Monitoring of complex nonstationary industrial processes

Rights statement
Awarding institution
  • University of Strathclyde
Date of award
  • 2020
Thesis identifier
  • T15555
Person Identifier (Local)
  • 201657429
Qualification Level
Qualification Name
Department, School or Faculty
  • Monitoring of complex manufacturing processes using multivariate statistical process control (MSPC) is becoming more important. However, classical MSPC is restricted to stationary data while most industrial processes are nonstationary. One way of addressing nonstationary data is to calculate the difference between consecutive time series data samples. However, this can cause loss of dynamic information, resulting in inadequate process monitoring or a reduced fault detection capability. Cointegration analysis has recently been adopted for process monitoring of nonstationary processes. However, the first applications considered only nonstationary variables whereas complex industrial processes contain both stationary and nonstationary variables. Furthermore, there is inefficiency in the modelling when dealing with higher level nonstationary time series. This particular issue can be solved by using common-trend residuals-based monitoring. However, the use of different models requires a number of control charts to be monitored by a data analyst. To solve these issues, a multi-level multi-factor model is proposed for the monitoring of complex continuous and batch industrial processes. The method uses a combination of principal component analysis (PCA), and cointegration and common-trend models at the 1st level, and then a PCA model at the 2nd level to monitor the combined stationary outputs from the 1st level. The method is tested with ramp and step type fault functions on continuous and batch process simulations, and compared with conventional PCA and cointegration based approaches. The findings show that the multi-level multi-factor model can provide better fault detection rates compared to conventional PCA and cointegration based approaches. In addition, a parameter tuning scheme based on the big-bang big-crunch global optimisation algorithm is used to select the optimum parameters for the multilevel multi-factor model when applied to continuous and batch processes. This not only improves the model’s performance but also assists with its practical application in an industrial environment.
Advisor / supervisor
  • Nordon, Alison.
Resource Type
  • This thesis was previously held under moratorium from 4th August 2020 to 4th August 2022.
Date Created
  • 2020
Former identifier
  • 9912916493502996
Embargo Note
  • This thesis is restricted to Strathclyde users only until 4th August 2025.