What originally started as machine-to-machine communication limited almost exclusively to the telecom industry, the Internet of Things is now everywhere. According to Statista, the number of devices connected to the internet will exceed 38 billion by 2025.
The figure is arguable though, for it is hard to draw the line as to what exactly an IoT device is. So, other reports suggest more restrained numbers. Think: around 16 billion devices in use by 2025.
The steep surge in the number of IoT devices will inevitably lead to an increase in the amounts of data collected. IDC reports that the volumes of IoT data generated globally will reach 73 Zettabytes by 2025. And that’s where it gets problematic. The collected information needs to be processed and analyzed to drive value. However, most enterprises fail at putting data to use, with between 60% and 73% of it going unused for analytics.
The good news is that enterprises can turn more of the generated data into business insights by leveraging the combined power of artificial intelligence and the Internet of Things.
In the article, we’ve covered everything you need to know about this potent mix, often referred to as the artificial intelligence of things, or AIoT. So, if you are considering jumping on an Internet of Things development bandwagon, carry on reading.
What is AIoT exactly?
An Artificial Intelligence of Things (AIoT) system is made up of two components: the Internet of Things (IoT) and Artificial Intelligence (AI).
In this robust combination, the role of IoT is to accumulate structured and unstructured data and enable the communication between the connected things and the user.
When amplified with AI — algorithms that can find complex interdependencies in huge amounts of data and describe, predict, as well as prescribe certain actions based on that — an IoT system gains human-like intelligence and can be applied to solve a wider variety of tasks. These could span “understanding” natural language, predicting users’ needs and adjusting a connected device’s behavior accordingly, and more.
The AIoT market is currently on the rise. Recent research estimates that it will reach $102.2 billion by 2026. And it’s perfectly clear why: AI adds value to IoT through improved decision-making, while IoT provides a platform for AI to drive value through connectivity and seamless data exchange.
How does the Artificial Intelligence of Things work?
AIoT systems may be implemented in two ways:
-
As edge systems running on connected devices.
The architecture of an AIoT system will vary depending on the implementation strategy.
Cloud-based AIoT
With the cloud-based approach, basic architecture of an AIoT solution looks like this:
Device layer: various hardware devices (mobility, tags/beacons, sensors, health and fitness devices, vehicles, production equipment, embedded devices)
Connectivity layer: field and cloud gateways
Cloud layer: data storage, data processing (AI engine), data visualization, analytics, data access via API
User communication layer: web portals and mobile applications
Edge AIoT
With edge analytics, the collected data is processed closer to the source — whether on connected devices or on field gateways.
Collection terminal layer: Various hardware devices (mobility, tags/beacons, sensors, health and fitness devices, vehicles, production equipment, embedded devices) connected to the gateway over existing power lines
Edge layer: facilities for data storage, data processing (AI engine), insights generation
Edge-focused implementations do not exclude the presence of the cloud, though. Cloud-based data storage can, for example, be used for collecting metadata about the system’s performance or contextual information needed for training or retraining edge AI (think: a paradigm for crafting AI workflows that involves the cloud and the edge, the latter made of devices outside the cloud that are closer to physical things.)
Top applications of AIoT in different sectors
Driven by a number of factors, such as the availability of new software tools, the development of simplified AI solutions, the infusion of AI into legacy systems, and advances in hardware upholding AI algorithms, the Artificial Intelligence of Things is creeping in many industries. Here’s a rundown of sectors that are already leveraging the opportunities provided by AIoT — with the most promising use cases spotlighted.
Healthcare
Diagnostic assistance
AI-based medical solutions are already surpassing human healthcare professionals in several diagnosis fields. Radiologists around the globe are relying on AI’s assistance for cancer screenings.
In a study published by Nature Medicine, AI outperformed six radiologists in determining if patients had lung cancer. The algorithm that was trained on 42,000 patient scans from a National Institute of Health’s clinical trial data records, detected 5% more cancer cases than its human counterparts and reduced the number of false positives by 11%. It’s worth mentioning that false positives present a particular problem in diagnosing lung cancer: JAMA Internal Medicine’s study of 2,100 patients states a false positive rate of 97,5%. Thus, AI helps address one of crucial diagnosis problems.
AIoT systems perform equally well when diagnosing breast cancer, skin diseases, and skin cancer. But the possibilities of smart, connected systems extend far beyond that.
Recent studies have shown that AI can detect rare hereditary diseases in children, genetic diseases in infants, cholesterol-raising genetic diseases, neurodegenerative diseases, and predict the cognitive decline that leads to developing Alzheimer’s disease.
Improving treatment strategies and tracking rehabilitation process
Following the same principle as in diagnosing patients, AIoT systems can help develop better treatment strategies and adjust them to the patient’s needs.
Combining data from treatment protocols, patient’s history, and real-time patient information from connected equipment and wearables, smart algorithms can suggest dosage adjustments, exclude the possibility of a patient developing allergies, and avoid inappropriate or over-treatment. Some of the essential areas where AIoT is facilitating treatment span:
-
More effective COVID-19 treatment
By monitoring patients who have been diagnosed with COVID-19 via AIoT-powered wearables that record patients’ vital body signs, doctors could offer patients due suggestions, thus providing more effective outpatient care.
-
Treating diseases accompanied by blood coagulation
Connected coagulation devices help measure the pace at which blood clots form, thus helping patients make sure the measurements are within their treatment range and reducing the number of visits to the office as the measurements can be communicated to healthcare providers remotely and in real time.
-
Better asthma and COPD management
Chronic respiratory diseases (COPDs) affect around 500 million patients worldwide. To mitigate the severity of these conditions, patients have to stick to a thorough routine, and using inhalers is an essential part of it. Still, many patients fail to adhere to the recommended treatment plans. AIoT-enabled inhalers that are bound to a mobile app help avert that, recording the time, date, and location of each use. The collected data can be used to set up automatic reminders for the next usage, predict asthma attacks, and identify trigger factors.
-
Optimized diabetes management
In the US alone, 30 million people are affected by diabetes. And for them, regular glucose measurements have always been a concern. AIoT-enabled wireless implantable glucose meters alleviate those concerns by notifying patients — and doctors — of changes in patients’ glucose levels.
Optimizing hospital workflows
AIoT can transform the way hospitals are run, improving daily workflows in the following key areas:
-
Reducing wait times
Automated bed tracking systems powered by AIoT can help hospital workers admit emergency patients as quickly as possible by notifying them when a bed is free. The experience of early adopters, such as Mt. Sinai Medical Center in New York, proves that technology can help reduce wait times for 50% of emergency department patients.
-
Identifying critical patients
Identifying patients in need of immediate attention is critical in providing quality care. To make the right decision, doctors need to analyze large amounts of information, while being under significant pressure. AIoT can give the medical staff a helping hand in prioritizing their efforts. Connected systems may analyze patients’ vitals and alert doctors of patients whose condition is deteriorating.
Several similar systems were tested out in intensive care units. For instance, the University of San Francisco piloted an AI solution that is able to detect early signs of sepsis, a deadly blood infection. The research results showed that patients whose treatments involved AI were 58% less likely to develop the infection; and the death rate was reduced by 12%.
-
Tracking medical equipment
With AIoT-enabled equipment tracking, hospitals can reduce the risk of losing critical medical equipment and make more informed equipment management decisions, thus tapping into $12,000 of savings per bed annually. Critical medical equipment can be tracked via RFID or GPS systems in and out of the hospital, while medical and administrative staff can use web and mobile applications to quickly locate the needed equipment.
Manufacturing
Enabling predictive maintenance
While basic analytics is often enough to detect equipment that is approaching a critical operating threshold, AI can predict anomalies in advance based on historical maintenance and repair data. As a result of predictive maintenance, according to a PwC report, manufacturers can improve equipment uptime by 9%, reduce costs by 12%, reduce safety risks by 14%, and extend the lifetime of their assets by 20%.
Improving asset performance management
With an AIoT system in place, manufacturers can get regular updates about how well their assets are performing and drill down the reasons for performance changes. The majority of IoT-based asset performance management systems allow getting automated alerts whenever a piece of equipment is deviating from the set KPIs.
The AI engine, in turn, helps dig into the reasons for performance deterioration, if there is any, and identify whether the measured KPIs are reasonable to track in each individual setting. Using performance management software, manufacturers optimize equipment utilization and improve the overall equipment effectiveness.
Boosting production planning with digital twins
According to Gartner, digital twins can help manufacturers gain a minimum of 10% improvement in production effectiveness. A digital copy of an asset, system, or process, an industrial, AIoT-enabled digital twin can help manufacturers gain an end-to-end visibility into the shop floor operations and help timely spot and even predict inefficiencies.
Manufacturing enterprises using digital twins state they could achieve lasting improvements, including an increase in reliability from 93% to 99.49% over two years, reducing receive maintenance by 40%, and saving $360,000 having predicted a power outage.
Automating shop floor operations via industrial robotics
Industrial robots have been a part of the shop floor for a long time. With manufacturing IoT solutions getting more accessible, robots are becoming smarter and more independent. Equipped with sensors and relying on AI, industrial robotics is now capable of making well-informed production decisions on the go, thus increasing the effectiveness of manufacturing units.
Automotive and transportation
Traffic management
The approach helped the city administration optimize traffic flow and ensure a safe and smooth driving experience.
Self-driving vehicles
Self-driving vehicles and advanced driver assistance systems (ADAS) are notable examples of AI algorithms interpreting and acting on real-time IoT data.
Self-driving, or autonomous cars create a map of their surroundings based on the data from a variety of sensors. Radar sensors, for example, monitor the position of nearby vehicles; video cameras detect traffic lights, road signs, other vehicles, and pedestrians; lidar sensors measure distances, detect road edges, and identify lane markings.
AI software then processes the sensor data, plots an optimal path, and sends instructions to the car’s actuators, which control acceleration, braking, and steering. Hard-coded rules, obstacle avoidance algorithms, predictive modeling, and object recognition help the software follow traffic rules and navigate obstacles.
Key AIoT implementation challenges, and how to solve them
Out of all IoT projects implemented globally, 76% fail, with 30% of them failing as early as in the Proof of Concept phase. To avoid directing investments into initiatives that are doomed to collapse, companies testing out AIoT waters should be aware of the common challenges that may hinder their AIoT implementations. The obstacles businesses encounter most often span:
-
Embarking on an AIoT journey with no clear goal in mind. Kicking off AIoT projects, organizations may get caught up in the novelty and fail to assess the feasibility of their ideas. That, in turn, could result in uncontrolled cost creeps at the latter stages of development and, ultimately, dissatisfied stakeholders. To avoid that, we advise starting your AIoT project with a discovery phase, wherein ideas can be vetted and weighed against the set business goals, customer expectations, and organizational capabilities.
-
Struggling to choose an optimal implementation strategy. As said earlier, AIoT solutions can be implemented as cloud, edge, or hybrid systems. While drafting the implementation strategy, carefully weigh out bandwidth, latency, and speed requirements for the future solution and map them against the set costs. The rule of thumb is going for edge deployments for time-critical systems spanning a large number of devices and relying on the cloud in case minimal latency and high bandwidth are less critical.
-
Slow deployment cycles with costs that are difficult to estimate. AIoT projects require a long-term commitment. Depending on a particular use case, the implementation process can take several months up to several years. With the technology landscape changing quickly, there’s a risk of a solution becoming outdated by the time it’s fully operational, as well as losing control over the implementation costs. To prevent that, businesses need to be agile enough to be able to introduce changes along the way.
-
The need to connect highly heterogeneous and complex systems. Depending on the scale of your future solution and the industry you operate in, you might need to connect highly heterogeneous legacy equipment to AIoT. Often a difficult task to accomplish, it needs planning and understanding the available options. For example, you can choose to attach sensors to the legacy machines, connect them via gateways, or even replace them altogether. No matter the approach, make sure to draft the feasible digitization scenarios early on.
-
Not having enough data for training AI algorithms. To generate reliable insights, AI algorithms need to be trained on vast amounts of data. If it is not available in sufficient amounts (or is available but cannot be used due to privacy reasons), you’ll have to use other strategies to compensate for the lack of data. Common ways include transfer learning (think: using an already trained neural network that solves a similar problem), data augmentation (modifying the existing samples to get new data entries), or resorting to synthetic data.
-
Struggling to achieve AIoT system’s sufficient performance. The performance of AIoT systems depends on an array of factors, including hardware capabilities, data load, system architecture, the approach to implementation, and more. To avoid performance issues in operation, plan out potential data loads in advance and tune the implementation strategy accordingly.
-
Addressing software and firmware vulnerabilities. Many AIoT projects fail because the security of data, devices, servers, and communication networks was not taken into account during the planning phase. If you deal with highly sensitive data, consider hybrid deployments, where the data is processed closer to the source, so the risk of it being compromised during transit or in the cloud is minimized.
