Support for Diagnosis: Enabling Diagnosis
From DSL
Enabling Diagnostic Features in GeNIe
Diagnostic features of GeNIe are grouped into the Diagnosis Menu and Diagnosis toolbar. These features are visible only if the Enable Diagnosis option in Network Menu is checked. In addition to the menu and toolbar, the following changes occur in GeNIe:
- . New options are added to the General Tab of Node Properties Sheet.
- . The Observation tab and Documentation tab are added to Node Properties Sheet.
You can enable/disable diagnostic features of GeNIe by selecting checking/unchecking the Enable Diagnosis option as shown below:
If the Diagnosis options are enabled the Diagnosis Menu will be visible in the Menu Bar and the Diagnosis toolbar should be visible.
You can selectively display/hide the toolbar by checking/unchecking the View/Toolbars/Diagnosis option.
Setting properties of the nodes for Diagnosis
Each defect, error message, symptom, and test is represented as a separate node of a graphical causal model that is at the foundation of a Bayesian network model. Since a general purpose Bayesian network model does not make a distinction between the meaning of different types of nodes, the modeler has to indicate which of them represent defects, which are observations, and which are possible tests. For example, nodes representing components can have states labeled Ok and Defective. For nodes representing error messages, the states can be Present and Absent. For test nodes, the states could be labeled Passed and Failed. Nodes are connected together in the Bayesian network using directed links. The links typically follow the causal direction, i.e., go from the nodes that represent possible faults to nodes representing observations, error messages, tests, and symptoms. A link from a given component to a symptom can indicate that the symptom can be caused by the defects of the component. A link from a given component to a test can indicate that the test can be used to determine whether or not the component is defective.
All nodes must be Chance type nodes for diagnosis.
The Node properties dialog tab can be used to enter new properties or alter current properties for a single node. Below is a list of descriptions of those tabs that are specific for diagnosis. Description of other tabs can be found in Node Properties Sheet section.
The tabs that are specific for diagnosis are General, Observation Cost, and Documentation. Each of the tabs has three buttons on the bottom, OK, Cancel, and Help. After making changes to any of the tabs, click on OK to save the changes. To continue on with the model without saving your changes, simply click on Cancel. Please note that some changes cannot be undone, in which case the Cancel button is grayed out. For any assistance in the Node Properties window, click on Help.
General tab
The General tab is one of the most important tabs among the Node properties tabs. When a new node is added or modified, many of the characteristics and properties of the nodes are specified here.
In the General tab, the following pieces of information must be filled out. First, the user must enter the Identifier, Name. Subsequently, states must be created and, for each of these, State Name and State ID has to be entered. The definitions of these terms can be found in the section Spreadsheet View.
The node Type must be chosen among Target, Observation, and Auxiliary.
Target nodes represent faulty or defective components. In Target nodes, at least one state must be designed as a target state. A target state is the defective state of a component.
Observation nodes represent error messages, symptoms, or tests.
Auxiliary nodes are rarely used and serve for modeling convenience.
In addition to the three node types, there are also two node subtypes: Ranked and Mandatory. Out of the six combinations of types and subtypes, only four are legal: Target Ranked, Observation, Observation Ranked and Observation Mandatory.
Target Ranked indicates that the target state of a node is to appear on the list of ranked targets in the Diagnostic window. Target Ranked components are ranked in terms of their likelihood of failure. Each of the target states of a Target Ranked fault will be listed along with the probability that it is present.
Observation nodes have several combinations of subtypes possible. However, only three will be discussed because they are most relevant to modeling using GeNIe.
Observation can be set to a default state, which is assumed to be the initial state of the observation, in which case the Observation is not ranked.
Observation Ranked nodes are typically used. They are ranked according to how informative they are with respect to the faults. They represent observations whose states are unknown in advance but that are useful in diagnosis. The purpose of this node is to find out how an observation ranks relative to other possible observations before it is performed by displaying a list ranked in terms of its effectiveness in troubleshooting. These tests are ranked in the Ranked-Observation pane of the Diagnostic window.
Observation Mandatory represents information that needs to be provided before the troubleshooting is to begin. Observation Mandatory may represent an action, a testing condition, or any other factor that needs to be performed or observed before the troubleshooting begins.
The user also has the option of selecting what state the node is in. For a target node, the target state can be selected. For a default node, the default state can be selected. The user must also specify the Special Name or Format that they have chosen for the node. All of the above can also be edited in the Spreadsheet window.
We need to set the distribution for the nodes appropriately. We can change distribution by using the Change Type option from the Node Menu.
All the above conditions are represented in a table below:
| Diagnostic Type | Typical use | Ranked/Mandatory status |
|---|---|---|
| Target | To represent faulty and defective components | Ranked |
| Observation | Error messages, symptoms or tests | None / Ranked / Mandatory |
| Auxillary | No specific use, only for convenience | None |
Documentation tab
The Documentation tab, shown below, is used to enter Description, Fix, Question, and Link for node and state documentation. Definitions of these terms are given in the section Spreadsheet View.
By right clicking on the node name in the Documentation tab, the following menu will appear.
The Node question option is enabled only for Observation type nodes.
The user can select the type of information they would like to add or modify. The documentation-editing window will appear for Description, Fix, or Question, and then the user can enter their text into the window. For the links, a Links editing window will appear. In the box labeled Title, the user must enter the name of the link. Then click on the button next to Link to browse through the computer contents for the link file in a conventional window. To view the file for an existing link, simply click on the Link icon in the Documentation tab.
Please see Documentation tab section of Node Properties Sheet for detailed information.
Cost of observation
The Observation cost tab allows for entering the cost of observing the value of the current node. The cost of performing a test can be expressed on some scale, e.g., time in minutes. It is entered in the node properties window under the Observation cost tab, similarly to entering probabilities or utilities. Costs can also be entered in the Spreadsheet View.
Note : The Group Cost check box is enabled only if the current node has more than one child. For more information on how to use Group Costs see Group Costs section of Cost of Observation.
See the Observation Cost section of Node Properties Sheet for more information on the Observation Cost tab.
For more information on how to use Cost of Observation in your network see Cost Of Observation.
