ORMAE

    Contact Us
    Contact Us
    Menu

    Failure Prediction of Petrochemical Equipment

     

    Machine Learning-driven AI Advisory Decision Support System to reduce failures and avoid unforeseen shutdowns

    About the customer

    A petrochemical giant with operations in several countries. Given the heavy-duty nature of the business and infrastructure reliability required, the company is highly dependent on the latest IT tools for consistent production and output.

    Business Problem

    The petrochemical giant was facing frequent drops of productivity due to a variety of reasons. Given the scale of operations and unpredictability of the business, the management wishes to have a accurate view into the future. The company approached ORMAE to:

    • Build an AI Advisory Decision Support System using Machine Learning models that predict failures.
    • Generate data to estimate failure probability and prepare for setbacks.

    Our Approach

    The solution that ORMAE envisaged for the client had to factor in faced in course of developing the solution pertained to scalability and data quality. Other aspects of the solution included:

    • Anomaly detection model to detect any variance from the trained normal ranges based on the tuning parameters.
    • Predictive model that predicts failures based on historical events that the equipment has gone through.
    • Exploratory Data Analysis is conducted to identify pre-failure patterns and the possible reasons for the failure.
    • Grouping together of equipment through clustering techniques and building of models upon these clusters.
    • Statistical models like Moving Average, FFT analysis, and others were built to detect anomalous behavior of vibration signals.
    • ML models like Gaussian Mixture Model, Residual analysis on regressor, Auto Encoder models, and others were developed and tested against known plant failures.

    The Result

    The resulting output exactly matched the expectations set by the client to ORMAE.

    • The cluster specific models were ready to be developed, validated and deployed in the client environment.
    • Automated training pipeline that re-trained the models after specific intervals. So in case of continuous high losses in prediction, the model can retrain itself.
    • A testing pipeline is deployed in such a way that proper alerts are sent prior to failures (predicted). This gives ample time for engineers to undertake needful corrective measures and avoid plant shutdowns.
    Get in Touch

    Have a project in mind?

    Looking for collaboration?
    Send an email to bdm@ormae.com
    for availability and enquires.

    Get in Touch

    Have a project in mind?

    Looking for collaboration?
    Send an email to bd@ormae.com
    for availability and enquires.

    ×