Artificial Neural Network-Based Vibration Data Analysis to Predict Failures of Powder Machines

Abstract

In the dynamic landscape of Industry 5.0, this research pioneers an advanced predictive maintenance methodology for a forced blower in PVC powder manufacturing at TCI Sanmar. Leveraging machine learning, this study reveals the Multilayer Perceptron (MLP) algorithm as a top performer, exhibiting exceptional efficacy in foreseeing machine failures. Following a comprehensive GridSearchCV optimization, the best-performing MLP variant demonstrated remarkable metrics, boasting an accuracy of 92.5%, Recall of 95.5%, Precision of 95%, an F-score of 92.7%, and a Matthews Correlation Coefficient (MCC) of 0.85. Furthermore, achieving consistently high Area Under the ROC Curve (AUC) values at 0.961, the selected MLP configuration outshines counterparts in the same industrial setting. This research extends beyond immediate scope, contributing crucial insights to predictive maintenance strategies. The results affirm the strategic importance of the MLP algorithm in the Industry 5.0 context, emphasizing its role in intelligent, interconnected manufacturing processes. This exploration optimizes equipment reliability, minimizes downtime, and provides valuable insights for industrial efficiency through proactive maintenance interventions.

Keywords

• maintenance strategy optimization
• industry 5.0 integration
• advanced machine learning applications
• equipment health monitoring
• intelligent manufacturing insights