XLT and Expert Systems
Expert Systems: One of the paths of the future is an Expert System approach to dealing with site alarm monitoring & maintenance. In the current and legacy RMCS systems, about 60% of alarms turn out to be false, or repetitive, or unimportant. But training humans to deal with this is not only problematic, it is expensive. The Expert System will receive an alarm with a computer, then do a healthy amount of decision making prior to calling in the efforts of a human operator. This decision process will be based on the collective wisdom of a reasonable group of highly experienced refrigeration/HVAC techs, distilled into code, or ‘Rules’. To make this practical, there must be real-time communication with the site allowing small random access questions at high speed.
- As an example, assume we received an alarm from Site DCA117 where the third circuit, fifth probe, has a temperature above alarm setpoint. An operator deals with the alarm, decides software to use (30 secs) calls back store (1 minute), walks through 3-5 screens (1 minute), then studies circuit screens (1 minute), to determine if the case is slowly recovering from defrost or if there is a real problem. Based on results, he may or may not call the store and/or it’s contractor. Result: 3 to 5 minutes, most of the process dependent on operator decisions and training, and a healthy part of this time unproductive work in the first place.
- With random access, when an alarm arrives, it contains protocol data to directly access the circuit in question. Before handing the alarm to an operator, the receiving computer connects directly to the site, asks for last defrost termination time, current value of probe, setpoint of the circuit, and probe trend (~3 seconds). Using this data, the receiver may choose to check back over 15 minutes, before involving an operator. It may decide to pass the alarm to an operator immediately. It may decide to call this a false alarm, and just toss it into a recorded bucket. It takes about 100 msec to make a decision based on established rules. When and if the alarm is passed to an operator, the pertinent data is painted on his right hand screen as fast as the alarm data paints on his left, along with an ‘Action Plan’ to follow in resolving the problem.
- The key to this operation is the random access and commonality discussed above. If an Operator must dial up, then log-in, main menu, secondary menus, etc, before getting to the heart of the matter, efforts bog down quickly. Also, if there is a different series of operations for Danfoss or CPC or ComTrol, the rules become a crazy quilt instantly. In the scenario above. You might say, “It’s only looking at the store once a minute, so that’s not difficult!’. Remember, however, that same system might be monitoring 5,000 sites, and these questions need to be answered in milliseconds, so we can get on to the next site doing the same thing. Further, the system can look at the past, see this has happened every third day for three weeks, and send a notice to ‘Site Efficiency Operations’ saying “You should look at this”.
