The Evolution of Industrial Analytics
By Isaac Brown
4 years ago, we hosted a series of events titled “The Evolution of Industrial Analytics”. It sounded cool, it got people to come to the events. But everyone had the same question: “What does ‘Industrial Analytics’ even mean?” Honestly, we weren’t totally sure either.
Over the past 4 years, the use of AI/ML and predictive analytics has become more prevalent within industrial operations, although we’re not where investors and vendors hoped we’d be when they predicted widespread AI-driven operations 4 years ago. That said, some forward-thinking industrials are scaling AI across their operations.
When I use the term “AI” with my parents, they think of robots or drones or self-driving cars (they’re old). Millennials often lean towards marketing/advertising analytics or Netflix movie recommendations when they think of AI – so here I’d love to share some really cool, high-value use cases for AI within the context of industrial operations. Lightning round…
Predictive maintenance is the approach of monitoring the conditions of a machine to look for indicators of upcoming machine health issues. It’s reasonably standard to monitor the vibration, temperature, our power parameters of critical equipment – AI can give this approach a huge boost by developing more advanced algorithms that improve accuracy and lead time for machine health issues. Predictive maintenance is among the most widely applied use cases for AI in industrial operations.
While Sustainability is often just a tab on the corporate website, large industrials now see sustainability as a cost avoidance strategy. Continuous process manufacturing is a great playground for sustainability AI, since those processes often require massive amounts of heat, pressure, electricity, water, and steam – and hefty fines for waste and compliance violations. Through AI and real-time process control integrations, operators can meaningfully peak shift and load balance to save money, both on power bills and compliance. Meanwhile real-time predictive steam systems can integrate data from boilers, pumps/valves, and meters to reduce water/steam consumption by substantial amounts.
We need AI to become more prevalent within industrial operations, because we’d all obviously benefit from operations that are more reliable, sustainable, secure, predictable, diverse, safe, and of higher quality.
Industrial cyber security is a maturing area where AI is being used for anomaly detection – these solutions baseline the normal industrial process/network behavior, and then look for deviations that might indicate a breach. Now that industrial control systems are being targeted maliciously, the space is developing quickly – the industrial malware events of 2017 (WannaCry and Petya) were major accelerators, and now nearly all industrial enterprises have bought a solution or have it high on their roadmaps.
Scheduling has historically been a highly-manual process for manufacturers, requiring staff to spend tedious hours making plans for materials, equipment, and labor. Novel AI solutions can integrate both internal and external data to rapidly produce optimized production plans. Manufacturers are leveraging AI-based scheduling tools to save big on production time, on-time delivery, reduced inventory, machine utilization, setup times, labor flexibility, and more.
I never thought of Hiring as a cool AI use case, but I’ve seen a few different tools on the market than leverage AI to better understand skillsets based on resumes and prior experience, in order to make recommendations about who would excel in different roles. Much of this is focused around hiring people based on specific skills/competencies for managing equipment and executing extremely specialized tasks. Some tools go as far as to normalize applicant profiles to filter out a range of biases.
Safety is a wholesome use case for AI in industrial operations. Slips, trips, falls, sprains, strains, and tears lead to untold lost hours and worker’s compensation. There are several AI solutions on the market (often including wearables) that can help industrial workers predictively avoid these kinds of injuries. Forklifts alone account for tens of thousands of serious injuries annually in the US – there are image analysis systems that can identify a person, and then shut down the forklift. Accurately identifying a human vs. another object is important to avoid the system becoming a nuisance and constantly shutting down the forklift.
Since we’re already talking about Image Analysis, let’s move along to some of its use cases in quality and reliability. Image systems can leverage AI to build a baseline of what a product should look like (especially high-volume discrete products) and then identify quality defects based on deviations from the baseline. Some systems can even watch people manually assembling discrete products to ensure people are doing their jobs optimally. Meanwhile many operators point AI camera systems at critical equipment, and the systems can learn to visually detect anomalies that indicate upcoming machine health issues (a cool take on predictive maintenance).
I’ll stop there since this isn’t an encyclopedia article… but I certainly haven’t cornered the market on industrial AI use cases, so please post with additional use cases that you think are valuable! And if any of the aforementioned categories are relevant for you, please feel free to reach out and I can share some insight, or connect you with potential partners in the space.
That all said, we need more industrial companies to adopt AI at scale. The use cases described above are not in production at most industrial companies, just at the more innovative ones. If you look at recent studies from the likes of PwC and McKinsey, most of them report that less than 25% of industrial companies are using AI at scale in production operations (many of the studies say less than 10%).
We need AI to become more prevalent within industrial operations, because we’d all obviously benefit from operations that are more reliable, sustainable, secure, predictable, diverse, safe, and of higher quality. Data acquisition hardware continues to become cheaper, while AI tools continue to become more available to non-data scientists that can focus on novel operational use cases – but the tech is the easy part… we need more believers, more cultural change, and more proof of ROI.