by Kevin Madden
“Machine learning” (ML) is a popular term used by Industry 4.0 enthusiasts. But while there is undoubtedly excitement surrounding the phrase and its potential implications for the manufacturing industry, many people have a difficult time explaining exactly what ML is. It is crucial that manufacturers have a high-level understanding of this concept, so they can realize the benefits of ML in their organizations.
What is machine learning?
ML is the branch of artificial intelligence that deals with analyzing big data. Specifically, it uses algorithms that enable computers to gather insights from data without explicitly being programmed to do so. With the seemingly endless flow of information in today’s world, having machines that are able to process and analyze this data on their own is extremely useful.
The two primary types of ML are supervised learning and unsupervised learning. Supervised learning is best thought of as a ML by example. It occurs when a computer takes in labelled data, uses an algorithm to determine the relationship between different attributes of the data, and then predicts future outcomes based on this learned information. With supervised learning, the more data the computer has access to, the better its predictive capability becomes because it is consistently able to learn from the data. Unsupervised learning is when a machine can group together clusters of unlabelled data based on characteristics that it can detect between similar observations.
The most applicable process with respect to manufacturing is supervised learning, as manufacturers tend to have access to a large amount of labelled data. Advanced sensors in manufacturing plants record real-time data on thousands of attributes down to the millisecond, all of which have a distinct label; this gives computers countless “examples” to learn from in order to refine their inferencing capabilities. As the computer takes in these examples, it better understands the relationships between different data points.
Applications of machine learning in advanced manufacturing
There are many important applications of ML in advanced manufacturing, but two in particular are worth mentioning: first, ML can be used to detect and predict problems in a company’s manufacturing system; and second, ML can be used in image recognition, reliably scanning images of products to check for problems, product quality, and other notable aspects of the product.
One of the biggest challenges that manufacturers have always faced is identifying problems in complex manufacturing systems in a timely manner; every minute spent with the plant not operating at its full capacity corresponds to lost profit for the business. ML can help reduce this time by spotting problems before they occur and quickly determining what the problem is when an issue arises. Through supervised learning, ML enables computers to learn every detail of the entire manufacturing process. The computers can then monitor the wear and tear of various tools and machines and general plant conditions, and detect patterns in the data. Using this set of data inputs, the computers are able to predict where and when failures will occur. If a computer fails to predict a problem before it arises, it can also be effective in identifying and resolving the issue quickly by understanding which conditions correspond to which problems. Further, the longer a computer uses ML, the more data it has access to and the better it becomes at performing these tasks.
In addition to detecting problems in the manufacturing system as a whole, ML can be used to monitor product quality through image recognition. Manufacturers today are increasingly using cameras to take high-quality pictures of products at various stages of production, with the hope of increasing quality assurance. ML can be leveraged to classify products as damaged (or not) by identifying irregularities in the images. By integrating ML into manufacturing systems in these ways, companies can gain greater control over their product quality.
What does ML mean for Ontario manufacturing?
While the increased use of ML seems to contribute to the fearful narrative that machines will eventually steal humans’ jobs, the reality is that humans are still invaluable to the manufacturing process because of our diverse background knowledge. ML provides companies with massive computational power to record and analyze big data, but humans are required to interpret the results ML is giving us and apply them to the big picture—to refine the algorithms, formulate new hypotheses, and undertake data-driven decision making. The real benefits of ML lie in combining human knowledge with the computational power of machines to make better, more informed decisions.
Ontario is fortunate to possess a wealth of manufacturing knowledge, which has been developed over many years. Understanding how ML works should help manufacturers intertwine human knowledge with the increased efficiency offered by ML to make Ontario’s manufacturing sector even more competitive on a global scale.