Multi-Objective Prediction of Pipe Shape in Oblique Rolling Piercing of Seamless Steel Pipes

Literature Overview

Jia Shiying, Wang Qinghua, Wang Zhenyan, Hu Jianhua, Shuang Yuanhua, and Zhou Xinliang from Taiyuan University of Science and Technology and Taiyuan Heavy Industry Co., Ltd. present a data-driven modeling study for predicting the pipe shape (wall thickness and outer diameter) of seamless steel pipes produced through oblique rolling piercing. Published in Forging Technology in 2022, Volume 47, Issue 10, this paper addresses a persistent challenge in seamless pipe manufacturing: the difficulty of accurately controlling blank tube dimensions through the complex, multi-parameter oblique rolling piercing process. The research was funded by the Shanxi Province Science and Technology Major Project (20191102009).

Process Background and Technical Challenge

Oblique rolling piercing is a critical forming process in seamless steel pipe production, where a heated billet is pierced through rotating rolls with a plug to form a hollow tube (blank tube). The process involves complex three-dimensional plastic deformation with significant interaction between multiple process parameters. The resulting blank tube dimensions—outer diameter and wall thickness—are critical for downstream processing and product quality. However, the nonlinear relationships between process parameters and output dimensions make precise prediction and control challenging.

The key process parameters influencing blank tube shape include:

Methodology: Least Squares Support Vector Regression

The study employs Least Squares Support Vector Regression (LS-SVR) for building the prediction model. This choice is motivated by several practical considerations: the dataset is relatively small (a common limitation in industrial process data), the input-output relationships are highly nonlinear, and there exist cross-correlations among the input parameters. LS-SVR is particularly well-suited for small-sample, nonlinear regression problems and avoids the local minima issues that plague traditional neural network training.

The modeling approach follows a systematic workflow:

1. Data collection from actual production records of the oblique rolling piercing process.
2. Input parameter screening using Grey Relational Analysis (GRA) to identify the most influential process parameters.
3. Construction of a multi-input multi-output (MIMO) LS-SVR model with five inputs and two outputs.
4. Model training and validation using actual production data.
5. Simulation experiments to verify model accuracy and robustness.

Grey Relational Analysis for Parameter Screening

The application of Grey Relational Analysis for parameter screening is a methodologically sound approach. GRA evaluates the degree of correlation between each input parameter and the output parameters by calculating grey relational grades. Parameters with higher relational grades are retained as inputs to the prediction model, while less influential parameters are excluded. This screening step is crucial for maintaining model accuracy and avoiding overfitting, particularly given the limited dataset size.

The selection of five parameters (forward extension, roll gap, guide plate gap, plug diameter, and billet diameter) as model inputs reflects the physical understanding of the piercing process combined with the statistical screening results. Each parameter has a distinct mechanical influence on the deformation process, and their combined effect determines the final blank tube geometry.

Multi-Objective Prediction Framework

The MIMO LS-SVR model simultaneously predicts both wall thickness and outer diameter, capturing the inherent coupling between these two geometric parameters. This multi-objective approach is more physically realistic than separate single-output models, as wall thickness and outer diameter are mechanically coupled through the deformation process. The model architecture can be represented as:

Engineering Practice Implications

The prediction model developed in this study has direct applications in seamless pipe production optimization:

• Process parameter adjustment: Operators can use the model to predict the effect of parameter changes on blank tube dimensions, enabling proactive process control rather than reactive quality correction.
• Quality assurance: The model can serve as a virtual gauge, predicting blank tube dimensions before they are measured, allowing early detection of dimensional deviations.
• Process optimization: The model can be integrated with optimization algorithms to identify parameter combinations that minimize dimensional deviation from target specifications.
• New product development: When transitioning to new pipe grades or sizes, the model can accelerate the trial-and-error process by providing initial parameter recommendations.

The practical value of this approach is enhanced by its ability to handle small datasets and cross-correlated inputs, which are characteristic of industrial manufacturing data. Traditional regression methods often struggle with these conditions, leading to unreliable predictions.

Defect Analysis and Process Control

Understanding the prediction model's outputs in the context of common piercing defects provides additional engineering value:

Key Questions and Reflections

A significant consideration is the model's applicability across different production conditions. The model is trained on data from a specific production line with particular equipment characteristics, material properties, and operating conditions. When applied to different rolling mills, different steel grades, or significantly different temperature ranges, the model may require recalibration or retraining. Engineers should validate the model's predictions against actual measurements before relying on it for production decisions, particularly when operating conditions deviate from the training data range.

Another important question concerns the model's ability to capture transient process phenomena. Oblique rolling piercing involves dynamic deformation with time-varying contact conditions, temperature gradients, and material flow patterns. The LS-SVR model, being a data-driven regression approach, captures the average statistical relationships but may not fully represent the dynamic variability inherent in the process. Process monitoring systems that provide real-time feedback would complement the prediction model by detecting deviations from expected behavior.

Summary and Concluding Remarks

This study demonstrates the effectiveness of LS-SVR for multi-objective prediction of blank tube dimensions in oblique rolling piercing. The integration of Grey Relational Analysis for parameter screening with the MIMO LS-SVR model provides a practical and accurate prediction framework that addresses the challenges of small datasets and cross-correlated inputs. The model offers significant value for process optimization, quality control, and new product development in seamless pipe manufacturing. Engineers should view this prediction model as a decision-support tool that augments, rather than replaces, their process expertise and judgment. The combination of data-driven prediction with physical process understanding represents the most effective approach to seamless pipe production optimization, and this study provides a solid foundation for further development of intelligent process control systems in the tube manufacturing industry.