Sarcouncil Journal of Engineering and Computer Sciences

Abstract: Plastic pipe production involves multiple interconnected stages, such as extrusion, cooling, and cutting. Each of these stages is prone to various types of failures, whether mechanical or environmental, which can disrupt workflow, reduce product quality, and lead to economic losses. Traditional maintenance methods, including reactive and preventive approaches, are often costly and inefficient. In this study, a predictive maintenance approach is proposed using a Long Short-Term Memory (LSTM) neural network optimized with the Penguin Search Optimization Algorithm (PeSOA). LSTM networks are particularly suited for analyzing time-series data from sensors due to their ability to learn long-term dependencies. PeSOA is applied to optimize key hyper-parameters such as learning rate, dropout rate, and number of LSTM units. Experimental results demonstrate that the LSTM + PeSOA model outperforms both the baseline LSTM and the LSTM optimized using Particle Swarm Optimization (PSO), achieving higher accuracy and reliability. This model shows promise in enabling early failure prediction and reducing unplanned downtime, offering a more intelligent and efficient maintenance solution for plastic manufacturing environments