Evaluation in Relation to the Brief

Overall, the final embedded system met all the requirements of the brief.

The embedded system was able to automatically collect environmental data using analogue and digital inputs, process the data with the FWI model and output the processed data. The integration between the hardware, software and model functioned well, and the high carbon level alert was effective.

One particularly successful requirement was the implementation of the FWI model. This was successful because it was developed and tested independently early on, allowing for accurate code to be written before being integrated with the rest of the system. Additionally, the usage of data sourced from the Met Eireann API allowed the model to be reliable but less demanding of the hardware.

One challenging requirement was making the embedded system's data collection reliable. I had issues with sensor reading consistency, communication dropping and having to use several interface types. To fix this, I added better error handling and logic, and validated data before inputting.

Evaluation in Relation to End User Needs

The system also successfully met the stakeholder and end user requirements.

The system provided a practical method of monitoring wildfire risk for the end users. It operated automatically, which makes it easy to use for users of all backgrounds

The high carbon level alert was useful for early detection of fires, which allows end users to quickly respond.

Limitations

Although the project met all of the brief's requirements, the final embeddded system is still a very rough prototype. In my opinion, the breadboard and development board assembly, makes it unrealistic to remotely deploy it in a forest for long-term observation. In its current state, the system is not durable, weatherproof or mechanicaly stable.

The system currently uses LTE, which I initially beleived to be a sound choice due to LoRa's need for proprietary gateways. However, after using LTE, I have found it to be unrealiable and to have high power consumption. This makes LTE unfeasible for this project, at least in the current form.

I also had issues with the GPS module, which was very unreliable in getting a signal.

The server side, while functional, would have to be heavily refined before being deployed. It lacks advanced features such as data visualisation, alerts and scalability.

Improvements and Future Iterations

To improve the project and make it feasible to be deployed as part of a real wildfire sensor network, I would implement the following improvements:

Hardware

• Design a custom PCB which would use the underlying chips and sensors from the development boards. This would make the embedded system cheaper, more relaible, less power-heavy and smaller.
• Enclose the PCB in a durable and waterproof enclosure. This would make the embedded system more reliable.
• The system could be redesigned to use LoRa instead of LTE. A LoRa network would provide long-range, low-power communication without depending on existing infrastucture. This would make the system more relaible.
• Add a passive or active GPS antenna to the embedded system. This would help the reliability of the GPS connection.

Fire Risk Model

• Attempt to use a shorter-term fire risk model, which could help localised prevention
• Implement more advanced AI/ML based modelling. This could allow for much greater accuracy than conventional mathematical models and for prediction accuracy to improve over time.

Server

• Perform fire-risk calculations server-side with the data received from the sensors. This would reduce the hardware requirements of the embedded system and make the network cheaper.
• A more advanced web-dashboard could be implemented to make the system more useful for stakeholders.
• Use a better database such as MySQL, MongoDB, MariaDB or PostgreSQL.