Zhuojin Pipe Fitting Co., LtdMaterial Performance Parameter Identification in Intelligent JCOE Forming Control for Large Fittings
Literature Overview
The paper by Sun Honglei, Zhao Jun, Yin Jing, and Li Jian, published in the Journal of Yanshan University (2011, Vol. 35, No. 3, pp. 223-227), addresses the identification of material performance parameters in the intelligent control system for JCOE forming of large-diameter pipe fittings. The work was supported by the Hebei Provincial Natural Science Foundation (E2011203008) and the Qinhuangdao Science and Technology Research and Development Plan Project (201001A030). The authors developed a neural network model using the Levenberg-Marquardt algorithm to identify material performance parameters in real time during the JCOE forming process. This research is highly relevant to engineers working on the manufacturing of large-diameter line pipe and fittings, where material variability poses significant challenges to process control.
Technical Background
JCOE forming is a process used to manufacture large-diameter steel pipes and fittings from steel plates. The process involves rolling a flat plate into a J-shape, then a C-shape, then an O-shape, and finally expanding the O-shape to the final diameter with an E-shape tool. The process is widely used for manufacturing line pipe with diameters ranging from 508 to 2438 millimeters, as specified in API 5L and related standards. The forming quality is highly sensitive to the mechanical properties of the steel plate, including yield strength, tensile strength, elongation, and strain hardening exponent.
JCOE Forming Process Parameters
| Process Stage | Key Parameters | Critical Material Properties |
|---|---|---|
| J-forming | Roll force, roll angle, feed rate | Yield strength, elastic modulus |
| C-forming | Roll force, roll angle, gap | Yield strength, strain hardening exponent |
| O-forming | Roll force, roll angle, gap | Yield strength, tensile strength |
| E-forming | Expansion force, expansion rate | Tensile strength, elongation, strain hardening exponent |
Material Variability Challenge
The fundamental challenge addressed by this paper is the variability in steel plate properties from batch to batch, even within the same nominal grade. Factors such as chemical composition fluctuations, rolling mill conditions, and heat treatment variations cause differences in mechanical properties that can lead to forming defects if not accounted for in the process parameters. Traditional approaches rely on pre-production material testing and fixed process parameter sets, which do not adapt to real-time material condition changes.
Neural Network Model Design
The authors designed a neural network model to identify material performance parameters from measurable process signals during the forming operation. The model architecture and training approach are described as follows.
Network Architecture
The neural network consists of an input layer, one or more hidden layers, and an output layer. The input layer receives process signals such as roll force, roll angle, and material deformation measurements. The output layer produces the identified material performance parameters, including yield strength, tensile strength, and strain hardening exponent. The hidden layers perform the nonlinear mapping between inputs and outputs.
Training Algorithm
The Levenberg-Marquardt (LM) algorithm was selected for network training due to its convergence speed and robustness. The LM algorithm is a hybrid of the Gauss-Newton method and the gradient descent method, providing efficient convergence for nonlinear least-squares problems. The authors used orthogonal experimental design to generate training data from 10 different material types, ensuring comprehensive coverage of the parameter space.
Training and Validation Results
The network was trained with 10 sets of sample data and validated against independent test data. The convergence accuracy was less than 1 per thousand, and the validation error was less than 3 percent. These results demonstrate that the neural network approach can accurately identify material performance parameters from process signals, enabling real-time process parameter adjustment.
Process Integration and Control Strategy
The material parameter identification system is integrated into the JCOE forming control system as follows.
Real-Time Monitoring
During the forming process, sensors measure roll forces, angles, and material deformation at multiple stations. These measurements are fed into the neural network in real time to identify the current material properties. The identified parameters are compared with reference values to detect any significant deviations.
Adaptive Parameter Adjustment
Based on the identified material properties, the control system automatically adjusts the forming process parameters, including roll force, roll angle, and feed rate, to maintain consistent forming quality. This adaptive control strategy compensates for material variability and ensures that each plate is formed with optimal parameters tailored to its specific mechanical properties.
Quality Feedback Loop
The system incorporates a quality feedback mechanism where dimensional measurements of the formed pipe are compared with target specifications. Any deviations are used to refine the neural network model and improve the accuracy of future material parameter identification.
Engineering Practice Considerations
The implementation of such an intelligent control system requires careful attention to several practical aspects. Sensor reliability is critical, as inaccurate measurements will lead to incorrect material parameter identification and potentially defective products. The neural network model must be periodically retrained with updated material data to maintain accuracy as production conditions change.
From a standards compliance perspective, the adaptive control system must ensure that the formed product meets the requirements of API 5L, EN 10216, or other applicable specifications. The system should include alarm and shutdown mechanisms that trigger when the identified material properties fall outside acceptable ranges, preventing the formation of non-conforming products.
Key Reflections
This paper represents a significant advancement in the automation of large-diameter pipe forming processes. The use of neural networks for material parameter identification bridges the gap between offline material testing and real-time process control. The approach is particularly valuable for high-volume production environments where material variability is a persistent challenge. However, the practical implementation requires investment in sensor infrastructure, data acquisition systems, and control hardware, which may be a barrier for smaller manufacturers.
The Levenberg-Marquardt algorithm is an appropriate choice for this application due to its efficient convergence properties. However, engineers should be aware that neural network models can be sensitive to the quality and diversity of training data. The use of orthogonal experimental design to generate training data is a sound approach, but the model's performance with unseen material conditions should be validated through extensive testing before full production deployment.
Summary
The development of a neural network-based material performance parameter identification system for JCOE forming represents a significant step toward intelligent manufacturing in the pipe and fitting industry. By enabling real-time adaptation of process parameters to material variability, this technology has the potential to improve forming quality, reduce scrap rates, and increase production efficiency. The demonstrated convergence accuracy of less than 1 per thousand and validation error of less than 3 percent confirm the technical feasibility of the approach. Engineers involved in large-diameter pipe manufacturing should consider adopting similar intelligent control strategies to enhance process robustness and product quality.