In today’s world of increasing speed and complexity, AI and machine learning technologies promise to empower manufacturers with the insights and capabilities they need to make their factories more predictive, adaptable, and efficient as they seek to secure a competitive advantage for the future. By Surya Kommareddy

Factories of the future are not so futuristic anymore. We already see factory floors teeming with robots that move relentlessly and synchronously to produce products, 24/7. Much of this automation, however, is still rudimentary, programmed to do a specific task, and inflexible to the changing needs of manufacturers.

But a new generation of productivity solutions, powered by artificial intelligence (AI), is now being developed that may ultimately change the very nature of production automation, and the definition of a factory of the future.

The Increasing Complexity of Manufacturing 

Automation in manufacturing operations is burgeoning into a complex “system of systems”, with many contributing factors that could affect the outcome of production.

With ever-more intense global competition and rising customer demands for faster delivery of better and cheaper products, manufacturers are increasingly deploying automation to be more productive and agile. But while automation has helped increase productivity, it has also increased the complexity of production. When issues arise in manufacturing operations and the supply chain, the increased speed of operations means more bad parts can be produced and shipped before the issue is caught. This can lead to large-scale and highly costly product recalls. A lack of skilled manufacturing labor also makes it harder for manufacturers to remediate issues at the heightened pace of today’s manufacturing operations.

Manufacturing operations are typically sequential build processes where raw material or semi-finished goods from suppliers are further processed to build the final product. In this process, a variety of resources in terms of Material, Manpower, Machines, Methods/Processes, and Management go into building the end product. Due to the dynamic nature of these factors, one cannot create an extensive set of rules to come up with a single universal formula for success. Considering the variability and uncertainty in each of these resources, it is a highly combinatorial problem to determine all factors that have led to poor quality, asset failure, or yield related issues. Figure 1 illustrates this highly combinatorial problem and the need for a solution that can handle this increasing complexity.

In this complex environment, companies often resort to designing models or other similar methods to troubleshoot and find root causes. But a typical troubleshooting process in complex operations could take anywhere from weeks to months. Stopping production or continuing to produce sub-par parts during this time could mean millions of dollars of lost revenue. Traditional triage methods and Business Intelligence (BI) tools do not provide the level of depth and insights needed to tackle these challenges rapidly.

More Data than Information 

With increased complexity and speed, of course, manufacturing operations can rapidly become data rich, but information poor. Manufacturing enterprises and supply chains are already replete with data that reflect the production and distribution of products, while automated equipment now includes a plethora of sensors. Since manufacturing data is produced at velocity, in a variety of formats, and in increasingly high volumes, human operators have begun to rely more on computer-based analytics to help them understand the performance and effectiveness of the systems and processes they use. It has become imperative that manufacturers embrace digitization so they can make sense of this data to uncover unforeseen insights that will help them deliver competitive advantage.

Manufacturing’s Digital Transformation 

That’s why the manufacturing industry is currently undergoing such a major transformation, moving away from the mechatronics-driven automation era, into a software-driven, data-rich, cyber-physical, Industry 4.0 era. With the latest innovations in IT such as Big Data, Cloud Computing, and AI and machine learning, along with cheaper and more capable sensors, a new generation of solutions and business capabilities are being developed.

Industry 4.0 enables a cyber-physical manufacturing environment where software digital tools bridge the gap between the physical and digital entities to create more powerful, digitally-enhanced, cyber-physical entities that can be integrated with the rest of a digitally-connected enterprise and business ecosystem. AI is becoming a key enabler for the advanced implementation of digital twins and other cyber-physical systems that help manufacturers create new business models to serve their customers better.

“It’s AI’s learning ability that makes it most effective in assisting humans in complex decision-making.”

What is AI? 

Artificial intelligence is the capability of a computer to imitate aspects of human intelligence and behavior related to learning, reasoning, and problem solving. Unlike the traditional “if-then-else” logic that computer programs are built upon, AI has the ability to self-learn and evolve by iteratively interacting with the data or environment.

It’s AI’s learning ability that makes it most effective in assisting humans in complex decision-making. The major advantage of AI algorithms in manufacturing is the ability to discover previously unknown insights and uncover the correlations of various influencing factors from the data sets. It is this ability to extract knowledge from complex, high-dimensional data, predict the future behavior of the systems, and provide results in a much shorter time compared to traditional methods, that make AI suitable for a host of manufacturing applications. The new insights provided by AI algorithms can then be used by process owners in speeding up their decision-making process, or harnessed by an autonomous agent for direct action.

AI Reduces the Cost of Prediction 

Prof. Ajay Agarwal, co-author of the 2018 book, Prediction Machines: The Simple Economics of Artificial Intelligence,posits that AI reduces the cost of prediction making it the single, most powerful transformative technology. The authors suggest business leaders use this economic premise to figure out the most valuable ways to apply AI in their organizations.

A number of recent industry reports also indicate that many companies expect both AI adoption and ROI to increase significantly in the near future. In a Boston Consulting Group Global AI 2018 survey, for example, 29 percent of the 1096 executive and production management participants considered AI as a very important driver of productivity improvement. The same group projected AI adoption would increase to 40 percent by 2030.

A 2018 KPMG Technology Industry CEO Outlook report, meanwhile, showed that 41 percent of the CEOs surveyed expect to see significant ROI from their AI investments over the next 3-5 years. The report also states that 51 percent have already implemented AI to automate specific processes.

And according to a recent report from McKinsey1, AI-supported predictive maintenance approaches would enable asset productivity increases of up to 20 percent, and overall maintenance costs would be reduced by up to 10 percent. The report also states that AI-enhanced supply chain management greatly improves forecasting accuracy, and potentially reduces errors anywhere between 20 percent and 50 percent; while reducing inventory levels between 20 percent and 50 percent.

Use Cases of AI in Manufacturing 

AI algorithms already have a myriad of applications in manufacturing where learning and predictive behaviors extracted from high-dimensional data can be leveraged in a dynamically changing environment – from design, to programming, planning, control, supply chain, and distribution.

Some of the prime use cases of AI applications include:

Robotic Process Automation (RPA) – RPA is a low-hanging fruit in the application of AI where business processes are automated using AI-driven, self-learning software robots or agents. Unlike the traditional way of utilizing APIs or scripting to automate a list of user interface or backend actions, RPA software agents observe human actions, learn from them, and then perform actions directly. RPA enables applications to automate highly repeatable, highly structured tasks across enterprise business software. When RPA is combined with AI-driven smart bots, the cognitive bots can handle ambiguity and make decisions just as humans would. Some prominent applications in manufacturing include automating invoice payment, order fulfilments, BOM, inventory management and reordering, and generic reporting.

3D Printing and Generative Designs – With 3D printing gaining traction in various industry verticals, it is now possible to print components with complex geometries that were previously unimaginable. Generative design is a new approach that leverages artificial intelligence and the power of cloud computing to churn and explore all possible combinations and select an optimal design that meets various design constraints and functional needs of the component. Increasingly aerospace, automobile, and other industries are adopting generative design software to come up with optimal component designs that are strong enough, yet very cost-effective to build. We will see more and more applications of AI in design, programming, and planning of production operations in the near future.

Predictive Maintenance – A Deloitte article2 on IoT applications in the oil and gas industry mentions that a single pump failure can cost from $100,000 to $300,000 a day in lost production. The same article also states that between 2009 and 2013, there were more than 2300 unscheduled refinery shutdowns in the United States alone, costing the equivalent of $20 billion per year. In this industry, a simple device or motor failure can lead to a huge catastrophe. As a result, finding the right time to service an asset based on its actual condition and utilization – as opposed to a fixed annual maintenance schedule – can greatly help increase asset utilization and performance. AI algorithms are being deployed to find correlations between equipment component state, external, and internal influencing factors to accurately predict when the asset is going to fail. The result is an increase in asset utilization, reduction of maintenance costs, and an increase in productivity due to a reduction in unnecessary disruption.

Digital Twins – A Digital Twin is a software replica that captures all the attributes and behaves exactly like its real-life physical entity. A Digital Twin can be of a product, a process, or even a person. It is a self-learning model that evolves as the physical device changes in real-time. Given this capability, its primary application is in implementing a cyber-physical system where the digital and physical entities interact to create a digital-physical-digital closed loop. The benefit of having a Digital Twin is in cost-effectively investigating the future by considering many “what-if” scenarios to determine an optimal solution to a production plan or a product issue.

Cobots – AI-driven robots are able to interact, learn, and autonomously make decisions in accomplishing assigned tasks. Cobots are a class of robots that accomplish their work while interacting with, and alongside, human workers. With the use of sensors and AI-based dynamic learning, Cobots enable a safe working environment while enabling a smooth interaction between machine and human operators, so increasing productivity and enabling new operational possibilities.

Vision-based Quality Checks – Product quality is a very important measure of production performance, yet quality control is still largely performed by humans with a highly developed vision and touch senses. However, humans are fallible when it comes to performing tasks that need repeated high attention, at high speed, for long periods of time. This is where AI has a strong role in vision-based quality checks. The latest generation of deep learning AI algorithms is well-suited for such applications, enabling automated, adaptable quality checks with very high precision, while working continuously 24/7.

“AI adoption is as much of a cultural shift as it is a digital transformation journey.”

The Path to AI Excellence 

Implementing AI in manufacturing is not straight-forward and manufacturing organizations need to understand how and where to apply the technology for best results. AI adoption is also as much of a cultural shift as it is a digital transformation journey. Manufacturing organizations must learn how to manage this shift proactively, build trust in delegating the decision-making process to a smart agent driven by AI, and build talent that can make this transition happen. The best AI-based solutions tend to shine where there is high dimensionality of data, variability, and uncertainty, and where there is a current limitation or lack of skilled staff to succeed.

There are many preceding steps that are required to achieving an AI-driven capability and maturity. The primary precursor is the availability of good data through the digitization process. Manufacturers should embark on their digital transformation journey as soon as possible in order to collect and store data from all entities in their supply chains as well as manufacturing operations. Once data is available, the next step is to identify and prioritize use cases where it makes business sense to deploy AI, either in an entirely automated fashion, or with a human in the loop. Finally, it is critical to decide whether to develop the data science capability in-house using open-source tools, or to leverage commercial solutions already available.

Are We There Yet? 

AI is clearly a fast emerging technology offering early adopters the promise of greater agility, adaptability, and new capabilities to help them compete more effectively. The current maturity level of AI technologies, and the lack of confidence and preparedness of many manufacturers, however, may not yet be conducive for fully autonomous manufacturing. But there is certainly huge potential. AI is already proving to be a valuable tool in making predictions and uncovering hidden insights from which humans can then act upon.

Where there is less need for human judgment, typically in repetitive, routine tasks, AI is also being deployed to function autonomously. AI smart agents, meanwhile, can also be valuable digital advisors that assist human operators in making manufacturing many types of activities more productive and profitable.

In the era of Industry 4.0, where technology innovation is happening at an exponential rate, those manufacturers that can leverage new AI technologies at an early stage are poised to gain maximum advantage. M