Fault diagnosis on bottle filling plant using genetic-based neural network M. Demetgul a,⇑ , M. Unal b , I.N. Tansel c , O. Yazıcıog ˘lu d a Department of Mechatronics Engineering, Technology Faculty, Marmara University, Istanbul, Turkey b Department of Electronic–Computer Education, Technical Education Faculty, Marmara University, Istanbul, Turkey c Mechanical Engineering Department, Florida International University, 10555 West Flagler Street, Miami, FL 33174, USA d ITICU, Department of Industrial Engineering, Uskudar, Istanbul, Turkey article info Article history: Received 5 May 2010 Received in revised form 13 May 2011 Accepted 30 June 2011 Available online 12 August 2011 Keywords: Neural network Genetic algorithm Bottle filling plant Pneumatic Fault diagnosis Back-propagation algorithm abstract Timely detection of the pneumatic system problems is important in industry. Many techniques have been employed to solve this problem. In this paper, Genetic Algorithm (GA) based optimal configuration of neural networks is proposed for fault diagnostic of bottle filling systems. Back-propagation is used for neural networks algorithm. The back-propagation algorithm had six inputs and one output. A fitness function was designed to the minimize execution time of ANN model by keeping the number of hidden layer(s) and nodes as low as possible while the mean square error of estimated output error is minimized. The designed GA–ANN combination and the graphical user interface (GUI) eliminate the trial and error process for selection of the fastest and most accurate configuration. The performance of the proposed sys- tem was evaluated by using experimental data collected at a pneumatic work cell which attach caps to the bottles. The sensory data was collected at normal operating conditions and a series of faults were imposed to the system such as missing bottle, attaching nonworking bottle caps at two different cylin- ders, two air pressure problems (insufficient and low air), and not filling water. The study demonstrated the convenience, accuracy and speed of the proposed GA–NN environment. It may also be used for train- ing for selection of ANN configurations at various applications. Ó 2011 Elsevier Ltd. All rights reserved. 1. Introduction Most of the modern manufacturing facilities are automated for minimizing the production costs to be competitive at the interna- tional markets and targets 24 h operation to maximize the output [1,2]. The cost of the unexpected downtimes due to machine fail- ures are extremely high since not only a specific machine require attention but the hole or partial production flow stops, trained spe- cialists involve and scheduled deliveries delay [2]. The importance of the condition based maintenance and health monitoring has been gaining importance for predicting the incoming problems, avoiding catastrophic machine failures and stopping the entire production lines unexpectedly [3]. It is required to detect, identify and then classify different kinds of failure modes that can occur within a machine system. Often several different types of sensors are installed to the system at various locations to detect and iden- tify the problems. The health monitoring system encode the signals of the sensors and estimate the problems [4]. Pneumatic systems have been widely used for automation of repetitive tasks if the speed of the movement of the parts and/or tools is not critical. They are cheap, clean, and easy to maintain. These systems repeat a programmed sequence many times. When the system encounters a problem, generally the manufactured parts will be wasted and the cost will increase. It is necessary to detect the problems and their source as quickly and accurately as possible to continue to operate the system with minimum inter- ruption. Monitoring the wear of the pneumatic components helps taking the right decisions in timely manner for the predictive maintenance. Air consumption of the pneumatic system directly correlated with the efficiency and health of the components. The stability of the operational parameters of pneumatic system and precision of the motions are directly related to the quality of re- leased production [5]. Many researchers prefer Genetic Algorithms (GAs) when the number of parameters are high and convergence to local minima is a serious concern. During the training of Artificial Neural Net- works (ANNs) having too many parameters and local minima are two important concerns. GAs worked effectively with ANNs for selection of the most compact model and/or estimation of the parameters to improve their performance [1,4,6,7]. Genetic pro- gramming (GP) has been used for preparation of computer pro- grams automatically and have been successfully applied to various problems [8]. The best known ANN approach is the back propagation algo- rithm [9]. It estimates the parameters by using gradient descent method. This approach has two major weaknesses. These 0965-9978/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.advengsoft.2011.07.004 ⇑ Corresponding author. Tel.: +90 216 4497109 655; fax: +90 258 2963263. E-mail address: mdemetgul@marmara.edu.tr (M. Demetgul). Advances in Engineering Software 42 (2011) 1051–1058 Contents lists available at ScienceDirect Advances in Engineering Software journal homepage: www.elsevier.com/locate/advengsoft