As the name suggests, AIoT is the combination of AI (Artificial Intelligence) and IoT (Internet of Things) that brings edge computing to the next level by making industrial and energy devices smarter. This combination is recognised as “Artificial Intelligence of Things” (AIoT).
Most organisations do not leverage the complete data generated by each industrial device because of the lack of a holistic AI-Cloud-based approach to ingest, train and process edge. Such an approach constitutes the foundation to create data products to achieve better data-based decision making and prediction.
IoT platforms have delivered many benefits to organisations. Integrating devices with sensorics have evolved to a key aspect of any business with embedded systems, control systems, edge devices, machinery, and other “things” that can connect and generate millions of data points. Centralising that generated data is the first milestone to oversee business activities and take action based on real device data or create new business models.
Real Examples in the Market
A common use case is the IoT Dashboard; managers or plant managers visualize Key Performance Indicators (KPIs) via a friendly interface. In this way, decision-makers or device users take action based on real facts and not (only) on intuitions. Quick identifications of outliers, such as production machinery with a low performance, help obtaining complete visibility of risks in the production. In some cases, even a machine learning algorithm predicts which machine needs maintenance.
Another case is the activation of car features. Volkswagen “We Connect” and “We Connect Plus” are easy to activate IoT digital services for certain cars. For example, the car’s report can be visualized by the driver showing the engine´s status, the door closed/open and park position—the value of these kinds of services increases, when connecting with other external systems. If the sensors “detect” a car accident, the car will “call” an ambulance. Another service is the online route calculation, which is very handy if you forgot your mobile. This last feature is part of “We Connect Plus”, the bundle of digital services that can be used -but is not for free.
From IoT to AIoT
When combining IoT with AI, the number of opportunities to innovate increases to a much higher level. A prime example from the automobile industry is Tesla´s autopilot features: navigating on autopilot, auto parking, traffic light and stop sign control and auto lane changing. Thanks to the auto lane changing, the car’s positions itself in the best lane and overtakes slow vehicles. For instance, this feature demonstrates how smart a car can be through software, traffic data, sensors, and camera sensors.
Incoming data from cameras can be processed intelligently with AI. A company from the energy sector uses artificial vision for detecting situations such as intrusions and facial recognition or even for identifying “things” such as materials, masks, and helms. An additional layer of AI lies in taking automatic actions proactively. For example, this is the case for a manufacturer that after processing vision and sensory data from the production uses an algorithm that automatically rejects defect bottles and continuously improves the production to reduce defect rate.
What AIoT brings to you
The text below is a posit to describe some achievable AIoT Benefits.
- Improving your products
Telemetric data, vision data, and sensor data are increasing exponentially in a world of connected devices. If industrial devices are streaming gigabytes of data in real-time, AI can help identify which information is important to solve your business problems. For manufacturing, this could mean identifying defective bottles or detecting quality issues in an automated way. This kind of initiative can have a positive impact on quality performance indicators.
- Intelligence in the Whole Production Lifecycle
Automation might go further than the data collection part. This is when AI is capable of changing the settings of a running machine to avoid defect production. A fictive example could be that the AI changes the production floor’s temperature to allow machines to avoid drastic temperature fluctuation that leads to overhit. In this scenario, performance indicators such as “Faults detected prior to failure” or “Overall equipment effectiveness” might be positively affected.
- Customer Experience goes up
Many technologies might improve customer experience or not- the value is when you are doing something that your customer wants to buy from you and recommend to her/his friends. A busy Tesla customer might use the automatic parking function because he/she is too busy making calls and talking with customers and cannot pay attention to the traffic. Another customer (a bad driver) might use the same feature just to be able to park the car in a better way. Ultimately, customers who like Tesla´s autopilot features might recommend it to their friends having a remarkable impact on Tesla Net Promoter Scores and in the long term on Net Profit Margins.
The Right Algorithms For Your AIoT Initiative
There are three types of algorithms that you might encounter in an AIoT project.
- Supervised Machine Learning
Supervised machine learning implies that the “target” is known. For example, you know that the temperature on the shopfloor influences machines’s effectiveness. Then, you use algorithms that calculate the perfect mix of variables that derive the perfect temperature threshold for your machines. Possible variables could be the weather, location, and size of the plants. The algorithm is in the position to use previous data sets and learn from them.
- Unsupervised Machine Learning
Unsupervised machine learning implies that the “target” is unknown. For example, you have 50 machines producing every day, and you would like to analyse them, but the outcome is unknown—a kind of exploration. After the analysis, you might discover new patterns that you never think of.
- Reinforcement Learning
Reinforcement Learning is about letting the algorithm learn from repetition and perform the most rewarding task—it’s just like training your pet. In the context of manufacturing production, time to market is usually a very important topic, and it can be measured by parameters such as ‘On-Time Delivery’. With Reinforcement Learning, machines can be trained through repetition to perform in different ways in reaction to a dynamic environment to produce fast at a certain level of quality. However, these kinds of projects can be complex because many factors such as the weather, pricing, location, and employee motivation might influence the production.
Image: Unsplash | @solimonster