journal article Mar 30, 2026

Hybrid Deep Learning Control of the Electrohydraulic Servodrive Position for the Repetitive Task

View at Publisher Save 10.1002/rnc.70472
Abstract
ABSTRACT
Electro‐hydraulic servo drives are fundamental to industrial applications that require high power density and precise positioning. However, significant nonlinearities, parameter uncertainties and friction complicate their control. Traditional controllers have difficulty accommodating these complex behaviors, while advanced methods are too computationally intensive for typical industrial PLCs. Furthermore, many existing neural network controllers lack formal stability guarantees, which poses risks in safety‐critical environments. It requires an advanced control architecture that can robustly adapt online to system nonlinearities, operate efficiently on industrial hardware with fast cycle times, and guarantee closed‐loop stability without requiring offline training or a pre‐existing plant model. This paper presents a hybrid deep learning controller (HDLC) designed to address these challenges. The HDLC features a novel architecture that integrates three learning paradigms: Kohonen maps (SOM), Hebbian learning and a globally stable adaptive law for a multi‐layer perceptron (MLP) controller. The HDLC is coupled with a diagonal recurrent neural network (Model‐DRNN) that performs online system identification, enabling the entire system to adapt in real time. At the same time, a Lyapunov‐based proof ensures stable operation. The HDLC has been implemented and experimentally validated on an industrial PLC‐controlled electro‐hydraulic servodrive with a cycle time of 2 ms. The HDLC demonstrated superior performance in comparative tests against a PID controller and a pre‐trained NN‐based direct inverse control (DIC). In a complex, repetitive tracking task, the HDLC achieved a 70% reduction in integral absolute error compared to the DIC. The HDLC demonstrated the fastest recovery time in a pressure disturbance test and maintained the lowest tracking error. Thus, confirming its robustness and effective online adaptation capabilities. Our research shows that the HDLC offers a practical and reliable solution for advanced control of nonlinear electro‐hydraulic systems.
Topics

No keywords indexed for this article. Browse by subject →

References
51
[5]
Możaryn J. (2022)
[19]
Borrás Pinilla C. (2019)
[34]
A.WinnickiandB.Guś “Neural Network Based on Direct Inverse Control for Electro‐Hydraulic Servo Drive ”146(2024):406–412 https://doi.org/10.1109/5.58325. 10.12693/aphyspola.146.406
[45]
Diagonal recurrent neural network based adaptive control of nonlinear dynamical systems using lyapunov stability criterion

Rajesh Kumar, Smriti Srivastava, J.R.P. Gupta

ISA Transactions 10.1016/j.isatra.2017.01.022
[49]
Gus B. (2023)
[50]
Mozaryn J. (2018)

Showing 50 of 51 references

Metrics
0
Citations
51
References
Details
Published
Mar 30, 2026
License
View
Cite This Article
Bartłomiej Guś, Jakub Możaryn, Arkadiusz Winnicki (2026). Hybrid Deep Learning Control of the Electrohydraulic Servodrive Position for the Repetitive Task. International Journal of Robust and Nonlinear Control. https://doi.org/10.1002/rnc.70472