<p>With revolutionary developments in Industry 4.0, cyber attacks have also grown in industrial systems. Data sharing in distributed training compromises data privacy, but to control cyber attacks in multi-node architecture, data sharing should be stopped. The main aim of this paper is to propose and develop federated learning based privacy preserved and increment training approaches for cyber attack classification using the concept of Honeynet in Industry 4.0. Proposed methodology included (a) physical systems environment setup, (b) acquiring and preprocessing of a real-time cyber-attack classification dataset, (c) identifying better deep learning approach out of available approaches including GRU, LSTM, BiLSTM, TD-CNN-LSTM, TD-CNN-BiLSTM and ResNet50-1D, (d) analysis of ResNet50-1D over federated learning ecosystem using IID and Non-IID, (e) incremental learning approach for federated learning. ResNet50-1D functioned as a better algorithm for deep learning as well as on FL. Incremental learning approach also implemented over federated learning and analyzed on parameters, viz., accuracy, precision, recall, and loss. Federated learning-based privacy-preserving and incremental modeling implementations also result similar to the base deep learning implementations. This proposal could be used in a real-time scenario and further analyze for vertical training models in the federated learning ecosystem.</p>