Lee, Namkyoung (Ph.D.)
Interpretable and Speed Adaptive Convolutional Neural Network for Prognostics
Faulty rotating machines exhibit vibrational characteristics that can be distinguished from healthy machines using prognostics and health management methods. These characteristics can be extracted using signal processing techniques. However, these techniques require certain inputs, or parameters, before the desired characteristics can be extracted. Setting the parameters requires skill and knowledge, as they should reflect the component geometries and the operational conditions. Using convolutional neural networks for diagnosing faults on a rotating machine eliminates the need for parameter setting by replacing signal processing with mathematical operations in the networks. The parameters that affect the outcomes of the operations are learned from data during the training of the neural networks. The networks can capture characteristics that are related to the health state of a machine, but their operations are not interpretable. Unlike signal processing, the internal operations of the networks have no constraints that guide the networks to transform vibrations into certain information, that is, vibrational characteristics. Without the constraints, there is no basis for understanding the characteristics in terms that can be associated with the physics of failure. The lack of interpretability impedes the physical validation of vibrational characteristics captured by the networks. This dissertation presents a method for changing the internal operations of a convolutional neural network to emulate a specific type of signal processing known as envelope analysis. Envelope analysis demodulates vibrations to extract vibrational signatures associated with mechanical impact on a defective rolling component. An understanding of envelope analysis, along with knowledge of the geometries of machine components and operational speeds, allows for a physical interpretation of the signatures. The dissertation develops speed adaptive convolutional layers and a rotational speed estimation algorithm to identify defect signatures whose frequency components change as the speed changes. The characteristics that are captured by the developed convolutional neural network are verified through a feature selection process that is designed to filter out physically implausible features. Case studies on three different systems demonstrate the feasibility of using the developed convolutional neural network for the diagnosis.
Nave, Gilad (Ph.D.)
Electrical and Structural Formation of Transient Liquid Phase Sinter (TLPS) Materials During Early Processing Stage
The growing demands of electrification are driving research into new electronic materials. These electronic materials must have high electrical conductivity, withstand harsh environments and high temperatures and demonstrate reliable solutions as part of complete electronic packaging solutions. This dissertation focuses on characterizing the initial stage of the manufacturing process of Transient Liquid Phase Sinter (TLPS) alloys in a paste form as candidates for Pb-free high-temperature and high-power electronic materials. The main objective of this dissertation work is to investigate the factors and decouple the multiple cross effects occurring during the first stage of TLPS processing in order to improve the understanding of material evolution. The work proposes, develops, and conducts in-situ electrical resistivity tests to directly measure material properties and analyze the dynamics at different stages of the material’s evolution. The research explores various factors, including alloying elements, organic binders, and heating rates, to understand their effects on the development of electrical performance in electronic materials. More specifically, the work examines the performance of Ag-In, Ag-Sn and Cu-Sn TLPS paste systems. Additionally, packing density and changes in cross-section are investigated using imaging techniques and image processing to gain insights into the early formation of the material’s structural backbone. An Arrhenius relationship together with Linear Mixed Models (LMM) techniques are used to extract the activation energies involved with each of the processing stages. The study then develops procedures to model different states of the TLPS microstructures at different heating stages based on experimentally observed data. Using these models, the study uses Finite Element Method (FEM) analysis to verify the experimental results and gain a better understanding and visualization into the involved mechanisms. This investigation not only sheds light on the material’s behavior but also has implications for robust additive manufacturing (AM) applications.
Yun, He (Ph.D.)
Manufacturability and Reliability of Additively Manufactured Planar Transformer Windings Using Silver-Based Paste
This dissertation is primarily concerned with the integration of additive manufacturing (AM) techniques into planar magnetics to achieve more efficient designs for power modules, which are in high demand. The two main focuses of this dissertation are: (1) the use of a paste-based AM technique called syringe-printing to create planar transformer windings without the need for pressure, using silver-based paste. The dissertation will address manufacturing considerations such as trace width, gaps, and heights that are printable, as well as the impact of electrical resistivity on the sintering process for the syringe-printed silver-based windings; and (2) the evaluation of the reliability of the syringe-printed silver-based windings, which will involve assessing adhesion performance between the metal/ceramic interface, conducting accelerated life tests (including thermal aging and thermal cycling tests), and identifying failure modes, failure sites, failure mechanisms, and developing degradation/failure models. In order to achieve the desired printing geometry in terms of width and gaps between segments, printing settings were studied parametrically by fitting targeted values with actual values. A low- temperature sintering profile was optimized, with a dwell time of 8 hours at 350°C resulting in a resistivity as low as 4.39 x 10-8Ω ∙ m, which was approximately 2.5 times higher than bulk silver. To improve bonding prior to syringe-printing the silver-based windings, it was suggested that an adhesive layer consisting of titanium (Ti) and silver (Ag) be deposited onto the alumina substrate. A degradation model was developed for thermal aging tests. Two batches of single-layer 7-turn syringe-printed windings were subjected to thermal cycling tests, and the corresponding failure modes and mechanisms were investigated. The failure data was used to combine with the strain- energy density extracted from the finite element simulation to develop the fatigue model, with the Coffin-Mason model being fitted for future comparison. A more conservative model could be recommended for real-world applications. Finally, the silver-based paste was syringe-printed onto a cooler with a limited footprint area, which served as the primary and secondary planar transformer board and was used in a 10 kW DC-DC full-bridge power converter with 97% efficiency. Corresponding thermal and electrical performance were discussed.
Kim, Changsu (Ph.D.)
Measurements of Effective Cure Shrinkage of Epoxy Molding Compound and Induced In-line Warpage during Molding Process
Cure shrinkage accumulated only after the gel point is known as effective cure shrinkage (ECS), which produces residual stresses inside molded components. The ECS of an epoxy-based molding compound (EMC) is measured by an embedded fiber Bragg grating (FBG) sensor. Under a typical molding condition, a high mold pressure inherently produces large friction between EMC and mold inner surfaces, which hinders EMC from contracting freely during curing. A two-stage curing process is developed to cope with the problem. In the first stage, an FBG sensor is embedded in EMC by a molding process, and the FBG-EMC assembly is separated from the mold at room temperature. The molded specimen is heated to a cure temperature rapidly in the second stage using a constraint-free curing fixture. The ECS of an EMC with a filler content of 88 wt% is measured by the proposed method, and its value is 0.077%. The measured ECS can be used to predict the warpage caused by molding processes. The validity of the prediction can be verified only by measuring the warpage during molding. A point-based measurement technique utilizing uniquely-generated multiple beams and binarization-based beam tracing method is developed to cope with the challenges associated with the warpage measurement during molding. The proposed method is implemented successfully to measure the warpage of a bimaterial disk that consists of aluminum and EMC as a function of time during molding process. Measurements are repeated to establish the measurement accuracy of the proposed method.
Sangepu, Lokesh (M.S)
Part Selection and Management Based on Reliability Assessment for Die-Level Failure Mechanisms
Electronic part manufacturers often provide reliability values in metrics such as Mean Time Between Failures (MTBF) and its inverse, Failures in Time (FIT). These metrics assume a constant failure rate and do not account for damage accumulation or wear-out phenomena, making the part selection and management based on this information meaningless. This thesis will report on the challenges associated with manufacturers' avoidance of sharing critical part information and how insufficient information hampers decision-making for part selection. This thesis uses four die-level failure mechanisms (Electromigration, Time-Dependent Dielectric Breakdown, Hot Carrier Injection, and Negative Bias Temperature Instability) as demonstration cases. It investigates the extent to which industry-published documents can be used to obtain the data necessary to simulate these mechanisms. It will report on methods of selecting an appropriate failure model based on the part technology level and identifying the required parameters for estimating the part's time to failure. Various scattered part information sources, literature, and industry-published documents may include the input parameters of failure models. The thesis provides insights into the complexity of understanding these information sources and various methods to obtain the required parameters to estimate the time to failure distributions. The methodology considers the susceptibility of parts to die-level failure mechanisms and compares components for reliability. A simulation template that facilitates practical implementation by enabling designers, engineers, and procurement teams to make informed decisions while selecting electronic parts for specific applications is introduced. The research findings and methodology presented provide valuable insights for users to improve the reliability and performance of electronic systems through effective part selection.
Kong, Lingxi (Ph.D.)
In-situ Investigation of Lithium Dendrite Growth and Its Interactions with a Polymer Separator in a
Lithium Metal Cell
Lithium dendrites are metallic structures that initiate and grow inside a lithium battery during charging. Lithium dendrite growth can negatively affect battery cycle life and safety. Observing the dendrite growth process and revealing its interaction with other components is necessary to improve battery safety. This study uses a transparent optical cell to directly observe the dendrite growth process, explore the lithium dendrite growth modes under various current densities, evaluate the interactions between the dendrite and separator, and explore the effect of electrolyte additives on dendrite growth behavior. The dendrite growth under different current densities showed the transition of dendrite morphologies from a dense structure to a porous structure. The examination of the dendrite-separator interaction regions showed that dendrites can deform and penetrate the separator. We show that additives can enhance the uniformity of lithium dendrite distribution compared with the dendrite formed in the electrolyte without additives.