Industrial systems are replete with sources of data generation.
Machines generate telemetry, sensors measure temperatures and vibration, vehicles indicate locations and movement, production systems capture events, and tracking platforms store information on assets and equipment.
The collection of this information tends to be straight forward, whereas the challenge is the transformation of distributed, physical world sources into something that can provide meaningful intelligence.
That is the space where AIoT (Artificial Intelligence of Things) comes into play.
What Does AI Bring to IoT?
A typical IoT architecture can be thought of as one that connects physical devices, collects the sensor data produced, transmits that information and makes it accessible to an application.
Something like:
Sensor → Gateway → Network → IoT Platform → Application
That architecture gives visibility into what is occurring.
The introduction of an AI layer, though, can add additional processing steps.
Physical Asset
↓
Sensors
↓
Connectivity
↓
Data
↓
AI / Analytics
↓
Insight
↓
Decision\Action
An AI layer can analyze information, and particularly leverages historical as well as current information to provide pattern detection, correlation, anomaly detection, and recognition of changes in behaviors.
The intent isn't so much for AI to provide additional data collection, but rather to allow greater extraction of meaning from the information that is present.
Context Enables More Meaningful Sensor Data
On its own, a sensor tends to capture a data point, but that rarely captures the whole information story.
Consider the example of an industrial machine indicating that it has experienced an increase in vibration.
The value of that measurement can be enhanced by additional information, such as:
Operating hours
Workload of the machine
Temperature
Location of the machine
Past maintenance
What and how much is being produced
Patterns from other sensors
Rather than a single data point about vibration levels, an AIoT approach could allow such data to be evaluated with other factors to determine if there is a potential indication of an underlying root cause.
This is especially true if other systems are present but haven't shared information, such as different systems capturing data on the same assets or equipment.
A tracking system might indicate location, whereas a maintenance system might capture when service was performed, and a production system might identify what was being produced.
Combining these sources of information could yield greater insight into what is occurring in the physical world.
From Monitoring Thresholds to Recognizing Anomalies
A traditional approach to monitoring is often to define a set of thresholds.
If a limit is crossed, then escalate it to the relevant business owners.
For instance:
IF temperature > threshold
THEN generate alert
This is a reasonable approach, but there are limitations in many industrial settings in that such a system is often only detecting changes that are known.
Many AIoT use cases involve industrial equipment, where historical patterns can provide additional indication of whether certain situations are abnormal.
That is to say, an AI-based analytics system can recognize variations from typical patterns and behaviors.
This allows the asking of questions like:
Is this machine operating differently than normal?
Has equipment changed its patterns over time?
Are there multiple systems that are indicating abnormal readings?
Might these signals indicate a situation worth investigating?
Such anomaly detection can be a critical component of predictive maintenance, equipment monitoring, manufacturing, logistics, transportation, construction, and energy systems.
AIoT Goes Beyond an AI Model
An AI model is one component of an AIoT solution, though.
A real-world implementation involves several other key components including sensing (recognition and collection of data), connectivity (data transmission), management of data (storage, cleansing, processing), analytics (discovery of patterns), AI (building a statistical model), integration (exposing information), and business process (interpretation and action by people).
That is the reason why successful application of AIoT tends to come from identifying business problems, and then working backwards to identify how these technologies can be applied, rather than starting with an AI model and looking for places for it to be deployed.
Physical World to Intelligence and Action
The overall idea behind AIoT is fairly straightforward: the physical world can be turned into data, which can be transformed into intelligence and result in action.
IoT provides the link to the physical world through devices and sensors, and AI provides the analysis and modeling of that information.
For those looking to learn more about how AI and connected systems can support industrial systems, Aperture Venture Studio has more information on AIoT and opportunities in the physical world.
The value of AIoT isn't just that machines can generate more data.
Its true power comes from taking information generated by connected devices and leveraging that context to understand what is occurring in an increasingly complex physical world, from manufacturers to supply chains and transport systems.
Top comments (0)