Climate change poses a considerable threat to humanity, and is not just a problem that should be left for future generations. If greenhouse gas emissions are not almost entirely eliminated by 2050, the world will face catastrophic consequences with regards to global temperature. While many industries have taken steps to reduce their emissions, there is still a reluctance to move in this direction because of the falsely presumed costs of looking for greener solutions.
Manufacturing, logistics, and building materials are some of the leading producers of greenhouse gas emissions. According to the Natural Resources Defense Council, the top 15 U.S. food and beverage companies generate more greenhouse gases each year than the entirety of Australia. However, with the adoption of machine learning (ML), the industry is also well-placed to make a positive climate impact.
ML has been heralded as a powerful tool for technological progress, but despite it being used to fight global food shortages, there has been a lack of effort to identify how it can be used for other environmental purposes. ML in factories, across various sectors, can contribute to reducing global emissions by driving energy efficiency, streamlining the supply chain, and improving production quality.
A common problem found in factories is that machines are operating at a capacity which is too high for the required output. So many factories could be so much more efficient by consuming less power from individual machines without affecting their performance. If a machine only needs 25% of its maximum power draw to run a line at a particular speed, a variety of ML techniques can highlight this by correlating power and production data, allowing you to optimize your power settings even if optimal settings change as machines get older.
In the case of power plants, ML can be used with thermal imaging to determine which parts of a plant are at excessive temperature levels and modulate how much power they apply to that part of the plant. If factories adopt a similar principle for electricity consumption, then this would further increase efficiency. One way of doing this could be streamlining a factory’s heating, ventilation, and air conditioning systems. Although it would be a potentially big investment, a factory could also use ML to simulate the production output and power consumption of their factories under different energy sources, making it easier to redesign industrial processes to run on low-carbon energy instead of coal, oil, and gas.
Aside from energy consumption, factories can also use ML to catch defective products before they are produced (for example, by using computer vision to spot product defects early in their lines, or using historical data to predict causes of error before they happen) significantly reducing their scrap. This has a variety of benefits: less time would be required to produce the same throughput, less time would be wasted on bad throughput, and fewer emissions would be produced on scrapped goods. If the scrap of a factory worth $100 million were to be reduced by 10%, it would have the equivalent effect on emissions of taking 2000 cars off the road for a year. This highlights the huge impact which seemingly minor changes in a factory could have on emission levels.
In many factories, products are overproduced or overstocked. This wastes resources through production but also leads to increased emissions from the toll on shipping and storage. ML can reduce this by forecasting the demand. In the example of the food industry, it could lead to decreased post-harvest losses by identifying when a product might be about to spoil, using either quantitative algorithms to track shelf life or even computer vision to track how changes in color mean food is getting closer to spoiling. If the strain on storage was reduced by streamlining the supply network in this way, then a higher percentage of products could be sold because the right products would be shipped when the demand was there. Theoretically, ML could also be used to help a factory set up a network of suppliers, based on categories such as geography and the age of the company, to build algorithms which help think through the decision-making process of which supplier to choose.
ML techniques such as computer vision also enable factories to "grade" and document the quality of their products. Performing these grades according to any widely accepted standard allows suppliers to provide a level of certification to their product, which gives confidence to potential customers and broadens the markets they can reach. As an example, due to heavy tariffs between the U.S. and China on steel goods, steel often gets shipped through third-party countries, diminishing an end customer's guarantee of final quality. ML-based inspection and certification, either on the supplier or customer side, makes it easier for American steel users to source steel from more countries.
Many sectors of manufacturing suffer from wasted materials and loss of energy in the production process. For example, in steel production, there is a lot of modification and heat transfer during forming which leads to substantial energy loss. 1.8 tons of CO2 are emitted across the supply chain per tonne of steel produced, and 9% of global greenhouse gas is emitted during cement and steel production. In factories where plastic is produced, there is a lot of waste from materials not being recycled because plastic cannot be recycled to the same extent as metals.
These two types of production could see the most drastic change in terms of a reduction in their scrap and waste. Rather than focusing on green solutions which many startups offer at additional costs, scrap and waste reduction should be the main incentive for these industries which in turn leads to energy savings and more sustainable production. Sustainability and clean energy solutions should not be seen as a luxury for factories that can afford them, but rather a byproduct of increasing efficiency. Factories can also increase their yield without going down a path of explicitly clean energy solutions — using less energy than needed is just good practice.
ML can help in the fight against climate change by refining manufacturing processes, which leads to improved efficiency, less energy consumption, and reduced emissions. In turn, these outcomes will allow factories to feel confident about shifting emphasis to cleaner production. Applying ML to tackle climate change can help to decarbonize the manufacturing sector, advance some of the techniques in ML which are still in their infancy, and benefit society as a whole.
Arjun Chandar and Hunter Ashmore are cofounders of IndustrialML Inc.
Timely, incisive articles delivered directly to your inbox.