Manufacturing industry based on dynamic soft sensors in integrated with feature representation and classification...

Area of Science:

• Automation and Control Engineering
• Artificial Intelligence
• Data Science

Background:

• Soft sensors are crucial for estimating difficult-to-measure industrial process variables.
• Accurate feature representation is key to developing effective soft sensors.
• Deep learning (DL) offers advanced capabilities for complex data structures in soft sensing.

Purpose of the Study:

• To propose a novel technique for dynamic soft sensor-based feature representation and data classification in industrial automation.
• To address common data issues like missing values and hardware failures through pre-processing.
• To enhance the accuracy and efficiency of soft sensors in manufacturing environments.

Main Methods:

• Data pre-processing to handle missing values and identify hardware/communication errors.
• Feature representation using a fuzzy logic-based stacked data-driven auto-encoder (FL_SDDAE).
• Classification of represented features using a least square error backpropagation neural network (LSEBPNN) to minimize mean square error.

Main Results:

• Achieved a 34% reduction in computational time.
• Improved Quality of Service (QoS) by 64%.
• Reduced Root Mean Square Error (RMSE) by 41% and Mean Absolute Error (MAE) by 35%.
• Demonstrated 94% prediction performance and 85% measurement accuracy.

Conclusions:

• The proposed FL_SDDAE and LSEBPNN approach offers a significant advancement in soft sensor technology for industrial automation.
• The technique effectively handles data complexities and improves key performance metrics.
• This method provides a robust solution for enhancing manufacturing process monitoring and control.