Machine Learning Aided Optimization of Non-Metallic Seals in Downhole Tools

Main Article Content

S. Pirayeh Gar
A. Zhong
J. Jacob


Machine learning-based optimization analysis is conducted to improve the performance and reliability of non-metallic seals that undergo dynamic loading. Non-metallic seals, mostly made from elastomeric materials such as rubber, are widely used in the oil and gas industry due to their combination of hyper-elastic behavior and lack of plastic deformation under pressure and temperature cycles. However, these elastomeric materials could soften at elevated temperatures or during cyclic loading which makes them more prone to extrude from the seal region. For simple seals such as O-rings, thermoplastic anti-extrusion backup rings can be used to resist rubber extrusion. For more complex molded or bonded seals subject to cyclic loads and temperature cycles, minimizing the rubber damage and maintaining the seal integrity becomes more challenging as there may be inadequate room for backup rings. Additionally, the rubber may also experience repeated long stroke lengths during service which can cause surface abrasion. Non-metallic seals are extensively used in Sand Control Tools, a class of completion equipment. In the development of these molded and bonded seals, advanced optimization analysis techniques are important to make the seal design sufficiently robust to meet the stringent requirements of industry codes and regulations. Machine learning-based optimization analysis allows for increasing the seal performance and reliability. The Mullins effect is included in the elastomeric material model to capture the material softening effects under cyclic loading. Machine learning-based optimization starts with a feasible design space that is defined based on the controlling design parameters related to the geometry of the seal. Comprehensive Finite Element Analysis is performed on each design sample under the target cyclic load and temperature cycles. The key response parameters of the non-metallic seal are selected as the principal strain and the contact stress in the rubber which together indicate the damage and seal-ability states of the design. A Performance Objective Index is defined as a function of the key response parameters. The FEA results covering the entire design space provide the simulation-driven training data to build the surrogate model using multi-layer neural networks. The optimum design solution is found from the surrogate model which correlates the Performance Objective Index to the design parameters. The results of this study reveal that the performance of the seal is greatly affected by the ratio of the elastomeric seal volume to the gland volume into which the seal is molded. As opposed to the base design, the optimum design shows a solid performance with no signs of extrusion when the principal strain of the rubber is kept below the elongation limit throughout the load cycles, and when sufficient contact stress is developed to maintain the seal.

Article Details

How to Cite
Machine Learning Aided Optimization of Non-Metallic Seals in Downhole Tools. (2023). Engineering Modelling, Analysis and Simulation, 1.

How to Cite

Machine Learning Aided Optimization of Non-Metallic Seals in Downhole Tools. (2023). Engineering Modelling, Analysis and Simulation, 1.


Seals and Sealing Handbook, 2nd Edition, Du Pont de Nemours International S.A.

Parker O-Ring Handbook, ORD 5700.

Xiaobin Betty Huang and Wayne Furlan (2019). Accelerating Non-Metallic Seal Development for Downhole Applications by Applying a Big Data Method in Material Data: International Petroleum Technology Conference (IPTC-19248-MS), Beijing, China.

Zhang Y, Zhang X, Yang L, Yu X. Optimization Design for Downhole Dynamic Seal Based on Response Surface Method. Advances in Mechanical Engineering. 2019; 11(2). doi:10.1177/1687814019828441

Dassault Systems (2009). Modeling Rubber and Viscoelasticity with Abaqus. SIMULIA Technical Manual.