- By Renee Bassett
- Cover Story
- If you are deploying IoT, deploy AI with it. Each has value alone, but they offer greater power when combined.
- Industrial AI applications fall into three categories: AI for assets, AI for processes, or AI for operational excellence and/or business agility.
- When starting a pilot project, aim for fairly soft outcomes and focus on worker augmentation, not worker replacement.
Here’s how to understand what industrial AI can do, how IoT feeds it, and how to start a pilot project of your own
By Renee Bassett
There is nothing "artificial" about the intelligence that can be gleaned from the detailed monitoring of machines, processes, and the people who interact with them. Ever since the time and motion studies of the efficiency experts of the early 1900s, industrial engineers have been turning real-time data into information and decisions that could improve productivity, efficiency, and profits. With the fourth industrial revolution upon us now, artificial intelligence (AI) technology is ready to go to work in ways that are not always obvious.
According to a Gartner Group forecast, The Business Value of Artificial Intelligence Worldwide, 2017-2025, AI and Internet of Things (IoT) "already work together in our daily lives without us even noticing. Think Google Maps, Netflix, Siri, and Alexa, for example. Organizations across industries are waking up to the potential. By 2022, more than 80 percent of enterprise IoT projects will have an AI component-up from less than 10 percent today" (2018).
The takeaway is clear, says data analytics software provider SAS: "If you're deploying IoT, deploy AI with it. If you're developing AI, think about the gains you can make by combining it with IoT. Either one has value alone, but they offer their greatest power when combined. IoT provides the massive amount of data that AI needs for learning. AI transforms that data into meaningful, real-time insight on which IoT devices can act."
AI and machine learning
Artificial intelligence uses a variety of statistical and computational techniques and encompasses a number of terms. Machine learning (ML), a subset of AI, identifies patterns and anomalies in data from smart sensors and devices without being explicitly programmed where to look. Over time, ML algorithms "learn" how to deliver more accurate results.
Because of this learning, "ML outperforms traditional business intelligence tools and makes operational predictions many times faster and more accurately than systems based on rules, thresholds, or schedules," according to SAS. "AI separates signal from noise, giving rise to advanced IoT devices that can learn from their interactions with users, service providers, and other devices in the ecosystem."
"The challenge is that people have not developed the level of trust in artificial intelligence and machine learning that they have in other technologies that automate tasks," says Oliver Schabenberger, COO and CTO of SAS. "People sometimes confuse automation with autonomy, he adds. But have no fear: "AI does not eliminate the need for humans, it just enables them to do their work more effectively," he says.
AI, around since the 1950s, is becoming a mainstream application as a result of the explosion in IoT data volume, high-speed connectivity, and high-performance computing.
Defining AI applications
Industrial AI can range from low-intelligence applications like automation to higher-end intelligence capable of decision making. It can also be controlled centrally or distributed across multiple machines. According to Gartner vice president and analyst Jorge Lopez, AI applications can be broken down into five levels of sophistication:
Reactors follow simple rules but can respond to changing circumstances within limits (such as basic drones).
- Categorizers recognize types of things and can take simple actions to deal with them within a controlled environment (warehouse robots).
- Responders serve the needs of others by figuring out questions and situations (driverless cars, personal assistants).
- Learners gather information from multiple sources to solve complex problems (IBM Watson, wholly automated military drones).
- Creators initiate a paradigm shift, such as inventing a new business model. They are not merely tools that people use; they have the potential to engineer actions harmful to humans. They will change humans' relationship to technology as well as people's roles within society and the economy, says Gartner. Therefore, "AI creator applications require profound thought before development."
These five artificial intelligence models have three types of organization, says Gartner: standalone, federation, or swarm. A standalone AI system is an individual entity that acts by itself to solve problems. The enterprise exercises centralized control over it by overseeing the entity as it performs.
In a federation structure, says Gartner, multiple versions of an entity work in the same way but on different problems (e.g., robo-advisors, personal assistants). The enterprise can exercise central control or give more autonomy to the entities. In a swarm structure, multiple entities work together on the same problem (e.g., Intel light show drones, Perdix drones). Control over execution is left to the machines entirely or requires only light human management.
Early AI adopters like retail and banking firms have reaped the benefits of AI, but it is not too late for fast followers, according to Petuum. AI has caught the attention of industrial innovators and naysayers alike.
Source: McKinsey & Company
More than automation
The most common place to start with AI is with automation, but experts say it is a mistake to stop there. The more powerful use of AI is to aid human decision making and interactions. Because AI can classify information and make predictions faster and at higher volumes than humans can accomplish on their own, those terabytes of data being produced by industrial IoT devices are being transformed into powerful tools today.
In a recent blog post for industrial AI startup Petuum, author Atif Aziz says, "Some industry leaders are zooming past the basics: digitization, cloud infrastructure, monitoring and dashboards. They are putting newly acquired data to good use through AI-driven advanced analytics (e.g., uncovering patterns through system of systems) and automating complex processes. Some early adopters are implementing as many as 100 digital transformation initiatives simultaneously or using AI to automate their core production processes across 30 or more plants," Aziz says.
On the other end of the spectrum, "some folks still need to understand how AI can provide real value and balance the ROI with their limited resources," says Aziz. "The breakneck speed of advancement in the Industrial AI/ML space over the last three years affords a unique advantage for these newcomers. They can skip many of the expensive intermediate steps (e.g., significant investments in data aggregation infrastructure, dashboards, and monitoring centers) and gain the same AI benefits as the savvier early adopters."
Aziz says most industrial AI initiatives fall into three categories. AI for assets includes equipment automation, equipment stabilization, and equipment health. AI for processes includes yield maximization through efficiency gains, automation and stabilization across multiple assets or spanning multiple flows, and quality improvement. AI for operational excellence and/or business agility includes energy cost optimization, predictive maintenance, logistics and scheduling, research and development, and more.
AI for assets
IBM Watson IoT helps organizations make smarter decisions about asset management by combining IoT data with cognitive insights driven by AI. IBM's Maximo enterprise asset management (EAM) system uses Watson IoT technology to make better decisions about critical physical assets in industrial plants-whether they are discrete machines, complex functional asset systems, or human assets.
One Maximo user, Ivan de Lorenzo, is outage planning manager for Cheniere Energy, a Houston-based liquefied natural gas producer. He says that, with the software, "we have better information on assets and maintenance activity, and more sophisticated tools and mechanisms for managing it all. The result is greater operational control and accountability, especially when it comes to planning and scheduling."
AI-based asset life-cycle and maintenance management solutions like Maximo use real-time data collection, diagnostic, and analysis tools to extend an asset's usable life cycle. Use of the software also improves overall maintenance best practices; meets increasingly complex health, safety, and environmental requirements; and controls operational risk by embedding risk management into everyday business processes.
IBM says EAM also helps "control the brain drain among employees facing retirement by [putting] into place proven workflows and enforced best practices that capture the knowledge and critical skills of long-time employees." Such a system also helps a reduced workforce to work more efficiently and cost effectively "by using the captured intellectual experience of skilled workers in a format easily dispersed in a wide range of languages."
AI for processes
AI systems are being used to improve whole processes as well as industrial assets. In an MIT Technology Review Insights publication produced in conjunction with IBM, Raytheon senior principal systems engineer Chris Finlay describes the benefits of replacing document-based information exchange with an AI-compatible digital platform to support engineering and design. "Once you start to capture things digitally, you can start to exploit machine learning or AI algorithms," he says. "You can start to reduce development costs because you can automate tasks that you were doing by hand."
Joe Schmid, director of worldwide sales for IBM Watson Internet of Things, says, "In the engineering process, you define what you want to do, design it, build it, test it, and prove that you've done it. The key is integrating those steps. But integrating is hard."
Customers that Schmid has worked with are often good at one part of the process, such as design, but they do not integrate design into the life cycle. "When they need to change goals or specs, it's all in people's heads," he says. "That doesn't work anymore with the complex systems we have today. One engineer can't have an entire system in their head. That's when errors pop up."
The goal of AI for engineering processes is to create an integrated "system of systems," a closed loop that runs from the requirements phase of product development to real-time monitoring of how consumers are using the product, and then deploy AI systems to analyze the data and leverage that knowledge to improve the product, says Dibbe Edwards, vice president of IBM Watson IoT connected products offerings.
In another example, global building materials company Cemex is on an industry 4.0 journey toward enhanced standardized operations using AI. The ultimate goals are increased efficiencies, reduced fuel and energy consumption, better quality, reduced costs, and improved decision making. The company announced in March that it had installed "AI-based autopilots" for its rotary kiln and clinker cooler systems that will "autosteer" its cement plants and enable autonomous, operator-supervised plant operations.
Cemex used OSIsoft PI systems to power Petuum Industrial AI Autopilot products. The two work with plant control systems to provide precise real-time forecasts for significant process variables, prescriptions for critical control variables, and a supervised autosteer function aligned with business objectives while staying within applicable static and dynamic constraints. The PI systems fuel real-time predictive and prescriptive recommendations.
Rodrigo Quintero, operations digital technologies manager for Cemex, says, "Petuum Industrial AI Autopilot helped us achieve something we didn't think was possible at this time: yield improvements and energy savings up to 7 percent, which is game changing for our industry. Additionally, this is a giant step in digital transformation toward safe, highly standardized operations, that will help us strengthen our high-quality products portfolio while also ensuring we meet our operational and sustainability goals, and minimize costs."
The Autopilot products can ingest data from a variety of sources, including unstructured, images, structured, time series, customer relationship management (CRM) data, enterprise resource planning (ERP) data, and others. The Petuum platform provides sophisticated data processing, data cleansing, and machine/deep learning pipelines to implement advanced AI that is sensitive to linear, temporal, long range, and nonlinear data patterns in a range of industrial use cases.
AI for operational excellence
Staying ahead of maintenance and production challenges to keep precision metals rolling out of its plants on time is a high priority for Ulbrich Stainless Steel & Specialty Metals. That is why the global company chose SAS Analytics for IoT to gain access to the latest suite of AI, machine learning, and streaming analytics available to analyze the data from plant sensors.
Jay Cei, COO at Ulbrich, says, "Collecting machine and sensor data from our factories and integrating that with ERP system data will help us understand the intricate relationships between equipment, people, suppliers, and customers.
Learning what their IoT data means is critical for understanding how the company can become more productive and efficient in the future, Cei says. DJ Penix, president of SAS implementation partner Pinnacle Solutions, says, "Streaming analytics will not only help Ulbrich understand what is happening now with their machines. It will also enable them to predict future events, such as when a machine needs maintenance before it breaks down."
The software provides a simplified way for any user to prepare stationary and streaming IoT data for analysis without specialized skills, says Penex. Whether a data scientist, business manager, or someone in between, they can use SAS Analytics for IoT to quickly select, launch, transform, and operationalize IoT data, he says.
Jason Mann, vice president of IoT at SAS, says companies can no longer afford to ignore the hidden signals in their IoT data. "To thrive, organizations need a solution that addresses data complexity and automates timely and accurate decision making," he adds.
Tips for AI pilot projects
According to a recent Gartner survey, 37 percent of organizations are still looking to define their AI strategies, while 35 percent are struggling to identify suitable use cases. Once you have developed a solid understanding of AI and its potential applications, it is time to make a case for a pilot. Here are some tips from Gartner for making the pilot project a success.
- Be realistic about a timeline. Once you have approval from executives, it can be tempting to think a pilot project will follow quickly. In fact, according to results from Gartner's 2017 Annual Enterprise Survey, 58 percent of respondents in companies currently piloting AI projects say it took two or more years to reach the piloting phase, and only 28 percent of respondents reported getting past the planning stage in the first year.
- Aim for fairly soft outcomes, such as improvements to processes, customer satisfaction, products, or financial benchmarking. Gartner Research Circle respondents urged others not to fall into the trap of seeking only immediate monetary gains. Aim initially for less-quantifiable benefits from which financial gains would eventually arise.
- Focus on worker augmentation, not worker replacement. AI's potential to reduce staff head count attracts the attention of senior business executives as a potential cost-saving initiative. A more informed expectation, however, is for applications that help and improve human endeavors, as AI promises benefits far beyond automation. Organizations that embrace this perspective are more likely to find workers eager to embrace AI.
- Plan for the transfer of knowledge from external service providers and vendors to enterprise information technology and business workers. External service providers can play a key role in planning and delivering AI-powered software, and knowledge transfer is crucial. AI requires new skills and a new way of thinking about problems. These include technical knowledge in specific AI technologies, data science, maintaining quality data, problem domain expertise, and skills to monitor, maintain, and govern the environment.
- Choose AI solutions that offer tracking and revealing AI decisions, ideally using action audit trails and features that visualize or explain results. To that end, Gartner predicts that by 2022, enterprise AI projects with built-in transparency will be twice as likely to receive funding from CIOs.
- Start small; do not worry about immediate return on investment. Digital transformation should begin with small experiments that are purely for learning, says Gartner. Use the time to pilot projects that employ a variety of technologies to assess which make the most sense for the business.
We want to hear from you! Please send us your comments and questions about this topic to InTechmagazine@isa.org.