Interval Valued PCA-Based Approach For Fault Detection In Complex Systems

The purpose of this article is to emphasize the importance of detecting process sensor faults using Principal Component Analysis (PCA). In practice, uncertainties in sensor data influence the system and introduce some problems into the control decision-making process, which results in an increased frequency of false alarms and inaccurate choices. As a consequence, a current method for expressing the influence of these uncertainties on sensors has been adopted, namely, an interval-valued data representation. Process modeling was carried out using PCA for interval-valued data, with four of the most well-known techniques being evaluated. To reduce false alarms, a threshold with a specified degree of confidence has been created for both Hotelling's $T^{2}$ and $Q$ statistics, along with a novel statistic Φ for detecting process problems. Finally, to verify the capacity of the suggested approach to reduce false alarms and missed detection rates, we tested cement rotary kiln data.