An integrated CRN-SVR approach for the quality consistency improvement in a diesel engine assembly process

Abstract

ABSTRACT As the last production link, the diesel engine assembly process (DEAP) significantly impacts the quality consistency of diesel engine products. Therefore, the quality consistency improvement of DEAP has become an urgent problem for academia and industry. The challenge is how to mine the causal relationship in DEAP and establish a reliable quality prediction model. This paper attempts to describe DEAP using a causal relationship network (CRN) and to provide an effective data-based scheme for improving quality consistency by integrating CRN with support vector regression (SVR). First, a two-step CRN learning method is proposed for describing the DEAP. In the first step, the association relationship network is developed by a hybrid direct association detection method of the maximal information coefficient and network deconvolution, which can accurately measure the data relations. In the second step, the information geometric causal inference is employed to determine the direction of the edges in the association relationship network, thus forming the CRN of DEAP. Then, an integrated CRN-SVR approach is proposed to realize the predictive modeling of the critical quality indicators in DEAP, which integrated SVR into CRN. At the same time, it also provides a feasible idea for the interpretability of existing machine learning techniques. Finally, the proposed approach is tested and verified in a real-world DEAP and the obtained RMSE is only 0.033. The results of this study provide theoretical support and technical guarantee for quality consistency improvement in DEAP.