Description
TitleAnomaly detection and process diagnostic in three-dimensional surface topography
Date Created2022
Other Date2022-10 (degree)
Extent1 online resource (187 pages) : illustrations
DescriptionThe recent development of optical measuring instruments has increased the use of surface topographic data for monitoring the quality of the engineered surface. Monitoring topographic variations in the surface is crucial for quality engineers since the change in the surface finish is closely related to the performance of products in use. However, several challenging issues such as the existence of spatial autocorrelation within the surface, and changes of topographic features such as position and shape of peaks and valleys across defect-free surfaces make it difficult to monitor variations in the surface. In addition, existing monitoring approaches fail to detect local changes in the surface. Moreover, surface inspection becomes more difficult when the surface contains local defects that are often concealed due to the high-dimensional data structure of the surface image.
This dissertation presents efficient anomaly detection approaches for monitoring topographic variations in the three-dimensional (3D) surface. We first propose a residual-based surface segmentation approach to effectively identify local surface changes. We obtain residuals from the surface through a fitted surface prediction model, which characterizes the generic behavior of defect-free surfaces. Then, the proposed approach binarizes the surface based on local residual properties and identifies the defective areas where residuals are spatially autocorrelated. By considering both deviated residuals and neighboring residuals, which have similar values from local regions, the approach effectively captures residual patterns in the defective areas, which are concealed due to the high-dimensional surface data structure. Then, a spatial randomness-based monitoring statistic is introduced to evaluate binarized surfaces in order to detect surface anomalies.
Second, we investigate unique surfaces that exhibit multiple in-control modes. Due to the complexities of modern industrial processes, surfaces of final products under the normal manufacturing process may have multiple modes, such that the surface consists of different topographic features from one in-control mode to another. In this case, existing monitoring approaches based on the single mode surface cannot characterize normal surfaces with multiple modes, and result in poor detection performance. To overcome this limitation, a new approach for monitoring variations in multimode surface topography is presented. We propose a multimode surface prediction model, which characterizes the generic behavior of normal surfaces with multiple in-control modes. Moreover, we present a mode-specific surface monitoring approach that identifies topographic variations on the surfaces based on the similarity between probability density function (PDF) of residuals from observed and normal surfaces obtained through the multimode surface prediction model. A novel probabilistic distance measure (PDM) is introduced to effectively measure the similarity between a single residual PDF and a set of residual PDFs under the same mode.
We then propose a novel approach for monitoring local topographic variations in the presence of multimode surface topography. We present a multimode surface binarization model that characterizes distinct topographic features of surfaces under different in-control modes and better identifies the local variations in the surface data. In order to systematically evaluate the deviations between the observed and defect-free surfaces with multiple in-control modes, we introduce a novel PDM that effectively compares spatial randomness between two binarized surfaces.
Lastly, we present a new approach for monitoring and diagnosing local topographic variations on the surface topography. Once the observed surface is considered as an anomaly, identifying the shape, location, or size of the defective area is essential for diagnosing the root cause of the abnormal surface. Thus we develop a variable selection-based fault detection and diagnostic approach that selects potentially defective areas on the surface by considering the spatial structure of the surface and residual information. Suspicious areas are then employed to detect anomalies and further utilized to identify defect size, location, and shape for fault diagnosis purposes.
The proposed surface monitoring approaches in this dissertation are efficient and effective in monitoring spatial patterns in topographic data. The proposed monitoring approach investigated in Chapter 3 is superior to existing surface monitoring approaches regardless of the defect size and the number of defects. In the presence of multimode in-control surfaces, the mode-specific surface monitoring approach developed in Chapter 4 outperforms existing single mode and multimode-based approaches in detecting global topographic variations. Moreover, local topographic variations in the presence of multimode surface topography are effectively captured by the proposed approach introduced in Chapter 5. Finally, the proposed approach presented in Chapter 6 identifies fault diagnostic information such as fault size and location, and it is sensitive to detecting local variations on the surface. The effectiveness of the proposed approaches is demonstrated through extensive numerical simulation studies and real-life applications of paper surface monitoring.
NotePh.D.
NoteIncludes bibliographical references
Genretheses
LanguageEnglish
CollectionSchool of Graduate Studies Electronic Theses and Dissertations
Organization NameRutgers, The State University of New Jersey
RightsThe author owns the copyright to this work.
Statistical Profile