American Journal of Intelligent Systems 2012, 2(4): 60-65 DOI: 10.5923/j.ajis.20120204.05 Back-Propagation Algorithm Used for Tuning Parameters of ANN to Supervise a Compressor in a Pharmachemical Industry D. Benazzouz * , M. Amrani, S. Adjerid Solid Mechanics and Systems Laboratory (LMSS), M’Hamed Bougara University, Boumerdes, UMBB Algeria dbenazzouz@yahoo.fr Abstract This paper presents the retro-propagation algorithm for tuning the parameter of Artificial Neural Networks used by pharmachemical industry. The obtained numerical test results on lubrication and air circuits shown that the proposal improves the performance in terms of number of iterations and reliability of the models. BEKER Laboratories production line, is a Pharmaceutical production company located at Dar El Beida (Algiers-Algeria), was kept as the main target of this study. After careful inspection, the weakest and the strongest points of the system were identified and the most strategic equipment within the line (the compressor) was taken as the equipment of focus. From this specific point, failure simulations are most adequate and from this selected target, the designed system will be better positioned for failure detection during the produc- tion process. The efficiency of this approach is its fast learning, and its accuracy of detecting failure which is of the order of 10 -3 . Keywords Artificial Neural Networks , Industrial Diagnosis , Industrial Monitoring, Gradient back, Propagation Algo- rith ms 1. Introduction Nowadays, the preventive maintenance domain has ten- dency to become an entire part in the market. The industrial systems became increasingly complex. For that, it is neces- sary to permanently supervise them in order to prevent any incident, to detect an eventual faulty in the equipment which allow a good quality of service. Emerging preventive main- tenance domain tends to establish itself as the sole market, mostly due to the more complex growing industrial systems. Hence, permanent industrial supervision is becoming more and more vital to maintain competitive production qualities. Due to the ease of their implementation and their high re- liability[1-3], the Artificial Neural Networks (ANNs) by their nature are most suited for extremely nonlinear proc- esses. Hence, they are quiet often found within the industrial monitoring systems[5-6]. Herein, we introduce an efficient neuronal approach, which was adapted to a pharmachemical industry from BEKER Laboratories. The main task was to determine and situate strong and weak points within the production line, based on the true data generated by the sensors; which are specific to the compressor. Notice that we want to recognize * Corresponding author: dbenazzouz@yahoo. fr (D. Benazzouz) Published online at http://journal.sapub.org/ajis Copyright © 2012 Scientific & Academic Publishing. All Rights Reserved if there is any failure and also what kind of failure the system has. We want to predict the system behavior while it is op- erating. Once done, this will make the automation of the diagnosis process doable[4]. The approach is based on the gradient back-propagation multilayer network because of it contains one or more hidden layers that can treat strongly nonlinear industrial systems, which we cannot treat with mathematical approach. Moreover, it is used for its fast learning and for its ability of generalization and classifica- tion. 2. Description of Beker Workshops Laboratory BEKER Laboratories is a pharmaceutical drugs company, established in Algiers Algeria since 2005. The main com- position structure of this company is as follows: Production Line Unit Quality Assurance Laboratory Research and Development Laboratory Workshop Unit. Our study was based on the production line structure as shown in Fig.1, as it had all the required elements that apply to the objective of this paper. Within this production line the air compressor constitutes the central unit that feeds all the other parts; therefore our focus was mainly oriented in the observation of this unit represented in Fig.1.