Thin Roll Defect Prediction Using Upstream Process Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In the production of continuous thin products wound in rolls, such as paper and nonwovens, defects like holes, chipped edges, and foreign bodies can occur during processing, leading to quality issues and production downtime, as these defects are often not detectable until later stages, causing unnecessary product loss and process halts.
Innovation Solution
A method using machine learning algorithms to predict product defects by analyzing historical process and product parameters from upstream processing steps, allowing for real-time adjustments to prevent defects in subsequent processing stages, implemented in a production plant with a database system and predictive model that correlates detected parameters with potential defects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional defect detection methods are used, then defects are detected after they occur, but this causes product loss and production downtime
Solution Approach 1:
The system performs preliminary defect prediction by analyzing process parameters from upstream production steps before the actual defects manifest in downstream processing. The machine learning model predicts potential defects based on historical data and current process conditions, allowing preventive actions to be taken before defects occur, thus eliminating production downtime and product loss associated with reactive defect detection
2Reliability
If traditional defect detection methods are used, then defects are identified in later stages, but this leads to unnecessary product loss
Solution Approach 1:
The system predicts defects in advance by monitoring process parameters from upstream production steps, enabling quality control actions to be taken before defects actually occur in the final product. This preliminary intervention prevents the need to discard large amounts of finished or semi-finished product, significantly reducing material loss while maintaining high quality standards
3Productivity
If real-time defect prediction is implemented, then production can be optimized and rejects reduced, but this requires complex machine learning models and database systems
Solution Approach 1:
The system uses a universal machine learning platform that can handle multiple defect prediction scenarios across different production lines and product types. The database system serves multiple functions including historical data storage, real-time parameter monitoring, model training, and prediction generation. This multi-functional approach consolidates what could be multiple separate complex systems into a single integrated solution, reducing overall system complexity while maintaining high productivity
Data Source
AI summary
Method for predicting the presence of product defects during an intermediate processing step of a thin product wound in a roll, which provides for—receiving a roll of thin product that has been assigned a unique identification code stored in a database system, this latter containing process and/or product parameters detected in the production steps of said thin product wound in said roll upstream of said intermediate processing step, associated with said unique identification code, —accessing said database system, —entering one or more of the process and/or product parameters associated with the unique identification code of said roll contained in said database system in a predictive model, which uses a correlation, created by means of machine learning logics, from historicized values related to the process and/or product parameters output from said intermediate processing step and historicized values related to process and/or product parameters of the same rolls detected in the production steps of said rolls upstream of said intermediate processing step, in order to predict product parameters output from said intermediate processing step, —comparing said aforesaid product parameters with respective predefined limit values, —generating predictive diagnosis information of thin product defects based on the result of said comparison.