Distillation Column Clogging Prediction with Synchronized ML Signals

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for predicting blockages in distillation columns of refineries are inefficient, resulting in high false positive and false negative predictions, leading to costly production slowdowns and unscheduled downtime due to their reliance on single predictors and inadequate data synchronization.

Innovation Solution

A multi-criteria prediction method using automatic learning that synchronizes real-time data to construct a learning model, incorporating various performance indicators and applying time offsets to improve prediction accuracy, reducing errors by 10-50% and enabling early detection of potential clogging.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional single predictor methods are used, then the prediction system is simple, but the prediction accuracy is low with high false positive and false negative rates

Engineering Contradiction:
Improveprediction accuracyVSAvoidprediction system complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The prediction system is segmented into multiple independent predictors (first predictor based on theoretical equations, second predictor based on temperature and differential pressure analysis) that operate separately and whose results are combined. This segmentation allows each predictor to focus on specific aspects of clogging detection, improving overall accuracy while maintaining manageable complexity through modular design.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Multiple prediction results from different predictors are merged into a unified prediction system. The combination of theoretical equation-based prediction, temperature-based prediction, and differential pressure-based prediction creates a more robust and accurate prediction model that reduces false positives and false negatives compared to single predictor approaches.

Inventive Principle:
Principle #5Merging (Combining)

2Reliability

If data from multiple sources is collected without synchronization, then more data is available for analysis, but the prediction reliability decreases due to temporal misalignment

Engineering Contradiction:
Improveprediction reliabilityVSAvoiddata processing complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

Data synchronization is performed as a preliminary action before prediction analysis. By aligning temporal references of data from multiple sources (process data, temperature data, differential pressure data) before combining them, the system ensures that all inputs correspond to the same operational moment, thereby improving prediction reliability without requiring complex real-time coordination during the prediction phase.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4121830B1Method for predicting clogging of distillation column(s) in a refinery, computer program and associated prediction system
Publication Date: 2024.08.28 TOTALENERGIES ONETECH
  • EP4121830B1 patent drawingFigure 1
  • EP4121830B1 patent drawingFigure 2

AI summary

The invention relates to a method (38) for predicting clogging of a distillation column using machine learning, comprising: - a phase of constructing and training (40) a machine learning module obtained from previously collected data, and a set of sensors, - an operational phase (42) of predicting clogging comprising: - collecting (56) a current data stream until a buffer is filled, - preprocessing (64) data from the data buffer through predetermined purging and classification, - synchronising (66) data in the current set of purged and classified data, - determining (70) a current variable value that is representative of at least a current performance of the at least one distillation column, - forming (72) a current set of transformed data by calculating predetermined derivatives, - predicting (74) the current state of the distillation column by applying the learning module to the current set of transformed data.