Author: Liam Critchley
The demands of the process industry are always increasing, with production efficiency, product quality, safety levels and environmental protection being the main areas of interest. The main way to achieve these higher efficiencies is through introducing more complex automation control systems which can analyse and measure a greater number of parameters.
Process control methods are largely dependent on the obtained data. This data can come from various points within a system using various types of sensors, and this data can then be used to develop advanced measuring systems that help to prevent plant failure, sensor failure, communication system failure and identify anomalous disturbances within a process.
What is Multivariate Process Monitoring?
Many companies within industry still rely on traditional statistical process control (SPC) methods, which only give statistical information for a single variable, such as the median or standard deviation. Traditional univariable methods often miss many of the underlying patterns with the process data and this is where multivariate statistical process control (MSPC) methods come in.
MSPC is an advanced statistical approach that identifies all the desired variables in a process and can spot specific underlying patterns within the data. In the most basic sense, MSPC methods are a series of algorithms that can extract information from multivariable data sets. An important factor of an MSPC method is that it can handle highly correlated, highly dimensional and noisy data.
All MSPC methods often work in a similar manner. The first step builds a data-based model using the data obtained when these processes work normally. These methods then give a projection about the new data based on the history of the data within the system so that it can judge whether any new data points are abnormal against the normal process data. If an abnormal point is found, then the MSPC method will identify any potential variables that could have caused this anomaly and ultimately determine the root cause of the abnormal point.
The most important aspect of multivariate process control methods is the ability to show abnormal process behaviour taking into account the relationships between different sets of variables. Overall, using these methods can give the user a much greater understanding and control over the different processes found in a manufacturing or process plant environment.
Advantages of MSPC over SPC
MSPC methods offer quite a few advantages over traditional SPC methods. Firstly, MSPC methods use fewer control charts so there is a smaller margin for error. MSPC simplifies the job of process operators by showing, in a single chart, information of all the relevant process variables, which means that a control chart is not needed for every single variable, unlike in traditional SPC methods.
The second advantage of MSPC methods is that you can see the full picture of the system, including any underlying, or otherwise hidden, patterns within the data. MSPC methods identify which variables interact with each other and enables any defects within the system to be identified. This process also shows which defects are related to each other, and which are independent of any other defects.
The third main advantage is the method enables process improvements to be used to deeper understand the system’s process behaviour. Normally, many strategies, often through trial and error, are used to improve process behaviour, but MSPC methods can model new strategies and predict their outcomes before they’ve been tested. This results in more efficient ways to implement new strategies, and ultimately reduces the time spent on new process strategies.
Benefits to Industry
The improved diagnostics, higher quality analyses, greater understanding of process behaviour, increased knowledge transfer processes and reduction in systematic and plant failures brings many benefits to industry, be it for a manufacturing site or a process plant. Some of the main benefits offered by MSPC methods enable these industries to reduce process costs, optimize processes, reduce the time it takes for a product to get to market, improve their product quality, reduce the usual problems that occur during scale-up processes and increase the overall efficiency of the equipment used within the plant. These benefits apply to many types of process and manufacturing industry, from chemicals, to plastics, glass manufacturing, mining, oil, electronics, food and drink, pharmaceuticals, and beyond.
Sources:
1) Slišković D., et al, Multivariate Statistical Process Monitoring, Tehni ki vjesnik, 19, (2012), 33-41.
Iowa State University: https://www.iastate.edu/
2) CAMO: http://www.camo.com/applications/process-control.html
3) Alberto Ferrer (2013) Latent Structures-Based Multivariate Statistical Process Control: A Paradigm Shift, Quality Engineering, 26:1, 72-91, DOI: 10.1080/08982112.2013.846093