A novel imputation based predictive algorithm for reducing common cause variation from small and mixed datasets with missing values

Abstract

Most process control algorithms need a predetermined target value as an input for a process variable so that the deviation is observed and minimized. In this paper, a novel machine learning algorithm is proposed that has an ability to not only suggest new target values for both categorical and continuous variables to minimize process output variation but also predict the extent to which the variation can be minimized.In foundry processes, an average rejection rate of 3%–5% within batches of castings produced is considered as acceptable and is considered as an effect of the common cause variation. As a result, the operating range for process input values is often not changed during the root cause analysis. The relevant available historical process data is normally limited with missing values and it combines both categorical and continuous variables (mixed dataset). However, technological advancements manufacturing processes provide opportunities to further refine process inputs in order to minimize undesired variation in process outputs.A new linear regression based algorithm is proposed to achieve lower prediction error in comparison to the commonly used linear factor analysis for mixed data (FAMD) method. This algorithm is further coupled with a novel missing data algorithm to predict the process response values corresponding to a given set of values for process inputs. This enabled the novel imputation based predictive algorithm to quantify the effect of a confirmation trial based on the proposed changes in the operating ranges of one or more process inputs. A set of values for optimal process inputs is generated from operating ranges discovered by a recently proposed quality correlation algorithm (QCA) using a Bootstrap sampling method. The odds ratio, which represents a ratio between the probability of occurrence of desired and undesired process output values, is used to quantify the effect of a confirmation trial.The limitations of the underlying PCA based linear model have been discussed and the future research areas have been identified.