LAUSR

Abstract Fault diagnosis is crucial for the printing quality assurance of a 3D printer. This paper presents a pre-classified reservoir computing (PCRC) method to diagnose the health condition of a 3D printer using the data collected by a low-cost attitude sensor. As the data from the low-cost attitude sensor often contain a large amount of interference information, it is difficult to accurately diagnose the printer condition. As such, a pre-classification strategy is proposed to reduce the intra-class distance by aggregating information labels of the same condition. Echo state network is then employed as a reservoir computing (RC) tool for applying data-driven based artificial intelligence to extract faulty features and to classify condition patterns simultaneously. The proposed PCRC method is evaluated using experimental data collected from the 3D printer. An optimal PCRC model is developed by tuning model parameters using experimental data. The advantages of the PCRC are demonstrated by comparing with other methods such as RC, random forest, support vector machine and spare auto-encoder. Due to the combined and compounded effect of the pre-classification strategy and RC modelling, the proposed method leads to the highest accuracy in fault diagnosis of the 3D printer with limited low-cost sensor data and relatively small datasets, without relying on physical domain knowledge.