Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Data-driven tool wear prediction in milling, based on a process-integrated single-sensor approach

Eric Hirsch, Christian Friedrich

TL;DR

This study tackles the challenge of predicting milling tool wear with a transferable, low-cost data-driven approach that uses a single accelerometer. By comparing ConvNeXt, LSTM, SVM, and Decision Trees on STFT-based inputs and handcrafted features, the authors demonstrate that a ConvNeXt backbone achieves top performance (up to 99.1% accuracy) when trained on data from just four tools. The work emphasizes data efficiency and transferability across processes, revealing that cross-machine generalization remains challenging and benefits from transfer learning. The resulting pipeline offers a practical pathway for industry-ready tool condition monitoring with minimal sensor complexity and rapid deployment potential.

Abstract

Accurate tool wear prediction is essential for maintaining productivity and minimizing costs in machining. However, the complex nature of the tool wear process poses significant challenges to achieving reliable predictions. This study explores data-driven methods, in particular deep learning, for tool wear prediction. Traditional data-driven approaches often focus on a single process, relying on multi-sensor setups and extensive data generation, which limits generalization to new settings. Moreover, multi-sensor integration is often impractical in industrial environments. To address these limitations, this research investigates the transferability of predictive models using minimal training data, validated across two processes. Furthermore, it uses a simple setup with a single acceleration sensor to establish a low-cost data generation approach that facilitates the generalization of models to other processes via transfer learning. The study evaluates several machine learning models, including transformer-inspired convolutional neural networks (CNN), long short-term memory networks (LSTM), support vector machines (SVM), and decision trees, trained on different input formats such as feature vectors and short-time Fourier transform (STFT). The performance of the models is evaluated on two machines and on different amounts of training data, including scenarios with significantly reduced datasets, providing insight into their effectiveness under constrained data conditions. The results demonstrate the potential of specific models and configurations for effective tool wear prediction, contributing to the development of more adaptable and efficient predictive maintenance strategies in machining. Notably, the ConvNeXt model has an exceptional performance, achieving 99.1\% accuracy in identifying tool wear using data from only four milling tools operated until they are worn.

Data-driven tool wear prediction in milling, based on a process-integrated single-sensor approach

TL;DR

This study tackles the challenge of predicting milling tool wear with a transferable, low-cost data-driven approach that uses a single accelerometer. By comparing ConvNeXt, LSTM, SVM, and Decision Trees on STFT-based inputs and handcrafted features, the authors demonstrate that a ConvNeXt backbone achieves top performance (up to 99.1% accuracy) when trained on data from just four tools. The work emphasizes data efficiency and transferability across processes, revealing that cross-machine generalization remains challenging and benefits from transfer learning. The resulting pipeline offers a practical pathway for industry-ready tool condition monitoring with minimal sensor complexity and rapid deployment potential.

Abstract

Accurate tool wear prediction is essential for maintaining productivity and minimizing costs in machining. However, the complex nature of the tool wear process poses significant challenges to achieving reliable predictions. This study explores data-driven methods, in particular deep learning, for tool wear prediction. Traditional data-driven approaches often focus on a single process, relying on multi-sensor setups and extensive data generation, which limits generalization to new settings. Moreover, multi-sensor integration is often impractical in industrial environments. To address these limitations, this research investigates the transferability of predictive models using minimal training data, validated across two processes. Furthermore, it uses a simple setup with a single acceleration sensor to establish a low-cost data generation approach that facilitates the generalization of models to other processes via transfer learning. The study evaluates several machine learning models, including transformer-inspired convolutional neural networks (CNN), long short-term memory networks (LSTM), support vector machines (SVM), and decision trees, trained on different input formats such as feature vectors and short-time Fourier transform (STFT). The performance of the models is evaluated on two machines and on different amounts of training data, including scenarios with significantly reduced datasets, providing insight into their effectiveness under constrained data conditions. The results demonstrate the potential of specific models and configurations for effective tool wear prediction, contributing to the development of more adaptable and efficient predictive maintenance strategies in machining. Notably, the ConvNeXt model has an exceptional performance, achieving 99.1\% accuracy in identifying tool wear using data from only four milling tools operated until they are worn.
Paper Structure (14 sections, 5 equations, 12 figures, 3 tables)

This paper contains 14 sections, 5 equations, 12 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Architecturedesign for wear prediction
  4. Problem description and Pipeline
  5. Data preprocessing
  6. Feature engineering
  7. Different models
  8. Training and Evaluation process
  9. Experimental results
  10. Setup
  11. Design of Experiment
  12. Evaluation
  13. Conclusion
  14. Tables of this research

Figures (12)

  • Figure 1: The influence of tool wear and the sensors that can be employed for its monitoring.
  • Figure 2: Pipeline for generating process data: (a) The milling force is converted into acceleration data $\ddot{x}$. (b) A plot of the recorded acceleration data is shown. (c) The result of the p2p function $\ddot{x}_{\text{p2p}_t}$ is displayed, along with the duration (green or red areas) where $\ddot{x}$ exceeds or falls below the threshold $th$. Using this analysis, the acceleration data is segmented into process data (d). The resulting process data $\ddot{x_p}$ is used for our classification pipeline.
  • Figure 3: (a) The diagram illustrates the transformation of acceleration data into different domains and the subsequently extraction of features using either a CNN architecture (i) or proprietary algorithms (ii). These features are then utilized by various classifiers (1, 2, 3, 4) to predict wear. (b) The architectures of the four classifiers used in this paper, categorised by their input type, are shown. Images from the following sources were used to create this diagram: westphal_machine_2021rungeReviewDeepLearning2021
  • Figure 4: Structure of the datasets used for the models: Each process dataset includes multiple machines, and each machine dataset contains several tool life cycles. A single tool life cycle represents the use of one tool until it becomes worn. Within each dataset, multiple workpieces are produced.
  • Figure 5: (a) Side view of the toolholder with the integrated 1-axis MEMS sensor. (b) Top-down view of the toolholder, illustrating how the sensor measures acceleration in the $xy$-plane, depending on the rotation angle $\varphi$.
  • ...and 7 more figures