Dual-Model Fusion Method Based on DBTBoost for Fault Diagnosis

Abstract

With the intelligent upgrading of manufacturing equipment, high precision and efficiency fault diagnosis improves the stability and productivity. To detect the fault state accurately, the fault diagnosis methods based on dual-model fusion are widely used. Actually, fault diagnosis needs to response fast, so high efficiency of fusion model requirements is imminent. The fusion model may have complex interrelationships and dependencies between multiple parameters, which impact on diagnostic accuracy. Therefore, a dual-model fusion method based on DBTBoost for fault diagnosis is proposed. First, based on the discrete gradient boosting method, multiple model weak classifiers are constructed, and the results of the classifiers are integrated through the Top-K mechanism to initially construct the dual-model fusion single-objective function. Second, based on the dynamic Bayesian multiparameters optimization method, the multiobjective parameters are adjusted globally to construct the optimal fusion model. Finally, the model performance is verified by model accuracy and efficiency evaluation indexes. The method is practically verified in the fault diagnosis of additive manufacturing (AM) equipment. The experimental results show that the dual-model fusion method based on DBTBoost achieves 99.45% accuracy, with a diagnosis time of 0.65 s and a training time of 503 s. The method improves the accuracy by 4.92% and efficiency by 41.4%.