The factory environment is a data scientist’s paradise: both highly multivariate and relatively quantifiable. And the increased use of large-scale data analysis in the manufacturing sector could mean good news—not just for recent graduates in statistics and computer science, but also for the U.S. economy.
After China overtook the United States to become the world’s largest manufacturer in 2010, some U.S. commentators worried that the trend toward building factories in countries with very low labor costs and more flexible supply chains was inevitable. But data-driven manufacturing, sometimes known as “smart manufacturing,” could provide a counterweight. Increased automation and more sophisticated robotics, driven by data science and fed with sufficient volumes of high-quality sensor data, could increase productivity dramatically and help regions with high labor costs stay competitive.
Automation and robotics are no longer solely engineering challenges; the increasing success of machine learning in applications such as computer vision, object manipulation and automated movement has made data science an increasingly relevant factor on the factory floor. Dedicated machine learning platforms for manufacturing have existed for several years, but such offerings remain relatively few in number and the full extent of the applications remains unexplored.
The increased adoption of these technologies may not be so far in the future since data is increasingly common in manufacturing. In addition to keeping ordinary financial and inventory databases, a manufacturer might take sensor measurements of dozens or hundreds of environmental and process variables inside a factory. Defense manufacturing giant Raytheon, for example, measures how many times a screw has been turned at its new plant.
In addition to driving automation, all that data can then be used to predictively model equipment failure rates, streamline inventory management, identify energy-inefficient components and even optimize factory floor space, among many other applications. Intel has deployed predictive analytics to prioritize its inspections of silicon chips, saving the company $3 million in manufacturing costs in 2012. And the applications need not even be particularly exotic; ordinary business intelligence software can be optimized for manufacturers in the form of manufacturing execution systems (MES). According to a 2013 report by Gartner, the global market for MES has increased by 50 percent since 2005.
But the primary obstacle standing between the wealth of manufacturing data and future applications may not even be technology; rather, it is the lack of data sharing and interoperability within individual companies. A 2013 joint report from Rockwell Automation and the UCLA Office of Information Technology noted that, while many manufacturers today use sophisticated software to optimize “each specific stage or operation of a manufacturing process… each is an island of efficiency.” After efforts to improve data quality and completeness, a strong push for data interoperability among departments within individual manufacturers is the next step toward data-driven innovation.
Data-driven innovation offers a major opportunity for the U.S. manufacturing sector, but it needs to garner support in federal, state and local governments if it is to permeate the industry rapidly and thoroughly enough to outpace the burgeoning efforts of countries such as China, India and Brazil. This support could come in many forms, including the president’s proposed National Network for Manufacturing Innovation (NNMI) which promises to be a useful platform for supporting and incentivizing entrepreneurs to build “smart manufacturing” plants. Government support through the NNMI and other avenues will result in cheaper, safer, more environmentally friendly factories and could improve the United States’ position in the global manufacturing economy in the process.
