Fault Detection and Knowledge Discovery

    Modern chemical plants have distributed control systems (DCS) that handle normal operations and quality control. However, the DCS cannot compensate for all fault events such as fouling or equipment failures. Process monitoring techniques can highlight trends in data and detect faults faster, reducing or even preventing the damage that faults can cause.

     

    Current interests include data visualization, data clustering, and fault detection & diagnosis. We apply data mining and machine learning algorithms to build systems that provide these functions.

     
       
Total Process Monitoring & Supervision

    The advent of faster and more reliable computer systems has changed the manner of monitoring and controlling industrial processes. These advancements resulted in the generation of larger amount of process data, yet the task of interpreting and analyzing this data is intimidating. Operators have neither the time, nor often, the expertise to effective process this information.

     

    The PSE group is investigating an automated support system, which can manage plant data and help in decision-making. The objective is to create the theoretical framework, to develop and implement an advanced Integrated Support System (ISS) for process monitoring, data analysis and interpretation, event detection and diagnosis as well as operations support for chemical and petrochemical industry.

       
Unsupervised Learning
       
Supervised Learning

Adaptive k-Nearest Neighbors (Ak-NN)

A distance-based approach for monitoring processes and detecting faults. The method is designed to adaptively track time-varying process behaviors and extend the continuous normal operation regions.

The Ignition platform was utilized to deploy the Ak-NN approach for monitoring

  • This platform enables creation of unlimited tags and addition of users as per requirement.
  • Applications can be deployed to any industrial or mobile device.
  • Serves as the central hub for all activities on the plant floor, ensuring complete system integration.
       
Deep Learning Image-Based Monitoring

Pyrolysis Reactor Application

On-line Image based sensor for CSD

       
Deep Reinforced Learning (DRL)

For Optimal Control

  • Example: control polymerization reactor

For Process Optimization

  • Example: Optimization of bio-reactor

Basic Control Tuning

  • Autotuning of PID controller
       
Supervised Learning - Online Monitoring Tool (OMT)
Geaux Alumni