Analysis and Prediction Model of Loudspeaker Defect Based on Knowledge Graph and Graph Neural Network
Bowen Yuan, Liqin Guo · Procedia Computer Science · 2025
In the speaker production process, defect analysis and prediction are key links to improve production efficiency and product quality. However, traditional defect analysis methods often rely on experience and have difficulty in handling complex production data and potential cause correlations. To address this problem, this paper proposes a speaker production defect cause analysis and prediction model based on knowledge graph and graph neural network (GNN). First, this paper constructs a knowledge graph that contains information on production processes, raw materials, production environment, worker operations, etc. Then, this paper uses the GNN model to process the constructed knowledge graph and uses the topological structure of the graph and the relationship between nodes to discover hidden causes of defects. In order to improve the prediction accuracy of the model, this paper combines a multi-layer graph convolutional network (GCN) to learn the features of nodes in the graph. Finally, the model is trained and verified using actual production data to optimize the accuracy and reliability of defect prediction. Experimental results show that the GNN-based defect cause classification accuracy is 88%, and the F1 value is 0.86. In addition, the AUC (Area Under Curve) value of GNN in defect prediction is 0.92, which is better than the LSTM (Long Short-Term Memory) model. Through multi-factor correlation mining, temperature and humidity are identified as the main influencing factors of defects in speaker production, providing an important scientific basis for production optimization. The model can effectively reveal the multi-factor correlation in the production process and provide effective support for quality control and process optimization.