Let’s start with an apology. A non-technical explanation of Automated Machine Learning is probably an oxymoron. If you continue reading further, you will come across terms such as “AutoML,” “anomaly detection” and “Artificial Intelligence”. Nevertheless, we will not delve into the specifics of how algorithms work. Instead, we’ll provide a high-level explanation of some of the concepts at the core of Industry 4.0.
Merely a few years ago, the field of Machine Learning was limited to academic research. This has changed significantly for three reasons:
- The cost to store, access and analyze Big Data has fallen significantly, thereby opening the potential for its use.
- Cutting-edge technological innovations including Artificial Intelligence applications can now be accessed by organizations of all sizes due to advances in cloud-computing.
- There is a push on the part of industry to cut costs and improve yield rates by reducing asset downtime.
Ultimately, it’s a combination of economic drivers and technological enablers that is bringing Machine Learning to the factory floor.
The Application of Machine Learning in the Industrial Sector
How is Machine Learning and Artificial Intelligence applied? Industrial plants have hundreds or even thousands of sensors that generate operational data. Traditionally, plants monitor signals such as machine temperature or vibrations in order to track the health of an asset. There are pre-determined manual control thresholds that have been set by engineers and if these controls are breached, then alerts are triggered.
There will always be a role for condition-based monitoring, but when we are dealing with complex and interconnected systems, it is a herculean task to isolate the root cause of a machine breakdown before it occurs.
With Artificial Intelligence, we are looking for abnormal data patterns before the control limits are breached. This is referred to as anomaly detection. Just like an irregular heartbeat or change in white blood cell count is used by medical practitioners to diagnose a patient, algorithms look for unusual behavior (or patterns of unusual behavior) within the data generated by machines’ sensors.
It is not commonly realized that many of the data scientists workstreams such as data cleansing are manually intensive and repetitive tasks. Although performed by skilled data scientists, there is an inherent risk of human error. With Automated Machine Learning, the Machine Learning tasks are performed by an algorithm, thereby increasing both speed and accuracy.
In the past, data scientists selected a particular algorithm to use in a given situation. With Automated Machine Learning solutions, such as SKF Enlight AI, there is an extensive library of algorithms that can be used. Instead of the data scientists making the determination which to use, the system itself selects the optimal algorithm for the data without the need for human input.
It is common for data scientists to select an algorithm and build a machine learning model for a particular problem and to use it indefinitely. However, models have shelf lives and over time they may drift away from the data they are supposed to be modeling and therefore become less useful and should either be recalibrated or be replaced by different model. With AutoML, there is an automated process of validating whether a given machine learning model is the best to use or whether it should be recalibrated or even replaced with a new one.
With advanced algorithms, we are improving the performance of the Machine Learning solution as well the ability to scale solutions broadly.
What does this mean for the industrial plant? A greater portion of industrial assets can be covered without the need to hire data scientists.
A small step for data science, but a big step for the industrial world.