Industrial Control Systems (ICS)s are crucial for monitoring and controlling critical infrastructure such as smart grids, the oil and gas industry, and transportation. Traditionally, ICSs were placed on isolated communications systems to keep them safe from cyber-attacks, where they were relatively inconspicuous and unknown to most attackers. The connection between ICS and wide area networks facilitates the online access, monitoring, and control of the systems remotely. The drawback of wide area network integration is increased attack surfaces in ICS and increased vulnerability to cyber-criminals. Considering the critical function of ICS within vital infrastructure operations, system compromise could result in substantial risks to both human life and the environment. Furthermore, the escalating frequency of cyber-attacks on ICS in recent years accentuates the pressing need for timely and effective security solutions. This thesis will propose an explainable deep federated multi-modal model for cyber-attack detection in ICS environments to secure the ICS networks and mitigate potential risks to human lives and the environment. Moreover, the proposed model will focus on the challenges of using machine learning techniques in the cybersecurity of the ICS, such as data privacy of training the models on the cloud. In addition, the proposed model analyzes sensor and network modalities of an ICS and detects cyber-attacks based on their joint abstract representation. Furthermore, a federated learning-based technique for the cyber-attack detection model will be proposed to build a powerful detection component based on several ICS without sharing clients' data to ensure data privacy.