How Machine Learning is Poised to Revolutionize Manufacturing

Machine learning empowers robotic systems to learn and grow with minimal human interaction

The manufacturing industry is on the path towards a fourth industrial revolution, thanks to the development of artificial intelligence and machine learning. This points to a future where many manufacturing tasks, such as predictive maintenance, supply chain management, and quality control, will be handled by complex machines and software solutions that require little human interaction.

Join us as we look at the growth of machine learning, how massive conglomerates like Siemens are utilizing it, and where the future may take us.

A Crash Course on Machine Learning

Machine learning is the application of human-like behavior to machine intelligence. It is, in essence, the use of learning algorithms to give A.I. programs the capacity to improve their efficiency and performance.

The algorithms do this by looking for patterns within the data they capture, and then use these patterns to make informed changes or recommendations. While pattern recognition was the origin of the concept, today’s machine learning algorithms go much further.

The Guts of a Machine Learning Server

Many machine learning solutions are powered by the cloud, as the hardware requirements are substantial. If you were to build your own PC for a deep learning algorithm, you would need to start with a GPU that has a lot of memory.

Data centers will use an NVIDIA Tesla P100. This is the world’s first AI super-computing data center GPU. It uses NVIDIA Pascal GPU architecture to create a unified platform for both HPC and AI. The Pascal architecture provides more than 21 teraflops of FP16 performance, while also offering 5 and 10 teraflops of double and single precision performance for High Performance Computering (HPC) workloads.

If you’re building something smaller, you can use the NVIDIA GeForce 900 series. Second, you’re going to need a lightning fast CPU, such as the Intel i7. At least 8GB of RAM will be required.

Cooling and power supply choices should be generous, especially if you’re overclocking the GPU to increase performance.

The hard drive you’re using should be a sold-state drive (SSD) for fast I/O with large data sets. When all is said and done, the costs come out to roughly $1,300 to build something capable of deep learning. This is why many machine learning solutions opt for AWS or Google Cloud servers to harness the hardware required.

How Machine Learning is Used Today

The ability to establish patterns and make changes that result in predictable outcomes quickly is the cornerstone of modern machine learning. While it has huge implications for manufacturing, this type of technology is already more prevalent than many of us realize:

• The medical industry uses machine learning to identify potential red flags that result in early diagnosis and improved treatment options
• Online advertisers use machine learning to understand which ads will appeal to you the most
• Machine learning is used to flag online transactions for potential fraud and identity theft
• Even things you interact with every day — like your phone’s personal assistant or the weather forecast — use this technology

Machine Learning in Manufacturing

Massive companies like Siemens, Intel, Kuka, Funac, and Microsoft are all placing significant investments into technology surrounding machine learning. This has given rise to the label “smart manufacturing.”

According to TrendForce, the global smart manufacturing market will cross $200 billion in 2018 and increase to $320 billion by 2020. Here are some examples of how this technology is changing the manufacturing landscape:

Siemens Mindsphere

Mindsphere, as described by Siemens, is a “smart cloud for industry.” It is used to monitor machine fleets for service purposes all around the world. The goal of this product is to monitor, record, and analyze everything in manufacturing.

By leveraging this type of big data, Mindsphere can find problems and suggest solutions long before humans ever knew they existed. Siemens often points to a specific case where their machine learning A.I. was able to best humans.

Norbert Gaus, Head of Research in Digitalization and Automation as Siemens, describes how the AI was able to improve emissions on gas turbines:

“Even after experts had done their best to optimize the turbine’s nitrous oxide emissions, our A.I. system was able to reduce emissions by an additional ten to fifteen percent.”

This had led to newer turbines that utilizes 500 sensors to monitor stress, pressure, temperature, and other data, which is fed to the neural network. Through machine learning algorithms, the network then achieves optimal conditions for combustion based on the data.

General Electric’s Brilliant Manufacturing Suite

GE is one of the largest manufacturing companies in the world, with a huge suite of products that range from industrial solutions to smaller home appliances. With 500 factories around the world, the company is taking steps to enable them with smart manufacturing technology.

The release of their Brilliant Manufacturing Suite offered a system for tracking and processing everything in the supply chain, to find issues before they emerge and pinpoint opportunities for higher efficiency.

This technology is fueled by Predix, an industrial internet of things platform. Predix combines machine learning with A.I. to leverage sensors that capture every step and monitor equipment constantly. Using deep learning algorithms, Predix spots both problems and solution.

GE expects Predix to process one million terabytes of data per day by 2020. It is already being utilized by seven GE factories that are serving as testing grounds for the product. Productivity has gone up, on-time delivery rates are higher, and unplanned downtime has been cut by as much as 20 percent.

A Recipe for Revolution

The robot revolution has already begun in the manufacturing industry. At this point, it’s only a question of how smart these robotic workers will become. As technology continues to grow, manufacturers are looking for ways to increase production and efficiency.

“Digital Twin” technology uses CAD models and simulations to predict potential issues or possible failures before they happen. Previously used by NASA to simulate spacecraft possibilities, it is now making the transition to industry.

Research firm Gartner predicts that Digital Twin technology will be used by half of industrial companies by 2021, resulting in a potential increase in efficiency of up to 10%.

Machine learning is also being used to provide software as a service (SaaS) that enhances supply chain management. IBM, for example, offers a Watson Supply Chain system that uses A.I to collect and monitor data from numerous sources — using machine learning to improve the exchange of information between all steps of the manufacturing process from warehousing to packaging.

The implications also extend into the physical space. With collaborative robot solutions appearing in factories all over the world, there’s a need for these robotic companions to become more intelligent. Machine learning can solve this problem.

Through the use of advanced algorithms, robots can learn through demonstrations and even improve their own processes through autonomous learning models. These robots can make informed decisions on how to improve their work simply by collecting and analyzing data, leading to better and more efficient quality control.

Modern uses of machine learning in robotic vision allow sensors to detect defects with incredible precision. This allows manufacturers to reduce waste and production delays.

The inclusion of machine learning in the manufacturing industry opens up numerous possibilities for optimization and efficiency.

A Glimpse into The Future of Artificial Intelligence in Manufacturing

Machine learning and artificial intelligence are the keys to the future of manufacturing. Michael Mendelson, a Curriculum Designer at the NVIDIA Deep Learning Institute explains the implications of machine learning:

“Without flexible algorithms, computers can only do what we tell them. Many tasks, especially those involving perception, can’t be translated into rule-based instructions. In a manufacturing context, some of the more immediately interesting applications will involve perception.”

Landing.ai, A startup founded by Andrew Ng, former chief scientist of Baidu, is already working on using utilizing advanced cameras and algorithms to spot microscopic issues with products like circuit boards. These types of defects are not possible to detect with the naked eye, but a properly equipped machine with this programming could train itself to detect them automatically.

Another step forward comes in the form of communication. For safety reasons, it would be ideal if humans could issue commands to their robotic partners. Current research is pointing towards the use of natural language to accomplish this feat.

The Robotics and Artificial Intelligence Lab at the University of Rochester was able to train a Baxter assembly robot to understand voice commands. It does this by taking the audio and transcribing it into text, which is then used to ascertain where the robot should go and what it should do.

These examples are just some of the ways that this technology is poised to change the landscape of manufacturing. Artificial Intelligence and machine learning are paving the way to a safer and smarter tomorrow.