Session: 03-20-05: Manufacturing: General

Paper Number: 139929

139929 - Optimizing Laser Cutting Parameters for Austenitic Stainless Steel: Insights From Gaussian Process Regression and Sensitivity Analysis. 

Laser cutting is a fundamental subtractive manufacturing process carried out and applied in various industries such as healthcare, manufacturing, and semiconductors. It involves the use of lasers to cut metals to obtain the desired product. This process can be executed through both experimental and computational means. However, the challenge in performing these techniques lies in knowing which input parameters (laser speed, gas type, laser power, and laser focus distance) has a greater influence on the output variable during this process. This study aims to construct a predictive model using gaussian process regression and rank the important cutting parameters for austenitic stainless steel via sensitivity analysis. We want to explore how varying the input parameters impacts the output, given data obtained from the experiments and simulations. The output of interest is kerf width, heat affected zones and surface roughness. We employ two primary methodologies for this study. Firstly, the Gaussian process regression technique is utilized. This method, a nonparametric Bayesian approach, is particularly adept at handling small datasets and can furnish uncertainty measurements for output predictions. This technique also helps in estimating the probability distribution across all admissible functions fitting the data. The radial-based function kernel and marten kernel are chosen as covariance kernel functions for their smoothness. Secondly, the sensitivity analysis approach. This uses the variance-based method to assess the significance of input factors by analyzing their respective contributions to the overall variance of the output quantity of interest. This comprehensive approach involves evaluating both first-order indices and total effect indices to gain a nuanced understanding of the importance of individual variables and their interactions in shaping the output variance. First-order indices provide insights into the direct influence of each input variable on the output variance. By isolating the effect of one variable while keeping others constant, these indices help in quantifying the impact of each factor individually. This allows researchers to prioritize variables based on their individual contributions to the overall variability of the output. In contrast, total effect indices offer a broader perspective by considering not only the direct effects but also the interactions between input variables. They capture the combined influence of a variable along with its interactions with other factors on the output variance. This holistic approach enables a deep understanding of how different variables interact and contribute actively to the variability observed in the output. The datasets used will undergo training and testing based on statistical and machine-learning models in Python to perform the predictive task. From our preliminary findings, laser power has an 85% influence on the laser cutting process, followed by laser speed, 10%, and laser focus distance has less than 5% cut influence. Overall, laser power emerges as the most significant input parameter, followed by speed, and focus distance. This methodology and analysis will identify the key parameters and interactions influencing laser cutting processes.

Presenting Author: Asonganyi Atayo Wichita State University

Presenting Author Biography: I am a Ph.D. candidate and graduate research assistant at Wichita State University pursuing mechanical engineering. My background includes a B.S. and M.S. in mechanical engineering from WSU in 2018 and 2020. Through my doctoral work, I have specialized in numerical investigations of manufacturing processes, leveraging uncertainty quantification to enable robust predictions in laser cutting models. Currently, I am expanding my research on thermal modeling and management of electrochemical energy storage, focusing on supercapacitors and lithium-ion batteries. To enhance predictive capabilities in these systems, I employ deep learning techniques. My diverse expertise in thermal modeling, manufacturing processes, electrochemical energy storage, deep learning and machine learning, provides a unique foundation to advance energy storage and manufacturing technologies through data-driven modeling.

Authors:

Asonganyi Atayo Wichita State University
Talha Khan Wichita State University
Rajeev Nair Wichita State University
Muhammad Rahman Air Force Institute of Technology