A Novel Sustainable Processing Mode for Burr Classified Prediction of Weak Rigid Drilling Process Using a Fusion Modeling Method

Siyi Ding, Xiaohu Zheng, Mingyu Wu, Qirui YangSustainability 2022 · DOI

TL;DR

This paper introduces a novel fusion model that combines traditional physics-based burr formation theory with a data-driven approach (TCN-DNN) to accurately predict burr quality in weak rigid drilling systems, like industrial robots. This approach significantly improves prediction accuracy, paving the way for more sustainable and efficient manufacturing by reducing costly deburring processes.

What Problem Does This Paper Address?

Industrial robots are increasingly used for precision tasks like drilling in sectors such as aerospace and military. However, their inherent "weak rigidity" (significantly lower stiffness than traditional CNC machines) leads to vibrations and unpredictable burr formation. These burrs degrade part precision and performance, necessitating extensive and costly deburring. Existing burr prediction models, relying solely on traditional theory, struggle to account for the complex dynamics and uncertainties of these weak rigid systems. Data-driven methods, while promising, can be overly context-dependent. The paper aims to bridge this gap by developing a hybrid model that leverages the strengths of both approaches for more robust and accurate burr prediction.

Methods

The core of this research is a fusion modeling method that combines a mechanism model with a data-driven error compensation model.

  1. Mechanism Model: The authors first develop a model based on the physics of burr formation during drilling in weak rigid systems. This model considers factors like material slip deformation, friction, and the unique dynamics of weak rigidity. It aims to provide an initial, albeit imperfect, prediction of burr characteristics.
  2. Data-Driven Error Compensation Model: Recognizing the limitations of the mechanism model, they introduce a data-driven approach to correct its predictions. This involves:
  3. Data Collection: Experimental data on drilling parameters and actual burr measurements are collected.
  4. Error Calculation: The difference between the mechanism model's initial prediction and the actual measured burr classification is calculated. This error serves as the "tag value" for training the compensation model.
  5. TCN-DNN Architecture: A Temporal Convolutional Network (TCN) combined with a Deep Neural Network (DNN) is employed. TCNs are chosen for their ability to capture long-term dependencies in time-series data (like drilling signals) without the vanishing gradient issues of traditional RNNs. The DNN then classifies the extracted features to predict the error.
  6. Fusion: The final burr classification is achieved by integrating the mechanism model's prediction with the error predicted by the TCN-DNN model.

Datasets

The paper describes experiments conducted on a custom-built weak rigid drilling platform. The dataset consists of:

  • Input Data: Machining condition parameters (cutting thickness, speed, rotation speed), system properties (elasticity modulus, stiffness, damping ratio, natural frequency), material properties (hardness, shear strength), tool geometry (rake angle, diameter, edge length), processing status signals (drilling force, acceleration), and workpiece information.
  • Output Data: Measured burr height and thickness, which are then classified into three levels (<5%, 5-7%, >7% burr height ratio to material thickness).

The paper does not mention using pre-existing public datasets for burr prediction. The experimental data collected appears to be specific to the developed platform and materials used.

Key Insights and Ideas

  • Synergy of Physics and Data: The most significant insight is that combining a physics-based model with a data-driven error correction mechanism offers a powerful way to overcome the limitations of each individual approach, especially in complex, dynamic systems.
  • TCN for Time-Series Burrs: The selection of TCN for error prediction is crucial. It effectively handles the time-series nature of drilling data, capturing subtle temporal patterns that traditional models might miss, leading to more accurate error estimation.
  • Weak Rigidity's Impact: The paper clearly demonstrates how weak rigidity in systems like industrial robots fundamentally challenges traditional burr prediction, highlighting the need for novel approaches.
  • Sustainability Through Prediction: Accurate burr prediction directly translates to reduced deburring efforts, saving time, resources, and energy, thus contributing to more sustainable manufacturing practices.

Key Results

  • Improved Accuracy: The proposed fusion model achieved a prediction accuracy of 91.67%, representing a 25% improvement compared to traditional drilling theory (mechanism model).
  • Reduced Error: The average error between measured and predicted burr height using the mechanism model alone was 12%, with a maximum error of nearly 30%. The fusion model significantly reduced this inconsistency.
  • Optimizer Performance: The Adam optimizer was found to yield the best prediction accuracy (91.67%) among the tested optimizers (Adam, RMSprop, SGD).
  • Model Convergence: The TCN-DNN model demonstrated good convergence, with the lowest loss value compared to CNN and traditional DNN models.

Limitations and Future Work

  • Environment Dependency: The authors acknowledge that the fusion model is currently somewhat dependent on the specific machining environment (tool type, material).
  • Real-time Control: The current model is primarily for burr classification and prediction, not real-time control of the drilling process.
  • Future Directions:
    • Transfer Learning: Integrating transfer learning to enhance the model's generalization capability and reduce its dependence on specific processing environments.
  • Decision Information Base: Developing a decision information base and optimization model library to enable real-time parameter control.
  • Real-time Monitoring: Focusing on state monitoring during machining to facilitate real-time optimal control.

Summary updated

Abstract

Weakly rigid drilling systems, such as the industrial robot, are widely used in aerospace, military, and other fields due to its good flexibility and large scope of operation. However, the weak rigidity can easily cause burrs, seriously affecting the precision of parts and product performance. To reduce the heavy deburring process and to improve continuous production and sustainable processing capacity, accurate prediction of burr quality is a prerequisite. Traditional burr forming theory cannot accurately predict the drilling defects. Data-driven approaches can be independent of prior knowledge and discover relationships between process parameters and machining precision directly from the data structure itself. Therefore, to take advantage of both approaches, a fusion model was established for burr classified prediction. On the one hand, the drilling and burr forming process was firstly modeled, and preliminary classification results for burrs were calculated. On the other hand, according to the measured data, the errors between initial calculation results and actual classification results were obtained and selected as the tag values of dataset, which served as inputs for the error compensation model of burrs. Finally, by training the network of TCN–DNN using the drilling data, the burr classified prediction in a weak rigid hole-making system was realized. Experimental results showed that compared with traditional drilling theory, the prediction accuracy of the proposed model improved by 25%, reaching 91.67%. The results can provide a basis for judging the process of burr post-treatment, which has practical guiding significance. This method is beneficial to reduce the heavy deburring process and to improve sustainable processing capacity.
BibTeX
@inproceedings{ding2022novel,
  author = {Siyi Ding and Xiaohu Zheng and Mingyu Wu and Qirui Yang},
  title = {{A Novel Sustainable Processing Mode for Burr Classified Prediction of Weak Rigid Drilling Process Using a Fusion Modeling Method}},
  booktitle = {Sustainability},
  year = {2022},
  doi = {10.3390/su14127429},
}
Figure