Preventing Unplanned Shutdowns

Using Multivariate Data Analysis

By |2017-02-16T23:22:32-05:00February 16, 2017||

preventing unplanned shutdownsUnplanned shutdowns can have a significant impact on your business. Almost every plant loses at least 5% of its productive capacity from downtime, and many lose up to 20%1. Of the manufacturers that estimate their downtime, they usually underestimate the total downtime cost (TDC) by 200 – 300%1.

While unplanned shutdowns are often associated with unexpected equipment failure, there are also many outages that result from poor process operation. Analyzing the process data for fault signatures can identify root causes and prevent future downtime. ProSensus has successfully used multivariate analysis on historical data to diagnose:

  1. Reactor fouling in speciality chemical plants – Identified process conditions that lead to accelerated fouling.
  2. Equipment damage in steel manufacturing – Unstable operation in steelmaking processes can lead to significant damage (breakouts, broken cooling equipment, etc.). ProSensus was able to identify process signatures with a high probability of resulting in expensive downtime.
  3. Unacceptable raw material variations in pharmaceutical plants – Using historical batch data, ProSensus was able to identify variations in raw material data resulting in downtime in a pharmaceutical plant. Specifications were tightened to avoid this cause of downtime in the future.

Identifying Normal vs. Poor Operating Conditions

In most cases, a model is built on data collected from normal, good operation. The model is then applied to the poor operation data to identify the combination of measurements contributing to the shift in operation. For example in a low density polyethylene process a multivariate model is built using process variables to predict product quality. The last four observations (time-points) in the score plot below show that the process is beginning to drift. This could also be observed in the hotelling T2 plot.

Score plot preventing unplanned shutdowns

Ht preventing unplanned shutdowns

Using a contribution plot we are able to compare the operating conditions of the last four observations (where the drift occurred) to the average operating conditions. Based on this contribution plot we can see that the process variable Z2 is much higher, and Tmax2 is much lower compared to the average. This methodology provides important insights to diagnose an unplanned shutdown to prevent a similar one in the future, or to correct a process in real-time so it doesn’t continue to drift.

contributions plot unplannedshutdowns

To learn more about how to use multivariate data analysis to diagnose and prevent unplanned shutdowns contact us.


  1. LinkedIn. (2014, December 3). Retrieved from LinkedIn:

About the Author: Marlene Cardin

Marlene Cardin
Marlene Cardin, MASc
Director of Projects
Since 2007, Marlene has helped many Fortune 500 companies in the pharmaceutical, food & beverage, speciality chemical, and oil & gas industries get actionable insights from their big data. In addition to working closely with our consulting clients, Marlene oversees the growth of our ProVision division which offers fully turnkey machine vision systems for real-time quality monitoring & control. Marlene has a Bachelor’s degree in Chemical Engineering and a Master’s Degree in Applied Science from McMaster University (Hamilton, Ontario). Marlene completed her Master’s degree under the supervision of ProSensus founder, Dr. John F. MacGregor.