Skip to content
Photo: Melina Granberg
View original picture , opens in new tab/window

Big Data becomes difficult to analyse

Published: 19 December 2016

Today for many processes, companies collect a multitude of measurements at high frequencies, which in some cases may not be necessary or relevant yet result in large amounts of data that is challenging to analyse. The problem is ever expanding and the companies need novel methods to achieve better understanding of their complex processes. This is the main focus in a research project at Luleå University of Technology.

Steel, paper pulp, food and pharmaceutical industries are examples of industries where the production process takes place in a continuous flow. Inaccuracies in the manufacturing or production stoppages are therefore costly.

To try to monitor and control their processes, industries make frequent measurements, which in some cases may not be directly necessary or even relevant, resulting in large amounts of data that is challenging to analyse.

– The common perception is that data equal information; hence a lot of data means a lot of information. And if you have too much information about your process, you should understand and know everything about it. This implies better capability to control, improve and optimize it – but despite these large amounts of data, industries still struggle with understanding the inner workings of their processes, says Murat Kulahci, Professor of Quality Engineering at Luleå University of Technology who also works at the Technical University of Denmark.

Through observational data and/or controlled experiments, Murat Kulahci tries to get an overview of the data generated, evaluate if it carries enough information and find ways to improve the process from which it was collected. The goal is to make sense of the available data and if needed get more relevant data.

– We can measure the temperature of this room at 20 different locations at the ceiling every few seconds and easily collect thousands of temperature measurements in a short span of time because it now is easier and cheaper to do so ever then before. But how fast can the temperature of a room change or how different two measurements can be if they are obtained in very close proximity?  In that sense, it may make more sense to combine these measurements into a single value to determine the “temperature” of the room. This will reduce the dimensionality of the problem quite a bit by taking advantage of the correlation among the measurements.

– There are major problems with how Big Data are handled in the process industry and the problems escalate. Therefore this research is extremely relevant. By being able to understand the current state of the process through empirical evidence, we can improve and optimize it. This in turn will result in less waste, increased productivity and quality. But everything starts with a better understanding of the process through available data, says Murat Kulahci.

The research is carried out together with Professor Bjarne Bergquist, Senior Lecturer Erik Vanhatalo and PhD student Francesca Capaci from Quality Engineering at Luleå University of Technology.

In media