February 2, 2017
By: Lyat Avidor Peleg
Machine Learning Ignited the Industry 4.0 Revolution – What Does It Take To Keep On Fueling It?
Machine Learning – the driving force behind Industry 4.0
Machine Learning (ML) is a moving force behind Industry 4.0. The ability to analyze, in real-time, amounts of data collected by sensors is key to optimizing manufacturing processes. ML algorithms have the potential to bring greater predictive accuracy to every phase of production. It can contribute immensely to the growth of technologies on one hand, and to the adaption of companies to market changes on the other. By 2020, it is projected that over 1 billion connected objects will equip factories – a huge increase from the 237 million objects in place today (here), and Machine Learning is taking a leading part in this change. However, when it comes to failure detection and predictive maintenance in complex plants of the process industry, Machine Learning does not offer a complete solution.
What is Machine Learning?
In a nutshell, ML is a technique for learning things by experience without rules dictated by people; it changes its program’s behavior based on what it learns from the data. ML enables the prediction and detection of anomalies when sensor data patterns indicate behavior outside of normal. It enables predictions that arise from the patterns in the existing data. In cases of an abundance of history, ML would detect innumerable anomalies. The machine needs to have references in order to build a memory which will enable it make choices according to situations with particular contexts.
Machine Learning in complex plants
In complex plants, the subject of our discussion, due to the complexity of the plant and the fact that the plant is heavily connected, there are innumerable possible states of the plant. This fact causes the ML algorithms to produce a very large number of anomalies from the data. At any given moment, there is a large probability for the occurrence of an anomaly, hence we witness hundreds of anomalies a day. To the user, this is noise, especially since an expert has to examine the anomaly, which could be composed of thousands of sensors, and decide whether it represents a failure or not. It is impossible for the operators to deal with hundreds of anomalies a day. The problem is not only overuse of the experts, but due to their inability to address and decipher hundreds of alerts a day, the experts become accustomed to ignoring the alerts.
A mathematical model on top of the Machine Learning based solution
At Precognize the strategy is reversed. The software is designed to utilize the knowledge and experience of the experts in order to build a mathematical model. Precognize’s software uses ML to detect all the anomalies, and then asks the model instead of the experts to identify each and every anomaly. The answers received from the model are identical to the ones we would have gotten from asking the experts. The Precognize machine determines what is a failure and what isn’t, and singles out critical events. We reduce hundreds of alerts a day to three or four a week. Precognize saves precious time for the experts, letting them focus on only a small number of accurate anomalies that are alerts of true failures. The experts’ time is top priority, and thus the system’s implementation is swift: it takes only two weeks for a big plant to have the ML software up and running.
Machine Learning has been a key player in embracing the fourth industrial revolution, and has helped to digitize the industry. Now that the technologies have improved, it is important to seek out the ones that lower the noise the ML generates and focus on accurate and reliable solutions.