Browsing by Author "Rautenbach, Eric"
Now showing 1 - 1 of 1
Results Per Page
Sort Options
- ItemFramework for process improvement in manufacturing of metal packaging(Stellenbosch : Stellenbosch University, 2022-04) Rautenbach, Eric; Dirkse Van Schalkwyk, T.; Stellenbosch University. Faculty of Engineering. Dept. of Industrial Engineering.ENGLISH ABSTRACT: Due to increased competitiveness in the packaging industry, process improvement is important to give businesses an edge over their competition. This thesis represents a study of the application of machine learning for process improvement in metal can manufacturing. A five step process improve ment framework based on the Six Sigma process improvement methodology and the CRISP-DM data science framework was developed. The framework consisted of different steps that included steps used in the Six Sigma process improvement methodologies as well as steps used in data science processes.The five steps were; Define, Understand, Model, Evaluate and Deploy (DUMED). The DUMED framework was used in a case study that predicted the axial load resistance of 2-piece metal food cans during the manufacturing process. The objective is to understand how axial load resistance relates to other factors in the process with the outcome that any changes made in the process will still deliver cans with suitable axial load resistance. A predictive model on axial load resistance will give enhanced capability to control axial load resistance, and will lead to less rejections and therefore less waste. A predictive model on axial load resistance can also supply valuable information on the possible viability for light weighting of material, which will have a decreased cost of raw material as a result and therefore hold financial benefit for the manufacturer. Various data science and machine learning principles were applied during the study related to data understanding, data assessing, data preparation, data modelling and model assessing. The framework was successfully applied in the case study, with the exception of the fifth step, deployment. The deployment phase will be dependent on further improvement of the predictive model. Machine learning was successfully used in the case study to develop a predictive model; the axial load resistance could be predicted within 2.3% of the actual values. The best results were obtained from using feature selected data obtained from a random forest feature selection algorithm that was modelled by using a gradient boost ensemble regression model. Machine learning was successfully applied to a metal package manufacturing line to predict quality characteristics of the final product and possibly bring about process improvement.