset, the 2010 PHM Data Challenge dataset, the NUAA Ideahouse tool wear dataset, and an in-house performed end-milling of the Ti6Al4V dataset. We inspect the impact of input features, training data size, hidden units, training algorithms, and transfer functions on the performance of the proposed BRANN model and demonstrate that it outperforms existing state-of-the-art models in terms of accuracy and reliability.

Computational Engineering, Finance, and Science,Artificial Intelligence

What problem does this paper attempt to address?

### Problems Addressed by the Paper

This paper aims to address the issue of tool wear prediction in manufacturing. Specifically, while existing data-driven models (such as machine learning and deep learning models) can predict tool wear with relative accuracy, these models often lack generalization capability and require a large amount of training data to achieve high precision. Therefore, this paper proposes a new data-driven model—Bayesian Regularization Artificial Neural Network (BRANN)—to overcome the limitations of existing models in tool wear prediction.

#### Main Contributions
1. **Model Design**: BRANN combines the advantages of Artificial Neural Networks (ANN) and Bayesian regularization, using ANN to learn complex patterns while Bayesian regularization handles uncertainty and prevents overfitting.
2. **Input Parameters**: Process parameters (such as feed rate, cutting depth, etc.) and monitoring sensor signals (such as cutting force, vibration signals, etc.) are used as input parameters for BRANN.
3. **Experimental Validation**: The performance of the BRANN model is validated using four different experimental datasets (NASA Ames Milling Dataset, 2010 PHM Data Challenge Dataset, NUAA Ideahouse Tool Wear Dataset, and an internally conducted Ti6Al4V end milling experimental dataset).
4. **Performance Evaluation**: By comparing the impact of different numbers of hidden units, transfer functions, training algorithms, and training data sizes on the performance of the BRANN model, the superiority of BRANN in terms of accuracy and reliability is demonstrated.

In summary, this paper aims to improve the accuracy and generalization capability of tool wear prediction by introducing the BRANN model, thereby optimizing the manufacturing process, reducing costs, and increasing production efficiency.