Elsevier

Neurocomputing

Volume 232, 5 April 2017, Pages 69-82
Neurocomputing

Multi-stage cognitive map for failures assessment of production processes: An extension in structure and algorithm

https://doi.org/10.1016/j.neucom.2016.10.069Get rights and content

Abstract

In recent decade, fuzzy cognitive map has had significant applications in the systems analysis. But in majority of recent studies, a process perspective to different issues has not been considered in cognitive maps and the whole processes have been modeled separately or integratedly regardless of the relationships between processes. However, in complex systems that include the various sub systems, considering the process approach to modeling is necessary. In this study, to solve this problem, a multi-stage cognitive map has been introduced in which concepts are in various stages and any stage is associated to other stages with a series of causal relationships, and presenting a new learning algorithm based on the extended Delta rule to train cognitive map to reach the minimum of squares of errors. Furthermore, a new approach using multi-stage cognitive map and process failure mode and effects analysis are used to validate the new cognitive map. In this approach, calculating the score for prioritizing of failures is done based on severity, occurrence, and detection factors and causal relationships of each failure with other failures is carried out by using multi-stage cognitive map instead of conventional score of risk priority number. Also, for the presented approach, three learning algorithms including non-linear Hebbian, extended Delta rule algorithm and its combination with the differential evolutionary algorithm have been compared. The case study on automotive parts manufacturing unit, provides the ability of the proposed approach in prioritizing failures using integration of multi-stage cognitive map and new proposed learning algorithm for this purpose and the analysis of failure modes and the proposed algorithm.

Introduction

In the real world, the analysis of the effective factors on a phenomenon is very complex; so that, these factors can be influenced by many factors and can affect on many others. Cognitive mapping is among the many ways that exist to show this relationship. Cognitive mapping (CM) is a way of design to express one's view of cause and effect or expert about a particular domain and then it is used to analyze the effects of policies or decisions pertaining to the realization of certain objectives [3]. This method by linking facts, values and processes to goals and policies allows researchers to predict and analyze mutual influences and performances of complex events using what-if analyses [10]. Cognitive maps have many applications in simulation, modeling, and decision analysis. The new applications of this method can be cited in medical applications such as prediction of pulmonary infection [28], long-term prediction of prostate cancer [12], strategic marketing planning for industrial firms [19], renewables local planning [18], dynamic risks modeling in ERP maintenance projects [20], studying urban resilience and transformation [24], assessment and decision support in the emergency department [13], estimating system outputs, [15], intellectual capital evaluation [2] and integrated environmental assessment [21].

The nearest work to our paper has been done by [25]. But, it is not efficient in real and practical problems. They proposed a cognitive map that there are time lag between concepts. In their paper, dummy concepts have been used in order to homogenize lags. If there are a lot of concepts and the time lag between them is large, the number of dummy concepts increases which makes the model too big, especially in real models that have a lot of concepts and training model will be difficult and inaccurate. Another study in this regard has been done by [31]. They used a type of FCM called fuzzy grey cognitive map for reliability analysis based on FMEA technique. One of the traditional reliability analysis is the inability of correctly support functional assessment in industrial equipment, because of complex and huge nature of them. They identified twenty failure causes in the transformer's active part and assessed them also six failure scenarios are simulated.

In general, due to the ability of fuzzy cognitive map (FCM) in modeling complex systems with limited data, and also being unavailable or high cost of data collection, FCMs can be introduced as a very convenient tool for modeling. This method does not have a numerical prediction capability but can show what will happen in the system based on the relationships between concepts and initial state of concepts. On the other hand, in the real world it is important to note that, all activities are affected by the activities of the previous stages; that is why it is essential to consider the correlation between the stages. However, in previous studies, the majority of cognitive maps have not considered the multi-stage view to different issues. Accordingly, the introduction of a multi-stage cognitive map (Multi-stage CM) to cover existing gap in the study of a system from a process perspective is needed.

On the other hand, to draw the FCM also time series data and experts' opinions can be used. In the cognitive map drawing approach based on experts’ opinion, the main problem is the accurate estimation of map's weights by experts and in recent years in order to overcome this shortcoming, learning algorithms are used to increase the accuracy of weights obtained and map convergence. In fact, the learning algorithms are used to increase the reliability of the decision. As a result, by using these algorithms, the problem of convergence has been solved and the efficiency of cognitive map increases. But now, most learning algorithms used in the weights correction of cognitive maps, are presented based on Hebbian unsupervised learning rules or based on the meta-heuristic algorithms. The different versions of the Hebbian algorithm have been created, including Active Hebbian Learning (AHL) and Nonlinear Hebbian Learning (NHL), but these algorithms are suffering from a lack of convergence in some circumstances. Hence, in this study, a new learning algorithm is used for cognitive maps to cover the weaknesses of Hebbian category algorithms.

As noted, the aim of this study is to investigate the process of a system to be performed based on multi-stage cognitive map. That's why the use of these techniques in a practical issue can display the capabilities of this approach. Thus, a new approach using multi-stage cognitive map and Process Failure Mode and Effects Analysis (PFMEA) is presented to calculate the score for prioritizing failures. Using conventional methods, such as failure modes and effects analysis (FMEA) can have a significant impact on ensuring the competitive and flawless launch of product in the market. FMEA in the process is a systematic approach for identifying and preventing the problem in the product and its process and also is an analytical method that relies on the prevention law before the outbreak used to identify the potential causes of failure. Several researchers have used the FMEA in different areas and the use of this method can be noted in these cases: increasing patient safety in the hospital with respect to a reduction in medical errors [8], checking the reliability of wind turbine systems to generate electricity [1], supply chain risk management [9], creating an efficient health system for quality management in the food industry [33] and risk assessment in marine engineering systems [36].

Also, some researchers to obtain useful results in improving the performance of various industries and to prevent failures, have combined the FMEA with other methods. This issue is one of the main reasons for the use of PFMEA along with multi-stage cognitive map in this research. For example, to minimize the potential failures, a hybrid approach including FMEA and fuzzy theory is presented for information security risk management and then it used in an academic research group project [32]. In another study, ranking the automotive parts manufacturing processes and the failures identified in this process has been done based on the FMEA by combining this technique with interval data envelopment analysis (DEA) model and grey relations theory [4]. In most of these studies, identifying the failures and ranking them, have been done based on the rating risk priority number (RPN). However, due to the problems of this score, a new score is needed to prioritize the failures. In this regard, [14] to achieve trusted results, have calculated the imposed costs related to any failure in the system, and the failures have been prioritized by data envelopment analysis (DEA). The results of FMEA-DEA approach based on the cost at the stone industry, show the abilities of the proposed approach. As well as to improve RPN score, [6] proposed an integrated method, combining multi attribute failure mode analysis (MAFMA) and 2-tuple representation, called generalized multi attribute failure mode analysis (GMAFMA). This integrated method was used in a TFT-LCD product of a technology company in Taiwan, to examine the nature of problems stepwise and structurally. Also, by analyzing the published research on common domains of FCM and FMEA [26], it can be concluded that this research merely uses time variables (LINGUISTIC) which has been used in FMEA method, benefiting from FCM method that has been changed to quantitative, and SOD indices have not been used.

In most of the recent studies in the field of cognitive maps, the process perspective has not been considered in different issues, and all the processes have been modeled separately and/or seamlessly without considering the relationship between them. However, in complicated systems that include various subsystems, considering the process approach is mandatory to analyze such systems. In order to solve this problem, this research introduces multi-stage cognitive mapping in which concepts are in different stages and each stage has a kind of cause-effect relationship with other stages.

A new learning algorithm based on extended delta rule will be presented later in this article to train recommended cognitive mapping in order to reach the minimum sum of squares error (SSE). The reason is that, most of the learning algorithms used in correcting weights of cognitive maps have been presented based on Hebbian unsupervised learning rules or based on the meta-heuristic algorithms so far. Meanwhile, different versions of Hebbian algorithms undergo non-convergence in some conditions. Therefore, a new learning algorithm will be used for cognitive maps in order to cover weak points of Hebbian algorithm categories.

Furthermore, since in this research it has been tried to study the process analysis of system based on multi-stage cognitive maps, we have used this method in a practical problem to show its capabilities. Therefore, a new approach using multi-stage cognitive map and Process Failure Mode and Effects Analysis (PFMEA) is presented to calculate the score for prioritizing failures. In order to calculate the scores, rating of the failures has been presented. The reason is that most of the researchers in process failure mode and effects analysis have identified failures and ranked them based on risk priority number (RPN). However, with regard to the problems of this index, there is a need for a new index for ranking failures. One of the main problems of RPN index is that it does not consider the cause-and-effect relationship between failures which can lead to changing priorities to address failures; because in reality, some failures of a system affect and/or are affected by other failures. Therefore, in this research, the obtained ranking from the suggested approach based on multi-stage cognitive mapping method is recommended for ranking failures in order to increase accuracy in ranking. In the recommended approach, priority ranking of failures is calculated based on three criteria, severity, probability of occurrence, probability of detection and cause-and-effect relationships of each failure with other failures by using multi-stage cognitive map index instead of ranking priority number of failures (RPN) in a way that in combining multi-stage cognitive map and PFEMA, each stage is an index of production stage, or in other words, is an index of the place of failures in the system being analyzed. Generally, by comparing the results of multi-stage cognitive map and traditional index of RPN, it is revealed that prioritizing according to the suggested method is closer to reality, because when the relationships between failures are considered, there is a priority to address the failures which in addition to high RPN have more effect on other failures upon occurrence. These explanations show that considering SOD and failure interrelationship criteria simultaneously can provide the management with more realistic results. Also, for the suggested approach, three learning algorithms including nonlinear Hebbian algorithm, extended Delta rule, and their combination with differential evolutionary (DE) have been compared and it has been shown that the resulted scores from suggested learning algorithm based on extended Delta rule provide the management team with more capability of differentiation for prioritizing failures.

Briefly, as mentioned above, all methods related to the RPN, focus the improvement efforts on the mode of failure since having higher RPN might lead to less deterioration in comparison to other failures that have lower RPN. Another basic problem of RPN score is disregarding the internal relationships between failures which can lead to changing priorities to address failures. Because in fact, some of the failures are affected by other failures or influence the others. So, in this study, in order to enhance the prioritization accuracy of risk prioritization score, the score derived from the proposed approach based on the multi-stage cognitive map is suggested to prioritize failures. As well as, to validate the mentioned score, prioritizing the existing failures in automotive parts manufacturing processes that have been identified by systematic analyzing of failures, is done by the obtained score of the multi-stage CM-PFMEA and provided learning algorithm is based on the extended Delta rule.

The paper structure proceeds in this way in Section 2, application background of research methodology including the introduction of cognitive map and PFMEA is investigated. Then, in Section 3, the proposed approach of research is presented. Section 4 introduces case study and in Section 5, the results of the analysis of implementing the proposed approach for case study are stated. Finally, in Section 6, conclusions and suggestions for the future of the research will be presented.

Section snippets

Fuzzy cognitive map

Cognitive map is a way to reveal the structure and content of the mental process of individuals. In fact, this method simplifies the people's understanding that directs and controls the decision-making process at the individual level by presenting a model. Fuzzy cognitive map (FCM) is also a cognitive map that can use the relationship between the components of a "mental vision" to calculate the "effect power" causal relationships with a number in the range [0,1] or [−1,1] [17]. In fact, FCM to

Multi-stage cognitive map

Cognitive map is a technique that has been developed over time and during its use, the benefits of using it in operational researches have become evident in different fields. These fields include helping to solve the problem by organizing complex or mass data, helping interview process by increasing understanding and controlling the applications and managing large amounts of qualitative data. In most previous studies, cognitive maps have not been considered to investigate the various problems

Case study

In this section, we present a case study to validate the proposed approach in prioritizing the failures of production process. Since the PFMEA technique mainly focuses on failure modes effects in the process that take advantage of defects and imperfections of assembling and manufacturing process, it has great importance in the production process of automotive parts. Accordingly, the case study of this research is one of the providers of automotive parts in Iran that is active with private

Analysis of the results

In this section, the performance of multi-stage cognitive map and the new approach is presented to prioritize the failures and to implement them in the case study. As noted, this case study shows 11 stages of manufacturing process of a part of the automotive. In this section, the analysis of the proposed approach for prioritizing the failures is done in three modes of the cognitive map and three different algorithms of network training. By comparing these cases, the ability of the proposed

Conclusion

The purpose of this study, with respect to the existence gap in the literature review, was the evaluation of complex systems that are defined in the real world as a stage. Therefore, for the investigation of such systems, multi-stage cognitive map approach was introduced and then it was attempted to show the application and ability of this type of cognitive map. That is why a new approach using multi-stage cognitive map (multi-stage CM) and failure modes and effects analysis in the process

Mustafa Jahangoshai Rezaee received BSc in Applied Mathematics from Kharazmi University, Iran in July 2003, MSc in Industrial Engineering: Social-Economic Systems Engineering from Tehran University, Iran in July 2005, and PhD in this field from Iran University of Science and Technology in January 2013. Prior to his PhD, He worked in Bid-Boland Gas refinery and was a project planner and consultant in National Iranian Gas Company (NIGC) for several years. He is currently an assistant professor at

References (36)

Cited by (0)

Mustafa Jahangoshai Rezaee received BSc in Applied Mathematics from Kharazmi University, Iran in July 2003, MSc in Industrial Engineering: Social-Economic Systems Engineering from Tehran University, Iran in July 2005, and PhD in this field from Iran University of Science and Technology in January 2013. Prior to his PhD, He worked in Bid-Boland Gas refinery and was a project planner and consultant in National Iranian Gas Company (NIGC) for several years. He is currently an assistant professor at Urmia University of Technology, West Azerbaijan Province in Iran since 2013. He has published and reviewed the numerous papers in scientific journals in Elsevier, Springer and IEEE. His research interests include Expert Systems, Artificial Intelligence, Operations Research and Systems Modeling and Analysis.

Samuel Yousefi received the BSc and MSc degree in Industrial Engineering from Urmia University of Technology, Iran, in 2012 and 2014, respectively. He is now a research assistant in Urmia University of Technology, Iran. His research interests include Operations Research, Cognitive Maps, Performance Evaluation and Risk Measurement.

Mahdieh Babaei received her BSc degree in Industrial Engineering from Tabriz University, Iran in 2012, and her M.S. degree in Industrial engineering from Urmia University of Technology, Iran in 2015. She is currently a junior lecturer in Payame Noor University (PNU) of Urmia, Iran. Her research interests include Supply Chain management, Robust Optimization and Risk Measurement.

View full text