Variational Progressive-Transfer Network for Soft Sensing of Multirate Industrial Processes

Area of Science:

• Industrial process monitoring
• Machine learning applications
• Data science

Background:

• Soft sensors are crucial for predicting industrial process variables.
• Existing methods often struggle with multirate data, where variables are sampled at different frequencies.
• This limitation hinders accurate prediction in real-world industrial settings.

Purpose of the Study:

• To develop a novel deep-learning approach for soft sensor development in industrial processes with multirate data.
• To propose a variational progressive-transfer network (VPTN) that effectively handles data sampled at different rates.
• To enhance the performance of soft sensors by leveraging knowledge transfer across data streams.

Main Methods:

• A variational progressive-transfer network (VPTN) is proposed, separating multirate data into uniformly sampled chunks.
• A variational multichunk data modeling framework unifies the modeling of these chunks using deep variational structures.
• A progressive transfer learning strategy transfers model parameters from faster to slower sampled data chunks sequentially.

Main Results:

• The VPTN method demonstrated effective modeling of multirate industrial process data.
• Knowledge transfer between data chunks significantly improved the performance of the terminal soft sensor model.
• Validation on debutanizer column and coal mill datasets confirmed the method's efficacy.

Conclusions:

• The proposed VPTN method offers a robust solution for soft sensor development in industrial multirate processes.
• Progressive transfer learning is a viable strategy for enhancing soft sensor accuracy with varying data sampling rates.
• The approach provides a valuable tool for improving industrial process monitoring and control.