Brought to you by EarthWeb
IT Library Logo

Click Here!
Click Here!

Search the site:
 
EXPERT SEARCH -----
Programming Languages
Databases
Security
Web Services
Network Services
Middleware
Components
Operating Systems
User Interfaces
Groupware & Collaboration
Content Management
Productivity Applications
Hardware
Fun & Games

EarthWeb Direct EarthWeb Direct Fatbrain Auctions Support Source Answers

EarthWeb sites
Crossnodes
Datamation
Developer.com
DICE
EarthWeb.com
EarthWeb Direct
ERP Hub
Gamelan
GoCertify.com
HTMLGoodies
Intranet Journal
IT Knowledge
IT Library
JavaGoodies
JARS
JavaScripts.com
open source IT
RoadCoders
Y2K Info

Previous Table of Contents Next


5.1. WIRELINE COMMUNICATION

There are many existing automated network management systems containing AI modules for diagnosis, repair, and service dispatching. AT&T's ACE maintenance expert system was developed in the early 1980s (Liebowitz, 1988). Today, ACE has been sold and installed in more than 100 sites. ACE is a rule-based system that assists telephone engineers in maintaining the local loop. The local loop is the part of the telephone network that connects residential or business telephones with a local switching center. ACE is a background data analyzer that does its analysis by querying the database of daily test results stored in the Cable Repair Administration System (CRAS) and looks for patterns in the data that indicate where trouble may exist in the local loop. Each output of ACE is a classification or diagnosis of the problem, along with detailed support evidence from the CRAS system. ACE uses a forward-chaining rules strategy that breaks the overall problem into independent subproblems. Each subproblem can therefore be solved independently and the results assembled into a complete solution.

In contrast, NYNEX's MAX, developed in the late 1980s, is a telephone trouble screening expert that takes customer reports as input to initiate a local loop diagnosis (Rabinowitz, 1991). MAX works on one trouble at a time and communicates with the Loop Maintenance Operation System (LMOS), just like a human user sitting at an LMOS terminal. It uses forward-chaining rules to perform its diagnosis based on electrical measurements, customer's service class, weather, and network topology information, and enters the recommended dispatch instructions on the original LMOS screen. A goal of MAX is to reduce the number of double and false dispatches. MAX's rules can be customized to local conditions by a set of parameters. MAX is running in every residence-oriented maintenance center of NYNEX.

An even more proactive approach was used by TCAF (Silver, 1995), a rule-based expert system that performs 24-hour monitoring and surveillance of the local loop in GTE's telephone network. The aim of TCAF is to identify and fix developing faults before the customer detects any problem. At the same time, TCAF is designed to be quickly reactive to such problems that cannot be foreseen. Like MAX, TCAF diagnoses faults using several sources of information, including electrical measurements, customer's service class, and network topology (no weather information). Unlike MAX relying on customer reports, TCAF combines interrupt-driven events (switch alarms) that trigger measurements, and an intelligent polling algorithm (based on prior test results, customer fault history, and loop topology) to schedule measurements, for fault discovery. There is no human input to the whole process of trouble detection. TCAF can correctly detect cable cuts and coin faults over 90% of time. It is monitoring over 8 million GTE telephone lines (out of total 18 million).

Recently, Pacific Bell adopted a different approach for the same local loop problem. The Trouble Localization (TL) system (Chen, 1996) utilizes probabilistic reasoning techniques and logical operators to determine which component has the highest failure probability. This is achieved by building a topology of the local cable network and constructing a causal (Bayesian) network model. The model contains belief of failure for each component, given their current status, history data, cable pair distribution, and connectivity to other components. The resulting system can handle the poor quality of information in databases, perform nonmonotonic reasoning, and generate a ranked list of faulty components. The TL system is a crucial part of Outside Plant Analysis System that has been deployed statewide in California.

SSCFI (Special Service Circuit Fault Isolation) is a rule-based expert system that is in operation at all GTE's U.S. sites (Worrest, 1996). Special circuits are the telephone circuits other than the regular ones in the local loop (e.g., bank ATM, or any high-capacity, hard-wired, customized circuits). They are considered more complex than regular circuits. SSCFI diagnoses problems by recursively partitioning the circuit until the responsible fault is isolated. SSCFI reads and interprets trouble reports based on the design of special service circuits, conducts analog and digital tests via remotely activated test equipment, and routes the report to the appropriate repair group with the results of its analysis. This rule-based system also has a model-based component. SSCFI reads the target circuit's design to generate an internal circuit model to select tests that maximize diagnosis quality and minimize test time.

A telecommunications network can also be thought of as having two parts: the local loop and the interoffice facilities (IOF) network. The IOF network connects the local switching centers to one another. NYNEX has developed an expert system called Arachne for planning the IOF network (Alesi, 1996). Arachne's task is to ensure that NYNEX's investment in the IOF portion of its network satisfies the forecasted demand between switching centers, while achieving the maximum benefit per dollar invested. Arachne views the IOF network in terms of four layers of multiplexed signals: DS0, DS1, DS3, and Optical. It decomposes the planning task into two types of subtasks: (1) subtasks (in DS0 and DS1 levels) in which the size of the data is large, the variation in planning styles is great, and the equipment cost of decisions is small, and (2) subtasks (in DS3 and Optical levels) in which the data size is small and the equipment cost of decisions is high. Efficient heuristics are used to make the routing decisions in the former, while optimization techniques (dynamic programming) are used to optimize the routing decisions over the entire network for the latter. This cost-effective approach of combining heuristics and optimization techniques saves the company millions of dollars in IOF planning.

Telephone companies deal with customers on a daily basis. SAR, developed by Telesoft, is an expert system that supports salespeople in selling Intelligent Network services (Liebowitz, 1995). Intelligent Network services are a class of complex and flexible telecommunication services, whose configuration can be significantly personalized by salespeople in order to fit customer needs (using special circuits). Salespeople use SAR while they interact with customers to define all the information needed to set up a service configuration. SAR is a helpdesk application that provides congruency check, completeness check, cost estimation, and checking available resources for each service order. SAR uses forward-chaining rules and interacts with databases based on standard SQL queries. SAR has been fielded since 1993.

Telecommunications databases contain hundreds of millions of customer records. Controlling uncollectables falls into the larger risk management process. Predicting or modeling uncollectables is inherently probabilistic. Therefore, AT&T's APRI system (Ezawa, 1996) uses a special type of Bayesian network model for classifying uncollectible calls, which can be constructed efficiently from extremely large databases (reading a database just five times). APRI automatically constructs graphical probability models using the entropy-based concept of mutual information to select nodes and links of the Baysian networks. Given 800 million bytes of test data, the resulting models correctly classified 37% of the uncollectable calls, compared with only 10% by other approaches.


Previous Table of Contents Next

footer nav
Use of this site is subject certain Terms & Conditions.
Copyright (c) 1996-1999 EarthWeb, Inc.. All rights reserved. Reproduction in whole or in part in any form or medium without express written permission of EarthWeb is prohibited. Please read our privacy policy for details.