Year 2002
Petri Enden

On-line testing of a process monitoring system and analysis of results

The purpose of process monitoring is to detect abnormal process deviations. When developing faults are detected at an early stage, corrective actions can be performed and loss of quality or production can be avoided. The aim of this master’‘s thesis was to study the theory of process monitoring and to test a process monitoring system developed by the Laboratory of Process Control and Automation.

The literature part focuses on presenting different process monitoring methods. Special attention is given to Kohonen self-organizing maps and their industrial applications. Statistical process control, principal component analysis and partial least squares are also described briefly. Neural network based applications for blast furnaces and automated machinery are presented as industrial-scale examples.

The experimental part focuses on testing results from the Harjavalta copper smelting process. The aim in process monitoring was to detect concentrate aggregations in the concentrate burner and dust aggregations in the waste heat boiler. Process data for the monitoring was collected from the end of June 2001 until the end of November 2001. Also, data was collected from the actions taken in faulty situations to bring the process back to its normal operational state. The latter data was used to mark the faulty situations into the process data.

Different maps with different variables were trained to monitor both phenomena. Principal component analysis was also applied to the process data. Maps with 8 variables were selected for the final testing period. Most of the variables were computational rather than direct process measurements. The monitoring of concentrate aggregations was rather successful, but with dust aggregations, only satisfactory results were obtained.

Due to the fact that both of the phenomena were very hard to monitor, it can be stated that overall good results were obtained. The small number of on-line measurements prevented accurate monitoring of the concentrate burner. In the case of the waste heat boiler, the lack of reliable fault data was the main cause for the results.

This info last modified 27 Apr 2024 - contact webmaster