Monitoring Setup and Customer Handover SOP

Purpose

This procedure guarantees that a finished PV system arrives at the shop's service desk as a monitored asset: a site record named the way the shop names sites, a module map proven against the roof, alerts routed to a person, and a customer who has logged in at least once with the tech standing there.

An unproven map is worse than no monitoring, because it makes the shop confidently wrong. A transposed row sends a tech to the southeast corner for a fault in the northwest, gets a healthy module pulled and shipped on a rejected warranty claim, and costs a second truck roll on top of the first. An unclaimed customer account fails the same way from the other end: the customer asks why nobody told them the inverter had been offline for five weeks, and the honest answer is that the alert went to an address nobody reads.

Scope

Covers commissioning of the monitoring platform on new residential and small commercial PV from the roof-side serial capture through the customer's first successful login. Applies to module-level platforms (microinverters or DC optimizers) and to string-level inverter portals, with the mapping steps reduced to string identification on the latter.

Does not cover physical commissioning of the inverter or array, which the installation SOP owns, nor battery configuration and backup testing, which the battery storage commissioning SOP owns and which carries its own handover. Does not cover investigating a shortfall once monitoring is live; that is the production shortfall investigation SOP, and it assumes this one was completed.

Roles and responsibilities

Procedure

Step 1 - Capture serials and physical positions on the roof, at layup, not from the box labels. Photograph each module's serial label with its grid position written on tape beside it, or scan into the platform's layout tool row by row. Acceptance: one serial per grid position, count equal to the module count on the plan, and no position blank. Wrong looks like a serial sheet transcribed from cartons on the ground, where the install order and the carton order differ. Stop rule: if a serial is unreadable after installation, do not guess it from the sequence; flag the position and resolve it in Step 4 by cover test. Hazard: this is roof work, so the fall protection system in the job's plan is rigged before anyone steps onto the roof plane, per 29 CFR 1926 Subpart M for construction-classified work at 6 feet, and module glass is not a walking surface.

Step 2 - Build the site record to the shop's naming convention, with timezone and array parameters set before any data lands. Name the site by the convention (customer surname, street number, install year) so it sorts and searches; set timezone, tilt, azimuth and DC rating. Acceptance: timezone matches the site's local zone, not the office's, and the modeled specific yield in kWh per kW-DC per year appears on the record. Wrong looks like a default timezone left in place; correct it now, because every time-shaped diagnosis the shortfall SOP performs (a collapse after a fixed clock hour reads as shading) is invalidated by a wrong clock, and back-dated data does not re-stamp when you fix it later. Hazard: a customer's utility account details create a data handling duty, so store them only where the shop's records policy allows.

Step 3 - Assign the module map row by row, in the direction the sketch reads. Enter positions in one consistent direction and note that direction on the sketch. Acceptance: every captured serial lands on exactly one position and the unassigned list is empty. Wrong looks like a leftover serial in the unassigned bin, which means either a position is doubled or a module on the roof is not reporting; stop and resolve which of the two before continuing, because they look identical on the map and only one is a hardware problem. Hazard is procedural: a map that reads plausibly is the failure mode, so do not accept a layout because the picture looks like the roof.

Step 4 - Prove the map with cover tests, one per array and one at each far corner, minimum three. With the system running, place an opaque cover over one known module, wait the platform's reporting interval (commonly 5 to 15 minutes, and the platform states its own), and confirm the tile that drops is the tile you covered. Acceptance: the covered position drops and no other position drops, on every test. Wrong looks like the wrong tile dropping, which means that row or that array was entered in the opposite direction. Stop rule: fix the direction, then re-run every test on that array plus one additional, and do not hand over on a partially re-tested map. Hazard: reaching a module means roof access under the same fall protection as Step 1, and the module stays electrically live under the cover because covering it reduces output without de-energizing it, so nothing gets de-mated during this test.

Step 5 - Set the production baseline and the alert thresholds from the model, not from the first week. Enter the modeled monthly expectation and set a monthly shortfall threshold, 15 percent below the modeled month as a common starting point, and a per-module underperformance threshold at 85 percent of array median. Acceptance: thresholds saved and visible on the site record with the modeled figures they were derived from. Wrong looks like thresholds left at platform defaults, which are tuned to avoid support calls rather than to protect your customer. Hazard: an over-tight threshold buries the service desk in noise for a month and then gets muted, which is how a real fault goes unseen; tune once after 90 days rather than at commissioning.

Step 6 - Route the alerts to a named role and prove the route with a live alert. Set the recipient to the shop's monitoring inbox, not an individual's mailbox, and add the customer only to the alert classes they can act on. Then shut the system down and confirm the comms-loss alert arrives. Shut down in sequence: AC disconnect open first, then DC disconnect, because PV connectors are generally not load-break rated and opening DC under load draws a sustained arc; restart in the reverse order, DC then AC. Acceptance: the alert lands in the monitoring inbox carrying the site name from the convention, and the service desk replies confirming receipt. Wrong looks like no alert inside the platform's stated window; stop and fix the routing, because an unproven route is an unmonitored site. Hazard: opening and closing disconnects is electrical work, so follow 29 CFR 1910.333(b)(2) practices rather than 1910.147, which excludes electrical utilization work at its own (a)(1)(ii)(C).

Step 7 - Create the customer account, have them log in on their own device in front of you, and keep installer service access. Send the invitation to the customer's own address, watch them set a password and open the site, and confirm the shop retains its installer role. Acceptance: the customer completes a login on their phone while you are present, and the access list shows both the customer and the installer. Wrong looks like a tech typing the customer's password or leaving shop master credentials on a customer device; stop and reset, because shared credentials end with the shop locked out of its own fleet when a password changes. Hazard: nothing physical, but a customer who never logs in has no way to see the system they bought, which is the trust gap this procedure exists to close.

Step 8 - Walk the system at the kitchen table and record what was shown. Show the live power reading, the daily energy curve, where the module layout lives, what a cloudy day looks like, and the one number to call. Say plainly that the system produces nothing at night and that a cloudy week is not a fault. Acceptance: the customer states back, in their own words, what a normal day looks like and who to call. Wrong looks like a tech reciting features to a customer nodding along; slow down, because the callback rate on systems handed over that way is the reason the step exists. Hazard is commercial: statements made here about guaranteed output become expectations, so describe the modeled figure as a model and route warranty questions to the written warranty document.

The record this produces

One monitoring commissioning record on the job: platform and site ID, site naming string, timezone, tilt, azimuth and DC rating, modeled specific yield, the serial-to-position map with its version and date, the direction each row was entered, every cover test with the position covered and the position that dropped, threshold values with the modeled figures behind them, alert recipients, the live-alert proof timestamp and the service desk's confirmation, the customer account address, the timestamp of the customer's first login, and the signed handover acknowledgement.

Who reads it later: the shortfall investigator, whose first step is verifying this map and who can skip a re-map when the cover tests are logged; the warranty desk, which identifies a module by position; and the next owner of the house, because a map living only in one tech's memory becomes unserviceable at resale.

Worked pass

A 22-module system, 400 W modules, 8.8 kW DC, microinverters, split into a 14-module south array and an 8-module west array.

Step 1 captures 22 serials, one per position. Step 2 names the site to convention, sets the timezone to the site's zone rather than the office's (one zone east), and records the modeled specific yield for that tilt and azimuth at 1,320 kWh per kW-DC per year, which is this site's modeled figure and not a national constant. Annual expectation is 8.8 multiplied by 1,320, or 11,616 kWh. Step 3 assigns all 22 serials with an empty unassigned list.

Step 4 fails. Four cover tests are run: south row 1 first position, south row 2 last position, west first, west last. The first three pass. Covering the west array's eighth position drops the tile labeled west position one. Stop rule taken: the west array was entered in the opposite direction to the south rows, so all eight positions are reversed. The direction is corrected, the sketch annotated, and the tests re-run on that array plus one additional, four tests on the west array, all four passing. Seven cover tests logged in total: six passing on the final map, plus the failure that found the defect.

Step 5 sets the monthly shortfall alert at 15 percent below the modeled month and the per-module alert at 85 percent of array median. Step 6 routes to the monitoring inbox, then shuts down AC first then DC; the comms-loss alert arrives carrying the site name, the service desk confirms receipt, and the system restarts DC then AC. Step 7 gets the customer logged in on her own phone with the shop retaining installer access. Step 8 runs on a partly cloudy afternoon, the better teaching day, and she states back that a jagged curve is clouds and a flat zero all day is a call.

The first clear day after handover gives an array median of 1.86 kWh per module, and no module falls below 85 percent of that, which is 1.58 kWh, so the baseline is accepted with no flags.

References

• 29 CFR 1926 Subpart M and 29 CFR 1910.28 for fall protection, applied by activity classification rather than by building type.
• 29 CFR 1910.333(b)(2) for electrical work practices, with 29 CFR 1926.417 as the construction counterpart; 29 CFR 1910.147 excludes electrical utilization work at (a)(1)(ii)(C).
• The monitoring platform's own commissioning documentation for reporting intervals, alert classes and access roles, which differ by manufacturer.
• See related: the production shortfall investigation SOP, which begins by verifying the map this procedure produces, and the battery storage commissioning and handover SOP for storage-side configuration.