Industrial Cyber-Physical Systems (ICPS) are heterogenous computer systems interacting withphysical processes in an industrial environment. The high degree of distribution and interconnectedness poses significant security threats to ICPS, including Slow−Rate Attacks (SRA ). SRA often exploit system vulnerabilities arising from the inherent limitations of resource-constrained ICPS computers like programmable logic controller (PLC). Existing literature identify several challenges in detecting and classifying SRA in resource-constrained ICPS. In this article, we propose an optimised Online Sequential−Extreme Learning Machine (OSELM)-based, novel light-weight active security framework for SRA detection. We reduce the memory and space footprint of OSELM through optimisation techniques for deployment in resource-constrained ICPS. Also, we introduce a simple stratified k-fold cross training method to improve the binary and multi-class SRA- detection capability of the proposed framework, focusing on performance and accuracy of attack detection. Experimental results demonstrate that our proposed framework can effectively and efficiently detect binary and multi-class SRA with an accuracy of 0.975 and 0.96 with average detection times of 0.03 and 0.04 sec, respectively, using less than 3 MB of memory and 10-12% of CPU usage in PLC-based ICPS.