Bayesian network approach for ceramic shell deformation fault diagnosis in the investment casting process

Abstract

The dimensional precision of casting is threatened by the deformation of the corresponding ceramic shell in the investment casting (IC) processes. Therefore, the monitoring and root cause diagnosis for the deformation of ceramic shell play an important role in the continuous dimensional quality improvement of the final casting. However, under the condition of small batch production, the data set for root cause diagnosis is small and incomplete, which makes the traditional statistical methods difficult to perform process effectively. This paper applies the approach of systematic Bayesian to solve this problem. Firstly, it is advised to gain the initial structure and parameters through the Bayesian network (BN) mapping method based on process expert knowledge representation. Then, a mutual information approach is proposed to update the structure of the BN. With more accessible measurement date, the conditional probability is updated. Finally, the root causes can be identified with the monitoring signals as evidence. The industrial computerized tomography (CT) method is also creatively advised to measure the inner dimensions of shell. The proposed BN approach and the industrial CT measurement method have a great significance to the dimensional precision improvement for the casting in the IC process.