<p>Industry 5.0’s rapid smart manufacturing growth has increased demand for intelligent, resilient, and sustainable anomaly detection frameworks. Traditional machine learning methods require large labeled datasets, have low interpretability, scalability difficulties, and high processing costs, limiting their real-world usefulness. This paper suggests a football player optimization algorithm (FPOA)-tuned variational autoencoder (VAE) for robust industrial anomaly detection to address these issues. The 100,000 recordings with 13 sensor properties record machine operating conditions like vibration, temperature, pressure, energy consumption, downtime risk, and system statuses. To counter balancing the experimental apparatus, erroneous conditions such as overheating, over-vibration and pressure dip were indicated against base operations. The proposed framework was experimented on the basis of truthful and faulty environments with 70 training, 15 validation and 15 testing. The FPOA-tuned VAE performed better than typical models as well as the Optuna baseline with F1-score of 0.983, AUC of 0.991 and accuracy of 0.982 and stable convergence between training and validation loss. The comparative research studies revealed that the method identified rare anomalies with minimal false positives, which guaranteed predictive maintenance reliability. The framework fosters human-centricity, resilience, and sustainability of Industry 5.0, as opposed to accuracy. The proposed model integrates technical innovation, social and ecological objectives, and explainability (through SHAP-based feature attributions), resistance to noisy inputs, and predictive maintenance, which reduces resource wastage. These findings indicate that FPOA-tuned VAE is a light, interpretable, and potent anomaly detecting framework of Industry 5.0 manufacturing.</p>