MATLAB IoT development can support a range of product architectures, including devices that cannot depend on continuous internet access or a constantly powered operating system. For industrial automation engineers and product teams, the key is not simply connecting a sensor to the cloud. It is choosing where data is processed, how algorithms are validated and how proven logic moves into production hardware.
Skyradâs work across leak detection, heating diagnostics and in-home sensing illustrates how those choices can change as a product evolves. MATLAB and Simulink were used to develop and refine algorithms, while deployment approaches were selected according to each deviceâs power, connectivity and response requirements.
MATLAB IoT development beyond the cloud-first model
A common IoT architecture places a relatively simple sensor at the edge and relies on an internet connection to send data to cloud-based analytics. That approach can make sense when a device has dependable connectivity and sufficient power. It is less suitable when a product must operate on batteries, respond locally or continue working without a network connection.
Skyrad used cloud platforms for infrastructure where appropriate, but did not treat cloud processing as a requirement for every product. Instead, the team used different architectures for different needs. MATLAB and Simulink supported algorithm development, while the production logic could run in the cloud, on a mobile device or directly on embedded hardware.
This separation between developing an algorithm and deciding where it runs gives engineering teams room to adapt. A cloud-based proof of concept can help collect and analyse data, but the final device may need local processing to meet power, availability or response constraints.
LeakBot: adapting processing to connectivity and power
LeakBot estimates water flow by using the temperature difference that can occur when incoming water is cooler than the surrounding environment. Its algorithm examines the temperature signal over time. A sustained flow may indicate a leak, although other causes, such as a dripping tap or hose, can also produce flow. The source reports that the system detects 98% of leaks above 5 ml/min.
The first proof of concept used an IoT setup so the team could develop and adjust algorithms remotely. Once the approach was proven and the client approved further development, the architecture shifted toward the needs of a battery-powered product.
Starting with standalone operation
The first production version had no connectivity. It was designed for a five-year battery life and alerted users with a beep when it detected a leak. The algorithm developed in MATLAB was hand-coded in C for the embedded hardware.
To help improve later versions, the device logged its data. Plumbers could retrieve those logs with a tool and send them back to the team. Reviewing field data made it possible to investigate classification errors and adapt future code without requiring the device to maintain a network connection.
Adding connectivity without sending everything
When clients wanted access to device data, a later version added Sigfox connectivity. The networkâs limited capacityâ12 bytes per day in this applicationâmeant that raw time-series data could not simply be transmitted. The device pre-processed the data and sent a compressed representation to the cloud.
This architecture allowed the team to tune the algorithm remotely despite the bandwidth constraint. Initial cloud rules ran in C; more advanced methods developed in MATLAB ran on MPS. Empirical and machine-learning approaches could then be compared in the cloud.
A subsequent Wi-Fi version addressed a stalled Sigfox rollout in the UK. Wi-Fi offered more bandwidth but consumed more power, so it could be active only briefly each day. Fast leaks, such as those caused by burst pipes, were detected locally. Other data was buffered and sent to the cloud every 24 hours. Running the existing, tested algorithm on MPS helped limit the revalidation needed for that version. Proposed enhancements could be tested against device logs before being moved into operation.
Using Simulink and Embedded Coder for local processing
Project Pippa, an in-home target-tracking system using a thermal imaging sensor, brought together several lessons from LeakBot. The device still needed to run on batteries and provide connectivity, but it also needed to respond faster. The team wanted to develop algorithms offline and move them into production with minimal risk and retesting.
An initial proof of concept used a Raspberry Pi sensor connected over Wi-Fi to an AWS database. A Simulink algorithm running in desktop MATLAB analysed the collected data, wrote results to an AWS table, and allowed the Pi to provide feedback to the user. The prototype gathered data in around 10 homes over 12 months.
The final architecture had to preserve useful operation when connectivity was unavailable and reduce the likelihood that the consumer would need to manage a connection. The team chose to run the algorithm on the embedded device and send outputs by SMS. Simulink models were refined using logged data, and Embedded Coder generated C code for deployment.
The same code could be compiled for different targets: Raspberry Pi for user testing while final hardware was being developed, the eventual low-power device, and Android and iOS applications. The code was also optimised for memory use and compiled for the target hardware. This supported a consistent algorithm across development and deployment without requiring a separate reimplementation for each platform.
What the approach means for engineering teams
These projects show why IoT architecture should follow product requirements rather than a default assumption that all analysis belongs in the cloud. Teams can evaluate where processing belongs by considering:
- Power: Whether the device can sustain an operating system, radio use or continuous communication.
- Connectivity: Whether a reliable link is available, and how much data the network can carry.
- Response: Whether decisions need to be made locally, particularly when rapid detection is required.
- Validation: How algorithm changes can be checked against logged data before deployment.
- Portability: Whether the same developed logic can be deployed across prototypes, embedded hardware and mobile platforms.
Cloud services can still be valuable for infrastructure, trials and data analysis. In some projects, the time required to recode an algorithm for a cloud service may be justified; in others, local execution is essential. The architecture should reflect those trade-offs rather than forcing every product into one pattern.
Conclusion
MATLAB and Simulink can support an IoT workflow that spans algorithm prototyping, analysis of logged data and deployment to different computing targets. In Skyradâs experience, generating C with Embedded Coder helped carry developed logic from desktop work to Raspberry Pi, mobile applications and low-power embedded devices, while reducing the need to recreate the algorithm for each target.
For automation and product teams, the practical lesson is to design around the realities of the device: power budget, network availability, data volume and required response time. Cloud processing is one option, not a universal architecture. Matching the location of computation to the application can help build IoT products that remain useful when connectivity is limited and that can be tested across development stages.
Access Official Technical PDF