The emergence of Industry 4.0 has brought about rapid integration of cyber-physical systems (CPS) across various sectors, including manufacturing, energy management, and smart grids. As these complex systems become more distributed and networked, the need for novel monitoring, diagnostics, and distributed control technologies has become imperative. Additionally, this interconnectedness and heterogeneity pose several cybersecurity challenges, including heightened connectivity, increased attack surface, and diverse infrastructure. The consequences of any disruption to the availability of CIs can be catastrophic, considering their fundamental role in supporting societal needs. Thus, safeguarding these systems against potential threats is of utmost importance. Due to growing data and advancements in machine learning and deep learning, data-driven approaches have gained widespread application in CPS. This dissertation explores and presents data-driven approaches for threat detection and diagnosis in cyber-physical systems. Firstly, this dissertation introduces a deep learning framework for the detection and diagnosis of threats on Photovoltaic (PV) Systems. Extensive quantitative experiments are conducted, utilizing both waveform data and micro-PMU data, to validate the performance of the framework. Next, an adaptive hierarchical framework for the detection and localization of threats in distribution systems is presented. Additionally, techniques such as transfer learning and few-shot learning are investigated to enhance the utilization of labeled data samples, particularly in scenarios with limited or no training data. Furthermore, apart from addressing threats on PV systems, this research delves into the applications of characterization of distribution system flexibility, all within the context of cyber-physical systems. Lastly, a cyber-physical testbed is constructed for studying security in manufacturing systems. A dataset is generated by collecting network data, systematic data, and physical data. The performance of data-driven approaches for cyber-physical data fusion is analyzed.