University of Ljubljana Uses MATLAB to Optimize Production System Performance

LASIM Laboratory is part of the Department of Manufacturing Technologies and Systems at the Faculty of Mechanical Engineering, University of Ljubljana, and is led by Prof. Dr. Niko Herakovič. Researchers here focus on agile manufacturing systems and key enabling technologies for Factories of the Future. Core expertise includes process modeling, simulation, and real-time optimization; AI- and ML-driven control; digital twins; edge computing; and AR/VR integration. Strong industrial collaborations are maintained with companies such as Kolektor, Yaskawa, Krka, Polycom, Gorenje, Riko, and Danfoss, ensuring the effective transfer of academic knowledge into industrial environments.

The research presented here successfully applied machine learning models to predict and optimize the behavior of a hydraulic press in an industrial context. By combining advanced regression techniques and AI-based solutions, the team identified efficient, accurate models and substantially reduced the system’s response error. The results highlight the strong potential of integrating AI-driven approaches into control and decision-making algorithms for production systems. 

Challenge 

Industrial production systems, such as hydraulic presses, require accurate prediction of system behavior to improve efficiency and reduce error. Traditional methods lacked flexibility and adaptability to changing conditions. The challenge was to evaluate machine learning techniques for predicting the hydraulic press’s response error and to identify the most effective input parameters for improved decision-making and control.

Solution 

In the first phase, the study analyzed a comprehensive dataset of approximately 40,000 data points collected from sensors embedded in a hydraulic press. These sensors recorded variables such as hydraulic cylinder displacement, valve opening, pressure in the upper and lower chambers, and the force exerted by the hydraulic cylinder. These variables were used as input parameters for various regression models to predict the response error of the hydraulic press, defined as the difference between the measured and reference displacements of the hydraulic cylinder.

Multiple machine learning algorithms were implemented and compared, including Linear Regression (LR), Decision Trees (DT), Support Vector Regression (SVR), Gaussian Process Regression (GPR), and Neural Networks (NN). GPR and NN provided the most accurate and flexible predictions, while LR, DT, and SVR offered faster training and prediction times, making them more suitable for time-sensitive scenarios. The built-in Regression Learner MATLAB application was used to train and test these regression models. 

In a follow-up study, the researchers explored the use of machine learning combined with artificial intelligence techniques to identify the most relevant input parameters. Using the GPR model, 15 different model configurations were trained and evaluated based on various combinations of input features. Evaluation metrics such as RMSE, MSE, MAE, and R² were applied. The findings revealed that, in most cases, only two input parameters were sufficient to predict the response error with an accuracy of 95% or higher.

In the next phase, the team designed an expert system where a smart adaptive control framework integrated decision-making algorithms with machine learning. A Polynomial Regression method was introduced as a more computationally efficient alternative to GPR, offering comparable predictive performance.In the final evaluation of the expert system’s closed-loop control capabilities, the results demonstrated a significant real-time reduction of the hydraulic cylinder response error by more than 94%. The researchers confirmed that the proposed approaches are effective and offer strong potential for integration into other production systems. 

Result 

The expert system demonstrated a significant improvement in performance, achieving a real-time reduction of the hydraulic cylinder’s response error by more than 94%. The study confirmed that machine learning models, particularly when integrated with intelligent control frameworks, can significantly enhance the accuracy and efficiency of industrial production systems. 

Featured products

MathWorks® products:

• MATLAB
• Simulink
• MATLAB Campus-Wide License
• Statistics and Machine Learning Toolbox
• Curve Fitting Toolbox
• Regression Learner App

Learn more

The full in-depth analysis is available in the following open-access research articles:

• University of Ljubljana. Faculty of Mechanical Engineering (DME). Laboratory LASIM
  LASIM Laboratory is part of the Department of Manufacturing Technologies and Systems at the Faculty of Mechanical Engineering, University of Ljubljana, and is led by Prof. Dr. Niko Herakovič.
  Visit the university webpage
• Article: A data-driven simulation and Gaussian process regression model for hydraulic press condition diagnosis, Jankovič, D., Šimic, M., Herakovič, N.
  Advanced Engineering Informatics, 2024, Vol. 59, Article No. 102276.
  Read the article
• Article: A comparative study of machine learning regression models for production systems condition monitoring, Jankovič, D., Šimic, M., Herakovič, N.
  Advances in Production Engineering & Management, 2024, 1(19) pp 78–92, ISSN 1854-6250
  Read the article
• Article: Polynomial Regression-Based Predictive Expert System for Enhancing Hydraulic Press Performance over a 5G Network, Jankovič, D., Pipan M., Šimic, M., Herakovič, N.
  MDPI Applied Sciences 14(24)
  Read the article
• MATLAB Campus-Wide License Page
  Unlock the full potential of your institution’s academic pursuits with the MATLAB Campus-Wide License, providing access to a comprehensive suite of tools for computational analysis, modeling, and data visualization.
  Visit the page
• On-Demand Webinar: The Benefits of the MATLAB Campus-Wide License – Focusing on AI Applications
  Watch this video to get acquainted with all the tools and opportunities provided by the MATLAB Campus License, including artificial intelligence applications with MATLAB.
  Watch the webinar