CeleraOne’s growing team of mathematicians and software engineers has more than ten years of industry experience in Machine Learning. Two years ago, CeleraOne started its expansion into the steel industry and acquired deep technical knowledge of steel manufacturing processes - prerequisite for any successful application of Machine Learning to cost optimization in the steel manufacturing sector.
In ambitious projects with steel manufacturers, CeleraOne applies Deep Learning. In contrast to many analysis firms, CeleraOne has profound knowledge of this central technology. Through a complete transfer of knowhow to its customers CeleraOne provides a lasting competitive advantage.
Applications of substantial impact
Cross-plant integration of process, surface, and machine data into advanced Machine Learning models
Separation of good / bad products
Automated identification of complex root causes of quality deviations using Machine Learning models
Cross-plant integration of process, surface, and machine data into advanced Machine Learning model
Data of current production is processed in Real-time
Before start of next process step: predicion of quality deviations in all following steps
Deep Learning (high-performance models) of consumer sector is being adapted to surface images
Strong reduction of training effort, only small adaptive part needs plant-specific optimization
Live implementation in automated surface inspection systems
Automated identification of defect root causes
Show Case: Root Cause Analysis, steelmaking
Result: Determination of production paths, equipment configuration, process parameters leading to defects
Identifying the exact root cause of non-trivial quality deviations may take months or years. A multitude of production routes and processing steps need to be checked. Thousands of time series and surface inspection data need to be correlated.
CeleraOne takes advantage of the advanced automation and extensive process monitoring steel manufacturers have implemented during the last decades. Relevant surface inspection data and process signals will be cleaned, preprocessed, and merged into a common machine learning model for Root Cause Analysis. CeleraOne delivers a detailed identification of the root cause of quality deviations, even for position-related problems.
Real-time prediction of downstream quality deviations
Show Case: Prediction of optimal rolling parameters, cold rolling
Prior to cold rolling of each coil, the Machine Learning model predicts which rolling parameters have to be configured for reaching the desired properties of the finished product.
Result: optimal operation leads to shorter processing times, reduced costs, extended product range.
Quality parameters are often measured at the end of the production chain. Root causes of deviations may be located far upstream. Standard processing of such defective input material must be avoided, because its results are downgraded output or even scrap.
Based on live production data, CeleraOne's predictive analytics models forecast quality deviations before downstream processing starts. This enables early repurposing of the material and dynamic reconfiguration of the downstream process to counteract varying material properties inherited from upstream suppliers. Predictive Analytics can be applied at each point of the production chain.
Higher accuracy, less maintenance
Automated surface inspection systems (ASIS) need permanent re-trainings on new, manually labeled data. In practice, defect classification accuracy varies.
CeleraOne adds modern Deep Learning technology to existing surface inspection systems. This reduces the required re-training effort significantly. At the same time, classification accuracy is improved.