Vol.:(0123456789) 1 3 Arabian Journal for Science and Engineering https://doi.org/10.1007/s13369-019-04071-7 RESEARCH ARTICLE - SYSTEMS ENGINEERING A Novel FMEA Model Using Hybrid ANFIS–Taguchi Method Semra Boran 1  · Seda Hatice Gökler 1 Received: 22 February 2019 / Accepted: 26 July 2019 © King Fahd University of Petroleum & Minerals 2019 Abstract Failure mode and efects analysis (FMEA) is a useful method to analyze and then prioritize failure, but it has many draw- backs. First of them is risk factors, severity, occurrence and detection, which are considered equally important but their scores may be not equal in real-life applications. Another is that the risk factor values of risk priority number of failures are usually assessed by team member in FMEA method in incomplete information and uncertainty situations. The last one is expert’s experience which is not incorporated in efective automation of the risk assessment. In this study, it is aimed to use adaptive neuro-fuzzy inference system (ANFIS) that is a soft computing method to eliminate these drawbacks. However, there are many numbers of parameters that afect the accuracy of the prediction in ANFIS structure and training phase of the model. For this purpose, the parameter values were determined using Taguchi method. A novel FMEA model using hybrid ANFIS–Taguchi method and FMEA model using ANN were applied in furniture manufacturing, and the results were compared with traditional FMEA. The accuracy of the novel FMEA model was 100% while the FMEA–ANN model was 94.118%. It is recommended to use the novel FMEA model because this model is used with insufcient and imprecise data and needs only one expert. Keywords FMEA · RPN · ANFIS · ANN · Taguchi method · Failure classifcation 1 Introduction FMEA as one of the efective risk analysis methods has been widely adopted in various felds to improve the security and reliability of systems [1]. In an FMEA application, potential or known failures that could afect a process, product, system and service quality are identifed; then, RPN is calculated to classify each failure and to defne suitable corrective actions. The failure corrections and reliability improvement actions make FMEA to be a mechanism to know where, what and why a system’s functions may fail and how they can be cor- rected reliably, before its related failures occur [2]. Since its failure prevention actions are usually done before a product is produced, the chances of occurring potential failure are reduced. However, traditional FMEA is a simple and useful method to risk evaluation and prioritization, and it still has many drawbacks. The main drawbacks have been expressed in studies in the literature [3–7]. First of them is the risk factors of RPN which are accepted equally important in traditional FMEA method. But they have diferent weights in real-life applications. Another drawback is FMEA needs experts’ opinions. The risk preferences of members (experts) of FMEA team have a signifcant impact on the risk evalua- tion results in FMEA method [1]. They assign score to risk factors according to their past experience and judgments because of not recorded or missing risk assessment infor- mation. Ambiguous, qualitative or imprecise information, as well as quantitative data, can be used in the assessment. This way leads to inaccuracy and inconsistency in RPN value. It is impossible to use crisp numbers to assign values of three risk factors. Fuzzy logic is a useful tool to reduce or eliminate this drawback of FMEA method. Fuzzy FMEA has several advantages compared to traditional methods that use numeric values [8, 9]. It is provided to evaluate the risk associated with failure directly using the linguistic terms. It gives a more fexible structure for combining the risk fac- tors. As a result, it can be the most correctly determined * Semra Boran boran@sakarya.edu.tr Seda Hatice Gökler shgokler@sakarya.edu.tr 1 Department of Industrial Engineering, Sakarya University, 54050 Sakarya, Turkey