Plant Asset Failure Models With Automated Feature Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current failure detection and prediction methods in industrial processes face challenges due to the overwhelming amount of real-time and non-real-time data, complexity in selecting relevant process variables, presence of invalid data, and the need for domain-specific feature engineering, which hinders the development of effective machine learning models for predicting equipment and process failures.
Innovation Solution
An automated approach that includes data cleansing, feature engineering, and optimal input selection to generate a reduced and enriched dataset for building scalable failure models, using techniques like cross-correlation analysis and multivariate statistical models to identify key inputs and reduce dimensionality, thereby improving the predictability of process failures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all process variable tags are used as model input candidates, then the model may capture all potential failure indicators, but the model building and training becomes extremely time-consuming and challenging
Solution Approach 1:
The patent extracts and removes irrelevant or redundant process variable tags from the complete set of available variables. This is achieved through automated feature selection techniques that identify and eliminate variables that do not contribute significantly to failure prediction, thereby reducing the input space while retaining the most informative variables for model training.
Solution Approach 2:
The patent segments the large set of process variable tags into meaningful groups or categories based on their relationship to potential failure modes. This segmentation allows the model to focus on specific subsets of variables relevant to different failure types, reducing the overall complexity and training time while maintaining detection accuracy.
2Reliability
If user experience is relied upon for model development, then domain-specific insights can be incorporated, but the process becomes challenging for non-expert users and requires significant manual effort
Solution Approach 1:
The patent implements self-service capabilities where the system automatically performs data preprocessing, feature selection, and model training without requiring extensive user intervention. The automated system guides non-expert users through the model development process by making intelligent decisions about variable selection and parameter tuning based on the data characteristics and failure patterns.
Solution Approach 2:
The patent creates a universal model development framework that can handle different types of process data and failure modes through a single automated system. This multi-functional approach allows the same platform to serve both expert users who need advanced customization and non-expert users who require simplified workflows, making the technology accessible to a broader range of users.
Data Source
AI summary
A system that provides an improved approach for detecting and predicting failures in a plant or equipment process. The approach may facilitate failure-model building and deployment from historical plant data of a formidable number of measurements. The system implements methods that generate a dataset containing recorded measurements for variables of the process. The methods reduce the dataset by cleansing bad quality data segments and measurements for uninformative process variables from the dataset. The methods then enrich the dataset by applying nonlinear transforms, engineering calculations and statistical measurements. The methods identify highly correlated input by performing a cross-correlation analysis on the cleansed and enriched dataset, and reduce the dataset by removing less-contributing input using a two-step feature selection procedure. The methods use the reduced dataset to build and train a failure model, which is deployed online to detect and predict failures in real-time plant operations.