Data is fast becoming one of the most important assets for companies throughout industry. In particular, it holds the key for developing a new generation of machine learning technologies that will have the potential to deliver even higher levels of production and efficiency, while empowering employees to make faster, better informed decisions.
Machine learning is not a new concept – the name was coined by artificial intelligence pioneer Arthur Samuel six decades ago when he was working for computer giant IBM. Its practical applications, however, are far more recent, ranging from self-driving cars to speech recognition software and web search technology.
The machine learning technique, which uses statistical methods to give computer systems the ability to learn without being explicitly programmed, is currently being developed by SKF to improve the maintenance and condition monitoring of rotating equipment.
Success, however, depends on being able to gather, categorize and intelligently interpret large volumes of data. In the same way that it is difficult for humans to acquire new skills without information or examples to follow, so machines are incapable of learning without access to data; machine learning algorithms essentially learn models of behavior from the data sets that are presented to them.
One of the biggest challenges that SKF is addressing is the need to find methods of organizing and defining data sets, to enable maximum value to be extracted in each application.
The company has initiated discussions among its engineers about which labels should be placed on the data being collected, how the data is stored, how it is combined, and how to define the correct taxonomy (or classification system) to be able to compare like-for-like data across different applications, departments and countries.
Although the company has collected data from many millions of bearings, this has to be organized in common formats to facilitate the subsequent development of accurate and reproducible methods of machine learning. Data quality is equally important if this process is to be successful. In many instances, although only a small proportion of the data collected will be required for effective machine learning, there is considerable difficulty in identifying and separating the data that is of real value for each area of application. Similarly, data is often held in silos, sometimes in different formats, making it difficult to interrogate holistically.
This work on data classification is being carried out simultaneously with a program to develop the most effective methodology for interpreting and analyzing data within machine learning systems.
The approach that SKF has developed as part of its program to improve the performance of rotating equipment begins with anomaly detection – using data to establish that a problem exists with a component or machine system. The next stage is auto diagnostics, where data analysis allows the exact nature of the problem to be defined. This is followed by prognostics, with data being interrogated intelligently to determine the appropriate remedial action – and the timing – that will be required.
Although the fundamental step by step approach is similar to that used for existing predictive maintenance techniques, the new approach being developed by SKF adapts this methodology for use in machine learning solutions.
Anomaly detection, auto diagnostics and prognostics will in the future all therefore be carried out automatically. The precision, with which decisions are taken, along with the machine adjustments required to prolong operating life without affecting efficiency, will be constantly improved as machine learning systems gather increasing volumes of data; this will in time include analyses of the most recent adjustments, which will be fed back into the learning cycle.
The drive to reach new levels of rotating equipment performance through advanced analytics is in turn leading to the development of new bearing and sensor technologies, and software tools. For example, SKF is developing a powerful software platform to underpin system integration, data gathering and intelligent data analyses. Benefits for SKF customers will include more accurate and faster data analyses, with improved machine performance and efficiency, all with less human intervention.
It is worth noting that although these processes will reduce, or potentially eliminate, the need for human intervention in many routine processes, they are unlikely to lead to major redundancies. Instead, they will offer engineers and factory team’s greater insight about the way in which machine systems are running, leading to better and more informed decision making.
Advanced data analytics and machine learning, and the associated development of Industry 4.0 technologies, are already creating new opportunities for industrial companies to improve productivity and efficiency. In particular, they allow them to exploit the potential from existing systems and to increase capacity without significant capital investment. They also allow suppliers such as SKF to offer new services and methods of partnering with customers; for example, offering methods of charging based on machine availability, uptime or output.
This trend is set to continue in the future as companies build, develop and understand their data sets and analytical tools. In turn, this will open up new methods of manufacturing or systems operation, with as yet undreamed of opportunities for change, growth and profitability.