An embedded KPI-based advisory framework for monitoring and diagnosis of soft sensor degradation
Modern industrial processes rely on soft sensors to estimate process variables that are hard to measure due to expensive and sensitive hardware sensors. The performance of soft sensors deteriorates due to changing operating conditions, sensor drift, or unexpected process disturbances. For their simp...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-09-01
|
| Series: | Results in Engineering |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S259012302502585X |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Modern industrial processes rely on soft sensors to estimate process variables that are hard to measure due to expensive and sensitive hardware sensors. The performance of soft sensors deteriorates due to changing operating conditions, sensor drift, or unexpected process disturbances. For their simplicity and interpretability, Partial Least Squares (PLS)-based soft sensors are often used in industries, but most research focuses on maintenance strategies without clear guidance on how to respond to faults. This gap often misclassifies faults, requiring model retraining and wasting resources. The Advisory Soft Sensor Monitoring Index (ASSMI), a systematic, multi-KPI-driven framework, detects, classifies, and recommends fault-specific corrective actions. A two-layered diagnostic system using embedded KPIs over segmented windows supports this framework. Four embedded KPIs (Hotelling's T², Squared Prediction Error (SPE), R² recovery, and detection ratio) evaluate faults as short or prolonged process faults. A second layer examines the consistency of T² or SPE across windows to determine if the fault is process-related or model-related. A pilot-scale distillation column Aspen Dynamic Simulation case study with three fault scenarios validates this tool. The short-term faults had low detection rates (T²: 7.07 %, SPE: 8.07 %) and quick R² recovery, indicating self-correction and no model update needed. For prolonged process faults, higher detection rates (T²: 18.73 %, SPE: 19.10 %) and no R² recovery indicate the need for fault investigation. Model fault required retraining or feature revision due to inconsistent behavior, high persistent bias in SPE (66.67%), and R². Thus, the ASSMI framework allows timely, fault-specific advisory actions to avoid unnecessary interventions, lower maintenance costs, and support robust soft sensor performance in complex, dynamic industrial environments. |
|---|---|
| ISSN: | 2590-1230 |