First Commercial Bayesian Software for Intelligent Troubleshooting and Diagnostics
by Claus Skaanning
Dezide Advisor is the first system of its kind based on Bayesian networks - a modern approach for artificial intelligence, providing better solutions for complex and uncertain problem domains. The system is able to provide intelligent advice for solving complex technical problems efficiently and precisely, and no special expertise or skills are required to develop the underlying knowledge base.
Dezide's product, Dezide Advisor, implements state-of-the-art research within the area of intelligent troubleshooting and diagnostics, and is well suited for support and maintenance of complex industrial machinery.
Dezide is a small Danish company staffed by former employees from an R&D department of Hewlett-Packard that was 'spun off' in 2001. There are currently eight employees, including one doctorate and three Masters, all specialists in computer science with a focus on Bayesian networks.
Bayesian networks allow the structuring of complex information in a graphical network, where arrows depict the direction of causality, that is, from cause to effect. The exact computation methods of Bayesian networks then allow the user to fix any variable, and continually re-calculate the probabilities of other variables given the new observation. In comparison with fuzzy systems and other AI technologies, Bayesian networks are mathematically sound, allowing them to be used for far more complex and uncertain problems.
Dezide's software is currently being used for troubleshooting in many different areas, including Internet connection problems and email setup problems (TDC, Denmark's largest telco), network router problems (Telenor), and common IT problems (SAS). In addition to this, Dezide is currently involved in activities where the software is being used for the maintenance of complex bridge structures and diagnostics of windmill systems.
Dezide Advisor is based on Bayesian networks and implements seven patents in this field. In short, a field expert stores his knowledge of problems, resolutions and symptoms using a special authoring tool. This knowledge can then benefit end-users, technicians, and operators in trying to more quickly and precisely resolve problems. In addition, Dezide Advisor can help with detecting deviations from normal operating behaviour, can predict future difficulties based on current measurements, and can find the probable cause of a problem and then either sort it out itself or call for the necessary assistance.
Dezide Advisor is particularly well suited for troubleshooting and diagnosis in complex technical systems. In this case, Dezide Advisor will guide the user through an optimal sequence of steps leading to a solution or diagnosis in as few steps as possible. Dezide Advisor can run interactive question-answer sessions, but can also run fully automatically, embedded in technical systems.
The next step in an interactive question sequence is calculated based on all previous answers, probabilities that steps will help, and their costs (time, risk, money etc).
|
Figure 1: Troubleshooting printer problems. |
Dezide Advisor is an improvement over existing solutions in several ways. It is able to reach a solution or diagnosis in as few steps as possible and with the lowest possible cost, and when it concludes, it presents a list of the potential root causes sorted by probability. The system is very flexible, as the user is not required to follow a fixed path but can skip steps that he cannot answer, and jump directly to steps that he wants to answer straight away. In addition to this, the Advisor is self-adapting and adjusts itself to the situation over time.
Dezide Advisor is based on several breakthrough inventions. These include:
- an authoring tool that allows anyone to construct Bayesian networks efficiently for diagnostics and troubleshooting
- algorithms for calculating the optimal sequence of actions and questions towards solving a problem or finding a diagnosis
- validation tools for testing and quality assurance.
Dezide Advisor includes a diagnostic engine that can be embedded in diagnostics applications, desktop tools for constructing and maintaining the knowledge base, and a scalable, secure and robust server system for Web-based access by thousands of simultaneous users.
It is possible to learn to use the knowledge engineering tool in only two days, and no knowledge of Bayesian networks is required. The tool includes a graphical visualisation functionality to get an overview of the knowledge base (see Figure 2).
|
Figure 2: A graphical view of an example knowledge base. |
Dezide is now looking at improving its software to handle more complex problems in industrial plants. Together with ABB, SICS, Aalborg University, the University of Stuttgart, Renault and several other research and industrial partners, Dezide is a member of a consortium proposing an EU research project known as 'The Butler Concept'.
In the Butler project, Dezide will be involved in conducting general research for improved knowledge acquisition and troubleshooting algorithms within highly complex industrial plants. In particular, Dezide will work on methods for tuning the knowledge base based on expert feedback, generating easily understandable explanations of the system's suggestions and conclusions, and importing of diagnostic data from other formats such as log files, decision tress, case bases and so on.
Link:
http://www.dezide.com
Please contact:
Claus Skaanning, Dezide, Denmark
Tel: +45 9655 7370
E-mail: claus.skaanningdezide.com
|