The paper presents a novel technology of data exclusion of intrusions in the Military Industrial Internet of Things (MIIoT) area via the technique of federating deep learning. Through utilizing the Flower Framework entropy mechanism and a customized FedAvg strategy including noise and clipping techniques, we create an optimized model to fit the specific requirements of MIIoT protocols. This study scrutinizes a particular part of a wide ranged data set which covers mostly Modbus data for the control systems and the military industrial applications. Specifically, this domain stands out from the others and, surprisingly, few things have been undertaken to assess its importance. Utilizing federated learning as a pivotal component of a distributed MIIoT system helps to overcome privacy concerns and bandwidth constraints among other issues while bringing value to the system. At the same time, the system has the capability of potentially identifying security threats. Data suggests that federated deep learning prove to be a defensive tool against military network intrusions.