Nonstationarity-Driven Cointegration Analysis for Condition Monitoring and Fault Diagnosis of Industrial Processes
Authors:
- Nasir RazviNRNasir Razvi
- Debela AlemaDebela Alema
- Wieslaw J. StaszewskiWSWieslaw J. Staszewski
- Phong B. DaoPDPhong B. Dao
Preprint
- 19 March 2026
- preprint
- Published by Elsevier in SSRN Electronic Journal
Abstract
Modern industrial processes frequently exhibit strong nonstationary behaviour, where process variables evolve with time-varying statistical properties. Such characteristics pose major challenges for reliable condition monitoring and fault diagnosis, as conventional multivariate statistical process monitoring (MSPM) techniques are primarily designed for stationary data and therefore struggle to capture long-term dynamics in nonstationary systems. To address this limitation, this paper proposes a nonstationarity-driven cointegration framework for industrial process monitoring and fault detection. The novelty of the approach lies in explicitly accounting for variables with different orders of integration. First, the Augmented Dickey–Fuller (ADF) test is used to classify process variables according to their integration order. Johansen’s cointegration method is then applied to extract long-run equilibrium relationships within each group, transforming nonstationary measurements into stationary cointegration residuals. These residuals are subsequently monitored using Mahalanobis Distance–based statistics, while control limits are established from healthy operating data. Faults are detected when monitoring statistics exceed the predefined threshold. The proposed framework is validated through two case studies: a numerical simulation involving multi-order variables with induced faults, and the Tennessee Eastman benchmark industrial process. Results demonstrate that the method effectively captures hidden long-term dependencies and enables reliable fault detection under complex nonstationary operating conditions. Overall, the proposed approach provides a robust and interpretable alternative to conventional MSPM methods, offering improved monitoring performance for modern industrial systems characterized by heterogeneous and evolving process dynamics.Keywords
This publication has 29 references indexed in Scilit:
- Machine learning and cointegration for structural health monitoring of a model under environmental effectsExpert Systems with Applications, 2023
- Open benchmarks for assessment of process monitoring and fault diagnosis techniques: A review and critical analysisComputers & Chemical Engineering, 2022
- A cointegration approach for heteroscedastic data based on a time series decomposition: An application to structural health monitoringMechanical Systems and Signal Processing, 2019
- A regime-switching cointegration approach for removing environmental and operational variations in structural health monitoringMechanical Systems and Signal Processing, 2018
- Condition monitoring and fault detection in wind turbines based on cointegration analysis of SCADA dataRenewable Energy, 2018
- Monitoring Nonstationary Dynamic Systems Using Cointegration and Common-Trends AnalysisIndustrial & Engineering Chemistry Research, 2017
- A cointegration-based monitoring method for rolling bearings working in time-varying operational conditionsMeccanica, 2016
- Overview of PCA-Based Statistical Process-Monitoring Methods for Time-Dependent, High-Dimensional DataJournal of Quality Technology, 2015
- Cointegration Testing Method for Monitoring Nonstationary ProcessesIndustrial & Engineering Chemistry Research, 2009
- Statistical analysis of cointegration vectorsJournal of Economic Dynamics and Control, 1988