(107b) A Comprehensive Web-Based Platform for Advanced Analysis of Oscillations and Root Causes in Control Systems

In chemical process plants, maintaining stability in multivariate control networks is vital for operational efficiency and minimizing faults. External factors such as stiction in control valves, and hard-tuned parameters, can degrade performance, leading to problems like oscillations, increased energy consumption, and lower product quality. In large-scale plants with numerous interconnected control loops, these disturbances often propagate from one control loop to another[1, 2], making manual inspection challenging and resulting in undetected oscillations[3] and missed optimization opportunities. Traditional diagnostic approaches, which depend on plant-specific information such as topology or causal maps, face limitations due to the unavailability of data and high computational costs. Although frequency-domain techniques, such as spectral envelopes[4], can help isolate oscillatory faults but are not easily adaptable to other fault types. Fault detection using rule-based methods, which draw decisions based on parameters like IAE, decay ratios, or regularity, often face difficulties in accurately identifying faults with noisy industrial data that contains varying frequencies and amplitudes, leading to challenges in accurate fault detection. This leads to the requirement for an automated, data-driven diagnostic tool that can accurately detect oscillations and identify root causes, improving operational efficiency and maintaining system stability.

This article introduces a web-based platform for the diagnosis of faults in process industries encompassing root cause detection, oscillation detection, and oscillation characterization. The platform features an intuitive, user-friendly interface, built using Flask, a lightweight Python framework that allows users to upload process system data in Excel format (.xlsx) for analysis. The platform greets users with a homepage having multiple faults diagnostic options which leads users to an upload page for selected analysis. The web-based platform instantly processes uploaded data, classifies oscillations, estimates period and amplitude, and identifies root causes of faults in the uploaded dataset. Once the uploaded data is processed, the platform directs the user to a results page and displays the selected analysis results with an option to download the generated results in portable document format(PDF). Flask sessions are used to store intermediate results which let the user move between pages without losing progress, enhancing the user experience.

The root cause analysis (RCA) method in the platform uses a τ-metric based on cross-correlation with weighted lags. The RCA tool applies to both oscillatory and non-oscillatory faults, evaluates relationships between variables, and identifies the top 3 potential root causes of anomalies in the uploaded data. The RCA method implemented in the platform was tested on synthetic data for various interconnected control systems and gives an accuracy of 92.54% for non-oscillatory faults and 86.88% for oscillatory faults. The method has also been validated using synthetic industrial case studies, showcasing its effectiveness in practical applications.

For oscillation characterization, the tool utilizes a neural network trained on a large set of synthetic data to classify signals as non-oscillatory, regular oscillatory, or irregular oscillatory. It employs Fast Fourier Transform (FFT) and FFT of the autocorrelation function (ACF) for feature extraction. Prominent peaks based on a specified threshold for a number of periods in oscillations are obtained from frequency-domain analysis. Frequency domain features based on these prominent peaks reduce the input features by 80% for the neural network while maintaining high detection accuracy, decreasing computational effort. For the synthetic data, the tool achieves a classification accuracy of 96% and a recall rate of 0.95 in detecting oscillatory behavior. The oscillation time period evaluated based on frequency and amplitude derived from the FFT of the ACF gives an accuracy of 90% for both regular and irregular oscillations. Furthermore, an amplitude estimation method enhances the overall reliability of the approach.

The platform has an all-in-one analysis module where the results for oscillation characterization and root cause detection are compiled together, accelerating fault analysis, reducing dependency on manual troubleshooting, and helping operators make quick decisions. Users can either use the all-in-one analysis or independently run the modules for oscillation characterization and root cause analysis, as needed. The planned future updates aim to include a stiction detection module to further enhance the diagnostic capabilities of the platform. Developed to scale for larger interconnected control systems, the tool is planned to offer cloud-based access, offering users greater flexibility and availability. With continuous updates and integration with additional industrial systems and platforms, the tool’s scope and utility will expand, delivering a more seamless and efficient diagnostic experience for users.

References

1. Venkatasubramanian, V., et al., A review of process fault detection and diagnosis: Part I: Quantitative model-based methods. Computers & chemical engineering, 2003. 27(3): p. 293-311.
2. Thornhill, N.F., J.W. Cox, and M.A. Paulonis, Diagnosis of plant-wide oscillation through data-driven analysis and process understanding. Control Engineering Practice, 2003. 11(12): p. 1481-1490.
3. Choudhury, M., Automatic detection and estimation of amplitudes and frequencies of multiple oscillations in process data. ADCONIP 2014, 2014: p. 514-519.
4. Jiang, H., M.S. Choudhury, and S.L. Shah, Detection and diagnosis of plant-wide oscillations from industrial data using the spectral envelope method. Journal of Process Control, 2007. 17(2): p. 143-155.