Model Selection for Predictive Quality in Hydraulic Testing

Abstract

Manufacturing companies are confronted with enormous challenges such as increasing product complexity, shorter product life cycles and growing product diversity. Politically and socially, increased demands regarding sustainability and resource consumption are streaming into the focus of companies. One solution strategy to increase the productivity of existing production systems while ensuring existing quality standards is the application of data-driven analytical methods such as machine learning.Due to the frequent changes in production conditions, the analysis of real manufacturing data is linked to sophisticated data pre-processing. Changes in production data are manifested in trends and systematic shifts over time. Data pre-processing includes rule-based data cleaning, the application of dimension reduction techniques and the identification of comparable data subsets. Within the used dataset of hydraulic valves by Bosch, the comparability of the same production conditions in the manufacturing of hydraulic valves can be identified within certain periods. Machine learning methods can process large amounts of data, unfavorable row-column ratios and discover dependencies between the input data and the specified target variable as well as evaluate the multidimensional influence of all input variables on the target variable. For use cases in manufacturing, neural networks, support vector machines and tree-based methods have so far proved to be very successful.The use of cross-process production data along the value chain of hydraulic valves is a promising approach to predict the quality characteristics of workpieces. Within this research, machine learning methods with deep and shallow structures are applied to predict the internal leakage of hydraulic valves based on geometric gauge blocks from machining, mating data from assembly and hydraulic measurement data from end-of-line testing. Moreover, the most suitable methods are selected, and accurate quality predictions are obtained.