This dissertation proposes novel Domain Adaptation (DA) methods in real-world industrial settings, where the availability of labeled data is limited and test data can significantly differ from training data. Particularly, our research addresses key challenges in DA, including the applicability of DA methods in industrial settings, strategies to enhance performance in the target domain with limited samples, improving DA methods in the presence of significant domain gaps, and aligning source and target domains when the target dataset is highly imbalanced. The dissertation comprises three primary sections, each focusing on a specific aspect of DA in industrial applications. The first section introduces AVATAR (Adversarial Self-Supervised DA Network for Target Domain), a novel unsupervised DA method that combines adversarial learning and self-supervised learning techniques to enhance model adaptability and accuracy in unlabeled environments. AVATAR addresses complex Unsupervised Domain Adaptation (UDA) tasks by balancing domain-invariant feature learning with domain-specific adjustments, enabling effective adaptation to dynamically changing environments. The second section presents ReflowNet, a domain-adaptive Convolutional Long Short-Term Memory (ConvLSTM) neural network-based oven recipe optimization framework for reflow soldering in Printed Circuit Board (PCB) assembly. ReflowNet utilizes synthetic datasets generated from physics-based Computational Fluid Dynamics (CFD) simulations and real-world experimental trial data to predict optimal oven recipes based on process-specific spatiotemporal information. By leveraging DA technique and ConvLSTM networks, ReflowNet offers a novel and effective solution for optimizing reflow soldering oven recipes in PCB assembly. The third section introduces STARBEAR (Spectrogram Transformer-based Adaptive Recognition method for Bearing faults), designed to bridge the domain gap in industrial machinery diagnosis tasks. STARBEAR integrates domain adversarial learning, self-supervised learning, and the state-of-the-art spectrogram transformer architecture under a single training framework, enhancing the model's ability to discern and adapt to bearing faults under diverse operating conditions. The proposed model demonstrates superior performance on three UDA benchmarks tailored to bearing fault diagnosis, highlighting its potential as a robust, adaptive solution for predictive maintenance in industrial settings. Through extensive experiments on multiple benchmark datasets and real-world industrial applications, the methods proposed in this dissertation demonstrate superior performance compared to existing DA approaches, particularly in scenarios with limited labeled target domain data and significant domain gaps. The research outcomes provide new insights and solutions for tackling DA challenges in various real-world industrial applications, ultimately leading to increased efficiency, reliability, and cost-effectiveness. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://bibliotheek.ehb.be:2222/en-US/products/dissertations/individuals.shtml.]