Zhuojin Pipe Fitting Co., LtdIterative Self-Learning Method for Average Wall Thickness Control During Seamless Steel Pipe Drawing-Reduction Process
Literature Overview
The paper by Liu Shan, Wu Tiejun, Liu Yuwen, and Wang Zhiguo, published in Iron and Steel (2002, Vol. 37, No. 4, pp. 29–34), proposes an iterative self-learning control algorithm for the average wall thickness control during the drawing-reduction (张减) process of seamless steel pipe production. The research was conducted jointly by Zhejiang University and Baoshan Iron & Steel Co., Ltd. The core idea is to leverage the repetitive nature of the rolling process to improve tracking performance by using the deviation between actual and desired outputs to generate improved control signals for subsequent passes.
Technical Background
The drawing-reduction process is a critical step in seamless steel pipe production, where a hot-rolled pipe blank is drawn through a series of rolling stands to achieve the final dimensions, including the target wall thickness. The process involves multiple rolling stands arranged in a tandem configuration, and the wall thickness is controlled by adjusting the steady-state rotational speed distribution among the stands.
The challenge lies in the fact that the physical parameters of the process—such as material flow stress, friction coefficients, and stand gap settings—are subject to time-varying uncertainties and modeling errors. These uncertainties lead to deviations between the actual and desired wall thickness, which directly impacts product quality and yield.
Iterative Learning Control Algorithm
The iterative learning control (ILC) approach exploits the fact that the same type of pipe is rolled repeatedly in production. The algorithm operates as follows:
- Initialization: A nominal control signal (rotational speed distribution) is applied for the first pass.
- Measurement: The actual wall thickness before and after rolling is measured, along with characteristic data of the pipe (outer diameter, length, material grade, etc.).
- Error Calculation: The deviation between the actual and desired wall thickness is computed.
- Control Signal Update: The rotational speed distribution for the next pass is adjusted based on the error signal using the iterative learning law.
- Convergence: Over multiple iterations, the control signal converges to the optimal distribution that minimizes the wall thickness deviation.
| Parameter | Description | Typical Range |
|---|---|---|
| Number of rolling stands | Tandem configuration | 4–8 stands |
| Target wall thickness tolerance | Final product requirement | ±0.5 mm |
| Iteration count for convergence | Number of passes to achieve target | 3–5 passes |
| Speed distribution adjustment range | Per iteration | ±2–5% |
Online Adaptive Adjustment
A key feature of the proposed method is the online adaptive adjustment of the steady-state rotational speed distribution of each rolling stand. This compensates for the time-varying uncertainties in physical parameters and modeling errors that cause deviations in the rolling stand speed distribution parameters. The adaptive nature of the algorithm ensures that the system can respond to changes in material properties, stand wear, and other process disturbances in real time.
Technical Assessment and Engineering Relevance
The iterative learning control approach is well-suited to the drawing-reduction process because of its inherently repetitive nature. Unlike single-pass control strategies, ILC can progressively improve performance over multiple passes of the same product type. This is particularly valuable in production environments where the same pipe specification is rolled in batches.
The method addresses a fundamental limitation of traditional model-based control: the inability to accurately model all physical parameters of the rolling process. By learning from past performance, the algorithm implicitly compensates for unmodeled dynamics and parameter uncertainties. This is conceptually similar to the "learning from experience" approach used in many industrial control applications.
In my experience with seamless pipe rolling mills, the wall thickness control problem is indeed challenging due to the complex interaction between the material deformation behavior and the rolling stand kinematics. The authors' approach of using measured data from previous passes to improve subsequent control is pragmatic and effective. However, the convergence rate and robustness of the algorithm under varying production conditions—such as sudden changes in material grade or stand maintenance—should be carefully evaluated.
One practical consideration is the need for reliable online measurement systems for wall thickness. The accuracy of the ILC algorithm is directly dependent on the quality of the feedback signals. If the wall thickness measurement system has significant noise or bias, the iterative learning process may diverge rather than converge. Therefore, investment in high-precision measurement systems is essential for the successful implementation of this control strategy.
Summary
This paper presents a practical and effective control strategy for wall thickness control in the seamless steel pipe drawing-reduction process. The iterative self-learning approach leverages the repetitive nature of production to progressively improve control performance, compensating for time-varying uncertainties and modeling errors. The method is well-suited to industrial applications where batch production of the same specification is common, and it represents a significant advancement over traditional model-based control approaches.