We live in an age where connected devices and systems can absorb staggering amounts of information. Information is valuable, but data in and of itself is not an actionable solution to optimizing systems. We need solutions that allow us to act on the data we receive.
Artificial Intelligence and the IoT are two fields that support and supplement each other well, and can be combined in practical applications. The AIoT is one name given to this space of technological development.
There’s a common workflow to an AIoT system that looks something like this:
- Physical Environment
- Sensors
- Connectivity
- Data Processing
- AI Analysis
- Decision
- Action
The data from the sensors feeds the AI and provides it with information about the physical environment it’s observing. The AI produces information that can be synthesized into a decision and finally, an action.
Depending on the environment, each subunit has specific considerations to make. The AIoT’s implementation depends on factors such as latency, bandwidth, device density, connectivity, and a whole host of others.
Sensors collect information such as temperature, vibration, position, moisture, and more. Then the connectivity layer takes this data (usually numeric values from different sensors) and converts it into a usable form. The data processing stage can refine this information further, searching for patterns or outliers. At the same time, the AI uses the data it gets from the sensors or the data processing layer and identifies patterns, irregularities, or behaviors of interest. It will form conclusions about the state of the physical environment and the systems in it. These conclusions are refined into a decision that can be executed by a human or a dedicated machine.
An example could be a vibration sensor attached to a rotating piece of machinery in a factory. As the motor turns, the sensor will read values and the system will analyze them. If the values suggest an anomaly or a deviation from the norm, it can be flagged as a point of interest for a human operator to examine more closely. Once the human observes the anomaly and any surrounding contextual information, a determination can be made about whether maintenance is necessary.
Another example might be a temperature sensor in a home or commercial building that controls a thermostat. Abnormal readings can prompt the system to adjust the temperature or recommend that a technician visit to inspect the HVAC system for issues.
At the core of any AI implementation is pattern recognition and determining significance based on available information. Information becomes data when it can be interpreted, and data can be evaluated for relevance and usefulness in any given situation. Information is data that has been organized and can be further used to make a decision. AIoT is information from the physical world that has been processed and organized to determine if something abnormal or noteworthy is happening and why.
Practical Ways AIoT Can Be Used As Examples
There are many applications for AIoT depending on the industry and use case. Here are some examples of particular interest.
Predictive Maintenance Can Be Added to Industrial Equipment Management.
Machines produce information about their current state and behavior while in operation. Vibration, temperature, pressure, and power consumption are only some of the characteristics that can be helpful in determining how well a machine works and whether it requires any service.
These behaviors and properties can be evaluated by an AI system, which can flag equipment that exhibits patterns suggesting abnormal operation. These alerts can be examined by technicians and double-checked against the known condition of the machines in the field to assess the accuracy of the AI.
Asset Tracking Can Be Used Across Many Industries and Scales of Operation
Keeping track of inventory and materials can be a challenge, particularly for businesses with distributed operations, warehouses, and supply chains. Devices for tracking can be implemented on a large scale, and the data collected can be helpful both in resolving specific issues and noticing overall trends.
Data from these devices can be analyzed alongside logistical and inventory software to observe and optimize material flow. The results and suggestions can be implemented on a smaller scale or expanded depending on the results.
IoT Devices and Analytics Can Aid in Environmental Monitoring at A Construction Site
There are many aspects of a construction site that change constantly and are difficult to monitor, equipment inventory, environmental conditions, weather states, and overall coordination of people and resources.
Devices and the information they collect in situ can be helpful in understanding these conditions and behavior patterns. Analytics can streamline these into recognizable trends so that managers and supervisors on the ground can determine how to address and resolve issues.
This streamlines many processes, but only to the extent that physical conditions on the ground and the needs of personnel are understood.
Cloud or Edge Processing: The Best Choice Depends on the Situation
Edge and cloud computing are two options that can be used depending on the application. Edge processing refers broadly to computation that takes place closer to the data source. Using this method, some or all of the processing and analysis done in the cloud can be shifted to an edge node. This has the advantage of reducing latency and data transmission, required computation resources and power. It also improves reliability and security by reducing points of failure and attack vectors.
The cloud provides virtually limitless storage capacity, allowing us to collect and store data from many sources. It’s also well suited for computation – intensive processing, training, and modeling of neural networks, and overall coordination of applications and users.
In practice, both methods are often used together, with some signal processing taking place at the edge and higher-level processing and long-term planning taking place in the cloud. Each system has advantages and disadvantages that must be considered in relation to the application. Bandwidth, processing power, and costs must be taken into account alongside reliability, security, and data integrity.
Some of the other practical challenges of implementing an AIoT solution are:
- Data quality and availability
- Integration with existing systems
- Connectivity
- Security and data governance
- AI reliability and evaluation
- Operational auditing
Having realistic expectations about the capabilities of an AIoT system is important before designing and implementing it.
Putting It All Together, Iteratively
Rather than jumping in blindly and connecting everything possible to an AI system, it can be helpful begin with a question.
Perhaps there is an issue that needs resolving. Perhaps there is an area of operations that needs improvement, expansion, or optimization. In any case, a question worded in terms of something we can measure will help us decide what data to collect.
We can also consider where the data might come from and what it might tell us. After that, we examine if and how we can refine this information to get usable results.
Once we’ve got an initial approach, we can implement it as a test. We can measure how well it performs, use benchmarking to compare it to other systems, and modify or expand it as needed. Afterwards, we can plan for long-term maintenance and further refinement based on ongoing feedback and new information. The most useful type of AIoT implementation is one that we understand and control, rather than something that is too broad and complex to maintain.
In practice, these initial observations and tests are essential before moving to a production-level environment.
Talking About Something Bigger Than Connected Devices
We already mentioned that AI complements physical machinery in a new field that goes beyond the IoT: Physical AI.
Physical AI is intelligence embedded in and interacting with machinery, robotics, and control systems, enabling it to operate independently and dynamically in the physical world.
While AIoT can help bridge the gap between connected devices and processes, physical AI enables new technologies requiring higher levels of perception and action in the physical environment. Real physical AI applications would still require extensive engineering, testing, safety measures, and human expertise to optimize and safely implement.
For more opinions on AI and connected devices and technology for practical fields, visit Aperture Venture Studio.
The Bottom Line
The value of any AIoT implementation is defined by how useful the information it generates is in making decisions about real physical objects and systems.
A practical application always begins with a question and data that is relevant and applicable to it. Information becomes data when we can understand and interpret it, and data can be used to make decisions and take action. The AIoT can provide solutions in many fields and spans many industries, but most implementations are narrow in scope but impactful nonetheless.
If you had to pick just one, which of the three potential pitfalls for implementing AI do you think is easiest to recognize at the onset of an AIoT project?
Top comments (0)