The 02:14 Alarm: Dosing a Sump With Nobody on Site

The call came at 02:14 because the high-level float had been in alarm for forty minutes and the site had no night cover. By the time a driver arrived at 05:30, the sump had overflowed into the bund, roughly 3 cubic metres of leachate on the concrete, and the cleanup cost more than a month of polymer.

Sites that run unattended for ten or twelve hours a day need a different answer from sites with a shift chemist. Not a better dose, a different operating logic.

What an unattended site actually needs

Three things, in order. Enough storage to survive the longest unattended window plus a margin. A dosing method that cannot be got wrong by someone in a hurry at the end of a shift. And an alarm that reaches a person, not a panel in an empty control room.

Storage first, because it is the cheapest insurance. If inflow peaks at 12 m3/day and the site is unattended for 14 hours, the sump needs freeboard for at least 7 m3 plus the peak hour. A sump sized on average daily flow will alarm every wet week. In practice, adding a 20 m3 buffer tank costs less than two emergency callouts.

Batch-on-demand versus continuous trickle

The instinct on an unmanned site is to automate continuous dosing: a small screw feeder running whenever the transfer pump runs. It sounds elegant and it fails in a specific way. The feed rate is set for average strength, so weak night inflow gets over-dosed and strong first-flush gets under-dosed, and there is nobody to notice either.

A batch-on-demand arrangement is more robust. A level switch starts a transfer pump at high level and stops at low level, filling a 10 m3 batch tank. The batch is dosed as a whole by a fixed-weight feeder triggered by the same switch, with a setpoint set from the last jar test. One batch, one dose, one record. If the strength shifts, the batch still fails safe into a known volume rather than an unknown trickle.

The weakness is that a single fixed setpoint cannot follow a rain event. The mitigation is to set the dose for the strongest condition seen in the previous month and accept the slight over-dose. On an unmanned site, over-dosing by 8 percent is cheaper than a failure at 3 a.m.

Making failure visible without a person

An alarm that only lights a lamp is not an alarm. The minimum that works is a float switch wired to a dialler or a GSM module that calls two numbers in sequence and repeats until acknowledged. Add a second switch at a higher level that closes a valve or starts a standby pump, so a single missed call does not become a spill.

Instrumentation drifts, so the alarm needs a monthly test: lift the float by hand and confirm the call arrives. Sites that skip this discover on the night of the event that the SIM credit expired eight months earlier.

There is a counterintuitive point about level measurement. Ultrasonic sensors are popular and they lie in a sump with foam, condensation and a ladder in the beam path. A guided-wave probe or a float is duller and more reliable. On an unattended site, dull wins.

The handover sheet is part of the control system

If the last person on site at 18:00 doses a batch and leaves, the next person at 06:00 needs to know what was dosed, how much, and whether the sump was rising when they left. A5 sheet, five lines: level at handover, batch number, dose, powder remaining in the hopper, anything odd. Photograph it and put it in the shift group chat.

Powder remaining deserves its own line. Running out at 04:00 is the most common unattended failure, and it is entirely predictable from consumption rate and hopper capacity. Keep a physical minimum mark on the hopper, not a number in a spreadsheet.

The design question to settle before the next wet season is simple: if the site is empty for fourteen hours and inflow doubles, what happens at hour nine? If the answer involves a phone call and a two-hour drive, the storage or the automation needs work. For sites building this out, a grade that hydrates predictably at low mixing energy suits a fixed-setpoint feeder better than a fast-swelling one, and the hydration rate figures are the ones to compare.