Neuro-Symbolic Artificial Intelligence (NSAI) integrates symbolic reasoning with the neural network. This integration aims at using the neural network for pattern recognition and symbolic reasoning to convert engineering knowledge into systematic rules. The proposed work focuses on exploring the feasibility of NSAI for the industrial domain and analyzing its performance against a pure neural network system. The proposed system uses the RuleBank mechanism to update the confidence of the rules based on Bayesian learning. The experimental test of the proposed model uses the predictive maintenance application. The results demonstrate an accuracy of 95% compared to the pure neural model that gave a result of 97%. Despite that the results obtained does not outperform the pure NN model, it provide explainability results for each production made by the system. This highlights a provable competitive advantage compared to a baseline of a pure neural network, achieving respectable performance, while at the same time providing interpretable explanations for each prediction. The results highlight the potential of neuro-symbolic AI for reliable and explainable predictive maintenance in industrial environments.
Neuro-Symbolic AI for Industrial Systems: Bridging Deep Learning and Domain Expertise
Barakat, Ghena;Gupta, Harshit;Longo, Francesco;Merlino, Giovanni;Puliafito, Antonio
2026-01-01
Abstract
Neuro-Symbolic Artificial Intelligence (NSAI) integrates symbolic reasoning with the neural network. This integration aims at using the neural network for pattern recognition and symbolic reasoning to convert engineering knowledge into systematic rules. The proposed work focuses on exploring the feasibility of NSAI for the industrial domain and analyzing its performance against a pure neural network system. The proposed system uses the RuleBank mechanism to update the confidence of the rules based on Bayesian learning. The experimental test of the proposed model uses the predictive maintenance application. The results demonstrate an accuracy of 95% compared to the pure neural model that gave a result of 97%. Despite that the results obtained does not outperform the pure NN model, it provide explainability results for each production made by the system. This highlights a provable competitive advantage compared to a baseline of a pure neural network, achieving respectable performance, while at the same time providing interpretable explanations for each prediction. The results highlight the potential of neuro-symbolic AI for reliable and explainable predictive maintenance in industrial environments.Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.


