No abstract

Neural networks do not readily provide an explanation of the knowledge stored in their weights as part of their information processing. Until recently, neural networks were considered to be black boxes, with the knowledge stored in their weights not readily accessible. Since then, research has resulted in a number of algorithms for extracting knowledge in symbolic form from trained neural networks. This article addresses the extraction of knowledge in symbolic form from recurrent neural networks trained to behave like deterministic finite-state automata (DFAs). To date, methods used to extract knowledge from such networks have relied on the hypothesis that networks' states tend to cluster and that clusters of network states correspond to DFA states. The computational complexity of such a cluster analysis has led to heuristics that either limit the number of clusters that may form during training or limit the exploration of the space of hidden recurrent state neurons. These limitations, while necessary, may lead to decreased fidelity, in which the extracted knowledge may not model the true behavior of a trained network, perhaps not even for the training set. The method proposed here uses a polynomial time, symbolic learning algorithm to infer DFAs solely from the observation of a trained network's input-output behavior. Thus, this method has the potential to increase the fidelity of the extracted knowledge.

Failure Mode and Effects Analysis (FMEA) is extensively used to identify and eliminate the potential failures in products, processes, designs, and services. In this approach, the detectability, occurrence and the severity of each identified failure mode need to be determined usually by a FMEA team. This paper aims to prioritize the failure modes of rolling stocks by integrating Analytical Hierarchy Process (AHP) and Risk Priority Number (RPN). To reduce the uncertainties and ambiguities, all calculations are done in the fuzzy environment instead of the crisp values. The importance degree of each failure mode is determined and then the overall fuzzy RPN is calculated. In this way, the critical failure modes are defined to make an efficient maintenance decision.

Dragline's availability plays a major role in sustaining economic feasibility and operation of opencast coal mine. Thus, its reliability is essential for the production availability of mine. The dragline's reliability and maintenance optimization are key issues, which should seriously be considered. Draglines' unexpected failures and consequently unavailability result in delayed productions and increased maintenance and operating costs. The applications of methodologies which can predict the failure mode of dragline based on the historical dataset of failure are not only useful to reduce the maintenance and operating costs but also increase the availability and the production rate of mining machineries. In this research a historical failure dataset of a dragline has been utilized in order to analyze and conduct predictive maintenance. Authors have already utilized the K-Nearest Neighbors (KNN) algorithm in order to predict the failure mode; however, there was a chance of getting into local optimum by utilization of the mentioned methodology. In this case, combination of Genetic Algorithm and K-Nearest Neighbor algorithm (i.e. called enhanced K-Nearest Neighbors) was applied for the failure dataset, so the probability of local optimum has been decreased by application of Genetic Algorithm. In previous studies, the Artificial Neural Network methods and conventional method of K-Nearest Neighbor has been applied to the same dataset, yet the result from enhanced K-Nearest Neighbor reveals better regression analysis.

No abstract