ARTICLE

Machines excel at performing a large number of varying tasks. Computers, for example, excel at performing mathematical computations. Even today’s slowest personal computer has more computational power than the average human being by a wide margin. As an example, the Apple iPhone 4, released in 2010, had the computational power to perform roughly 1.6 Billion Floating Point Operations per second. As you can imagine, computational power has only increased since then. Of course, mathematics is not the only arena in which computers excel. For decades, machines have been defeating some of the world’s top chess players, the most famous example of this being the Deep Blue-Kasparov match that took place over 2 decades ago.

Although machines excel in certain circumstances, there are others in which they fail miserably. As it turns out, modern computers are only really good at operating within a set of formalized rules. Take the game of chess as an example of this. While there are hundreds or thousands of different strategies that a player can use throughout a game of chess, the rules of the game are set in stone. There are a limited number of pieces that can move in a limited space, and in a limited number of ways. These limits, the rules of the game, make it possible for a computer programmer to teach a computer how to best play the game. Similarly, mathematics is an arena where there is very little room for interpretation. Formulas and equations have only one correct answer (be it set of numbers or a set of output values), and getting to these answers is governed by a strong set of formalized rules.

But in areas where formalized rule sets are difficult or impossible to identify, human intelligence still reigns supreme. Ask a toddler to identify whether a photograph contains a cat or a dog, and most will be able to answer correctly. Ask the same of a machine and you are unlikely to get a coherent answer, let alone a correct one. Human beings learn differently than machines do. We apply our knowledge in different ways, and we excel at performing tasks that machines are incapable of for these reasons.

Deep Learning is a field of study within computer science that aims to equip machines with the capability to think more like human beings. Simply stated, its main goal is to provide machines with the ability to ‘think’ and to ‘learn’. As a theoretical practice, Deep Learning has been around for decades. Only in recent years, however, has computer hardware become fast enough to support real-world implementations of Deep Learning models.

A Deep Learning model is a software application that has been designed to mimic the way that human beings think about and process information. Like with traditional application development, the development of Deep Learning models requires developers (data scientists, statisticians, etc.) to identify variables in a data set that might have an effect on the outcome, or decision, made by the model. Unlike traditional software, however, the relationships between input variables, as well as the relationship between input variables and outcome(s) are not defined during initial development. Instead, Deep Learning models ‘learn’ these rules based on information found within the dataset itself. The method used by Deep Learning models to ‘learn’ the formal rules and relationships within a problem domain is complex and highly mathematical in nature.

Without getting bogged down in the intricacies of Deep Learning theory, we can still recognize that there are some advantages to this technology. Firstly, Deep Learning models are able to identify relationships without needing to define a set of formal rules and relationships at the time of development. This means that it is possible for Deep Learning models to identify relationships that human beings may have missed. It also means that Deep Learning models are more flexible than their traditionally developed counterparts. For a Deep Learning model, if a new data point contrasts with an existing rule, the data point and rule can be evaluated critically, and the model can be updated automatically if necessary.

Consider some concrete examples of how Deep Learning might directly change the manufacturing industry as we know it. Imagine a piece of software that knew how to determine the most optimal way to organize and schedule operations within your manufacturing environment. What if that same software had the capability to take in information and make decisions about whether or not scheduling changes needed to be made? Better yet, what if this same piece of software was able to automatically make adjustments to the schedule in real time based on data from the shop floor. What if it could update your suppliers, handle your logistics, and update your customers without requiring any human intervention whatsoever? With Deep Learning, this magical software is something that is a real possibility.

Here is another concrete example. Imagine that your production machines had the capability to determine when a defect might occur. Imagine that, before a defect was likely to occur, they could automatically take corrective action, thereby preventing the defect from ever happening. What if machines could think about their performance and automatically determine when they might need preventative maintenance? What if a machine could tell you what’s wrong with it and exactly how to fix it? Again, with Deep Learning, these things are not just a dream, but in fact a definite possibility.

Although the possible applications of Deep Learning are exciting and virtually endless, there are still some major hurdles to the successful adoption of Deep Learning in manufacturing. Firstly, while most manufacturers record and store large amounts of data, much of this data is not well labeled or classified. In order for Deep Learning to be truly effective, datasets must be uniform, well labeled, and understandable by Deep Learning applications. Integrating a new technology, like Deep Learning, into existing Manufacturing Operations is another major hurdle for the industry. How can the insights gleaned from the application of Deep Learning be used in existing processes? How can Deep Learning applications be integrated with software that is sometimes years, or even decades, old? These are questions all manufacturers must ask themselves today, because whether they choose to buy into it or not, their competitors surely will. The fact remains that the adoption of Industry 4.0, the Deep Learning concept and the technologies that come with it will determine the future leaders of industry.