Extending Process Monitoring to Simultaneous False Alarm Rejection and Fault Identification (FARFI)
A new framework for extending Statistical Process Monitoring (SPM) to simultaneous False Alarm Rejection and Fault Identification (FARFI) is presented in this paper. This is motivated by the possibly large negative impact on product quality, process safet
- PDF / 341,876 Bytes
- 15 Pages / 439.37 x 666.142 pts Page_size
- 33 Downloads / 173 Views
Abstract. A new framework for extending Statistical Process Monitoring (SPM) to simultaneous False Alarm Rejection and Fault Identification (FARFI) is presented in this paper. This is motivated by the possibly large negative impact on product quality, process safety, and profitability resulting from incorrect control actions induced by false alarms— especially for batch processes. The presented FARFI approach adapts the classification model already used for fault identification to simultaneously perform false alarm rejection by adding normal operation as an extra data class. As no additional models are introduced, the complexity of the overall SPM system is not increased. Two case studies demonstrate the large potential of the FARFI approach. The best models reject more than 94 % of the false alarms while their fault identification accuracy (> 95 %) is not impacted. However, results also indicate that not all classifier types perform equally well. Care should be taken to employ models that can deal with the added classification challenges originating from the introduction of the false alarm class. Keywords: Chemometrics · Statistical Process Monitoring (SPM) Fault Detection & Identification (FDI) · Batch processes
1
·
Introduction
Modern industry strives towards better-performing processes, resulting in safer, more profitable, and environmentally friendly production. Early detection of deviations from normal operation, as well as the identification of the root causes of these deviations, is of utmost importance [1–4]. Hence, a close monitoring is paramount, and Fault Detection and Identification/Isolation (FDI) has received much attention over the last few decades [2–5]. While first-principles models can be used for FDI, this approach is limited to well-known processes of small size [6]. Statistical Process Monitoring (SPM) c Springer International Publishing Switzerland 2016 P. Perner (Ed.): ICDM 2016, LNAI 9728, pp. 334–348, 2016. DOI: 10.1007/978-3-319-41561-1 25
Extending Process Monitoring to Simultaneous FARFI
335
aims at exploiting already available databases of process measurements for FDI in more complex processes such as (bio)chemical, steel, pulp and paper, or semiconductor industries. Extensive overviews of SPM are available in, i.a., [1–11]. SPM systems typically consist of three steps [12]. First, a model characterizing Normal Operation Conditions (NOC) is established. Next, fault detection references current process measurements against this model to detect deviations from NOC. Finally, fault identification identifies the root cause of the detected disturbance. Since fault detection in SPM is performed via statistical analysis, the occurrence of false alarms is unavoidable. As they trigger unnecessary— and in some cases incorrect—control actions, which in turn potentially affect process safety, product quality, and profitability, the influence of false alarms on the entire SPM system should be reduced as much as possible. Even so, the influence of these false alarms on fault identification is typically negl
Data Loading...
