next up previous contents
Next: Data Mining Up: Case-Based Reasoning Previous: Retain

An Example

Let us illustrate CBR with a simple example from the oil industry[*]. An operator is carrying a heavy pipe, and slips on oil spilled on the floor when he is bowing to pass below a lever on a platform. He falls, and his right hand is crushed between the pipe and the floor.

This accident can be said to have several causes. The lever should not be placed in a position so that it is not possible to pass it without bowing. The oil spilt on the floor should have been removed, and the operator carrying the pipe should have used gloves.

If the operator had the possibility of consulting a case-based reasoning tool with data on accidents on platforms before transporting the pipe, he might have found that accidents when carrying equipment has happened on other platforms, and that causes for injury mostly involve not using personal protection. Suggested solutions for the problem may be to find routes where it is easy to walk, and use safety equipment like gloves and suitable shoes. The obvious way to reuse this information would be to use gloves and find another route, and to retain it as a ``negative'' case (!) if the equipment transfer proceeded without injury.

If we also had some statistical background knowledge, that 60% of the accidents on platforms happen when only one operator performs a job, and that the most common injuries are simple physical ones that lead to a loss of an average of two working days per injury, the work could probably have been organized better by the management.

If we had background knowledge in the form of rules, we could examine cases more closely when retrieving them. Say that we have the following casebase:

ACTIVITY    DEVIATIONS                    INCIDENT    INJURY       WEATHER
==========================================================================
CARRY_PIPE  LEVER & OILSPILL & NO_GLOVES  OP_SLIPPED  HAND_CRUSHED FINE
CARRY_PIPE  LEVER & OILSPILL & NO_GLOVES  OP_SLIPPED  HAND_CRUSHED RAIN
CARRY_PIPE  OILSPILL                      OP_SLIPPED  FOOT_CRUSHED RAIN
...

When trying to retrieve cases to discover problem situations when carrying a pipe, a case-based reasoning system can use a set of rules, say:

WEATHER is unimportant BECAUSE OIL is slippery in all weather_situations.

So that even though the weather is fine the system will make use of case number two because weather is unimportant when it comes to slipperiness. This is the select subtask of the retrieve step in the CBR cycle. In this simple example, reuse , revise and retain would be useful if we store a solution together with the case. Solutions might be both preventive like ``always make sure that there are two people working, when carrying equipment on platforms'' or a solution that reduces the damage of the problem after it has occurred.


next up previous contents
Next: Data Mining Up: Case-Based Reasoning Previous: Retain
Torgeir Dingsoyr
2/26/1998