A randomized model ensemble approach for reconstructing signals from faulty sensors

Abstract

On-line sensor monitoring aims at detecting anomalies in sensors and reconstructing their correct signals during operation. The techniques used for signal reconstruction are commonly based on auto-associative regression models. In full scale implementations however, the number of sensors to be monitored is often too large to be handled effectively by a single reconstruction model. In this paper we propose to tackle the problem by resorting to a pool (ensemble) of reconstruction models, each one handling an individual group of signals. This approach involves two main technical steps: firstly, a procedure for constructing signal groups, and secondly a procedure for combining the outputs of the reconstruction models associated to the groups. For the signal grouping step, a wrapper optimization search is proposed to identify the optimal number of groups in the ensemble and the size of the groups. For the model output aggregation step, a simple arithmetic average is adopted. Ensemble accuracy and robustness is achieved by promoting diversity between the signal groups through the use of the Random Feature Selection Ensemble (RFSE) technique in combination with the Bootstrapping AGGregatING (BAGGING) technique for training data selection. The individual reconstruction models are based on Principal Components Analysis (PCA). The proposed approach has been applied to a real case study concerning 215 signals monitored at a Finnish nuclear pressurized water reactor. The results obtained have been compared with those achieved by an equivalent ensemble of models based on a grouping directly optimized by a Multi-Objective Genetic Algorithm (MOGA).

Robust nuclear signal reconstruction by a novel ensemble model aggregation procedure

“…, measured in the plant, a set of K groups of mn  signals are generated by randomly sampling the signals from the n available [12][13][14]. The procedure is simple, allows a direct and fast group generation suitable for large scale applications, guarantees high signal diversity between the groups (and thus high diversity between the models outcomes, beneficial to ensemble reconstruction) and attains high signal redundancy, upon a reasonable choice of the ensemble parameters m and K [14].…”

“…To do this, the data set X of N signal patterns available is partitioned into a training set X [12,14,20,21]; for the generic pattern 1, 2,..., ,( ) ( ) 1, 2,..., …”

“…The groups at the basis of the ensemble are generated by randomly selecting their signals [12][13][14]. This allows injecting high diversity in the group structure; for each group, a regression model based on Principal Component Analysis (PCA) [15][16][17][18] is developed.…”

Fault Detection in Nuclear Power Plants Components by a Combination of Statistical Methods

“…with overlapping, i.e., the same signal can belong to more than one group [12], [22]- [25], and without overlapping [26]- [28]. In practical applications, the latter strategy tends to be preferred because it allows for a smaller number of models to be developed, at a lower computational effort [28].…”

“…The 46-dimensional patterns (5798) have been divided into a set X M of 2798 patterns used to perform the hybrid signal grouping (Section 3), and a validation set X V of 3000 patterns used to validate the condition monitoring and fault detection scheme (Section 4). In [29], it has been shown that this hybrid method leads to reconstructions that are more tolerant to the fault propagation problem (that has been mentioned in the Introduction) than proceeding with the reconstruction based on the single group of all signals [19], [27], or on groups of signals defined by a filter CA approach only [19], [25], or by a wrapper GA-based approach only [28].…”