
In a TECH PERSPECTIVES column by Nan Johnson in Water Finance and Management, the author traces how automation in water and wastewater has changed. In the older framing, automation mainly improved equipment operation: pumps started and stopped on their own, tank levels were held in range, and treatment steps ran more consistently. The column's title states where the emphasis now sits: "Operational Visibility Has Become Just As Important As Equipment Control." For a utility, that shift has a practical consequence. Every new point of visibility also brings a limit, an alarm, and a decision about who acts on it.
Why the alarm list keeps growing
Alarm lists rarely shrink on their own. A project adds a level transmitter, a pressure sensor, or a flow meter, and the integrator configures high and low limits because the template requires them. The old alarms stay in place. Over several upgrades, a plant can end up with alarms on signals that no longer feed a control decision, alarms tied to setpoints nobody can justify, and alarms that chatter during normal events such as a pump start or a backwash. When operators see the same alarm many times a shift, they tend to acknowledge it without reading it. That habit is the real hazard, because it works against the one alarm that matters.
Sort by response owner first
An alarm is useful only when someone can name the action it triggers. Before changing any limit, build a table with one row per alarm point. Use columns for the signal, the current setpoint, the on-delay, the role that responds, the action available to that role, and the consequence if no one responds for one hour. Then sort the rows into three classes. Process-critical alarms need immediate action and a named owner on every shift. Operational alarms call for action within the shift, often a check or a valve adjustment. Informational alarms belong in the historian for trend review and should not reach an operator's screen at all.
The sort tends to expose two kinds of rows that cause trouble. The first is an alarm with no owner, where the implied response is "call the supervisor" and nobody has been told that is the response. The second is an alarm whose only available action is one the operator cannot take from the console, such as a field inspection that takes hours to schedule. Both kinds should be rewritten, reassigned, or removed.
The tradeoff in delays and deadbands
Longer on-delays and wider deadbands reduce nuisance alarms, but they also hide slow drift. A level that creeps toward its limit over several hours may never trigger an alarm if the delay is generous. Shorter delays catch drift sooner but multiply the alarms produced by transient events. No single value is correct for every loop. The right setting depends on process dynamics, hydraulic response time, and how quickly a bad condition causes harm downstream. That reasoning should be written into the engineering file rather than left in the configuration, where the next integrator will not see it.
Remote views add another layer. If a remote dashboard suppresses an alarm that the local operator interface shows, or the reverse, the on-call person and the operator may be looking at different states. Confirm that both views draw from the same alarm state before a new project goes live. Plants differ in their architecture, so the exact check depends on the vendor and the network design, and the integrator should document what each view displays.
What to document before the next upgrade
Several items are worth recording before a new sensor is added. Count alarms per shift over a representative period rather than estimating them from memory. List the ten most frequent alarms and their causes. Identify every alarm that lacks an assigned owner. Record the basis for each setpoint, whether it is a design calculation, a permit condition, or an operator's long-standing practice, since each basis implies a different review path. Note the retention period for alarm history, because a trend question that arises six months later depends on it. Finally, confirm whether the remote view and the local interface share one alarm state.
Visibility gained without this record does not reduce the workload. It moves the workload onto the people watching the screen, who are then left to decide which of many alarms deserves a response.