Subscribe to MEM Magazine

MEM Technology

Four Ways to Reinvent Industrial Processes with Machine Learning

Four Ways to Reinvent Industrial Processes with Machine Learning

By Swami Sivasubramanian, Vice President, Amazon Machine Learning, AWS

The industrial revolution bore countless inventions and new products unlike any age in human history. While we revere the loom, steam engine, electricity, or mass production pioneered by the Ford Model T, we often overlook the inspired mechanisms and processes that made such incredible products possible. Consider the humble innovation found within ubiquitous manufacturing practices like equipment maintenance, quality assurance, and supply chain optimisation. These inventions are as critical to industrial and manufacturing processes today as they were more than a century ago but doing these successfully at the scale and complexity required in the current global market is challenging. Thanks to the convergence of data and machine learning, these enduring practices of industrial manufacturing are now poised to be reinvented.

Every day, companies are generating huge troves of data at the edge, storing this information in the cloud, and using those assets to rethink virtually all of their processes. To derive more insights from their data and ultimately drive faster and more informed decisions, companies in manufacturing, energy, mining, transportation, and agriculture are leveraging new types of machine technology to improve industrial workloads like engineering and design, production and asset optimisation, supply chain management, forecasting, quality management, smart products and machines, and more.

From operational efficiency to quality control and beyond, here are four keyways that companies are using machine learning to rethink industrial processes:

Predictive maintenance of equipment

A common, but significant challenge many industrial and manufacturing companies face today is the ongoing maintenance of their equipment. Historically, most equipment maintenance has been either reactive (after a machine breaks) or preventive (performed at regular intervals to help avoid machines breaking), with both being costly and inefficient practices. The best solution, predictive maintenance, gives companies the ability to foresee when equipment will need upkeep. However, most companies lack the necessary staff and expertise to build their own solution.

Thankfully, for companies like GE Power – a leading provider of power generation equipment, solutions and services – predictive maintenance is finally within grasp. There are now end-to-end systems that use sensors and machine learning to detect and alert companies of abnormal fluctuations in machinery vibration or temperature, with no machine learning or cloud experience required. This type of technology helped GE Power quickly retrofit assets with sensors and connect them to real-time analytics in the cloud, moving from time-based to predictive and prescriptive maintenance practices. As they scale, GE Power can use these systems to remotely update and maintain their fleet of sensors, without ever having to physically touch them.

Computer vision powered anomaly detection

Just as important as ensuring that equipment is functioning properly is guaranteeing the quality of the products that the equipment produces. The visual inspection of industrial processes typically requires human examination, which can be tedious and inconsistent. To improve quality control, industrial companies are looking to computer vision to provide greater speed and accuracy in identifying defects consistently. Once again, complex barriers had prevented companies from building, deploying, and managing their own machine learning-powered visual anomaly systems. Now, companies can use high accuracy, low-cost anomaly detection solutions that are able to process thousands of images an hour to spot defects and anomalies, and then report the images that differ from the baseline so that appropriate action can be taken.

For example, Dafgards, a household food manufacturer in Sweden, uses computer vision in the production of their brand Billy’s Pan Pizza, a microwaveable pizza baked and packed at a speed of 2 pizzas per second. While they had previously installed a machine vision system to detect proper cheese coverage on their pizzas, it failed to detect defects on pizzas with multiple toppings. By using a new machine learning service that leverages computer vision, they were able to easily and cost-effectively scale their inspection capability. The venture was so successful that Dafgards expanded the use of computer vision to multiple pizza varieties as well as other product lines such as hamburgers and quiches to name a few.

Improving operational efficiency

Many industrial and manufacturing companies are also looking to apply computer vision to assist in their efforts to optimise efficiency and improve operations. Today, companies manually review video feeds across their industrial sites to authenticate access to facilities, inspect shipments, and detect spills or other hazardous conditions. But doing this in real time is not only a difficult task, but error prone and expensive. And while companies might seek to upgrade existing internet protocol (IP) cameras for smart cameras that have enough processing power to run computer vision models, this can be expensive and even with smart cameras getting low latency performance with good accuracy can be challenging. Instead, industrial companies can use hardware appliances that allow them to add computer vision to existing on-premises cameras, or even use Software Development Kits to build new cameras that can run meaningful computer vision models at the edge.

Global energy company BP is looking to deploy computer vision at its 18,000 service stations worldwide. They are working to leverage computer vision to automate the entry and exit of fuel trucks to their facilities, and to verify that the correct order has been fulfilled. And computer vision can help alert workers if there is a collision risk, identify a foreign object in a dynamic exclusion zone, and detect any oil leaks.

Forecasting for supply chain optimisation

Today’s modern supply chains are complex global networks of manufacturers, suppliers, logistics, and retailers which requires sophisticated methods of sensing and adapting to customer demand, fluctuations in raw material availability, and external factors such as holidays, events, and even weather. The repercussions of not prognosticating these variables correctly can be costly, resulting in either over or under-provisioning and leading to wasted investment or poor customer experiences. To help foresee the future, companies are using machine learning to analyse time-series data and provide accurate forecasts that help them to reduce operating expenses and inefficiencies, ensure higher resource and product availability, deliver products faster, and lower costs.

Machine learning helped Foxconn, the world’s largest electronics manufacturer and technology solutions provider based in Taipei, Taiwan, when it faced unprecedented volatility in customer demand, supplies, and capacity as a result of the COVID-19 pandemic. The company developed a demand forecasting model for its factory in Mexico to generate accurate net order forecasts. Using the machine learning model, they were able to increase forecasting accuracy by 8 percent, a projected savings of $553,000 annually per facility, while minimising wasted labor and maximising customer satisfaction.

To live up to the potential that machine learning can provide industrial environments, manufactured products as well as logistics and supply chain operations, companies are increasingly looking to machine learning to make processes easier, faster, and more accurate. By using a combination of real-time data analysis in the cloud and machine learning at the edge, industrial companies are steadily turning their aspirations into realities and spurring the next industrial revolution.

Share this post

Featured MEM Technology

Subscribe to MEM Newsletters!