Wireless vs wired energy monitoring for retrofits

Choose the data path point by point. If the main incomer already has a Modbus meter on a working RS-485 bus, read it there. If a distribution board three floors away has no meter and no cable route, a wireless meter avoids the cable, the containment and the fire-stopping at each floor.

This guide is part of the metering and CTs series.

Three decisions for each point

For each measurement point, decide three things:

1. Measurement. What quantity, and which meter or sensor measures it?
2. Power. Does the device take power from the circuit that it measures, from a separate supply, or from a battery?
3. Data path. How do the readings reach the gateway: a radio network, an RS-485 bus, Ethernet or a pulse input?

The choices constrain each other. A battery device cannot report often, and on a Zigbee network it does not route traffic for other devices. A mains-powered meter can do both. The data path does not change the measurement. Accuracy comes from the meter and its current transformers: for example class 1 to IEC 62053-21 or class 0.5S to IEC 62053-22 for the meter, and IEC 61869-2 for the CTs. Many meter datasheets give the class for the meter alone, and the CT error adds to it.

Compare the installation work

The meter and CT work is the same in both columns. The difference is the data cable. For each point, count the cable length and containment, each riser or fire-compartment wall that needs fire-stopping, the switch port and IT approval for an Ethernet meter, out-of-hours access in an occupied building, and commissioning time.

An RS-485 bus is daisy-chained. One cable serves up to 32 unit loads over up to 1,000 m at 9,600 baud on AWG 26 or thicker cable (Modbus serial line guide). The wired cost therefore rises with the distance between points more than with their number. M-Bus (EN 13757), the two-wire bus on many heat and water meters, is also shared by every meter on one cable. Ethernet needs one run and one switch port for each meter. The energy monitoring payback calculator puts the hardware and installation cost against the expected saving.

Choose the network for the workload

Zigbee runs at 2.4 GHz, and every device on one network shares a 250 kbit/s channel. A report of about 60 bytes is on air for about 2 ms, and each hop costs about 4 ms of channel time once the acknowledgement and the random back-off are added. Take 200 meters that report every minute over an average of three hops: they use about 40 ms of every second, or 4% of the channel. At 5-second reporting the same network needs about 48%, and a CSMA-CA network loses frames long before the channel is full. Keep most points at 1 minute or slower, and use faster reporting only where the analysis needs it. Mains-powered devices normally act as routers and relay traffic for their neighbours. Battery devices are end devices and do not. The mesh therefore grows with the mains-powered installation and can route around a failed router. Steel is the main obstacle: a closed steel enclosure with the radio inside it can stop the link completely.

LoRaWAN uses sub-GHz bands (863 to 870 MHz in Europe) and a star topology. Each device talks directly to one or more LoRaWAN gateways. Range is long and airtime is scarce. The default EU868 channels sit in sub-bands with a 1% duty-cycle limit under ETSI EN 300 220-2. A reading with a 10-byte payload is on air for about 62 ms at SF7 and about 1.5 s at SF12. At SF12 the device must then wait about 147 s before it transmits again in that sub-band. The public The Things Network sandbox also limits each device to 30 s of uplink airtime a day, which is about 20 readings a day at SF12. LoRaWAN suits a 15-minute or hourly reading from a remote meter. It does not suit 1-minute power data. The Zigbee and LoRaWAN guide compares the two in more detail.

A polled RS-485 bus has a worst case that you can calculate. Take 12 meters on one bus at 9,600 baud with even parity, which gives 11 bits a character. The master reads 20 registers from each meter with one request. The request is 8 bytes (9.2 ms) and the response is 45 bytes (51.6 ms). Add a 20 ms meter response delay and the 3.5-character frame gap (4.0 ms). One exchange takes about 85 ms, and one cycle of the bus takes about 1.0 s. At 19,200 baud the cycle falls to about 0.63 s. A meter that does not answer costs the full timeout on every attempt: with a 1 s timeout and two retries, one dead meter adds 3 s to each cycle. The Modbus RTU timing calculator does this calculation for your bus, and the RS-485 wiring guide covers termination, bias and addressing.

What happens when a link fails

For each point, decide what the data must show when its link fails.

Late data needs three checks at the receiving system. It must keep the measurement timestamp, not the arrival time (source time and arrival time). It must remove duplicates when a buffer resends. It must recalculate any interval totals that it has already published.

Test each failure at the pilot site: one device off, the gateway restarted, and the upstream link down for longer than the reporting interval. Record what the data shows. The data storage guide explains the difference between a local record and a delivered one.

Security and joining

Zigbee 3.0 encrypts network traffic with AES-128 under a network key. A new device can join only while the network accepts joins. Open joining for a few minutes, pair one device, check its data, and close joining again. LoRaWAN devices hold their own AES-128 root keys and join through a LoRaWAN network server, which then holds the session keys. Decide who owns that server and its accounts. Modbus RTU has no authentication or encryption. Any device on the bus can read and write registers, so control physical access to the cable and to the gateway.

Plan the pilot

1. List every point with its measurement, power, data path and reporting interval.
2. Choose the hardest location for the pilot: the basement plant room, the metal-clad board, the far building. A pilot in an open office does not show how the mesh behaves through a basement slab or a closed steel panel.
3. Survey the radio at the real device positions, with the panel doors closed. IEEE 802.15.4 requires a 2.4 GHz receiver sensitivity of −85 dBm or better, and most current chips reach about −100 dBm. Keep at least 15 dB between the received signal and the chip's sensitivity, because doors, people and stock move. Link quality indicators are relative and differ between radio chips, so do not accept a link on one reading. Count the reports received against the reports expected over at least 24 hours at the final reporting interval.
4. Commission every point from the conductor to the report, and test the failures above.
5. Record what must change before rollout, and who owns each change. The guide to choosing an energy monitoring system turns the results into a comparison between suppliers.

The commissioning checklist lists the evidence to keep.

Wired and wireless with EpiSensor

The ZGW-20 Gateway runs Edge and brings both kinds of data into one system. One Gateway carries 250 wireless devices. EpiSensor wireless devices report over its Zigbee mesh. They link up to 50 m apart indoors and 300 m outdoors, and each mains-powered device adds about 1,000 m² of floor coverage on average. The ZEM electricity monitor is rated class 0.5S to IEC 62053-22 for the meter and its current sensors together.

Edge reads wired Modbus meters over Modbus TCP or RS-485, and polls BACnet/IP equipment on the building network. LoRaWAN sensors reach Edge through a LoRaWAN network server. Where a wired meter is far from the Gateway, a ZMB Modbus interface polls up to 30 of its registers over a short local RS-485 link and sends them over the mesh. A ZHM M-Bus interface does the same for one wired M-Bus heat meter.