Data Analytics Models and Methods for Fault Identification and Prognosis in Mechanical Structures and Manufacturing Processe

The accurate detection and efficient prognosis of faults in engineering systems are of great practical importance. The systems concerned encompass a broad spectrum of human-made structures and processes, including civil, mechanical and aerospace structures and various manufacturing processes. The precise detection of faults involved in the systems is critical in avoiding structure deterioration, performance degradation, productivity loss and even loss of lives. Prognosis is the ability to predict accurately the future condition of the systems, such as degradation status and remaining useful life. The prognosis helps to carry out the optimal maintenance scheduling for structures and smart operation management of manufacturing processes.;The rapid development of sensor techniques makes it possible for data collection in a quick and accurate manner. Quantitative analysis based on physical model or statistical model applying on the large amount of collected data provide great opportunities for achieving precise fault detection and prognosis. However, significant and fundamental challenges exist in fully exploiting the available data to achieve this goal. For example, the identifiability of a fault based on collected data is essential and should be addressed before any fault identification efforts. Specifically, the commonly used finite element model (FEM) has not been validated for its identifiability in the application of structural damage identification. The induced bias due to linearization is often ignored for damage estimation, which may lead wrong fault identification. Also, efficient methods to predict the progression of structural properties based on finite element models are lacking. Furthermore, various data types require specific data modelling and analysis techniques for fault detection beyond the traditional statistical monitoring methods in manufacturing processes. These issues are being studied in this dissertation.;Specific contributions of this thesis are made in fault identification and prognosis in mechanical structures and manufacturing processes. In mechanical structures, the identifiability of FEM, the bias reduction by measurements selection and the prognosis of structural property degradation are addressed. In specific:;• A quantitative framework is proposed to address the identifiability of structural damage identification based on finite element models.;• A measurement selection algorithm is proposed for bias reduction in damage estimation.;• A hierarchical Bayesian degradation model is proposed to efficiently estimate the trend of damage growth in structures.;In manufacturing processes, two specific methods are proposed for fault identification of untraditional data type. Specifically,;• Defects with specific spatial patterns on semiconductor wafer are recognized by converting the original pattern recognition problem as point matching problem using Hough Transformation.;• Variations of acoustic attenuation curves are being quantified by linear mixed effect model and permutation tests to provide the guidelines on the quality inspection in nanocomposites manufacturing.;Besides the aforementioned challenges, there are other issues need to be addressed. For example, the integration of piezoelectric transducer circuitry network into mechanical structures enhances the performance of frequency-shift-based damage identification method. However, a quantitative analysis on the tuning variable of the network is lacking of studies. The quantitative study will not only enhance the understanding of such integrated network, but also provide iv guidelines on tunings to achieve the optimal fault identification. Also, the location of the integrated network significantly influences the performance of the fault identification. Analysis on the optimal allocation of the transducers leads the most sensitive system response due to the structural damages, in which provides the most accurate fault detection.