Realizing IoT’s potential with AI and machine learning

The key to getting more value from industrial internet of things (IIoT) and IoT platforms is getting AI and machine learning (ML) workloads right. Despite the massive amount of IoT data captured, organizations are falling short of their enterprise performance management goals because AI and ML aren’t scaling for the real-time challenges organizations face. If you solve the challenge of AI and ML workload scaling right from the start, IIoT and IoT platforms can deliver on the promise of improving operational performance.

Overcoming IoT’s growth challenges

More organizations are pursuing edge AI-based initiatives to turn IoT’s real-time production and process monitoring data into results faster. Enterprises adopting IIoT and IoT are dealing with the challenges of moving the massive amount of integrated data to a datacenter or centralized cloud platform for analysis and derive recommendations using AI and ML models. The combination of higher costs for expanded datacenter or cloud storage, bandwidth limitations, and increased privacy requirements are making edge AI-based implementations one of the most common strategies for overcoming IoT’s growth challenges.

In order to use IIoT and IoT to improve operational performance, enterprises must face the following challenges:

• IIoT and IoT endpoint devices need to progress beyond real-time monitoring to provide contextual intelligence as part of a network. The bottom line is that edge AI-based IIoT / IoT networks will be the de facto standard in industries that rely on supply chain visibility, velocity, and inventory turns within three years or less. Based on discussions VentureBeat has had with CIOs and IT leaders across financial services, logistics, and manufacturing, edge AI is the cornerstone of their IoT and IIoT deployment plans. Enterprise IT and operations teams want more contextually intelligent endpoints to improve end-to-end visibility across real-time IoT sensor-based networks. Build-out plans include having edge AI-based systems provide performance improvement recommendations in real time based on ML model outcomes.
• AI and ML modeling must be core to an IIoT/IoT architecture, not an add-on. Attempting to bolt-on AI and ML modeling to any IIoT or IoT network delivers marginal results compared to when it’s designed into the core of the architecture. The goal is to support model processing in multiple stages of an IIoT/IoT architecture while reducing networking throughput and latency. Organizations that have accomplished this in their IIoT/IoT architectures say their endpoints are most secure. They can take a least-privileged access approach that’s part of their Zero Trust Security framework.
• IIoT/IoT devices need to be adaptive enough in design to support algorithm upgrades. Propagating algorithms across an IIoT/IoT network to the device level is essential for an entire network to achieve and keep in real-time synchronization. However, updating IIoT/IoT devices with algorithms is problematic, especially for legacy devices and the networks supporting them. It’s essential to overcome this challenge in any IIoT/IoT network because algorithms are core to AI edge succeeding as a strategy. Across manufacturing floors globally today, there are millions of programmable logic controllers (PLCs) in use, supporting control algorithms and ladder logic. Statistical process control (SPC) logic embedded in IIoT devices provides real-time process and product data integral to quality management succeeding. IIoT is actively being adopted for machine maintenance and monitoring, given how accurate sensors are at detecting sounds, variations, and any variation in process performance of a given machine. Ultimately, the goal is to predict machine downtimes better and prolong the life of an asset. McKinsey’s study Smartening up with Artificial Intelligence (AI) – What’s in it for Germany and its Industrial Sector? found that IIoT-based data combined with AI and ML can increase machinery availability by more than 20%. The McKinsey study also found that inspection costs can be reduced by up to 25%, and annual maintenance costs reduced overall by up to 10%. The following graphic is from the study:

Above: Using IIoT sensors to monitor stock and vibration of production equipment is a leading use case that combines real-time monitoring and ML algorithms to extend the useful life of machinery while ensuring maintenance schedules are accurate.

• IIoT/IoT platforms with a unique, differentiated market focus are gaining adoption the quickest. For a given IIoT/IoT platform to gain scale, each needs to specialize in a given vertical market and provide the applications and tools to measure, analyze, and run complex operations. An overhang of horizontally focused IoT platform providers rely on partners for the depth vertical markets require when the future of IIoT/IoT growth meets the nuanced needs of a specific market. It is a challenge for most IoT platform providers to accomplish greater market verticalization, as their platforms are built for broad, horizontal market needs. A notable exception is Honeywell Forge, with its deep expertise in buildings (commercial and retail), industrial manufacturing, life sciences, connected worker solutions, and enterprise performance management. Ivanti Wavelink’s acquisition of an IIoT platform from its technology and channel partner WIIO Group is more typical. The pace of such mergers, acquisitions, and joint ventures will increase in IIoT/IoT sensor technology, platforms, and systems, given the revenue gains and cost reductions companies are achieving across a broad spectrum of industries today.
• Knowledge transfer must occur at scale. As workers retire while organizations abandon the traditional apprentice model, knowledge transfer becomes a strategic priority. The goal is to equip the latest generation of workers with mobile devices that are contextually intelligent enough to provide real-time data about current conditions while providing contextual intelligence and historical knowledge. Current and future maintenance workers who don’t have decades of experience and nuanced expertise in how to fix machinery will be able to rely on AI- and ML-based systems that index captured knowledge and can provide a response to their questions in seconds. Combining knowledge captured from retiring workers with AI and ML techniques to answer current and future workers’ questions is key. The goal is to contextualize the knowledge from workers who are retiring so workers on the front line can get the answers they need to operate, repair, and work on equipment and systems.

How IIoT/IoT data can drive performance gains

A full 90% of enterprise decision-makers believe IoT is critical to their success, according to Microsoft’s IoT Signals Edition 2 study. Microsoft’s survey also found that 79% of enterprises adopting IoT see AI as either a core or a secondary component of their strategy. Prescriptive maintenance, improving user experiences, and predictive maintenance are the top three reasons enterprises are integrating AI into their IIoT/IoT plans and strategies.

Above: Microsoft’s IoT Signals Edition 2 Study explores AI, digital twins, edge computing, and IIoT/IoT technology adoption in the enterprise.

Based on an analysis of the use cases provided in the Microsoft IoT Signals Edition 2 study and conversations VentureBeat has had with manufacturing, supply chain, and logistics leaders, the following recommendations can improve IIOT/IoT performance:

• Business cases that include revenue gains and cost reductions win most often. Manufacturing leaders looking to improve track-and-trace across their supply chains using IIoT discovered cost reduction estimates weren’t enough to convince their boards to invest. When the business case showed how greater insight accelerated inventory turns, improved cash flow, freed up working capital, or attracted new customers, funding for pilots wasn’t met with as much resistance as when cost reduction alone was proposed. The more IIoT/IoT networks deliver the data platform to support enterprise performance management real-time reporting and analysis, the more likely they would be approved.
• Design IIoT/IoT architectures today for AI edge device expansion in the future. The future of IIoT/IoT networks will be dominated by endpoint devices capable of modifying algorithms while enforcing least privileged access. Sensors’ growing intelligence and real-time process monitoring improvements are making them a primary threat vector on networks. Designing in microsegmentation and enforcing least privileged access to the individual sensor is being achieved across smart manufacturing sites today.
• Plan now for AI and ML models that can scale to accounting and finance from operations. The leader of a manufacturing IIoT project said that the ability to interpret what’s going on from a shop-floor perspective on financials in real time sold senior management and the board on the project. Knowing how trade-offs on suppliers, machinery selection, and crew assignments impact yield rates and productivity gains are key. A bonus is that everyone on the shop floor knows if they hit their numbers for the day or not. Making immediate trade-offs on product quality analysis helps alleviate variances in actual costing on every project, thanks to IIoT data.
• Design in support of training ML models at the device algorithm level from the start. The more independent a given device can be from a contextual intelligence standpoint, including fine-tuning its ML models, the more valuable the insights it will provide. The goal is to know how and where to course-correct in a given process based on analyzing data in real time. Device-level algorithms are showing potential to provide data curation and contextualization today. Autonomous vehicles’ sensors are training ML models continually, using a wide spectrum of data including radar to interpret the road conditions, obstacles, and the presence or absence of a driver. The following graphic from McKinsey’s study Smartening up with Artificial Intelligence (AI) – What’s in it for Germany and its Industrial Sector? explains how these principles apply to autonomous vehicles.

Above: Autonomous vehicles’ reliance on a wide spectrum of data and ML models to interpret and provide prescriptive guidance resembles companies’ challenges in keeping operations on track.

Real-time IoT data holds the insights needed by digital transformation initiatives to succeed. However, legacy technical architectures and platforms limit IoT data’s value by not scaling to support AI and ML modeling environments, workloads, and applications at scale. As a result, organizations accumulating massive amounts of IoT data, especially manufacturers, need an IoT platform purpose-built to support new digital business models.