New information technology and telecommunications, distributed computing in parallel, reduced processing and data storage costs, as well as cheaper devices, are fostering the emergence of a new paradigm for businesses.

The number of companies that can access vast amounts of information for their business is increasing, as well as their ability to monitor their processes in real time and obtain valuable information about their clients, etc. The aim is clear: to obtain a competitive advantage, whether it be to save costs, improve the quality of a process/product, improve customer satisfaction, etcetera.

However, the best technology alone does not guarantee success – it is also crucial to practice some self-criticism and ask ourselves: are we exploiting our data?

A real-life example: industrial process monitoring

For example, during industrial process monitoring, significant investment is made in measuring devices, servers, licences, etc. to achieve an infrastructure that can monitor thousands of process variables with a high sample rate. However, in many cases, companies continue to use traditional statistical controls which are no table to exploit the potential of the data obtained.

Traditionally, statistical process controls have been based on Shewhart, CUSUM and EWMA control charts and approaches to monitor a few key quality and/or process variables. Such control charts are really useful when handling just a few variables, as they can detect abnormal situations and carry out corrective actions once the possible deviation has been detected.

However, what happens to the processes that have mass amounts of stored data as a consequence of the high sample rate and a large number of variables? There is no sense in having hundreds of charts monitoring each variable as the human eye cannot cope with that.

What’s more, by individually monitoring each variable we cannot observe the breaks in the correlation between them, which is usually indicative of a change in the process.

In this context, multivariate process monitoring is an efficient statistical tool for process controls, as it reduces the number of control charts to a few, as well as considering all of the variables. In this regard, it is not only capable of identifying whether an anomaly has occurred, but it can also identify which variable(s) are responsible for it, and this makes it an effective diagnostic tool.

Multivariate analysis also helps to better understand our data, as it allows us to analyse complex processes (many variables and many samples) by showing the relationship between individuals and the relationship between variables through charts.

To conclude, in the current context of cheaper data acquisition systems and information processing systems, it has become popular to obtain vast amounts of information even among small and medium-sized businesses. It is viewed as a strategic issue to review procedures to make the most of the data.

As indicated in [1], [2] and in numerous studies, the effectiveness of multivariant analysis tools is clearly proven. By missing the boat, this would mean a strategic disadvantage that we cannot afford to lose out on.

In future articles, we will talk about this powerful tool being applied to other non-industrial fields.

[1] Nomikos, P. & MacGregor, J. F. Multivariate SPC charts for monitoring batch processes Technometrics, Taylor & Francis, 1995

[2] Ferrer, A. Multivariate statistical process control based on principal component analysis (MSPC-PCA): Some reflections and a case study in an autobody assembly process Quality Engineering, Taylor & Francis, 2007.