# Introduction

In 2015, renowned engineer and economist Klaus Schwab wrote that “we stand on the brink of a technological revolution that will fundamentally alter the way we live, work, and relate to one another.” The era to which he was referring is Industry 4.0, or the Fourth Industrial Revolution. The First occurred with the implementation of water and steam-powered machinery. The Second was ushered in by steel and electricity while the Third began with the use of computers and a shift to digital technology. Now, as breakthroughs like machine learning are seeing widespread adoption in manufacturing, we are witnessing the birth of an industrial age powered by the Internet of Things (IoT) and advancements in artificial intelligence (AI).

The benefits of implementing AI are already being taken advantage of by manufacturing companies around the globe, with major players like GE and General Motors successfully infusing AI into their production processes. The triumphs in AI for manufacturing exhibited by these and other companies are becoming more common as manufacturers seek out methods to increase efficiency and expand capabilities. With leaps forward in these technologies happening at an unprecedented rate, AI enhancing, or in some cases upheaving, manufacturing processes across industries is projected to become the new norm.

via GIPHY

# Use Cases

## Predictive Maintenance

A manufacturer likely isn’t making money if its machines aren’t running. In fact, it’s estimated that machine downtimes are responsible for USD $647 billion in losses globally each year. Throw in costs for labor, repairs, and replacements and machine maintenance and calibration become elements of operation with extreme financial influence. Traditionally, keeping machines up and running has relied heavily on preventative maintenance, or scheduled maintenance and calibration based on factors like manufacturer recommendations. An everyday example of preventative maintenance might be changing the oil in your car after the number of miles suggested in the manual. In the manufacturing world, employees might also use handheld sensors to collect data on factors like temperature and runtimes to get a better idea of when a machine needs attentions. It would be very difficult for a human worker to assess the amount of data at play in determining when a machine component might break, however, and and there is still much guesswork used to determine what information should be considered a key indicator of failure. This guesswork coupled with routine maintenance schedules can lead to both missed opportunities to fix a simple problem before a complete machine breakdown happens and functional machines being pulled from production before they actually need service. Enter AI for predictive maintenance, or the ability to monitor machine components to determine if a problem is looming before it happens. Manufacturers can use a combination of machine learning software and sensors to monitor machine parts more accurately than ever before. Sensors can send real-time information about how a machine is running and report issues, while the software can look at a component’s data and historical factors to predict that a problem will arise. A variety of sensors, including pressure, vibration, and current, can be implemented to effectively determine maintenance needs. Sound and vibration sensors are even being used to determine when tools are dulling out since sharp tools sound different than worn ones. Determining that a part is likely to break or need calibration before it causes a problem saves time and money, while the ability to know that a component does not yet need attention despite recommended maintenance schedules reduces downtime. The benefits of predictive maintenance expand into reduced waste, more efficient part management, and beyond. By 2025, predictive maintenance could have as much as a USD$630 billion per year economic impact.

## Quality Assurance (QA)

In a manufacturing world dominated by competition, consumer trust and consistent products are defining elements of success. Virtually all major manufacturers have QA systems in place to ensure that the products being made are up to standards and that no faulty items are being sent out into the world. While some of these systems are advanced and multi-faceted, most still rely on a human worker’s ability to catch flaws along each step of the process.

Traditional quality assurance processes typically require an employee to pull individual products and perform visual scans or functional tests. This process is repetitive and slow, allowing factors like worker fatigue to impact accuracy. Even the most skilled employee is subject to human error, making the implementation of AI an attractive alternative to increase efficiency.

With AI for quality assurance, a system can be taught what qualifies as a good product and what a defect looks like. Machine learning once again plays an important role as systems continually add information to their intelligence, expanding their knowledge on quality indicators. Computer vision technologies and cameras allow these systems to know how a proper product should appear and to see when something is awry.

AI for quality assurance also works hand-in-hand with AI for predictive maintenance. Since the latter can pick up anomalies in the manufacturing process, it can also alert manufacturers to potential production errors before they happen. This capability also saves time and resources as products that are flawed can be pulled from production before advancing further in the manufacturing process.

## Reinforcement Learning for Maximizing Throughput

Reinforcement learning refers to AI software that can make decisions on the best way to achieve a goal. The algorithms at play are rewarded for making correct decisions and penalized for making the wrong ones, hence the “reinforcement” part of the term. Famous instances of reinforcement algorithms have been trained to play video games and even to pit machine against human in competitions. Paired with deep learning, reinforcement algorithms can rapidly expand their skills and become better and better at achieving their goals.

Of course, reinforcement learning algorithms are good for more than just gaming and can greatly improve manufacturing operations. Reinforcement learning is particularly beneficial for simulation modeling. Software can create 2D or 3D renderings of areas such as warehouses and apply algorithms and equations to determine the best flow of traffic or manufacturing layouts to optimize efficiency if, for example, instructed that congestion is to be avoided. It can also be employed to determine whether or not a proposed system or component will actually work without the potential material and time wasted by making it in the real world. Beyond simulation modeling, reinforcement learning can maximize the efficiency of a machine or component by determining how to achieve factors such as maximum output while remaining within specified temperature or power requirements.

Reinforcement learning in manufacturing offers vast improvements to operations. Utilizing these algorithms can not only improve efficiency and safety, but also dramatically increase throughput.

# Market Growth

As AI technologies become more readily available, manufacturers across industries are turning to these advancements to lower operating costs, improve efficiency, and reduce time to market. Just how quickly this adoption of AI in manufacturing is happening is evident by notable projected market growth.

In 2017, the global AI in manufacturing market size was USD $513.6 million. That number is expected to hit USD$15,273.7 million in 2025, demonstrating a compound annual growth rate (CAGR) of 55.2%. This CAGR suggests that AI in manufacturing represents the future of production. Factory implementation of AI for predictive maintenance alone is expected to increase by 38% between 2017 and 2021, giving those manufacturers adopting the technology a competitive edge and less machine downtime.

# Results

With predictive maintenance keeping machines running, AI for quality assurance increasing output and consistency, and reinforcement learning optimizing both facilities and products, the benefits of AI in manufacturing are already being felt. One company reported seeing a 35% reduction in calibration and test time by using these technologies for determining calibration needs and avoiding production bottlenecks.

McKinsey reports that implementing predictive maintenance can increase asset productivity by 20% and reduce maintenance costs by 10%, while using AI for quality assurance can increase productivity by 50% and defect detection rates by 90%. Cost and time savings can also be seen in areas such as better employee morale, increased planning time, and more continuity. Overall, implementing AI and IoT applications in factories could unlock as much as USD $3.7 trillion in value by 2025. # Conclusion Implementing AI in existing factory settings may just be the beginning as a push for Smart Factories is also being seen. These facilities would rely on artificial intelligence and robotics to run autonomously and may even have the ability to self-correct, dramatically increasing production and efficiency while giving human workers the opportunity to focus on other tasks. In 2018, LG announced its plans to invest USD$524.60 million in a Smart Factory in Changwong City. The facility, which will implement technologies including artificial intelligence and big data solutions, will be able to produce 50% more appliances per year than its current facility.

As we move further into Industry 4.0, manufacturers will likely continue to adopt AI in nearly all aspects of their business. Regardless of whether implementing AI in existing factories or the launch of a new generation of Smart Factories represents what is to come, it is clear that AI is the future of manufacturing and those hesitant to embrace these technologies may fall behind the competition.

## Why Skymind?

• Our software deploys to edge devices that can be used to monitor factories, assembly lines and heavy equipment.
• Experience working with major automakers.
• Machine-learning models that produce state-of-the-art accuracy.
• Focus on producing business value.