Operational Resilience in Industrial Systems: Latency-Aware and Adaptive Architectures

Abstract

Operational resilience in industrial systems requires continuous adaptation to dynamic loads, uncertain disturbances, and strict timing constraints. Existing intelligent automation approaches emphasize accuracy but often overlook latency-aware decision-making and resilience under fluctuating industrial conditions, creating limitations in time-critical environments. This research aims to design Latency-Aware Deep Learning (DL) Architectures for Operational Resilience in Industrial Systems using a Cat Swarm Algorithm-driven Dynamic Deep Neural Network (CSA-DDNN). The purpose of the proposed method is to enable low-latency, fault-tolerant, and reliable industrial decision-making under dynamic operating conditions. Data pre-processing employs normalization and noise filtering using Min–Max normalisation and adaptive Kalman filtering to ensure temporal consistency. Feature extraction integrates Principal Component Analysis (PCA) to derive compact and discriminative representations. The model systematically acquires data, performs preprocessing, extracts salient features, and feeds them into a latency-aware learning module for resilient decision-making. The CSA-DDNN model utilizes CSA for optimal hyperparameter tuning and adaptive weight selection, enhancing exploration–exploitation balance, while the DDNN adjusts its structure in response to latency variations and system uncertainties for robust inference. This architecture explicitly models latency constraints, enabling timely responses and sustained operational stability. The Experimental result demonstrates an accuracy of 98.4%, an F1 score of 98.1%, parameters of 3.8 M, FLOPs of 250 M, Latency of 17 ms, and Energy of 6.6 mJ, which is implemented through Python. The approach enables scalable, low-latency, and resilient industrial intelligence with efficient resource utilization and consistent operational continuity.