with the collaboration of Iranian Society of Mechanical Engineers (ISME)

Document Type : Research Article

Authors

1 Department of Mechanical Engineering, Faculty of Engineering, Shiraz Branch, Islamic Azad University, Shiraz, Iran

2 Department of Mechanical Engineering, Hormozgan University, Bandar Abbas, Iran

Abstract

Introduction
Automation of agricultural and machinery construction has generally been enhanced by intelligent control systems due to utility and efficiency rising, ease of use, profitability and upgrading according to market demand. A broad variety of industrial merchandise are now supplied with computerized control systems of earth moving processes to be performed by construction and agriculture field vehicle such as grader, backhoe, tractor and scraper machines. A height control machine which is used in measuring base thickness is consisted of two mechanical and electronic parts. The mechanical part is consisted of conveyor belt, main body, electrical engine and invertors while the electronic part is consisted of ultrasonic, wave transmitter and receiver sensor, electronic board, control set, and microcontroller. The main job of these controlling devices consists of the topographic surveying, cutting and filling of elevated and spotted low area, and these actions fundamentally dependent onthe machine's ability in elevation and thickness measurement and control. In this study, machine was first tested and then some experiments were conducted for data collection. Study of system modeling in artificial neural networks (ANN) was done for measuring, controlling the height for bases by input variable input vectors such as sampling time, probe speed, conveyer speed, sound wave speed and speed sensor are finally the maximum and minimum probe output vector on various conditions. The result reveals the capability of this procedure for experimental recognition of sensors' behavior and improvement of field machine control systems. Inspection, calibration and response, diagnosis of the elevation control system in combination with machine function can also be evaluated by some extra development of this system.
Materials and Methods
Designing and manufacture of the planned apparatus classified in three dissimilar, mechanical and electronic module, courses of action. The mechanical parts were computer-generated by engineering software in assembled, exploded and standard two-dimensional drawing required for the manufacturing process. Carrier and framework of control unit and actuator mainly designed to have the capability to support and hold the hardware and sensor assembly in an easy mountable fashion. This arrangement performed feasibility of the movement and allocating of control unit along the travel length of belt above the conveyor unit.
In this work a multilayer perceptron network with different training algorithm was used and it is found that the backpropagation algorithm with Levenberge-Marquardt learning rule was the best choice for this analysis because of the accurate and faster training procedure. The Levenberg-Marquardt algorithm was an iterative technique that locates the minimum of a multivariate function that was expressed as the sum of squares of nonlinear real-valued functions. It has become a standard technique for non-linear least-squares problems, widely adopted in a broad spectrum of disciplines. LM can be thought of as a combination of steepest descent and the Gauss-Newton method. When the current solution was far from the correct one, the algorithm behaves like a steepest descent method: slow, but guaranteed to converge. When the current solution is close to the correct solution, it becomes a Gauss-Newton method. The Levenberg algorithm is:
1. Do an update as directed by the rule above.
2. Evaluate the error at the new parameter vector.
3. If the error has increased as a result the update, then retract the step (i.e. reset the weights to their previous values) and increase l by a factor of 10 or some such significant factor, then goes to (1) and try an update again.
4. If the error has decreased as a result of the update, then accept the step (i.e. keep the weights at their new values) and decrease l by a factor of 10 or so.
Results and Discussion
The study of multi artificial neural network learning algorithm by using base Levenberg–Marquardt was the best choice to estimate function experimental data convergence. Artificial neural networks databases were generated by experimental measurement data condition scales.
It has been observed that the artificial neural networks could be used in height control. The function estimation problem with parameters in Levenberg–Marquardt algorithm showed a high performance and has a high speed, the error in the most cases were decrease and show a high convergence. Sum square error between ANN predictions and experimental measurements was less than 0.001 and correlation coefficient is above 0.99.
Conclusions
ANN method was capable to predict and capture the behavior of experimental measurements.
ANN method can easily be used to determine new results with considerably less computational cost and time. Results show that the back-propagation method with Levenberg-Marquardt learning rule was suitable for training the networks.
The Sum square error between ANN predictions and experimental measurements was less than 0.001 and the correlation coefficient is above 0.99.
Replacement of the identity matrix with the diagonal of the Hessian in Levenberge-Marquardt update equation has great advantages in convergence and computation time.

Keywords

Main Subjects

1. Bird, J. 2003. Mean and root mean square values, Fourth Edittion, USA: Elsevier Limited: 457-460.
2. Cetinkunt, S. 2006. Mechatronic with Experiments, USA: John Wiley & sons Inc.: 358-372.
3. Cetinkunt, S., U. Pinosopon, C. Chen, A. Egelja, and S. Anwar. 2001. Positive flow control of closed center electro hydraulic implement by wire systems for mob equipment applications. Mechatronics 14: 403-420.
4. Darold, W., Z. Ming, and S. Vasaraghavan. 2005. An ultrasonic/optical pulse sensor for precise distance measurements. Sensors for industry conference Houston, Texas, USA.
5. Dragoi, C., N. Vasiliu, and C. Calinoiu. 2010. Experimental identification of the electrohydraulic servo systems used in speed governors for hydraulic turbines. A Doua Conferinta 1: 661-672.
6. Elsadiq, A. A. B., and A. A. B. Mohammad. 1996. Evaluating the accuracy of laser for engineering surveying, Journal of Kingsoud University Engineering Sciences 8: 121-131.
7. Figueroa, F., and A. Mahajan. 1994. A robust navigation system for autonomous vehicle using ultrasonic, Control Engineering Practice 2 (1): 49-59.
8. Hojjat, Y., A. R. Ghane, and Sh. Mirzamohamadi. 2013. Design and fabrication of a transit time ultrasonic flow meter. Modares Mechanical Engineering 13 (5): 153-156. (In Farsi).
9. Khadem, S. M. R., S. M. Khadem, A. Afsari, H. Dokoohaki. 2010. Development of a Machine-Road Simulator for Performance Investigation of Height Measuring Sensors in Earth Moving Machinery, Proceeding of first international conference on sensor device technologies and applications, Venice, Italy 10-15.
10. Levenberg, K. 1994. A method for the solution of certain nonlinear problems in least square. Quarterly of applied mathematics 2 (2): 164-168.
11. Marquardt, D. W. 1963. An algorithm for least-squares estimation of nonlinear parameters. SIAM journal of applied mathematics 11: 431-441.
12. Massa, D. P. 1987. An Automatic Ultrasonic Bowling Scoring system. Sensors 4 (10).
13. Massa, F. 1992. Ultrasonic in industry, Fiftieth Anniversary issue Proc IRE, Ireland.
14. McNamee, J. M., and V. Y. Pan. 2013. Root-Squaring Method, Studies in Computational Mathematics 16: 139-214.
15. Patterson, D. W. 2005. Artificial Neural Networks Theory and Applications. Prentice-Hall International, UK, London.
16. Shirley, P. A. 1989. An introduction to Ultrasonic Sensing. Sensors 6 (11).
17. Shraga, S., and B. Johann. 2001. Using coded signals to benefit from ultrasonic sensor crosstalk in mobile robot obstacle avoidance. 1EEE international conference on robotics and automation, Seoul, Korea.
18. Wilamowski, B., M. lplikci, S. Kaynak, and O. Efe. 2001. An algorithm for fast convergence in training neural networks, proceedings of international joint conference on neural networks, Washington DC, USA.
19. Wobschall, D., M. Zeng, and B. S. Vasaraghavan. 2005. An ultrasonic/optical sensors for precise distance measurements, in The SI con/05-Sensors for industry conference, USA, Texas, Huston.
20. Zeb, J. 2004. Mathematical modeling of a position control electro hydraulic servo-system. In second international conference of bhurban, Pakistan, bhurban.
CAPTCHA Image