Throughout the Industrial Revolution, manufacturing processes transformed entire socioeconomic systems in Europe and the United States, while simultaneously accelerating factory productivity, raising working wages, and even increasing life expectancy in children and adults as parents were finally able to lead more balanced lives at home. Nobel Prize-winning Economics Professor Robert E. Lucas, Jr. advocates that quality of life has been the most measurable impact of automating industrial manufacturing. “For the first time in history, the living standards of the masses of ordinary people have begun to undergo sustained growth,” Lucas famously stated.
In the 21st century, machine learning (ML) software has vastly changed the way modern companies approach industrial planning, design, distribution, maintenance, waste, and energy management. Adaptability is key for industrial companies implementing lean manufacturing methods to boost production through statistical analysis, which is why companies like Fero Labs are merging historical data with predictive software to build an even smarter Fourth Industrial Revolution.
Fero Labs, a Newlab member company and frontrunner in AI-driven industrial improvement, marries machine learning and actionable insights. Their ML software automates and interprets industrial processes to offer targeted insights on how to boost industrial output, prevent costly machine breakdowns, reduce waste, improve product quality, and lower production costs. While algorithmic data collection already exists, Fero’s ML utilizes a unique “white box” protocol that makes manufacturing recommendations on how to enhance key performance indicators based on explainable changes to manufacturing processes.
Inside The White Box
Facebook and Google have long used behavioral data & algorithms to predict which advertisements have the highest click-through rate; GPS-based map systems use traffic data to determine the quickest driving route. Most companies rely on these “black box” models for developing ML applications, which have seen widespread adoption in the technology sector.
However, Fero’s “white box” approach relies on ML technology that provides actionable insights. This kind of ‘interpretable’ machine learning reveals the relationships between inputs and outputs that not only produces reliable predictions but also explains the decision process entirely, whereas “black box” models only make predictions based on historical data.
“It’s very expensive to make mistakes in a factory,” explains Fero Labs’ CEO & Electrical Engineer Dr. Berk Birand. “Our software tells clients how confident they are in their predictions, which is very valuable to the engineers. We can tell them why these decisions should be made, and why things went wrong with predictive confidence that can be mapped out.”
While a black box model might work for Amazon’s product recommendations, it presents no clear way to understand how the algorithm came to decide specific optimal settings. Dr. Birand expounds further: “Our white box ML learns the physics and reactions that are taking place in the system, detecting patterns from data in real time, evaluates the historical data, and then makes recommended corrections to the manufacturing process.”
Historical vs. Predictive Goals
Fero Labs’ software operates through a browser-based web-tool that reports yields and other quality control data. Dr. Birand details, “each company has its own nature. Each wants to understand whether their data is right and valuable. Our software pairs the historical data with the new set of predictive goals to determine the best processing refinements.”
Smart factory processing, particularly leveraging predictive goals, is essential to preventing product loss, manufacturing errors, and recalls. For instance, by using data sensors, an African gold mine identified a problem with oxygen levels during leaching. Once fixed, they were able to increase their yield by 3.7 percent, which saved them $20 million annually. The American Society for Quality recently estimated quality issues, such as product recalls, induce costs in the range of 40 percent of total operations.
Yet the rate of adoption for ML software and automation technologies has been surprisingly slow for large or small companies. Fero hopes to resolve that issue with an “altruistic and accessible” software. Smart factories will generate as much as $3.7 trillion in value by 2025, according to one estimate, but while many manufacturers have sensors in place, few use them to increase revenue-driving insights.
ML applications enable industries to “consume the massive raw data proactively and apply a machine learning algorithms,” offers Fero Labs’ Chief Scientist and Adjunct Columbia professor, Dr. Alp Kucukelbir. He describes how their software “establishes a confidence interval: a range of expected values based on data-driven metrics which optimize production, improve the quality of product and service, and predict machine failures. Modern ML can identify patterns with huge data sets across a variety of verticals, unraveling those patterns alongside human expertise.”
Eco-Friendly Industry 4.0
Dr. Kucukelbir co-founded Fero Labs with Dr. Birand after they noticed a huge need for smart automation in industrial manufacturing. “Problems in the industrial sector looked really interesting; it was the perfect time for the application of ML to solve problems that actually matter,” describes Dr. Kucukelbir.
Once only possible for enterprises with huge budgets, robotics & ML applications are now more affordable and available to organizations of every size, which also means a safer environment and improved energy usage. “For instance, steel has one of the most inefficient CO2 emissions,” Dr. Birand explains, “but a steel company recently using our software analyzed and developed a new alloy that could reduce their carbon imprint significantly and save cost on raw materials.”
A key component of Industry 4.0 – a buzzword blanketing the current trend of automation and data exchange in manufacturing technologies – is IoT. From internal operations to the cloud environment where data is stored, resources and operations in tandem can be optimized by leveraging the insights of how equipment is used daily, if not moment to moment.
Dr. Birand hopes companies large and small will adopt ML programs sooner rather than later, admitting: “Technology companies are often times, software companies. But the industrial sector has stopped and said: ‘hold on, we’re not software experts.’ A steel company knows it’s not a software company, nor should they be, but we knew their excellent engineers needed software to make stronger, higher-grade steel faster and more efficiently. We want to allow all users to do their jobs better by using interpretable data that’s actionable in minutes versus days or months.”
2019 poses to be a crucial year for Fero Labs with expansion into new market sectors. ML software has positioned itself as an integral component to operating high, efficient production rates, whether it’s in a booming steel and chemical industry to carbon-conscious ventures in renewable energy. Virtually any industrial manufacturer wishing to future-proof and streamline processes through data optimization is a prime candidate for Fero Labs’ software, which ensures the longevity of smarter protocols long into the 21st century.