Manufacturing, Sustainability And Profitability: How Deep Learning Can Improve our Environmental Impact
This article was originally published on Forbes
In the mid-1700, the first industrial revolution changed the face of our society by upgrading the way goods were produced. At the core of this were coal-fueled machines. Fast-forward three hundred years, the manufacturing sector represents a huge chunk of human-caused carbon emissions. Within the United States, it is estimated that one-third of carbon emissions are originating from the industrial sector.
While we can’t change the past, our focus is now shifting to ensure a cleaner future. In 2018, the Intergovernmental Panel on Climate Change issued a stark warning, setting the ambitious goal to stabilize global warming at 1.5 degrees Celsius above pre-industrial levels to prevent reaching a point of non-return.
One way in which the US government is taking action is with the Inflation Reduction Act (IRA) of 2022, with a strong push in domestic energy investments and manufacturing, with the goal to slash carbon emissions by 40% by 2030. An integral part of this act is a $6 billion investment for manufacturers to cut emissions in steel, aluminum, cement and other energy-intensive industrial processes.
Swapping clean energy for carbon-emitting ones is surely an important task, with the IRA devoting $386 billion for climate energy spending and tax breaks, including a healthy $50 billion targeting clean-energy manufacturing.
What happens “beyond the power outlet,” though, is as important, if not more crucial. A wasteful, inefficient production floor will not magically improve by swapping coal with solar-powered energy. Radical changes are required to make sure manufacturers become less wasteful.
Just how inefficient are our production floors? A proxy for this ephemeral notion is provided by the number of product recalls, namely failure in production quality inspections. The picture is not encouraging:
• 15M: the record-setting number of vehicles recalled for electronic component defects
• $10M: Average costs to a food company for a product recall
• 2000: Number of products recalled by the U.S. Food & Drug Administration (FDA) (40% of which were related to medical devices)
• 256: Number of U.S. Consumer Products Safety Commission (CPSC) recalls, a figure including sports and recreation products, toys and children’s products, electronics, and appliances and furniture.
To counteract the monumental wastes outlined above, whose environmental impact won’t be affected by simply swapping energy sources, adding additional workers to check those products seems to be the right move.
Unfortunately, this won’t be the case. In the past 10 years, the US has faced an acute shortage of skilled labor. The very discrete process of assessing the quality of production absorbs today some 35 million workers—a workforce foreseen to be hit by a decrease, numbering tens of thousands in this decade alone.
Given that “throwing more people” at the issue is not a viable option, the industrial world is turning to AI and automation, leveraging these techs to embed human-level AI that can extract actionable insights directly on the production floor, augmenting human’s ability to tell good from bad production.
The Promise: No Product Will Leave The Factory Unchecked
Let’s take semiconductor manufacturing—a key sector today under pressure and causing huge cascades of issues in so many products, from the vehicles we drive to our microwave ovens. Building integrated circuits (IC) has an enormous carbon footprint, beginning with mining the raw materials all the way to chip fabrication. Wasting a chip is wasting “carbon” that is simply not replaced by using more clean energy. To reduce waste, IC manufacturers use a combination of techniques and resources, from visual checking performed by human operators to x-ray inspection, to functional testing, with the cost of the component growing with the amount of testing. Today, AI software can either replace or complement existing quality checks, indicating the type of failure/defects such as solder traces not complete, components missing or components in the wrong orientation.
How more wasteful we can get than throwing away food, whose production—think of meat alone—is a massive source of carbon emissions? A study released by the UK government has highlighted how human error on the production line is at the root of the 10.2 million tons of food that is wasted annually in that country. It is no surprise that global manufacturers are now quickly turning to technology to fix the issue, with AI at the forefront. Visual AI software is integrated today into existing infrastructure (inexpensive cameras, industrial PCs) to inspect raw materials such as meat for foreign contaminants (e.g., plastic particles) as well to inspect case packing with overwrap correctness of canned foods to ensure the palletizing process is not interrupted by poor overwrapping. Another food production process traditionally performed only intermittently via human oversight is the production of baked goods, such as cookies, bread and frozen food, with AI today enabling to check 100% of the products. All the way down the production line, AI can also be used to compare the barcodes of powdered milk with their packaging barcodes to verify that the correct expiration date is assigned to the right carton, preventing waste and recalls right before shipping.
And technology does not need to be expensive: enterprises want to embrace a greener future without giving up … the other green, namely their bottom line! The good news is that today, AI software platforms are available to make complex AI problems simple, provide integration hooks, hardware flexibility, ease of use by non-experts and the ability to work with little data.
To reach our ambitious climate goals, we need all the tools we can use. AI, with its multiplicative effect on the human workforce, is a mature technology that is already at the forefront of a smarter green revolution.