Automated Decision Support for Customer Support Operations Troubleshooting Networked Printer Systems

It is a complex task to troubleshoot a printing system that consists of several components, like the application the user is printing from, the printer driver, the network connection, the server controlling the printer, the printer itself, and all the subcomponents of these components, etc. The printer industry spends millions of dollars a year on troubleshooting operations. A joint research effort (SACSO) between Aalborg University and Hewlett-Packard has resulted in complete troubleshooting systems for several HP printers. Given observed symptoms, these troubleshooting systems can compute optimal sequences of troubleshooting steps, and thereby reduce the expected cost of repair to a minimum.

Troubleshooting as e-Service Applications
The troubleshooting systems are offered to customers as a web-based e-service application. Thus, when encountering a problem with his printer system, the customer engage in an interactive session with the e-service, providing information to the service regarding the symptoms observed from his malfunctioning printer system. The service replies with a most cost-efficient troubleshooting action given the information provided. Having performed the recommended action, the customer returns with new information resulting from the action, and the service replies with the next action, and so on. This process continues until the problem has been solved or the troubleshooting system recommends service call, and all the information gathered so far will be transferred to a support agent who will continue the troubleshooting. The introduction of such e-services is expected to significantly reduce the need for customer support operations provided by human call agents, and when such agents are needed they can take advantage of all gathered information by skipping steps already performed. The troubleshooter systems may be extended with data probes that automatically gather information from the customer s environment (printer, PC, network, etc.) without involving the customer.

The Troubleshooting System
The troubleshooting systems are based on Bayesian-network models of the printer system, including all components involved in the printing process, i.e., from the application the user is printing from to the printer itself. The models describe cause-effect relationships between the components of the printer system, symptoms that can be observed given the various causes, and troubleshooting steps that can be performed to either solve the problem or gather information that can possibly indicate the cause of the problem. Thus, these troubleshooting steps can either be repair actions or questions. To each troubleshooting step is associated a cost of performing it. Given such Bayesian-network models and cost estimates, it is possible to compute which sequence of troubleshooting steps that gives rise to the lowest expected cost of repair.

A pice of the Bayesian network model - Click to enlarge (391kb)

Building a Troubleshooting System
The traditional knowledge acquisition bottleneck (i.e., the actual construction of the Bayesian network models) has been overcome in the SACSO project by construction of a knowledge acquisition tool that requires no knowledge of Bayesian-network terminology or probability calculus. In fact, after only a few hours of training with this tool, printer engineers have been able to construct a complete troubleshooting system in less than two weeks

Using a Troubleshooting System
When interacting with a troubleshooting system, the user also does not get to see the underlying Bayesian-network models, as a special-purpose GUI has been developed for the troubleshooting systems.

The BATS application - Click to enlarge (102kb)

Maintaining the Troubleshooting Models
To validate and maintain the system it is possible to simulate a user's behavior and generate a number of realistic random cases. The domain expert then evaluates these cases to check if the model is behaving correctly. If the model does not solve the problem in specific situations, or solves the problem too slowly, the domain expert can make modifications to the model.

Perspective
The SACSO project has focused on development of methods for interactive troubleshooting of complex electro-mechanical systems, construction of a generic troubleshooting environment, construction of Bayesian-network models of concrete electro-mechanical systems, and construction of a tool supporting the construction of these models. The target applications of the project have been networked printer systems, but the methodology developed is applicable for wide range of electromechanical systems and processes.