Determining optimal quality distribution of latex weight using adaptive neuro-fuzzy modeling and control systems q Osman Taylan ⇑ , Ibrahim A. Darrab King Abdulaziz University, College of Engineering, Industrial Engineering Department, P.O. Box 80204, Jeddah 21589, Saudi Arabia article info Article history: Received 23 May 2009 Received in revised form 31 July 2010 Accepted 2 May 2011 Available online 8 May 2011 Keywords: Fuzzy logic control Quality control Carpet industry ANFIS Latex weight abstract This paper introduces a systematic approach for the design of an adaptive neuro-fuzzy inference system (ANFIS) for latex weight control of level loop carpets. In high production volume of some industries, man- ual control could lead to undesirable variations in product quality. Therefore, process parameters require continuous checking and testing against quality standards. One way to overcome this problem is to use statistical process control by which a complete elimination of variability may not be possible. Fuzzy logic (FL) control is one of the most significant applications of fuzzy logic and fuzzy set theory. Fuzzy if-then rules (controllers) were developed in a systematic way that formed the backbone of the neuro-fuzzy con- trol system. The developed ANFIS was able to produce crisp numerical outcomes to predict latex weights. The neuro-fuzzy system behaved like human operators. ANFIS outcomes were encouraging because they provide a more efficient and uniform distribution of latex weight and seemed to be better than the other statistical process control tools. FL controllers provide a feasible alternative to capture approximate, qual- itative aspects of human reasoning and decision making processes. Ó 2011 Elsevier Ltd. All rights reserved. 1. Introduction Fuzzy logic control (FLC) has gradually been recognized as the most significant and fruitful application of fuzzy logic and fuzzy set theory. Indeed, for complex-ill defined systems that are not easily subjected to conventional automatic control methods, fuzzy logic controllers (FLCs) provide a feasible alternative since they can capture the approximate, qualitative aspects of human reasoning and decision making processes (Jang & Sun, 1995). However, with- out adaptive capability, the performance of FLCs relies exclusively on two factors: the availability of human experts, and the knowl- edge acquisition techniques to convert human expertise into appropriate fuzzy if-then rules and membership functions. On the other hand, investigation into using neural networks in auto- matic control systems did not receive much attention until the back-propagation learning rule was introduced and the number of neural controllers design methods have been proposed (Jang, Sun, & Mizutani, 1997; Pedrycz, 1993). Manufacturing industries are in the midst of a new technological revolution, characterized by the increasing advances and application of information and communication technology for both automatic control and deci- sion support systems (Swanepoel, 2004). To construct a fuzzy con- troller, one needs to perform knowledge acquisition which takes a human operator’s knowledge about how to control a system and generates a set of fuzzy if-then rules as the backbone for a fuzzy controller that behaves like the original human operator (Chowdh- ury & Li, 1998; Labib & Yuniarto, 2005; Mendel, 2001). Usually we can obtain two types of information from a human operator: lin- guistic information and numerical information (Jang & Sun, 1995). In this study, numerical data were obtained and recorded by the domain engineer for the desired input–output data pairs. These data pairs were used as a training data set in construction of fuzzy controllers. As it is well known, considerable work related to the product and process control is still carried out manually in most of the manufacturing industries. Manual product and process control is difficult, stressful, unreliable, subjective, and time con- suming. In a very high production volume of carpet industry, man- ual control of process parameters might lead to undesirable variations in the evaluation of product quality. However, the carpet industry requires continuous checking and testing of process parameters against quality standards. In the carpet industry, inspections are usually made manually for tufting, pile height, yarn type, primary specifications, carpet design, and latex weight, etc. There are usually three types of carpet groups: The Tip Shear, level loop carpet group and 1000 carpet group. Latex weight is the sub- ject of this study and is an important quality parameter affected by certain input parameters. Latex weight is the last stage to finish the carpet production process where the printed and dyed carpet is coated. The main raw materials used in backing process are latex compounds and secondary materials. Latex compound is needed 0360-8352/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.cie.2011.05.002 q This manuscript was processed by Area Editor E.A. Elsayed. ⇑ Corresponding author. Tel.: + 966500031056. Computers & Industrial Engineering 61 (2011) 686–696 Contents lists available at ScienceDirect Computers & Industrial Engineering journal homepage: www.elsevier.com/locate/caie