The functionality of manufactured components is intricately linked to their surface topography, a characteristic profoundly shaped by the fabrication process. Repeatable quantitative characterization of surfaces is essential for detecting variations, defects, and predicting performance. However, the plethora of surface descriptors presents challenges in optimal selection of the correct assessment metric. This work addresses two of these aspects: automatic selection of surface descriptors for classification and an application-specific approach targeting scan path strategies in laser-based powder bed fusion (LPBF) additive manufacturing.
A framework, titled Surface Quality and Inspection Descriptors (SQuID), was developed and shown to provide an effective systematic approach for identifying surface descriptions capable of classifying textures based on process or user-defined differences. Using a form of univariate analysis rooted in signal detection theory, the predictive capability of a discriminability value, d', is demonstrated in the classification of mutually exclusive surface states. A discrimination matrix that offers a robust feature selection algorithm for multiclass classification challenges is also introduced. The generality of the approach is validated on two datasets. The first is the open-source Northeastern University dataset consisting of intensity images from six different surface classes commonly found in cold-rolled steel strip operations. The application of signal detection theory's measure, d', proved successful in quantifying a texture parameter's ability to discriminate between surfaces, even amidst violations of normality and equal variance assumptions regarding the data.
To further validate the approach, SQuID is leveraged to classify different grades of surface finish appearances. ISO 25178-2 areal surface metrics extracted from bandpass filtered measurements of a set of ten visual smoothness standards obtained from low magnification coherent scanning interferometry are used to quantify different grades of powder-coated surface finish. The highest classification accuracy is achieved using only five multi-scale descriptions of the surface determined by the SQuID selection algorithm. In this case, spatial and hybrid parameters were selected over commonly prescribed height parameters such as Sa, which proved ineffective in characterizing differences between the surface grades.
Expanding surface metrology capabilities into LPBF additive manufacturing, additional studies developed a methodology to comprehend the relationship between scanning strategies, interlayer residual heat effects, and atypical surface topography formation. Using a single process-informed surface measurement, a critical cooling constant is derived to link surface topography signatures directly to process conditions that can be calculated before part fabrication. Twelve samples were manufactured and measured to validate the approach. Results indicate that the methodology enables accurate isolation of areas within the parts known to elicit heterogeneity in microstructure and surface topography due to overheating. This approach provides not only a new surface measurement technique but also a scalable parameterization of LPBF scan strategies to quantify track-to-track process conditions. The methodology demonstrates a powerful application of surface texture metrology to characterize LPBF surface quality and predict process outcomes.
Overall, this thesis contributes a systematic approach for identifying discriminatory parameters for surface classification and a novel process-informed surface measurement for predicting track-scale overheating during LPBF-AM of a nickel superalloy.