Key Issues to Be Considered while Dealing with Artificial Intelligence Fault Management - Literature review Example

Network Management Intelligent Fault Management Group Members: SID Name 18245006 Ashish Sharma 18253812 Minh Khanh Le 17295736 Amin Dehghani Firoozabad 18054154 Rajwinder singh 16890549 Sahil Dost Table of Contents 1.0.Abstract 2 2.0.Introduction 3 3.0.Artificial Intelligence Intelligent Fault Management 4 3.1.Approaches in Fault Management Using Artificial Intelligence 4 3.2.The Process of Identifying and Diagnosing Faults 6 3.3.Application of Neutral Networks for Correlation of Alarms 7 3.4.Process of Filtering and Correlating Alarms 8 4.0.Conclusion 9 1.0. Abstract Intelligent fault management continues to grow and this means that approaches to network management has to be looked into from different perspective. This development has been complicated by the fact that computer networks are continuously growing in complexity and size. For instance, at any given time, issues such as enterprise network will consist of different hardware platforms or architectures, user interface technologies and communication protocol making the need for intelligent fault management essential. On the other hand, performance requirements and high availability have placed a high demand on proper management of such networks. There is currently need to increase efficiency which have often resulted in the decline in the ration between the network elements and network operators that are managed. This has further caused loss in service and network availability. In such cases, it has been essential to integrate or bring together an expert system into the network management architecture so as to transform events of network into value-added information that enable the operators of the network to properly manage the network. Based on this information, this paper presents critical assessment of intelligent fault management. 2.0. Introduction As studies have noted, there have been traditional network management activities like fault management that need different approaches (Varga, Jennings & Cockburn 1994). These managements have been performed with the involvement of human actions. However, as Sterritt et al. (2000) also note, these are activities that are becoming more contentious and demanding and data intensive, as a result of the heterogeneous ways of and increasing size of networks currently seen. It is for this reason that a need to artificial intelligence is required in such management. As Korbicz et al. (2012) argue that artificial intelligence will play a significant role in the process of solving problems as well as reasoning techniques that are used in the process of fault management. Just like Ramchurn et al. (2012) noted, expert systems have been reported to have been successfully applied to some types of fault management. However, there have been number of concerns that have been raised. First, enterprise computer networks are rapidly increasing in size and complexity. A typical enterprise network consists of numerous hardware architectures, communication protocols, and user interface technologies. High availability and performance requirements place a high demand on the management of such networks. The need to increase efficiency often results in the decline in the ratio between the network operators and the network elements being managed, resulting in a loss of network availability and service. In such scenarios it is essential to integrate an expert system into the network management architecture to transform network events into value-added information enabling the network operators to efficiently manage the network. In this paper we present details of an intelligent network management system which is successfully being used to manage a large Internet service provider network. Secondly, these systems have been found not to be flexible enough with regard to today’s evolving needs in network (Strasser et al. 2015). It is therefore because of this reason that there is need to propose Artificial Intelligence (AI) solution that applies both the neutral networks and case-based techniques of reasoning for the management of faults for networks. 3.0. Artificial Intelligence Intelligent Fault Management 3.1. Approaches in Fault Management Using Artificial Intelligence The more complex processes of fault management include alarm filtering and correlation, fault identification, and correction. Many of these functions involve analysis, correlation, pattern recognition, clustering or categorization, problem solving, planning, and interpreting data from a knowledge base that contains descriptions of network elements and topology. Artificial intelligence technologies are ideal for these types of functionalities. Currently, most systems employing AI technologies for fault diagnosis are expert or production rule systems (Corn et al., 1988; Joseph et al., 1989; Yamahira, Kiriha & Sakata, 1989). Many of these systems are well developed; however, they have their limitations. Generally speaking: Expert Systems (ESs) cannot handle new and changing data. Rules are brittle and not robust when faced with unforeseen situations (e.g., a new combination of alarms due to changing network topology). They cannot learn from experience (i.e., they cannot use analogy to reason from past experiences or remember past successes and failures in the context of a current problem). The rules that are incorporated at development time cannot easily adapt as the network evolves. They do not scale well to large dynamic real world domains. It is difficult, especially for technicians or operators not familiar with AI, to add new rules without a comprehensive understanding of what the current rule base is and how a new rule may impact the rule base. They require extensive maintenance when the domain knowledge changes; new rules have to be added and old rules adapted or deleted. They are not good at handling probability or uncertainty. Fuzzy logic can be employed to create fuzzy rules. However, fuzzy expert systems still have the problems discussed above. They have difficulty in analyzing large amounts of uncorrelated, ambiguous and incomplete data. The domain must be well understood and thought out. This is not entirely possible in domains such as fault management. From the issues raised above, it is apparent that there is need for application of different AI technologies that will help technicians or other individuals to overcome these difficulties. According to Travé-Massuyès (2014), these problems can also be solved as an enhancement of ESs. We therefore argue that even probabilistic approaches that are currently there such as neutral networks or to some extent Bayesian belief network can be essential in handling this case. On the other hand, studies have also mentioned that another concern for fault management is where one has AI technologies that have a positive effect, in such case; intelligent planning system or fault correction can be applied (Ferreira et al., 2016). 3.2. The Process of Identifying and Diagnosing Faults There have been tremendous growths in terms of networks as it has already been mentioned. As a concern, the process of identifying and diagnosis of faults form a critical step in fault management or application of AI in such management. As studies such as Sterritt et al. (2000) have noted, it is now essential for a fault management system to act such that is can be able to make adjustment to different changes that are likely to take place within the network elements and typology. Based on this understanding, there could be a need for the training of a neural networks (NN) or just like Sterritt et al. (2000) observed, engineers can build Bayesian belief networks (BNN) so that filtration, process of correlation and identification of general categories of faults. This is one perspective of looking at the issue. There are some cases where there is need to identify the exact location of a problem or a fault making a given procedure essential especially where there is need for appropriate tests to pinpoint the location of fault. This approach will entail a series of different decisions that are premised on human expertise meaning that it may not be possible for it to be implemented using BNN or NN alone. This understanding according to Varga, Jennings & Cockburn (1994) mean that it will be appropriate to construct a BBN or a NN that will help in the correlation of alarms and as a result, recognize different patterns in faults thus using symbolic processing like case-based reasoning (CBR). CBR may be essential in this case as it will further analyse different datasets, run different sets and where need be point exactly the location of problems for analysis. Most studies that have taken this procedure have affirmed that it is essential to integrate a hybrid Artificial Intelligence system as one of the most ideal options as it offers the diverse nature of the management of fault tasks (Travé-Massuyès, 2014). It therefore means, from a different perspective that instead of performing the whole task by just applying one technique that may not be ideal for all aspects, it is recommended that an application of different techniques be used to be able to provide the most essential solution for the problem that can be noted or found. This will in turn ensure that there is understanding of the strength of every technique and that those strengths are taken note of while the weaknesses are dealt with by the other technique. However, there are challenges in using a hybrid AI and one of such is that the process of acquiring knowledge which must be undertaken twice and in different ways. Putting this within a given context (giving example) there is need to train a neutral network with large amount of output/input data pairs and a CBR system will therefore need to be seeded with initial cases that are drawn from experts or other symbolic sources of data. 3.3. Application of Neutral Networks for Correlation of Alarms There are methods that have already been proven to be efficient and effective if other fields of studies and application that if applied in this case will enhance the process of fault management. A good case is the feedforward neutral networks that are good diagnostic approach as far as AI fault management is concerned. As a result it is possible that NNs can be a solution for problems that it has solved in fields such as medicine. Therefore relating it to issues such as network management it will serve effectively. Additionally, there have been other characteristics of multilayer feedforward neutral networks that make them perfect fit for this case. These are as follows: They have the ability of approximating any given function provided that they have been given enough neutrons They are able to relay or provide efficient method needed for analysis of incoming alarms They can deal with imperfect or ambiguous data Just like Travé-Massuyès (2014) further noted, feedforward networks are essential and necessary in this case since they have the ability to approximate any function, given enough neutrons. Additionally, their ability to generalize or learn from samples of input-output pairs. The process of learning however will be accomplished by strategically adjusting the connection weight with respect to input-output pairs. 3.4. Process of Filtering and Correlating Alarms There are about four processes that define the process of filtering alarm. These include but not limited to count, compression suppression and generalization. These process were suggested by Travé-Massuyès (2014) but have further been looked into differently by other studies. For instance, Ramchurn et al. (2012) take compression as the process of reducing multiple occurrences of the same alarm into a given unit, count is the substitution, suppression being the inhibition and generalizations being the process of referring to alarm by its superclass in such case, the superclasses are estimated by domain experts. Additionally, after filtering, the correlation of the alarms will be essential. Ramchurn et al. (2012) noted that the process of correlating the alarms may be challenging or difficult procedure especially noting that they already have some inborn ambiguity. Even when there is large volume of data, there is a possibility of significant amount of uncertainties or some inconsistencies. That is why Ramchurn et al. (2012) noted that in most cases, there can be more than one plausible reasons or explanations for the main cause of group of alarms. Giving an example, where there is non-occurrence of a remote event, there is a chance that it may cause a device waiting for that event, to time-out, with other possible causes for this lack of response. 4.0. Conclusion This report has presented key issues that need to be considered while dealing with AI fault management. Due to growing options and discussions that have been presented by scholars, the report highlighted some areas noting that there is need for comprehensive analysis so that the actual process of AI fault management can be understood. Due to the growing concerns on the best AI approach to be used, we suggest that an application of probabilistic AI technology will be essential for correlating issues such as BBNs and NNs. In as much, additional problem solving need to be applied so as the process of analyzing issues such as correlated faults, perform tests and the exact point of faults can be made easy. This study believes that with the application of probabilistic and symbolic AI techniques the process of fault management will be flexible and ability to diagnose faults will be equally easy and efficient. References Corn, P.A., Dube, R., McMichael, A.F., & Tsay, J.L. (1988) An autonomous distributed expert system for switched network maintenance. In Proceedings of IEEE GLOBECOM’88 (pp. 1530-1537). Ferreira, V. H., Zanghi, R., Fortes, M. Z., Sotelo, G. G., Silva, R. B. M., Souza, J. C. S., ... & Gomes, S. (2016). A survey on intelligent system application to fault diagnosis in electric power system transmission lines. Electric Power Systems Research, 136, 135-153. Joseph, C., Kindrick, J. Muralidhar, K. So, C. & Toth-Fejel, T. (1989) MAP fault management expert system. In Meandzija, B. & Westcott, J. (Eds.) Integrated Network Management, I. North-Holland: Elsevier Science Publishers B.V. Korbicz, J., Koscielny, J. M., Kowalczuk, Z., & Cholewa, W. (Eds.). (2012). Fault diagnosis: models, artificial intelligence, applications. Springer Science & Business Media. Ramchurn, S. D., Vytelingum, P., Rogers, A., & Jennings, N. R. (2012). Putting the'smarts' into the smart grid: a grand challenge for artificial intelligence. Communications of the ACM, 55(4), 86-97. Sterritt, R., Marshall, A. H., Shapcott, C. M., & McClean, S. I. (2000). Exploring dynamic Bayesian belief networks for intelligent fault management systems. In Systems, Man, and Cybernetics, 2000 IEEE International Conference on (Vol. 5, pp. 3646-3652). IEEE. Strasser, T., Andren, F., Kathan, J., Cecati, C., Buccella, C., Siano, P., ... & Marík, V. (2015). A review of architectures and concepts for intelligence in future electric energy systems. Industrial Electronics, IEEE Transactions on, 62(4), 2424-2438. Travé-Massuyès, L. (2014)"Bridging control and artificial intelligence theories for diagnosis: A survey." Engineering Applications of Artificial Intelligence 27: 1-16. Varga, L., Jennings, N. R., & Cockburn, D. (1994). Integrating intelligent systems into a cooperating community for electricity distribution management. Expert Systems with Applications, 7(4), 563-579. Yamahira, T., Kiriha, Y. & Sakata, S. (1989) Unified fault management scheme for network troubleshooting expert system. In Meandzija, B. & Westcott, J. (Eds.) Integrated Network Management, I. North-Holland: Elsevier Science Publishers B.V Read More