Industrial IoT initiatives can capture huge amounts of telemetry data, but the process involves more than getting the information. What most of us who design solutions for a living forget to address are the decisions we must make to get actionable information.
We can think of a practical AIoT system as encompassing five different layers.
1. Sensors and connected devices
An AIoT system always begins with data sources such as temperature sensors, vibration readers, RFID, and machine controllers.
Each type of sensor has different requirements in terms of sampling rate, precision, connection, and power. The choice of devices should be based on the specific needs of the application rather than the coolness of the technology.
2. Edge processing
It is sometimes unnecessary to send all the raw data to a faraway platform.
Edge devices can filter data, perform calculations, keep statistics, detect events, and save results locally or send them to the cloud over a more reliable connection. It is also useful for time-critical processes running in the local environment to have some processing power available.
3. Connectivity and ingestion
The data must reach some system capable of storing or analyzing it. This part of the system depends on the physical infrastructure, which may involve using an industrial protocol, a gateway, or a message broker.
The people building the system should consider such practical details as intermittent connections, device authentication, data validation, and field consistency.
4. Storage and analytics
Time-series databases, along with other technologies, can store the data for later analysis or make it available for operational use.
AI models can analyze this information to identify patterns or anomalies and predict future trends. These results must be interpreted in the context of expected values and compared to determine whether false positives or negatives are not occurring too often.
5. Applications and human decisions
The last step in the system is to provide an interface that displays the results, notifies human operators, or connects to maintenance and business systems. Anomalies in scores rarely provide enough information to make a decision; it is often necessary to analyze the characteristics that make up that score and understand what is happening.
Design considerations that are easy to overlook
Data quality : Missing or incorrect data can cause analyzes to produce faulty results.
Security : Devices must be authenticated, data protected, and only the right people can see the results.
Observability : Data pipelines and AI models must be monitored to ensure that they are doing what they were designed to do.
Interoperability : It is important to design systems around existing industrial infrastructure.
Human oversight : Not all decisions should be fully automated, especially when it comes to safety.
Start with a measurable use case
Rather than attempt to build something big and unproven, it is better to choose a specific use case, such as finding abnormal vibrations in a certain category of machines. Then, a model can be developed, data collected, and an assessment made of how useful it is in practice.
Aperture Venture Studio thinks strategically about industrial innovation at the intersection of AI and IoT. Our perspective can give you a valuable insight into the creation of connected-technology solutions.
You can learn more about us on our website at apertureventurestudio.com .
Engineering consideration: A practical AIoT system goes beyond raw information and analyzes it as a means of making better organizational decisions, connecting the physical world to the business.
Top comments (0)