Monitoring and Diagnosis of Dimensional Deviation in Assembly Process Using Tensor Regression

Abstract

The monitoring and diagnosis of dimensional deviation is an important means for the continuous improvement of assembly capacity in manufacturing systems. As optical scanning measurements become more common, statistical models based on 3D point cloud data have gained widespread attention. This paper introduces a general tensor regression method that builds a linear regression model between input and output. First, voxel technology is used to transform the 3D point cloud with deviation information into tensor form. Secondly, a Tensor-on-Scalar regression model is established for quality monitoring, and a Tensor-on-Tensor regression model is established for fault diagnosis. Then, tensor decomposition is used to reduce the number of parameters to be estimated in the regression model. Finally, the learning of parameter values is achieved by combining alternating least squares and gradient descent or coordinate descent. A case study of the door inner panel assembly process evaluates the predictive performance of a tensor regression model. The results show that the proposed method outperforms most existing methods in terms of prediction accuracy. Hence, the tensor regression model can achieve high-performance quality monitoring and fault diagnosis.