As growing numbers of internet-connected sensors are built into cars, planes, trains and buildings, businesses are amassing vast amounts of data.
Tapping into that data to extract useful information is a challenge that's starting to be met using the pattern-matching abilities of machine learning (ML) -- a subset of the field of artificial intelligence (AI).
Firms are increasingly feeding data collected by Internet of Things (IoT) sensors -- situated everywhere from farmers' fields to train tracks -- into machine-learning models and using the resulting information to improve their business processes, products and services.
One of the most visible pioneers is Siemens, whose Internet of Trains project has enabled it to move from simply selling trains and infrastructure to offering a guarantee its trains will arrive on time.
Under the Internet of Trains project, Siemens has embedded sensors in trains and tracks in select locations in Spain, Russia and Thailand, and then used the data to train machine-learning models to spot tell-tale signs that tracks or trains may be failing. Having granular insights into which parts of the rail network are most likely to fail, and when, has allowed repairs to be targeted where they are most needed -- a process called 'predictive maintenance'. That, in turn, has allowed Siemens to start selling what it calls 'outcome as a service' -- a guarantee that trains will arrive on time close to 100 percent of the time.
One of the earliest firms to pair IoT sensor data with machine learning models was thyssenkrupp, which runs 1.1 million elevators worldwide and has been feeding data collected by internet-connected sensors throughout its elevators into trained machine-learning models for several years.
These models provide real-time updates on the status of elevators and predict which are likely to fail and when, allowing thyssenkrupp to target maintenance where it's needed, reducing elevator outages and saving money on unnecessary servicing. Similarly, Rolls-Royce collects more than 70 trillion data points from its engines, feeding that data into machine-learning systems that predict when maintenance is required.
The application of machine learning to Industrial Internet of Things (IIoT) data is not all about predictive maintenance. For agricultural equipment maker John Deere, the computer vision made possible by deep learning is allowing it to experiment with herbicide sprayers whose built-in cameras can distinguish between weeds and plants. The aim is to apply insight to each stage of the farming process, eventually producing planters and harvesting equipment that can adjust how they operate on the fly in order to maximize crop yields.
In a recent report, IDC analysts Andrea Minonne, Marta Muñoz, Andrea Siviero say that applying artificial intelligence -- the wider field of study that encompasses machine learning -- to IoT data is already delivering proven benefits for firms.
"Given the huge amount of data IoT connected devices collect and analyze, AI finds fertile ground across IoT deployments and use cases, taking analytics level to uncovered insights to help lower operational costs, provide better customer service and support, and create product and service innovation," they say.
According to IDC, the most common use cases for machine learning and IoT data will be predictive maintenance, followed by analysing CCTV surveillance, smart home applications, in-store 'contextualized marketing' and intelligent transportation systems.
That said, companies using AI and IoT today are outliers, with many firms neither collecting large amounts of data nor using it to train machine-learning models to extract useful information.
"We're definitely still in the very early stages," says Mark Hung, research VP at analyst Gartner.
"Historically, in a lot of these use cases -- in the industrial space, smart cities, in agriculture -- people have either not been gathering data or gathered a large trove of data and not really acted on it," Hung says. "It's only fairly recently that people understand the value of that data and are finding out what's the best way to extract that value."
The IDC analysts agree that most firms are yet to exploit IoT data using machine learning, pointing out that "a large portion of IoT users are struggling to go beyond a mere data collection" due to a lack of analytics skills, security concerns, or simply because they don't have a "forward-looking strategic vision".
The reason machine learning is currently so prominent is because of advances over the past decade in the field of deep learning -- a subset of ML. These breakthroughs were applied to areas from computer vision to speech and language recognition, allowing computers to 'see' the world around them and understand human speech at a level of accuracy not previously possible.
Machine learning uses different approaches for harnessing trainable mathematical models to analyze data, and for all the headlines ML receives, it's also only one of many different methods available for interrogating data -- and not necessarily the best option.
Dan Bieler, principal analyst at Forrester, says: "We need to recognize that AI is currently being hyped quite a bit. You need to look very carefully whether it'd generate the benefits you're looking for -- whether it'd create the value that justifies the investment in machine learning."
Most companies will need to turn to one of the major cloud platform providers -- Amazon, Microsoft, Google, Alibaba Cloud or IBM, for example. These firms offer a range of services for storing IoT data and preparing it for data analytics, as well as for training and running machine-learning models and for creating dashboards, graphs and other easy-to-grasp layouts for visualizing the information these models generate.
However, setting up such a system is by no means straightforward, and in-house expertise expertise will be required to determine which data should be collected, which patterns should be looked for, and why.
Working this out will typically need collaboration between in-house data scientists and staff with a deep understanding of the goals of the business or specific department.
"The most important thing from a recruitment perspective is to have a very small team of data analysts who can talk to the business division to understand what are the business requirements, the customer pain points, and the problems they are trying to solve with big data," says Forrester's Bieler.
"You can have all the data analysts in the world, but if they don't know what to do with the data it will never provide any sort of meaningful value to the business. You need to avoid putting the cart before the horse."
Bieler recommends that companies have a clear business goal in mind -- predictive maintenance, as in the case of thyssenKrupp and Rolls-Royce -- before starting any such project.
"You need the data scientists with the domain expertise, as well as the software developers to develop the models, says Gartner's Hung.
"This is not a three-month project, and the folks who have adopted this earlier are the ones who have quite a bit of historical data they can use that to jump-start the process. To some extent it's a case of trial-and-error to work out which machine-learning algorithm is best suited to your application," Hung says.
That said, there are some industry-specific offerings that are starting to offload some of this work, such as those offered by start-ups like analytics specialist Uptake.
While the services available from the major cloud providers for managing IoT devices and ingesting and analysing IoT data are largely comparable, it's worth detailing the offerings from the two largest players.
Amazon's AWS IoT Analytics offers a service for collecting, processing and storing data collected by IoT devices, allowing users to query data and run analytics on it, including applying machine learning to that data using hosted Jupyter Notebooks. AWS IoT Core offers a cloud service for managing IoT devices that allows devices to securely connect to AWS cloud services for processing and storing their data, while AWS IoT Greengrass is designed to carry out machine learning inference on devices near to the edge of the network.
Similarly, Microsoft offers Azure IoT Edge, a service on its Azure platform for deploying trained machine-learning models to IoT and edge computing devices, alongside its Azure IoT Hub service for managing IoT machines, and Azure Sphere for securing and servicing IoT devices, all of which are packaged under its Azure IoT Central offering.
Before embarking on a project to extract information from IoT data, it's worth checking what your competition is up to. The effectiveness of machine-learning models depends heavily on having large quantities of good-quality data, so it may be hard for smaller organizations to compete with established players, according to Forrester's Bieler.
"One of the more worrying sides of big data is you could potentially argue this is a self-reinforcing cycle, where the leading companies today are very likely to be the leading companies of tomorrow because they have gathered more data and accordingly have more insights that helped them to pull away from competitors," he said.
Bieler gave Uber as an example: "Uber has much more data available than any small start-up about passenger likes, driver performance, popular routes; they might be able to send cars to certain areas where they can predict there will be heavy demand three hours from now because they know that on Monday after a football game these kind of patterns will emerge."
"You need to understand where you want to apply machine learning -- you can't tackle everything."
5G and the IoT: The UK's first smart factory just switched on
Internet of Things sensors and applications on trial at two smart factory sites
Nokia wants to make your IoT project easier to set up and run
Four IoT packages aim to make enterprise deployments of sensors and analytics easier.
AI, IoT, and voice in 2019
Lynne Capozzi, chief marketing officer at Acquia, tells Tonya Hall about where AI, IoT, and voice are expected to go in 2019.
Ericsson extends IoT portfolio
Ericsson now has four different Internet of Things segments across Massive IoT, Critical IoT, Broadband IoT, and Industrial Automation IoT.
How to differentiate between AI, machine learning, and deep learning (TechRepublic)
Tech leaders need to put AI and its subcategories into practice--and into common business vocabulary that everyone can understand.
Google, Amazon, Microsoft: How do their free machine-learning courses compare? (TechRepublic)
There are an increasing number of options for those wanting to get a grounding in the field via free courses provided by the major tech firms.