Optimizing IoT Architectures for AI

Harnessing the power of AI in IoT remains a huge interest for many data professionals attempting to both integrate disparate platforms and leverage real-time insights. Achieving this advancement in IoT comes with many obstacles, ranging from the complexities of infrastructure setup to data streaming, cost-efficient scalability, and more.

Anais Dotis-Georgiou, lead developer advocate at InfluxData, joined DBTA’s webinar, Unleashing the Power of AI in IoT with InfluxDB, to examine how enterprises can create AI-powered solutions engineered for IoT applications with the advanced capabilities of both  InfluxDB and other third-party platforms.

Understanding how to achieve AI for IoT implementation begins by exploring the nature of data pipelines, according to Dotis-Georgiou. As data is generated from a sensor or application, the way it moves to its destination is through pipelines.

HiveMQ, an MQTT broker, allows data workers to take information as a message to a topic which then allows other devices to subscribe to the topic, access the data, process it, and write it to their databases, such as InfluxDB, with a variety of extensions that allow you to connect sensor data directly to HiveMQ Client or to HiveMQ Edge, HiveMQ’s edge gateway. HiveMQ acts as the piping that connects all  your devices and data stores together—as well as any ETL engines or processors—at extreme scale with security in mind.

For a modern data pipeline, time series data is of high importance, especially for IoT implementations. By indexing data, storing data chronologically, and identifying high throughput, time series data presents significant value for industrial IoT (IIoT) sensor data, according to Dotis-Georgiou.

InfluxDB 3.0 is well-suited for times series data, where its columnar format specifically compresses repeating data values—which can often occur when tracking IoT sensor data. InfluxDB’s columnar format is also capable of identifying global maximums/minimums and first/last values, without having to iterate across every single column in a data table, ultimately increasing the efficiency of querying.

Putting this into practice, Dotis-Georgiou demonstrated how a data professional might integrate data pipelines in application architectures with the help of InfluxDB and HiveMQ. HiveMQ will collect all sensor data then write the data directly into InfluxDB. From there, data can be processed and enriched, as well as visualized with tools from vendors like Tableau or Grafana.

Another third-party vendor—Quix—allows this pipeline to be fully completed in combination with HiveMQ and InfluxDB. Quix is a complete solution for building, deploying, and monitoring event streaming applications using Kafka under the hood with a focus on Python-based plugins so users do not need to understand Kafka to leverage Quix. Quix, when integrated into the HiveMQ and InfluxDB architecture, allows users to easily perform data processing for tasks like anomaly detection and machine learning (ML).

Dotis-Georgiou then walked webinar viewers through real-world challenges that, through this HiveMQ-InfluxDB-Quix architecture, can be optimized for IoT use cases. She further added that the future of incorporating AI into IIoT can present a variety of opportunities, specifically through clustering algorithms that identify the condition of machines, leveraging LLMs as real-time processors, application and outcomes assistants, and more.

For an in-depth discussion of IoT architectures and its potential with AI, you can view an archived version of the webinar here.