In June 2022, Siemens Smart Infrastructure (SI) announced one of the largest M&A deals in the history of IoT software by acquiring Brightly Software for $ 1.75 billion. Brightly is a software-as-a-service company offering smart asset management solutions.
So how does Brightly leverage AI and IoT capabilities?
1. Capture data from assets in real-time
2. Analyze data and model trends
3. Implement intelligent automation at scale
Pepsi Bottling Ventures in Garner, North Carolina saved 20 hours per week by using Brightly solution for automating work order processing as soon as an error is detected in an asset.
While IoT enables proactive monitoring of assets, the data generated is processed with the help of AI algorithms to detect errors automatically and initiate action without human intervention.
Consider a production line in a factory where all the machines are IoT-enabled. Each machine has sensors, actuators, and microcontrollers (MCUs) and can connect to a shared network.
Sensors generate data based on changes in the environment. Sensors are deployed to detect changes in temperature, pressure, humidity, air quality (gas leaks), and level of substance within the machine (oil levels). There are also optical sensors to detect visual anomalies and proximity sensors to detect nearby objects.
The actuators convert the electrical signals from sensors into action, such as motion, triggering an alarm, or stopping the machine.
Microcontrollers (MCUs) contain a Central Processing Unit (CPU), memory to store data, and peripherals such as ethernet ports, Wi-Fi, and General-Purpose Input/Output (GPIO) that connects sensors with the CPU.
Until recently, the MCUs did not have the compute power to run AI/ML algorithms autonomously. Data is sent to a digital twin, a replica of the shop floor hosted on the cloud, for processing. The digital twin on the cloud is equipped with Machine Learning (ML) programs to understand the data, detect patterns, and trigger an alert in case of an anomaly.
1. Latency – when IoT data is sent to the cloud, there is a delay between data generation and processing because of network latency.
2. Network failure – timely action is not initiated in case the network is down and the digital twin cannot reach the IoT endpoints.
3. Data Breaches – possibility of data breach at the server farm or the cloud.
4. Size of data – directly proportional to the number of sensors generating data.
5. Power Consumption – energy requirements and battery capacity define when and how often to send/receive information from the cloud.
Gartner’s Hype Cycle for Artificial Intelligence in 2021 has Edge AI plotted at the top of the curve, which means ‘peak of inflated expectations.’ Combined with the prediction that 70% of companies would have shifted from ‘big’ to ‘small and wide’ data by 2025 indicates that IoT as edge devices will be able to do more than just collect data and send it to the cloud for processing.
AI, Machine Learning and Deep Learning programs are run on cloud infrastructure because these algorithms require substantial computing power to process data.
On the contrary, Edge AI refers to running AI programs on devices known as edge nodes where the data is generated. IoT devices are great examples of edge nodes.
Distributed computing, an important enabler for AIoT, refers to connecting multiple Graphical Processing Units (GPUs) or Central Processing Units (CPUs) together to increase computing capacity. In other words, distributed AI refers to creating a grid of IoT Edge Devices so that computing resources can be pooled together.
Resource-intensive ML algorithms can run on this grid and process data as soon as it is generated to detect patterns and respond to events in real-time. However, traditional ML programs are not suited for embedded systems like IoT Chipsets.
TinyML has emerged as the new Machine Learning paradigm for embedded systems. TinyML is targeted towards battery operated edge nodes and enables IoT devices to perform on-device data analytics at low power. It can process all types of sensor data such as video, audio, biomedical, and process variables.
Google’s TensorFlow Lite, which is based on the TinyML framework, allows developers to deploy ML models on IoT microcontrollers. These on-device ML models make accurate predictions in real-time based on the input data.
1. Cobots
Collaborative robots, known as cobots, are designed to work alongside humans in manufacturing and assembling units. With the help of data generated by IoT devices and AI programs such as computer vision, cobots assist humans in the following areas:
a. Production – loading and unloading materials from conveyor belts
b. Assembly – lifting and positioning heavy parts in automobile assembly
c. Quality Control – measuring, testing, and inspecting products
d. Packaging – counting, packaging, labelling, and placing finished products in cartons
Cobots depend on IoT data to work efficiently and the reduced latency of AIoT devices enable cobots to manage human-machine interactions.
2. Smart Industry
AIoT applications in smart industry goes beyond conventional manufacturing, supply chain, and quality control. It extends to predictive maintenance and workplace health and safety.
Condition-Based Monitoring and Predictive Maintenance: tanks storing hazardous material are equipped with AIoT devices that monitor variables such as pressure, temperature, and levels. These AIoT devices are trained using real-time data to detect patterns and raise an alarm in case of unusual events. Predictive maintenance alarms and refilling alerts can be initiated without human intervention.
3. Energy Management
AIoT inferences from input data can reduce energy consumption and automatically turn on or off HVAC systems based on multiple parameters such as ambient light, temperature, number of people in an enclosed space, and timing. AI algorithms are trained in virtual mode using training data before getting deployed in IoT edge nodes.
The convergence of AI and IoT along with faster network connectivity through rollout of 5G services will enable smart factories and smart cities. As AI programs reside in IoT chipsets at the edge, there is no need to send data to server farms located far away, thereby resulting in robust security and reduced energy consumption.
Let us conclude by looking at the strategic benefits of implementing AIoT:
1. Manufacturing Process Automation
2. Asset Tracking and Maintenance
3. Product Quality and Engineering Efficiency
4. Fleet Management and Telematics
5. Interconnectivity of IoT Devices
6. Data Security and Regulatory Compliance
This article is brought to you by Softura.