Artificial intelligence, or machine learning, describes any task performed by a software program or machine in a human-like manner. AI systems share human characteristics, such as planning, learning, reasoning, analyzing, perceiving, and decision-making (inferencing). Currently, AI is experiencing tremendous growth as the IoT expands rapidly. In particular, AI provides the IoT with a crucial component: data analytics.
Traditionally, AI analyzes IoT data in the cloud or the central server. First, data is collected at locations away from the cloud or at the edge. Then the datasets are transferred to the cloud for AI analytics. After inference is reached, the decision will be sent back to the edge for execution. In this model, data-collecting sensors and terminal devices like simple computers, mobile phones, or tablets are located at the edge. Meanwhile, central servers, or the cloud, handle data processing and storage.
But the model is changing.
Cloud vs. Edge Computing
Despite the many benefits of IoT cloud computing, significant issues exist in the speed and expense of data transfer.
All the data collected by sensors must head to the central cloud for computing. While the speed of data transfer has increased significantly, so has the amount of data that needs to be transferred. As a result, there will be a time lag that is large enough to prevent real-time analytics. In addition, after inference is reached in the cloud, the decision needs to be transferred back to the edge for execution, thus adding to the lag. While delays may not be consequential in some sectors, the delays will carry enormous ramifications in industries like oil and gas, energy management, autonomous vehicles, customs control, traffic engineering, and law enforcement.
The second factor is the expense of transfer, which is proportional to the size of data transfer. Because most of the data collected at the edge is irrelevant to analytics, transferring all of it to the central cloud is inefficient as well as expensive.
The issues in cloud computing have, therefore, given rise to AI edge computing. Data analytics and storage are distributed closer to the edge in edge computing. Data is collected and analyzed at the edge devices for real-time inference. Later, only the data that has been determined to be relevant will be moved to the cloud. As a result, edge computing has much smaller, if any, lag time. Reducing data transfer amounts also yields significant cost savings. In addition, edge computing can be independent of the central server and even run offline.
The increasing availability of high-performance as well as low-cost processors and data storage enable most analytics done at the edge using smart devices or intelligent single-board computers (SBCs). Applications of AI edge computing are popping up across many sectors, such as the examples described below.
Smart Buildings
When people hear building maintenance, they often think of heating, ventilation, and air conditioning (HVAC); fire safety; lighting; security surveillance; and parking.
Sensors that are installed at various locations inside and outside a building can now collect different kinds of data, such as temperature and foot traffic. With insights from AI analytics, building owners and managers can make decisions that reduce costs and increase building security.
AI facial recognition can be used to restrict access to a building or identify suspicious entries. Data can be collected on the amount of sunlight or wind on the building’s sides during different seasons and various times of the day. Armed with this information, shading, heating, or air conditioning, even in unoccupied rooms, can adjust to minimize energy consumption. In addition, foot traffic data can help establish room occupancy patterns, and lights can come on in sync with those patterns. Monitoring elevators makes scheduling maintenance predictively to prevent expensive emergency repairs possible. Bathroom monitoring allows cleaning to be scheduled in response to varying foot traffic volumes to ensure cleanliness and, at the same time, avoid wasting either supplies or staff resources.
And the applications of AI computing go beyond those just noted.
For example, instead of waiting until the end of a lease, a business can use cloud computing to understand in real time where space is underutilized so that it can adjust quickly. Research has tied employee wellbeing to productivity and creativity. Data from various sensors can help employees become healthier and more productive. Wearable sensors can track how much employees move around. Also, sensor data can help manage carbon dioxide levels so that employees can stay alert and focused. There are discussions that cross-team interactions may contribute to higher productivity — a company can collect related data to verify that. Cloud computing can also provide amenities like guiding employees to available conference rooms or an open parking spot.
In conventional buildings, there are separate controls for heating, lighting, security, and parking. In smart buildings, integrating all the control systems will streamline management and achieve cost savings.
Smart buildings can even talk to one another. Managing blocks of buildings together will facilitate better communication with the grid, help balance energy supply fluctuations, and reduce overall demand.
As building blocks of a smart city, smart buildings can help achieve energy efficiency for the entire infrastructure. In a future blog, we will discuss the development and adoption of smart buildings in various countries and regions.
Smart Oil and Gas Platforms
Oil and gas exploration involves tremendous risks and high costs in drilling and equipment maintenance and material transport. Also, worker safety is a concern. Analytics can be used to help increase operational efficiency as well as reduce risks and costs.
Drilling becomes more dangerous and less productive as the rig digs deeper. Injections are used to help extract more material at each drill, but incorrect injections can be counterproductive or even dangerous. Also, too much pressure on the wellhead can result in breakdowns that cause delays and generate costs. Analytics on the data collected from the wellhead or pumps can help workers determine optimal vacuum pressure, for example, as well as the best time to do the injections.
Equipment maintenance is another area where AI edge computing can make a difference. Situated in remote locations and harsh environments, oil and gas platforms need to withstand corrosion by wind, water, and salt. Material leakage damages the environment and must be fixed quickly to reduce the danger of combustion, racking up delay-related costs, or potential code violations. While visual inspection by workers can identify large cracks, this also means that problems will go undetected until they become visible and significant. Today, most shutdowns caused by faulty equipment are unplanned and expensive. But going forward, sensor data will be able to help schedule predictive maintenance and reduce the risk and cost of uncertainty.
The transport of combustible materials such as oil and gas requires close supervision. The tankers need to be monitored for leaks, tank level, temperature, and pressure to prevent leaks or explosions. An IoT system enables a manager to track the exact location and condition of each oil and gas asset.
Lastly, facial-recognition AI can help ensure compliance with required actions, such as wearing safety gear. Also, facial-recognition AI can identify fatigued or unalert workers.
Smart Factories
When people hear smart factories, it might call to mind the image of robots working assembly lines, but such robots are only one way that a factory uses data analytics to make itself smart. Product design and testing, production, collaboration, security, equipment maintenance, quality control, worker safety and training, and post-production support are among the other factory elements that can benefit from AI edge computing.
For example, sensor data from field tests can help product design engineers understand how different real-world configurations will affect performance. Having the analytics makes it possible to resolve design issues with fewer tests or prototypes. As a result, the time to market can be significantly shorter.
During factory production, historical and comparative data from sensors installed on various equipment can help engineers identify areas that need improvement. Data analytics can also help workers identify and resolve points of failure quickly. In addition, data from the factory floor helps optimize the layout of the floor to minimize collisions and congestion. Moreover, data analytics can help optimize the number of operators needed at each step, balance the workload on each station, and decrease the amount of non-value-added work.
Data analytics also facilitates collaboration among stakeholders. Each of them can gain a better understanding of the process, solve problematic scenarios effectively, or achieve the optimal quality and productivity.
As with smart buildings, facial-recognition AI can be used to restrict and control access to different areas in a factory or identify suspicious persons.
While machine breakdown is inevitable, reactive repairs are unpredictable and costly. In contrast, predictive maintenance scheduled as a result of analytics removes uncertainty in the process. Predictive maintenance not only saves money on emergency repairs, it eliminates the cost of routine but unnecessary service while extending the life of equipment.
AI can create a safer workplace for workers. Sensor data such as a worker’s body temperature, pulse, activity levels, or eye movements can help managers identify workers who are too fatigued or unfocused to work and rebalance the workload within a group accordingly. In addition, AI can help enforce safety codes such as wearing helmets, goggles, work boots, or earmuffs. Sensor data can also help monitor the temperature or the level of radiation and toxic fumes in the factory.
Helping employees work more efficiently is another task that AI can assist with. Collaborative robots can work alongside humans — the robots will handle the heavy-duty, monotonous work so that humans can focus on complex tasks that require nimble problem-solving. Sensor data can be used to better fit tools to workers and to provide workers with the right components at the right time to reduce errors.
Post-assembly, AI machine vision with high-resolution cameras and image processing can enhance quality control. The analytics can even detect and predict if and when the quality will trend toward more defects.
Smart Supply Chain
Each supply chain has multiple components and spans many locations. The pain points that can be improved with AI analytics include the tracking of inventory, assets, asset quality, asset authenticity, and driver wellbeing.
The traditional way of checking inventory is slow, labor-intensive, and error-prone. Warehouse mistakes or shortages will propagate and escalate through the entire supply chain, affecting downstream stakeholders and damaging business relationships. Data analytics can feed into automatic replenishment of inventory, speeding the process with fewer errors and reduced effort.
During transport, the wellbeing of the driver, vehicle, and goods are all critical. Data analytics can help different stakeholders like manufacturers, contract conveyers, and customers know where each vehicle in a fleet is at any given point, on-time delivery status, and if inventory has been misplaced or lost.
In the event of a vehicle breakdown, analytics can also help locate emergency services to minimize downtime. Together with IoT, AI edge can also be relied upon for optimizing shipping or trucking routes to reduce fuel consumption or save time. In addition, on return trips, analytics can connect empty shipping containers to new revenue opportunities.
Some goods, such as food or pharmaceutical products, must be stored within a specific range of temperature, humidity, or acidity at all times. With sensor data, transportation companies can maintain an unbroken chain of custody that prevents spoilage and quickly identifies any problems that do occur. This level of visibility will address one of the biggest pain points in the supply chain, which is ensuring that goods arrive intact and usable.
Lastly, because truck drivers work long hours in strenuous yet monotonous conditions, sensor data and facial-recognition analytics can help fleet managers monitor the level of alertness or exhaustion in the drivers.
Looking Forward
AI edge computing is still in its nascent stage and needs to resolve several pain points that are specific to it. For example, because most of the analytics and storage occur at the edge, the hardware and software at the edge must meet heightened processing and data storage requirements. Also, the edge hardware needs to offer high processing power while maintaining energy efficiency, as it’s usually installed in remote places that make battery replacement a challenge.
In our next blogs, we will discuss AI edge computing in greater detail. In addition, we will consider the application of edge computing in specific sectors, such as smart buildings, factories, and energy management, and explore possible solutions.
