میزان شکست شناسایی محصول بر اساس طبقه بندی شبکه های بیزی مشروط
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|29116||2011||8 صفحه PDF||سفارش دهید||محاسبه نشده|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Expert Systems with Applications, Volume 38, Issue 5, May 2011, Pages 5036–5043
To identify the product failure rate grade under diverse configuration and operation conditions, a new conditional Bayesian networks (CBN) model is brought forward. By indicating the conditional independence relationship between attribute variables given the target variable, this model could provide an effective approach to classify the grade of failure rate. Furthermore, on the basis of the CBN model, the procedure of building product failure rate grade classifier is elaborated with modeling and application. At last, a case study is carried out and the results show that, with comparison to other Bayesian networks classifiers and traditional decision tree C4.5, the CBN model not only increases the total classification accuracy, but also reduces the complexity of network structure. Research highlights ► CBN introduces the conditional independence relationships among attribute variables. ► CBN provides an effective approach to classify the failure rate rank of products. ► CBN increases the classification accuracy. ► CBN makes an acceptable balance between classifier complexity and performance.
In recent years, maintenance has been playing a more and more important role in industrial fields due to the high demand for system safety, operational efficiency and life cycle cost control. In China, we cooperated with some aircraft corporations to develop a maintenance management system for years. This maintenance system daily collects bulky failure data during the airplane operation which is in different formats. The challenge faced currently is how to discover the potential failure knowledge from these data for prediction and decision making. Data mining, which is also referred to as knowledge discovery, means the process of extracting nontrivial, implicit, previously unknown and potentially useful information from databases (Witten & Frank, 2005). Depending on the types of knowledge derived, mining approaches may be classified as association rules mining, clustering, classification, prediction and others. In the area of product failure data mining, it has been used widely for the purpose of failure prediction, failure classification, and failure association. Al-Garni, Jamal, Ahmad, Al-Garni, and Tozan (2006) developed an artificial neural network (ANN) model for predicting the failure rate of De Havilland Dash-8 airplane tires utilizing the two layer feed-forward back-propagation algorithm. Using 6 years of data, the results show that the failure rate predicted by the ANN is closer to the actual data than the failure rate predicted by the Weibull regression model. Chen, Tseng, and Wang (2005) defined the root-cause machine set identification problem of analyzing correlations between combinations of machines and the defective products and then proposed the Root-cause Machine Identifier (RMI) method using the technique of associating rule mining to solve the problem efficiently and effectively. Han, Kim, and Sohn (2007) applied sequential association rules to extract the failure patterns and forecast failure sequences of Republic of Korea Air Force (ROKAF) aircrafts for various combinations of aircraft types, location, mission and season, which could improve the utilization of aircrafts by properly forecasting the future demand of aircraft spare parts. Because of the variety of each failure dataset and the diversity of each knowledge discovery mission, researchers have to build proper data mining models and processes according to the characteristic of target dataset and request. In this study, we limit the focus to product failure rate classification. Traditional product failure rate enactment is used to theoretically calculate the system reliability thanks to a static mathematical formula that ignores the actual application of each batch of products. Using the historical product failure data, we could provide a more accurate and effective classification of failure rate according to the configuration and operation. With such results, this model could satisfy the expectations of maintenance scheduling, spare parts supply chain management and product operation optimization. From recent classification literature, with the characteristics of causality and conditional independence, the Bayesian networks (BN) have been recommended as a comprehensive method of indicating relationships among and influences of variables in system reliability domains (Boudali and Dugan, 2005, Langseth and Portinale, 2007, Mahadevan et al., 2001, Muller et al., 2008 and Weber and Jouffe, 2006). It is a powerful technique for handling system uncertainty and it shows a high performance in prediction and classification tasks. Friedman, Geiger, and Goldszmidt (1997) evaluated approaches for inducing classifiers from data, based on the theory of learning general Bayesian networks (GBN) and put forward a tree augmented Naı¨ve Bayes (TAN) method, which outperforms Naı¨ve Bayes (NB), yet at the same time maintains the computational simplicity and robustness that characterize NB. Cheng and Greiner (1999) learned BN augmented Naı¨ve Bayes (BAN) and GBN using a conditional-independence (CI) based BN learning algorithm and evaluated the algorithms with NB and TAN. Experimental results show that the obtained classifiers are competitive with (or superior to) the other two classifiers. Madden (2002) introduced a new partial Bayesian network (PBN) and describes its constructing algorithm. The algorithm constructs an approximate Markov blanket around a classification node and the results indicate that PBN performs better than other Bayesian network classification structures on some problem domains. Because of the variety of collected data and application domains, researchers also have to focus on the individual case and choose the most effective classifier and modelling process. Baesens et al. (2004) compared and evaluated several Bayesian network classifiers with statistical and artificial intelligence techniques for the purpose of classifying customers in the binary classification problem. The experimental evidence showed that Bayesian network classifiers offer an interesting and viable alternative for customer lifecycle slope estimation problem. This paper is organized as follows. In Sections 2.1 and 2.2, we discuss the principle of Bayesian networks and common Bayesian network classifiers. To deal with the weakness of present BN classifiers, a new conditional Bayesian network (CBN) classifier and its modeling process are described in Sections 2.3 and 2.4. In Section 3, the case study, the performance criteria and the comparison results are presented. Finally, Section 4 concludes the paper.
نتیجه گیری انگلیسی
The paper proposes a new kind of conditional Bayesian network classifier on the basis of traditional NB, TAN and GBN model. The principle, algorithm and modeling of CBN are described in details to guide the application of identifying the product failure rate. Because it considers the conditional independence between attribute nodes and target node’s Markov Blanket given the target node, the CBN could provide a more effective classification result. The case study shows that, with comparison to the BN classifiers and decision tree C4.5 classifier, the CBN got the highest performance in the all criteria of total accuracy, AUROC, Gini index and mean Lift. Although the GBN classifier had the simplest network structure, the CBN made an acceptable balance between network complexity and promotion of classification performance. It may satisfy the expectations for maintenance decision making, spare parts supply chain management and product configuration optimization.