A fault isolation strategy for industrial processes using outlier-degree-based variable contributions.

Area of Science:

• Industrial Process Monitoring
• Fault Diagnosis
• Data Analysis

Background:

• Analyzing industrial process faults is challenging due to large data volumes and the smearing effect in traditional methods.
• Existing contribution analysis methods suffer from interdependencies between variable isolation indices, reducing accuracy.

Purpose of the Study:

• To propose a new fault isolation method that enhances accuracy by addressing the limitations of traditional approaches.
• To effectively isolate true fault variables from extensive process data.

Main Methods:

• A novel fault isolation approach is developed, integrating the local outlier factor and an improved k-nearest neighbor rule.
• Nearest neighbors are identified along specific variable directions to calculate outlier-degree values, serving as variable contributions.
• An isolation threshold is established by selecting the maximum contribution across all samples.

Main Results:

• The proposed method demonstrates improved fault isolation accuracy in numerical and Tennessee Eastman process case studies.
• Effectiveness in identifying dominant fault-causing variables in real-time industrial monitoring was evaluated.

Conclusions:

• The developed local outlier factor and k-nearest neighbor-based method effectively overcomes the smearing effect in fault isolation.
• This approach offers a more accurate and reliable solution for identifying fault variables in industrial process monitoring.