Experimental investigation of process parameters for pack boronizing of SS410 using anova and machine learning approaches

Abstract

In recent years, applications of Machine Learning and Artificial Intelligence are gaining momentum to the production researchers to analyze the complex interdependencies present in the production dataset. The manufacturers have started to incorporate machine learning approaches to the production process & predictive algorithms to fine-tune the quality of the product. The objective of the proposed work is to apply classification and regression algorithms to analyze the input process parameters for the pack boronizing process of SS410. To prepare the dataset, 9 experiments were carried out and the test specimens having ø 55 mm & thickness of 10 mm are pack boronized using the boronizing agent 325 mesh size. This process is carried out in 4.5 kW 'INDFURR' electric furnace with varying input parameters of temperature, time and gas pressure. The output parameters are boronizing thickness and microvickers hardness. The SEM and optical microscopic images of the specimen confirm the formation of the boronizing layer. To find the influence parameters, it is analyzed using ANOVA and Decision Tree algorithm. Both the techniques confirmed that time as the most significant parameter for boronizing thickness. For surface hardness, time & temperature are the major influencing parameters. Various regression models from machine learning were formulated to find the relationship between variables. Among these models, multilayer perception produced maximum correlation co-efficient & minimum root mean square error.

Investigation of surface roughness and material removal rate of WEDM of SS304 using ANOVA and regression models

“…Balasundaram et al [22] proposed a step-by-step technique for using the decision tree algorithm to solve a twomachine flow shop scheduling problem. Ramakrishnan and Balasundaram [8] applied regression models to the boronizing process of SS310. From their results, it was concluded that multilayer perception had the maximum correlation coefficient and it also has a minimum root mean square error.…”

“…Design of experiment (DOE) is widely used as a statistical tool in various domains of research, particularly to aid the design, development, and optimization of processes [8]. A set of methods based on statistics are used in design of experiments to identify and quantify relationships between variables and outputs.…”

Study on Optimization of WEDM Process Parameters on Stainless Steel

“…The analysis is carried out by several authors utilising analytical and statistical approaches. The MRR, kerf width, and surface roughness are S/N ratiobased metrics of process performance [28].…”

“…According to a report on WEDM, the SR increases if T on , V, I and reduces if T off increases [19]. The various L9 and L27 orthogonal arrays 4 Journal of Nanomaterials of Taguchi approaches are used on the austenitic stainless steels for finding the best processing inputs and the optimal values for optimization technique like GA [27,28]. The genetic algorithm is adopted for cutting the SS347 using laser beam machining, and the parametric optimization is adopted to find the best model [57].…”

Surface modification method of duplex type stainless steels by the pack boriding process

“…Thus, to find the wear resistance, in this work the effects of abrasive wear was investigated. The abrasion loss was calculated from the difference between the final weight after the wear test and the initial weight before the test [27] and the results are shown in Figure 8.…”