Feature extraction and fault detection scheme via improved locality preserving projection and SVDD

Abstract

Manifold learning is widely adopted for the fault detection of industrial processes. However, the quality of low-dimensional embedding coordinates can be adversely affected by ill-constructed graph Laplacian. An improved locality preserving projection (ILPP) scheme is proposed. ILPP is built on a geometrically inspired Laplacian, and the Riemannian metric is used to find the suitable bandwidth parameter. The proposed approach combines the advantages of ILPP in preserving manifold data structures and those of support vector data description (SVDD) in handling complex process data distributions. Case studies on helix data, hot strip mill, and Pensim benchmark processes demonstrate the utility and feasibility of the proposed approach. The average fault detection rate for proposed ILPP is 99%, which is higher than locality preserving projection (LPP; 87.8%), local tangent space alignment (LTSA; 74.9%), and principal component analysis (PCA; 90.6%).

Incipient fault detection and condition assessment in DFIGs based on external leakage flux sensing and modified multiscale poincare plots analysis

“…where ξ i is the relaxation variable that enables some of the training samples to be outside the θ; The C represents the regularization parameter used to adjust the number of misjudged samples and the size of R to prevent the model from being damaged by individual extreme samples [19]. According to the existing research [16,19,25,26], the relaxation variable requires ξ i >0, and the regularization parameter generally requires 0 < C ⩽ 1.…”

“…According to metric-based distance and statistics results of the SVDD [16,25], the earliest anomaly point, the accuracy, the fault detection rate (FDR), and the false alarm rate (FAR) are descriptions of healthy conditions [25], FDR and FAR are defined as follows: FDR = Number of detected fault samples Total number of fault samples × 100% FAR = Number of healthy samples identified as fault Total number of healthy samples × 100%…”

“…In this manuscript, the relaxation variable ξ i was 0.04, and the regularization parameter C was 0.5. In this case, the hypersphere obtained using gaussian kernel functions was more closely surrounded by the normal state samples and more sensitive to the abnormal state samples, and it is easier to monitor the occurrence of the abnormal state of the automaton [16,19,25,26,39].…”

“…In addition, the modified MSPP and GMPOP [19] have similar and better performances in various aspects and are all better than PP [23,24], and other methods. Moreover, although using the 24 feature indicators in this manuscript based on the recommendations of [41] and [42] combined ILPP [25] can also detect the incipient fault and have a good FDR (%) and accuracy (%), it has higher FAR (%) for false alarms than the proposed method.…”

“…Mechanical and electrical equipment systems, particularly high-powered wind generators, often have easily accessible data for normal operating conditions [16,19,25,26]. By contrast, acquiring data on fault conditions may be more difficult and expensive.…”