In manufacturing, as in many other areas of life, technology has long played a leading role in setting trends - quite normally. However, for some time now, this leading role has been focused on software technologies. Recent concepts such as Cloud, Big Data, Data Mining, Internet of Things, Augmented Reality and finally Industry 4.0 are all trend structures and terms built on software technologies. The latest trend has entered our lives in three blocks; Artificial Intelligence, Machine Learning and Deep Learning.
Although artificial intelligence as a concept has been with us for a very long time, the possibility of programming an artificial intelligence that has the ability to learn on its own in real terms emerged with Machine Learning. At the stage of turning the possibility into reality, artificial neural networks were created inspired by the human brain to ensure completely autonomous learning by eliminating the need for human intervention, and thus we met Deep Learning. Here, we can say that Machine Learning is a process that aims to realize Artificial Intelligence, while Deep Learning is a sub-method created to autonomize Machine Learning.
Another remarkable thing about near-term trends is that all these technologies are clearly applicable to production systems. In fact, the logical link between production and technology trend concepts is so tight that these concepts can give the impression of belonging directly to the production methodology, as in the case of machine learning. This intertwining of manufacturing and software technologies also makes it feasible to adapt continuous improvement efforts to current technologies on the shop floor.
When the concepts of manufacturing and machine learning are considered together, the first innovation that comes to everyone's mind is predictive maintenance. As is known, predictive maintenance is an application based on the logic of monitoring and analyzing instantaneous machine operation data (vibration), which is key data, and then, when a mismatch is detected in this key data, identifying the source component of the mismatch and performing maintenance-repair before the failure occurs.
To better illustrate the difference between traditional Machine Learning and Deep Learning, the following example about predictive maintenance can be given:
The concept of key data in this definition is a good example to better illustrate the difference between traditional Machine Learning and Deep Learning. As mentioned earlier, the difference between Machine Learning and Deep Learning is the need for human intervention. Here, human intervention means the process of determining the key data for the machine. In traditional Machine Learning, key data such as vibration, temperature, oil values to be monitored are determined by us and algorithms are built accordingly. In Deep Learning, on the other hand, machine operation data is not given key data status and a higher meaning by an external factor. All operation data is distributed to artificial neural networks and the relationship-pattern between these data is learned and revealed by the Deep Learning algorithm by itself. In this way, connections are established between components and data that are not normally thought to be correlated, and these and similar correlations between artificial networks are taken into account during observation. Thus, as an external factor, any incompatibility in a seemingly insignificant operation data, which humans would not normally give key importance, can activate the warning mechanism and predictive maintenance can be realized.
In addition to predictive maintenance, it is of course possible to realize many more innovation studies on the adaptation of artificial intelligence methods to the production field. In particular, we are witnessing that process development ideas focused on the use of image processing technologies and autonomous vehicles in the production area are frequently voiced among teams that attach importance to continuous improvement. As such technologies based on Machine Learning diversify and evolve, the distance between us and the dark factory concept will gradually decrease.
Selçuk SÖZERİ
Software and Integration Specialist