IoT can detect physical occurrences in an environment. AI can help determine the significance and possible outcome of the information.
Together they are becoming more and more important in industrial environments where sensors, machines, assets, and operational procedures are involved.
Getting more data is certainly valuable, but what is the point of collecting more information if that data is not used somewhere.
IoT Infrastructure
The IoT infrastructure, usually, consists of some sort of sensors, machines, gateways, tracking systems, etc. that house or collect the information.
Depending on the case there may be information collected about equipment status and performance, location and movement of assets, temperatures, production insights, inventory movements, employee or asset data, and others.
That information can be transferred via gateways or pipelines in order to be processed, stored, and/or analyzed.
Now, that brings us to a more essential question.
What is the point of collecting that information.
Applying AI
AI can help in finding patterns or irregularities that would lead for an inference about the state of the environment or system.
Such applications can be:
Anomaly detection
Predictive maintenance
Demand and/or inventory forecasting
Discovery of patterns
Optimization of processes
Recognition and evaluation of risks
For instance, an anomaly could exist when the vibrations or temperatures change for a machine. By analyzing such data in conjunction with the information about the current state of the equipment, an AI model can detect irregularities or patterns that would point toward an issue.
This is how the interpretation of certain information is made actionable.
An Example of AI Infrastructure
A very basic depiction of AI infrastructure could be seen as:
Sensors → IoT Gateway → Data Pipeline → AI/ML → Application → Operational Procedure
Each step implies a different set of operations.
While the gateway collects data from connected devices, the pipeline routes, filters, transforms, and prepares the data for further processing and operations.
AI/ML components process, analyze, detect, predict, and advise. The actual application relays information to operators, engineers, and other systems. This is one of the key points of an AI solution: to connect it with other procedures as the value proposition of AI is in its application.
An operational procedure or decision using the information from an AI model is what ultimately makes it valuable and helpful. If nobody looks at the pretty graphs, it might as well not be there.
Edge, Cloud, or On-Premises?
When it comes to AI, the placement and architecture matter.
The processing of such data can happen anywhere depending on the needs: in edge devices, on-premise, in a public or private cloud, or via a hybrid platform in a combination of places.
Some of the components, especially at the lowest level, may have to process some information closer to the source in order to reduce latency or data size. Others may rely on the global view provided by a central system.
The infrastructure setup depends on the use case and should be prioritized accordingly.
A Working Example
Let's imagine a few machines in a factory. Each machine has a set of connected sensors that collect information about the state of the equipment and the environment. The information from those sensors is gathered and sent to another system where it can be processed and analyzed.
By analyzing the data from these sensors in comparison to the general performance of the equipment an AI model can detect certain patterns or irregularities suggesting the possibility of an anomaly.
In this example, we are not only collecting data about the equipment, but utilizing it in order to improve our ability to predict possible problems.
This could look something like this: Machine → Sensors → Gateway → Data Pipeline → AI/ML → Alert → Maintenance Process
It is important to note that such a process is a great example of how AI implementation does not only make sense as another analysis step, but rather is integrated into the process itself.
It becomes a part of the operational procedure.
But Data Is Important Too
Adding AI into the mix does not eliminate the importance of data quality. In fact, it adds to it.
In cases where the information provided by a sensor is inconsistent there is little to be gained by trying to interpret it. Before rushing to implement some fancy machine learning model, it may be worth considering such factors as sensor quality and data consistency, synchronization, preprocessing, model quality, integration possibilities, security, and scalability.
There are scenarios in which improving the data prior to processing can contribute far more than tweaking the ML model.
IoT + AI Final Notes
AI + IoT is more than an exercise in trying to apply AI to any system in order to extract additional value.
A successful system will feature integration with the physical environment as well as a well-designed data pipeline and application infrastructure which marries the data and the business logic as a system to fulfill a particular set of requirements.
Top comments (0)