Search results

Filters

  • Journals
  • Authors
  • Keywords
  • Date
  • Type

Search results

Number of results: 4
items per page: 25 50 75
Sort by:
Download PDF Download RIS Download Bibtex

Abstract

Automation of data processing of contactless diagnostics (detection) of the technical condition of the majority of nodes and aggregates of railway transport (RWT) minimizes the damage from failures of these systems in operating modes. This becomes possible due to the rapid detection of serious defects at the stage of their origin. Basically, in practice, the control of the technical condition of the nodes and aggregates of the RWT is carried out during scheduled repairs. It is not always possible to identify incipient defects. Consequently, it is not always possible to warn personnel (machinists, repairmen, etc.) of significant damage to the RWT systems until their complete failure. The difficulties of obtaining diagnostic information is that there is interdependence between the main nodes of the RWT. This means that if physical damage occurs at any of the RWT nodes, in other nodes there can also occur malfunctions.

As the main way to improve the efficiency of state detection of the nodes and aggregates of RWT, we see the direction of giving the adaptability property for an automated data processing system from various contactless diagnostic information removal systems. The global purpose can be achieved, in particular, through the use of machine learning methods and failure recognition (recognition objects). In order to improve the operational reliability and service life of the main nodes and aggregates of RWT, there are proposed an appropriate model and algorithm of machine learning of the operator control system of nodes and aggregates. It is proposed to use the Shannon normalized entropy measure and the Kullback-Leibler distance information criterion as a criterion of the learning effectiveness of the automated detection system and operator node state control of RWT. The article describes the application of the proposed method on the example of an automatic detection system (ADS) of the state of a traction motor of an electric locomotive. There are given the test data of the model and algorithm in the MATLAB environment.

Go to article

Authors and Affiliations

Bakhytzhan Akhmetov
Valeriy Lakhno
Ayaulym Oralbekova
Zhanat Kaskatayev
Gulmira Mussayeva
Download PDF Download RIS Download Bibtex

Abstract

This paper presents how Q-learning algorithm can be applied as a general-purpose selfimproving controller for use in industrial automation as a substitute for conventional PI controller implemented without proper tuning. Traditional Q-learning approach is redefined to better fit the applications in practical control loops, including new definition of the goal state by the closed loop reference trajectory and discretization of state space and accessible actions (manipulating variables). Properties of Q-learning algorithm are investigated in terms of practical applicability with a special emphasis on initializing of Q-matrix based only on preliminary PI tunings to ensure bumpless switching between existing controller and replacing Q-learning algorithm. A general approach for design of Q-matrix and learning policy is suggested and the concept is systematically validated by simulation in the application to control two examples of processes exhibiting first order dynamics and oscillatory second order dynamics. Results show that online learning using interaction with controlled process is possible and it ensures significant improvement in control performance compared to arbitrarily tuned PI controller.
Go to article

Bibliography

[1] H. Boubertakh, S. Labiod, M. Tadjine and P.Y. Glorennec: Optimization of fuzzy PID controllers using Q-learning algorithm. Archives of Control Sciences, 18(4), (2008), 415–435
[2] I.Carlucho, M. De Paula, S.A. Villar and G.G.Acosta: Incremental Qlearning strategy for adaptive PID control of mobile robots. Expert Systems With Applications, 80, (2017), 183–199, DOI: 10.1016/j.eswa.2017.03.002.
[3] K. Delchev: Simulation-based design of monotonically convergent iterative learning control for nonlinear systems. Archives of Control Sciences, 22(4), (2012), 467–480.
[4] M. Jelali: An overview of control performance assessment technology and industrial applications. Control Eng. Pract., 14(5), (2006), 441–466, DOI: 10.1016/j.conengprac.2005.11.005.
[5] M. Jelali: Control Performance Management in Industrial Automation: Assessment, Diagnosis and Improvement of Control Loop Performance. Springer-Verlag London, (2013)
[6] H.-K. Lam, Q. Shi, B. Xiao, and S.-H. Tsai: Adaptive PID Controller Based on Q-learning Algorithm. CAAI Transactions on Intelligence Technology, 3(4), (2018), 235–244, DOI: 10.1049/trit.2018.1007.
[7] D. Li, L. Qian, Q. Jin, and T. Tan: Reinforcement learning control with adaptive gain for a Saccharomyces cerevisiae fermentation process. Applied Soft Computing, 11, (2011), 4488–4495, DOI: 10.1016/j.asoc.2011.08.022.
[8] M.M. Noel and B.J. Pandian: Control of a nonlinear liquid level system using a new artificial neural network based reinforcement learning approach. Applied Soft Computing, 23, (2014), 444–451, DOI: 10.1016/j.asoc.2014.06.037.
[9] T. Praczyk: Concepts of learning in assembler encoding. Archives of Control Sciences, 18(3), (2008), 323–337.
[10] M.B. Radac and R.E. Precup: Data-driven model-free slip control of antilock braking systems using reinforcement Q-learning. Neurocomputing, 275, (2017), 317–327, DOI: 10.1016/j.neucom.2017.08.036.
[11] A.K. Sadhu and A. Konar: Improving the speed of convergence of multi-agent Q-learning for cooperative task-planning by a robot-team. Robotics and Autonomous Systems, 92, (2017), 66–80, DOI: 10.1016/j.robot.2017.03.003.
[12] N. Sahebjamnia, R. Tavakkoli-Moghaddam, and N. Ghorbani: Designing a fuzzy Q-learning multi-agent quality control system for a continuous chemical production line – A case study. Computers & Industrial Engineering, 93, (2016), 215–226, DOI: 10.1016/j.cie.2016.01.004.
[13] K. Stebel: Practical aspects for the model-free learning control initialization. in Proc. of 2015 20th International Conference on Methods and Models in Automation and Robotics (MMAR), Poland, (2015), DOI: 10.1109/MMAR.2015.7283918.
[14] R.S. Sutton and A.G. Barto: Reinforcement learning: An Introduction, MIT Press, (1998)
[15] S. Syafiie, F. Tadeo, and E. Martinez: Softmax and "-greedy policies applied to process control. IFAC Proceedings, 37, (2004), 729–734, DOI: 10.1016/S1474-6670(16)31556-2.
[16] S. Syafiie, F. Tadeo, and E. Martinez: Model-free learning control of neutralization process using reinforcement learning. Engineering Applications of Artificial Intelligence, 20, (2007), 767–782, DOI: 10.1016/j.engappai.2006.10.009.
[17] S. Syafiie, F. Tadeo, and E. Martinez: Learning to control pH processes at multiple time scales: performance assessment in a laboratory plant. Chemical Product and Process Modeling, 2(1), (2007), DOI: 10.2202/1934- 2659.1024.
[18] S. Syafiie, F. Tadeo, E. Martinez, and T. Alvarez: Model-free control based on reinforcement learning for a wastewater treatment problem. Applied Soft Computing, 11, (2011), 73–82, DOI: 10.1016/j.asoc.2009.10.018.
[19] P. Van Overschee and B. De Moor: RAPID: The End of Heuristic PID Tuning. IFAC Proceedings, 33(4), (2000), 595–600, DOI: 10.1016/S1474- 6670(16)38308-8.
[20] M. Wang, G. Bian, and H. Li: A new fuzzy iterative learning control algorithm for single joint manipulator. Archives of Control Sciences, 26(3), (2016), 297–310. DOI: 10.1515/acsc-2016-0017.
[21] Ch.J.C.H. Watkins and P. Dayan: Technical Note: Q-learning. Machine Learning, 8, (1992), 279–292, DOI: 10.1023/A:1022676722315.
Go to article

Authors and Affiliations

Jakub Musial
1
Krzysztof Stebel
1
Jacek Czeczot
1

  1. Silesian University of Technology, Faculty of Automatic Control, Electronics and Computer Science, Department of Automatic Control and Robotics, 44-100 Gliwice, ul. Akademicka 16, Poland
Download PDF Download RIS Download Bibtex

Abstract

This article accounts for the development of a powerful artificial neural network (ANN) model, designed for the prediction of relative humidity levels, using other meteorological parameters such as the maximum temperature, minimum temperature, precipitation, wind speed, and intensity of solar radiation in the Rabat-Kenitra region (a coastal area where relative humidity is a real concern). The model was applied to a database containing a daily history of five meteorological parameters collected by nine stations covering this region from 1979 to mid-2014.
It has been demonstrated that the best performing three-layer (input, hidden, and output) ANN mathematical model for the prediction of relative humidity in this region is the multi-layer perceptron (MLP) model. This neural model using the Levenberg–Marquard algorithm, with an architecture of [5-11-1] and the transfer functions Tansig in the hidden layer and Purelin in the output layer, was able to estimate relative humidity values that were very close to those observed. This was affirmed by a low mean squared error ( MSE) and a high correlation coefficient ( R), compared to the statistical indicators relating to the other models developed as part of this study.
Go to article

Authors and Affiliations

Kaoutar El Azhari
1
ORCID: ORCID
Badreddine Abdallaoui
2
Ali Dehbi
1
ORCID: ORCID
Abdelaziz Abdalloui
1
ORCID: ORCID
Hamid Zineddine
1

  1. Moulay Ismail University, Faculty of Sciences, Zitoune, 50000, Meknes, Morocco
  2. University of Oxford, Mathematical Institute, Oxford, United Kingdom
Download PDF Download RIS Download Bibtex

Abstract

Low-cost sensor arrays are an economical and efficient solution for large-scale networked monitoring of atmospheric pollutants. These sensors need to be calibrated in situ before use, and existing data-driven calibration models have been widely used, but require large amounts of co-location data with reference stations for training, while performing poorly across domains. To address this problem, a meta-learningbased calibration network for air sensors is proposed, which has been tested on ozone datasets. The tests have proved that it outperforms five other conventional methods in important metrics such as mean absolute error, root mean square error and correlation coefficient. Taking Manlleu and Tona as the source domain and Vic as the target domain, the proposed method reduces MAE and RMSE by 17.06% and 6.71% on average, and improves R2 by an average of 4.21%, compared with the suboptimal pre-trained multi-source transfer calibration. The method can provide a new idea and direction to solve the problem of cross-domain and reliance on a large amount of co-location data in the calibration of sensors.
Go to article

Authors and Affiliations

Feng Tianliang
1
Xiong Xingchuang
2
Jin Shangzhong
1

  1. College of Optical and Electronic Technology, China Jiliang University, Hangzhou, Zhejiang 310018, China
  2. National Institute of Metrology, Beijing 100029, China

This page uses 'cookies'. Learn more