Machine Learning-Based Diagnosis of Thermal Barrier Coating Process Quality

Abstract

Based on machine learning algorithms, a method is proposed for quality diagnosis of atmospheric plasma spraying (APS) processes used in thermal barrier coatings with determined coating materials and processes, aiming to swiftly evaluate the quality of APS coatings. First, the three-dimensional morphology of the coating is reconstructed through surface interpolation fitting, employing one-dimensional morphology standards and abnormal training set samples of the plasma-sprayed thermal barrier coating. This algorithm enables the extraction of cross section data of the coating at any angle. The mapping relationship between the characteristic parameters of the Gaussian peak and the process and coating characteristics is thoroughly analyzed, and the 12-dimensional characteristic parameters are utilized to effectively represent the one-dimensional morphology samples. Subsequently, principal component analysis (PCA) and K-nearest neighbor (KNN) algorithms are employed for accurate prediction and classification of the process quality of coating samples. Additionally, an exploratory factor analysis (EFA) model is established to comprehensively depict the relationship between plasma spraying process parameters, the process, and the three-dimensional morphology of the coating. The experimental results show that the machine learning algorithm has high accuracy in quality diagnosis, and its robustness is further verified by K-fold cross-validation. When combined with the EFA model, the proposed method facilitates rapid feedback on process quality, enabling real-time evaluation. Overall, this innovative approach presents a novel solution for the quality diagnosis of atmospheric plasma spraying processes. The incorporation of machine learning techniques and the establishment of the EFA model contribute to enhanced efficiency and accuracy in the evaluation process, paving the way for advancements in thermal barrier coating applications.