Making sense of big data with graph technology, machine learning

Feature articles | August 16, 2016

By Rich Pell

Sensing / Conditioning MPUs/MCUs Data Acquisition Software & Embedded tools

Modern manufacturing equipment has been advancing rapidly; plants are filled with sensors to monitor equipment performance. The number of sensors that allow devices to connect to the internet is growing and so too is the volume and complexity of data available to plant managers. The collection, storage and analysis of this data is vital in unlocking the benefits big data can provide.

Graph databases
Traditionally, data has been stored in table-structured relational databases, but development in this field has led to the introduction of the next generation of relational databases, graph databases, a type of NoSQL database. In a graph database, information is stored and represented with nodes, edges and properties. Nodes represent individual entities, edges are lines that connect nodes to each other and properties represent information relevant to the nodes. Unlike relational databases, which form a square structure, graph databases are much more flexible.

Graph databases can be used to quickly access information and identify trends in large data sets, such as supply chain patterns, logistics and new business leads. The system is naturally adaptive, allowing new nodes to be easily added. The analysis can be done in real time to address problems in manufacturing.

Machine learning
Machine learning is a concept that has been around for many decades. In machine learning the computer doesn’t rely on rule-based programming, rather the algorithms can adapt and learn from the data. This means that manufacturers using this software don’t need to rely on the time and expense of dedicated data analysts to find patterns and make predictions. Companies like Amazon have also used cloud based machine learning to make warehouse logistics more efficient by being able to quickly and seamlessly adapt to changes in inventory demand at peak times and during seasonal highs and lows.

Machine learning can incorporate hundreds of causes, effects and non-linear responses. This model can adapt itself over time to continually improve the quality of predictions. Machine learning can be combined with graph databases to gain valuable insights into processes. Whether it’s condition monitoring or predictive maintenance of a process plant, demand forecasting in automotive manufacturing or digital twinning — a type of virtualisation — machine learning facilitates better decision making in an increasingly complex business environment.

Machine learning is commonly used for predictive analytics, which can give insight not only into customer intentions, but also into the state of machines on the factory floor. Information analysed from the sensors can relay any potential suboptimal performance that may lead to unplanned down time if left unaddressed. This leaves plant managers time to order replacement parts, such as an obsolete or refurbished part from EU Automation, or perform other necessary maintenance to prevent system failure.

Machine learning is particularly useful in largely automated systems, where equipment is required to make its own decisions. The continuous learning process makes data more reliable, analysis techniques more repeatable and ultimately improves the human input into any system. Aside from predictive analytics it can also be applied to optical part sorting, failure detection, analysis and product testing.

Although machine learning and graph technology both offer a powerful way of analysing the ever increasing volume of data available to us, much of the technological potential is yet to be realised. To gain the most valuable insights, it’s important that business leaders embrace a thoroughly modern form of analysis. In doing so, it may come as less of a surprise that competitive advantage is less than a mere six degrees of separation away.

About the author:
Jonathan Wilkins is marketing director at EU Automation – www.euautomation.com