Machine Learning Approaches for Phase Identification Using Process Variables in Batch Processes
Research output: Contribution to journal › Review article › Contributed › peer-review
Contributors
Abstract
Specialty and fine chemicals are often manufactured in multipurpose batch production plants. Compared to continuous production, these plants offer increased flexibility at the cost of operational complexity. A recipe defines the sequence and process parameters of different batch phases that are needed to transform raw materials into the desired product. In some plants detailed information about the executed recipe is not always captured by data acquisition systems. Knowledge of these phases is essential for optimizing quality and throughput. State-of-the-art data-driven machine learning techniques can recognize recurrent patterns in noisy time series data, enabling automatic labeling of batch phases based on widely available sensor data. In this review paper, we provide an overview of several machine learning approaches that can be used in an industrial setting.
Details
Original language | English |
---|---|
Pages (from-to) | 989-1002 |
Number of pages | 14 |
Journal | Chemie-Ingenieur-Technik |
Volume | 95 |
Issue number | 7 |
Publication status | Published - 26 Apr 2023 |
Peer-reviewed | Yes |
External IDs
Mendeley | 25ad0e77-0a2f-3d94-b115-527603d11886 |
---|---|
WOS | 000976059700001 |
ORCID | /0000-0001-5165-4459/work/142248277 |
Keywords
Research priority areas of TU Dresden
ASJC Scopus subject areas
Keywords
- Active learning, Batch process, Data analytics, Phase identification, Time series