Break a wireless sensor network and find the fault

Start with one question: do all nodes go quiet at the same time? If yes, look at the gateway, its power or its uplink. If only one node goes quiet, look at that node or its radio path. The gateway measures the signal strength (RSSI) and the time of each packet. Each node sends a packet counter and its battery voltage. These values give each fault its own mark:

• Gateway uplink lost. The server gets no data from the gateway, but the gateway still receives the nodes and keeps the readings. When the link is back, the data arrives with its time stamps.
• Gateway without power. All counters show the same gap when the gateway starts again. The readings sent in that time are lost.
• Weak link. The RSSI of one node goes down and its counter shows gaps.
• Node reset. The counter of one node is back at 1. Once is an event. If it repeats, check the battery or the supply.
• Low battery. The voltage of one node goes down. Near the end, the node resets when it sends, then goes silent.

Run the Gateway power cut scenario and compare it with Battery runs down.

The dew point is the temperature at which the water in the air condenses. The simulator computes it from the temperature and the relative humidity (RH) with the Magnus formula, and shows it for each node. When the air cools at night, the RH goes up and the gap between the air temperature and the dew point gets small. Any surface colder than the dew point gets wet: walls, roofs, metal and goods.

So a limit on RH alone is not enough. This simulator also warns when the dew point of an air node comes within 2 °C of its temperature for 3 readings. In a real site, compare the dew point with the temperature of the coldest surface, for example a roof or a steel wall.

In a warehouse, part of the night rise also comes from outside air that enters through doors and vents with more water in it. Run A day and a night on the greenhouse and watch the dew point at 03:00.

Each battery node has a temperature and humidity sensor and a low-power radio. It wakes up, reads the sensor, sends one small packet and sleeps again. A GateBerry FLEX gateway receives the packets. The radio receiver can be a board on one of its four USB ports, as in the warehouse case, or the LoRa expansion board with your own LoRa protocol. For LoRaWAN sensors, use a LoRaWAN gateway and read its MQTT feed.

On the gateway, a program in Node-RED or Python checks each packet: counter, RSSI, battery and limits. It writes the data to local storage and sends it with MQTT to a broker, over Ethernet or the 4G expansion board. A database stores the data and a dashboard shows it. The GateBerry FLEX takes one expansion board: if you use it for LoRa, send the data over Ethernet or a USB modem; if you use it for 4G, put the radio receiver on USB.

The gateway has a real-time clock with a battery, so each reading keeps a correct time stamp. Its UPS with supercapacitors shuts Linux down safely when the power fails.

The nodes, the gateway and the rules stay the same. What changes is the radio path, the place of each point and the limits.

• Warehouse. A large floor with racks and dock doors. Metal racks and goods change the radio path when the stock moves. The points near the doors change first.
• Greenhouse. High humidity at night and condensation on the crop and the roof. Put the sensors at crop height, in a shield against direct sun.
• Grain store or silo. Temperature cables hang in the grain and find a point that warms up while the others stay steady. The radio node sits at the top of the silo and reads them: grain and steel walls weaken radio signals a lot.
• Field storage pile. Probes go into the pile on a cable. The radio node stays on the surface or on a stake. The inside of the pile changes slowly, the surface follows the night air and the frost. The gateway stands at the edge, with 4G for the data.

The limits of the warehouse in this simulator come from the case. The limits of the other sites are examples, not recommendations. At the grain store and the pile, the limits apply to the probes in the product, and the air nodes only warn on condensation. Set yours from your product, your process and the rules of your country.

• Warehouse temperature and humidity: 11 battery radio nodes read temperature and humidity every minute on a floor of 450 x 250 m. A Raspberry Pi gateway sends the data with MQTT over LTE to a cloud server. The case explains the radio frames, the map of the floor and the link checks. This simulator adds the gateway faults (4G link and power), the alarm rules with delay and hysteresis, the dew point and three more sites. Read the case study

What you see in this simulator is a general pattern. The layouts, the radio values and the thresholds are examples, not the design of a real site.

1. How often each node reports

Each reading costs energy. In the warehouse case each node wakes every 60 s, and two AA cells are expected to last more than a year. A longer interval gives a longer battery life, but the gateway needs more time to find a silent node.

2. A counter in every packet

A counter that only goes up costs a few bytes and tells you three things. A gap means lost packets, the same number means a duplicate, and a counter back at 1 means a reset. Without it, you only see that a value is old.

3. When is a node silent?

This simulator raises the alarm after 3 minutes without a packet, that is 3 missed readings. Do not alarm on one missed reading: a lost packet now and then is normal on radio.

4. A delay and a hysteresis on each limit

The alarm comes when a value stays outside the limit for 3 readings in a row. It clears only when the value is 0.5 °C or 2 % RH back inside the limit. So a value that moves around a limit does not raise and clear the alarm every minute.

5. Compare each point with the others

A limit catches a floor that is too warm. A hot spot in grain, or a heater near one point, shows earlier as a difference from the other points. Here the rule is: more than 3 °C away from the median of the points of the same group for 10 minutes. When a point is already close to its limit, the limit can come first. The comparison is what finds a hot spot that stays inside the limits.

6. Keep the data when the uplink fails

The gateway writes each reading to its storage first and sends it with MQTT after. When the 4G link comes back, it sends the buffer with the original time stamps. On the server, one alarm says that the gateway is offline, instead of one alarm for each node.

7. Tell a gateway fault from a node fault

When all nodes stop at the same time and their counters show the same gap, the cause is the gateway or its power, not the nodes. The rules hold the node alarms while the gateway is offline.

8. Warn early on the battery

With lithium AA cells the voltage stays almost flat and falls fast at the end. Alkaline cells fall slowly over their whole life. In both cases the voltage drops during each radio transmission, so a node near the end can reset when it sends. In cold sites, use lithium cells. This simulator warns at 2.4 V for two AA cells, an example value. Check the minimum supply voltage of your radio and your sensor in their datasheets.

To switch fans, heaters or vents from the same values, add a PLC next to the equipment: the GateBerry FLEX is the gateway, the PLC drives the outputs. See the greenhouse climate simulator.

Full specs: GateBerry FLEX · GateBerry datasheet.

These examples cover parts of this pattern. Some use an ESP32 PLC or a Raspberry PLC: the logic is the same.

• Central Mosquitto MQTT broker for a device fleet
• Node-RED MQTT concentrator
• Offline-first MQTT data logging with an SD card
• MQTT to SQL Server bridge in Python
• Payload bit packing for small radio packets
• FreeRTOS tasks with a LoRaWAN watchdog
• Hardware watchdog for unattended devices
• Remote site telemetry simulator: alarms from remote sites in a water network

More examples: PLC Solution Library · all simulators.