Mohd Herwan Sulaiman obtained his B. Eng. (Hons) in Electrical-Electronics, M. Eng (Electrical-Power) and PhD (Electrical Engineering) from Universiti Teknologi Malaysia (UTM) in 2002, 2007 and 2012 respectively. He is currently serves as an Associate Professor at Faculty of Electrical & Electronics Engineering, Universiti Malaysia Pahang (UMP). His research interests are power system optimization and swarm intelligence applications to power system studies. He has authored and co-authored more than 100 technical papers in the international journals and conferences and also has been invited as a Journal reviewer for several international impact journals in the field of power system, soft computing application and many more. He is one of the inventors of the new nature inspired algorithm namely Barnacle Mating Optimizer. He is also a Senior Member of IEEE. His website can be accessed through .
RESEARCH INTERESTS
Power system optimization and swarm intelligence applications to power system studies
EMAPlus-optimized adaptive convergence prescribed performance control for high-precision steering of rack steering vehicles Addie Irawan, Norsharimie Mat Adam, Mohd Iskandar Putra Azahar, Mohd Zamri Ibrahim, Mohd Herwan Sulaiman Engineering Research Express, 2026 This paper presents an optimal Adaptive Convergence Prescribed Performance control cascaded with Anti-Windup PI (ACPPC-API) controller for steering-position control of rack steering vehicles (RSV) operating on cornering paths, optimized using the proposed Enhanced Evolutionary Mating Algorithm Lite (EMAPlus). The ACPPC framework regulates steering-error evolution through dynamically shaped convergence envelopes, while the Anti-Windup PI (AW-PI) inner loop stabilizes actuator behavior under saturation constraints. EMAPlus is employed to jointly tune the ACPPC and AW-PI parameters, enabling fast, stable, and computationally efficient optimization compared with the original Evolutionary Mating Algorithm (EMA), Ant Lion Optimizer (ALO), and Grasshopper Optimization Algorithm (GOA). Simulation results demonstrate that the EMAPlus-tuned ACPPC–API controller achieves the highest steering-tracking accuracy, reducing overshoot by up to 64%, suppressing residual ripple to within ±0.02 radians, and improving settling time by 20%–35% relative to the benchmark optimizers. These performance gains translate into superior vehicle-level responses, including 30%–60% lower curvature-tracking error, 45%–65% smaller sideslip deviation, and smoother lateral–yaw coordination during cornering maneuvers. Actuator-level and ride-quality indicators further reveal 35%–60% reductions in peak road-wheel rate and lateral jerk. Energy analysis confirms that more than 80% of the lateral–yaw kinetic energy is effectively directed into productive lateral-velocity motion with a shortened transient duration. The results establish the EMAPlus-optimized ACPPC–API controller as an efficient and robust steering solution for high-precision RSV cornering applications.
Tool wear classification in CNC machining via metaheuristic optimization of discrete neural network configurations Mohd Herwan Sulaiman, Zuriani Mustaffa, Mohd Razali Daud Engineering Research Express, 2025 Tool wear detection is essential for predictive maintenance in CNC machining systems, enabling early identification of worn tools to reduce defects, minimize unplanned downtime, and improve production efficiency. Traditional approaches, often relying on manual inspection or fixed thresholds, suffer from limited accuracy and adaptability. This study explores the use of metaheuristic optimized feedforward neural networks for automated tool wear classification using a publicly available CNC milling dataset. Three nature-inspired algorithms, namely Barnacles Mating Optimizer (BMO), Particle Swarm Optimization (PSO), and Artificial Protozoa Optimization (APO), were employed to optimize discrete neural network parameters, including the number of hidden neurons (ranging from 5 to 100), hidden layer activation functions ( tansig , logsig , ReLU ), and output layer activation functions ( purelin , tansig , logsig ). Model performance was evaluated using accuracy, precision, recall, F 1 score, and AUC across five independent runs. The BMO-NN model achieved the highest average results, with an accuracy of 92.49 percent, precision of 91.86 percent, recall of 93.92 percent, and F 1 score of 92.88 percent. The best performing BMO-NN configuration used 100 hidden neurons with tansig activation functions in both layers. These findings highlight the potential of BMO based neural networks for robust and accurate tool condition monitoring in intelligent manufacturing.
