Safety & Reliability
How weather, incoming power, roof states and independent emergency chains protect the equipment.
Weather and mains safety
What “safe” actually means
An unattended roll-off-roof observatory must interpret weather, power and hardware state together. AAG CloudWatcher Solo evaluates sky/cloud conditions and other environmental parameters. N.I.N.A.’s Safety Monitor also reads the UPS/incoming-mains status. If mains power is lost, the monitor becomes Unsafe even when Solo weather is still Safe and starts the N.I.N.A. safety sequence to park the mount and close the roof using the UPS reserve, before that reserve can be depleted. Home Assistant independently reports UPS and power conditions and checks that the roof reaches the closed limit switch. A separate Hydreon RG-11 rain sensor, wired to a Shelly digital input, provides another emergency path.
The roof itself has at least two end-position observations: fully closed (FCCL) and fully open (FCOP). During travel neither end contact necessarily indicates a completed position. A closed command is not proof of a closed roof; a software notification alone is not proof either. The mechanism also exposes a motor-activity input. The physical STOP button is part of the electrical roof-control circuit and interrupts movement independently of N.I.N.A., Home Assistant and their software.
Home Assistant presents these facts separately: the Overview can show the Solo as UNSAFE while the roof is independently confirmed CLOSED, alongside alarm, UPS, door and individual weather states. The weather verdict and the roof’s physical end position are different observations.
Normal Unsafety: N.I.N.A. responds first
During regular imaging, N.I.N.A. has a When Unsafe recovery path. An Unsafe transition caused by weather or loss of incoming mains can interrupt the exposure in progress; a ten-minute Hα exposure is not necessarily allowed to finish. The astronomical sequence then carries out the planned safety operation: move/park the mount as required, close the roof and remain in a waiting state. On mains loss, this allows the roof to close while the UPS still has reserve power.
When the N.I.N.A. Safety Monitor returns to Safe after weather and power conditions recover, a continuous 600-second (10-minute) confirmation is required in the N.I.N.A. sequences before the reopening path is considered. A new Unsafe condition during confirmation prevents treating a brief improvement as stable recovery. The reopened sequence then reconnects/repositions the equipment, performs the necessary preparation and returns to the still-relevant imaging blocks.
This ten-minute Safe-to-reopen qualification is different from Home Assistant's five-minute closure-backup delay. One prevents opening too early; the other detects that the ordinary closing procedure has failed.
Reopening is inhibited when fewer than approximately 30 minutes remain before nautical dawn, preventing a lengthy reopen-and-refocus cycle close to morning.
Independent Home Assistant backup
Home Assistant's Unsafe backup: one five-minute deadline
A dedicated Home Assistant automation supervises Solo Unsafe events. It has four kinds of trigger: the Solo becomes Unsafe; the roof leaves its closed limit switch; the physical STOP is released; or remote maintenance is disabled. The extra triggers matter because conditions may already be unsafe when a manual exclusion is removed or the roof begins to move.
At the initial check, all of the following must apply:
- CloudWatcher Solo is reporting Unsafe (
0in Home Assistant). - The roof's closed limit switch is not asserted.
- The physical STOP is released.
- Remote maintenance mode is not active.
Home Assistant waits five minutes regardless of whether N.I.N.A. is taking pictures. N.I.N.A. can interrupt the current exposure on Unsafe and normally has time to park the mount and close the roof before the independent backup intervenes.
After five minutes, HA reads the conditions again. If the weather is now Safe, the automation ends. If the weather remains Unsafe but the roof has reached FCCL, no emergency restart is needed and an “Unsafe control completed” notification can be sent. If Unsafe persists, the roof remains not confirmed closed and the two maintenance exclusions remain clear, HA requests the emergency script.
The Home Assistant rule uses restart mode. A new trigger can restart the five-minute timer. The rule also relies on recently updated limit-switch information.
Direct rain detection
Separate fast-path emergency: RG-11 rain detection
The Hydreon RG-11 is not the Solo's rain channel: it is an optical rain sensor physically installed inside the observatory, connected to a Shelly Plus i4 input. Water detected inside means it has entered the enclosure and the instruments may already be getting wet, so this is treated as an immediate equipment-protection emergency. In this installation OFF corresponds to the rain-alarm condition. Home Assistant starts emergency closure as soon as OFF has persisted for 10 consecutive seconds, provided the roof is not confirmed closed. Shorter disturbances are filtered by this persistence condition.
When both conditions apply, HA sends a high-priority RAIN INSIDE Telegram message and starts the emergency script immediately; it does not wait for the normal five-minute Solo Unsafe backup. The RG-11 has its own internal operating modes and relay-hold behavior, so a two-minute electrical alarm does not prove that rain continuously fell for two minutes.
The RG-11 processes optical wetness events using its internal relay timing. The 10-second HA filter avoids reacting to brief electrical disturbances; once it is met, the internal sensor's location makes any alarm a direct indication that water may be reaching equipment. The sensor relay may remain active longer than the water contact itself.
Roof movement and personnel safety
Mount-position interlock and movement warning
Before any roof movement commanded by N.I.N.A.—in either the main sequence or the emergency sequence—the sequence activates a laser aimed at the 10Micron mount's parked position. When the laser detects the parked mount, it changes a Denkovi DI; that DI activates a Denkovi relay, whose contacts close an electrical permissive circuit for the separate relay that sends the roof-motion pulse. If the laser does not detect the mount in its parked position, this circuit remains open. Thus a roof pulse requested from either N.I.N.A. or Home Assistant cannot reach the roof motor controller unless the laser permissive is electrically present; it is not a software condition that either application can override. The interlock remains in force during rain emergencies: the roof must not strike the telescope, even if water may enter while the mount is being brought safely to park. N.I.N.A. deactivates the laser after roof travel.
An independent IR safety barrier is installed inside the observatory, parallel to the roof rack and at approximately the height of an operator's head. The moving bar that engages the roof end stops and motor limit switch could strike someone standing in its path; if the barrier detects a person in that area, its electrical circuit interrupts roof movement. The barrier and its safety circuit use the roof motor's 24 VAC movement supply through a dual-pole relay wired with the motor and sensor circuit, so they are energised when the roof moves. An indoor buzzer and an exterior flashing light also warn that the roof is in motion; they switch off when it stops. These personnel protections are electrical and independent of application software.
The laser interlock protects the telescope by physically enabling or blocking the roof pulse; the IR barrier, physical STOP and audible/visual warnings protect people during travel. These electrical safety paths do not depend on N.I.N.A., Home Assistant or network software. An application or network software failure does not bypass the electrical permissive; roof movement still requires the hardware conditions to be met.
Loss of controlled outputs
Independent fast-path emergency: all Denkovi DO off
A separate HA entity aggregates the state of the Denkovi digital outputs. If all outputs are reported off for 30 consecutive seconds, and the roof is not confirmed closed, Home Assistant sends an alert and requests the same emergency script.
This is intended to catch an abnormal loss of essential controlled outputs while the roof is open. The condition does not mean that the controller itself is necessarily dead; a communications failure could also leave HA with an outdated view. The health of the Denkovi Python bridge must therefore be evaluated alongside any such indication.
The direct rain and all-output emergency paths do not use the maintenance and STOP exclusions that are applied to the ordinary Solo Unsafe backup. They remain separate, higher-priority triggers.
Emergency procedure and roof confirmation
Emergency execution chain
The HA emergency script requests an action through the server's HASS.Agent. HASS.Agent launches a preconfigured Windows scheduled task, which stops the normal N.I.N.A. process, pauses briefly and starts a fresh N.I.N.A. instance with a dedicated emergency sequence. This separates the hard emergency from the normal imaging-sequence branch, which might itself be blocked.
The emergency sequence follows this priority order:
- Load the required equipment profile and establish the necessary controller and mount connections.
- If the 10Micron mount is powered down, power it up and allow its startup period. In the sequence the conditional start branch contains waits of 10 and 60 seconds, in addition to real connection time.
- Ensure the mount is put into the required safe parked state. The sequence uses Unpark/Park, with short pauses, because the mount may be parked in a position unsuitable for the intended roof safety operation.
- Command CloseDomeShutter. The instruction is configured with up to five attempts; actual success depends on the device driver, motor mechanism and limit switches.
- Check the shutter state before sending a positive confirmation. In the emergency sequence, a conditional distinguishes closed-shutter state from a failure and can send the explicit warning that the roof remained open.
- Only after the roof-closing phase does the sequence handle accessory cover closure, remaining electrical operations, disconnections and the LIGHT-file report script.
Non-essential movements and flat-cover operations are placed after the roof-closing step, reducing time before the motor can operate. The mount must reach a mechanically safe parked position before closure begins.
Closure timing
The closing time depends on the initial mount status, connection latency, mount parking and the physical roof motor. Programmed waits specify only part of the delay; confirmation occurs when the closed-position limit switch is reached.
Telephone escalation
Telephone escalation: when Twilio is activated
Twilio is not called for every weather change or every opening/closing movement. In Home Assistant, its trigger is the transition of the emergency script itself from off to on. It then waits another five minutes and calls only if the roof's closed limit switch is still off. Telegram accompanies this escalation.
That rule applies to emergencies originating from RG-11 rain, all DO off, the five-minute Solo Unsafe backup or a direct manual invocation of the same HA script. Thus, for a persistent Solo Unsafe with failed normal closure, a telephone call is expected at roughly 5 + 5 = 10 minutes from the HA Unsafe trigger, provided no timer restarts occur and the script activation is observed. For direct rain detection the alarm starts after 10 seconds and escalation follows five minutes after the script starts.
Escalation boundary: the call is triggered by the emergency action changing from inactive to active. It does not independently watch roof motion. If that state transition is missed, or if FCCL is stale, telephone escalation is not guaranteed.
Electrical protection and maintenance
Mains supply, UPS and controlled recovery
Home Assistant monitors the power infrastructure in multiple ways. UPS status indicates operation on mains or battery; other signals indicate the presence of incoming mains and protected circuits. In the configuration, an input voltage above 253 V for five seconds generates a Telegram warning, and a return below the threshold for 30 seconds generates a recovery notification.
Separate alerts watch the observatory UPS and the house UPS. A Shelly-controlled supply feeding the observatory PC triggers a direct “power off” warning, five-minute reminders while off, and a “power restored” message. This matters because the PC can continue temporarily on its own battery even after the mains-fed supply disappears.
There is also a supervised post-outage restart path. When UPS status returns from battery operation to Online, HA waits one minute, verifies that the observatory network is available and that the observatory PC remains unreachable, then attempts to press its power button using a SwitchBot. A Telegram message records the attempt and the SwitchBot battery level is monitored separately. This is a recovery mechanism, not evidence that the PC will always start successfully.
Maintenance, reporting and real limits
The physical STOP and HA remote-maintenance flag prevent the Solo Unsafe backup from executing while maintenance is authorised. A periodic reminder is sent while a maintenance exclusion is still active. The separate urgent rain and all-DO paths should be considered independently before anyone works on the roof or mounted equipment.
Tests should distinguish a successful command, movement of the roof motor, confirmation at FCCL, and the subsequent telephone alarm if FCCL is not reached. A roof-close command might have been issued even if the mechanism did not move. Likewise, “server ping works” does not prove that N.I.N.A., HASS.Agent or the Denkovi bridge is healthy.
The aim is not to remove every pause but to put delays where they serve a measurable purpose: prevent contact bounce, allow a normal N.I.N.A. close, qualify a safe reopening and escalate a genuinely failed operation. The installation's safety architecture has been improved through repeated real tests, but it remains a system that needs periodic end-to-end verification.