SIAM Journal on Scientific and Statistical Computing
Trainable method of parametric shape description
Image and Vision Computing - Special issue: BMVC 1991
Active shape models—their training and application
Computer Vision and Image Understanding
A non-linear generalisation of PDMs using polynomial regression
BMVC 94 Proceedings of the conference on British machine vision (vol. 2)
Non-linear point distribution modelling using a multi-layer perceptron
BMVC '95 Proceedings of the 1995 British conference on Machine vision (Vol. 1)
Multivariate locally adaptive density estimation
Computational Statistics & Data Analysis
Adaptive Principal component EXtraction (APEX) and applications
IEEE Transactions on Signal Processing
Probability density estimation from optimally condensed data samples
IEEE Transactions on Pattern Analysis and Machine Intelligence
| Hi-index | 0.00 |
The ability to effectively model dimensional variation of stampings and assemblies is an important tool for manufacturers to investigate, assess and control quality levels of their products. Statistical Process Control (SPC) and Six-Sigma approaches use the assumptions of statistical independence and normally distributed data to create quality process control guidelines which are predominantly used in industry. Multivariate statistical techniques such as Principal Components Analysis (PCA) have been more recently applied to automotive body assembly analysis in order to capture the surface co-linearity present in the dimensional variation of stampings and assemblies. This paper combines the Point Distribution Model, which is based on PCA, and Kernel Density Estimation to provide a statistical shape model (the KDE-PDM) that can deal with high dimensional data sets, represent correlated variation modes, and provide accurate estimates of the underlying shape distribution. Examples from FE simulation and production case studies are presented to highlight the advantages of the KDE-PDM over two other statistical shape models: the univariate shape model, and the original PDM. The KDE-PDM's capabilities make it particularly suited to variation monitoring and diagnosis of high dimensional measurement data sets made available by optical measurement devices, and some suggestions for its implementation are also presented.