The automatic detection and rectification of surface and aesthetic defects in the production of wooden panels

Abstract

The trend in production of furniture, flooring, housing parts and similar is to repair and upgrade defective feedstock rather than scrap it. The corresponding task of rectification is mainly based on manual labour with a clearly visible need for automation to reduce costs. This thesis shows that it is possible to automatically carry out defect rectification on softwood panels by first detecting unwanted defects of various kinds, second assessing them correctly also by pre-determined aesthetic aspects and thirdly generating instructions for correspondingly an aesthetically acceptable repair.

A novel approach based on pixel-wise registered multi-dimensional images, cascaded unsupervised and supervised learning and an expert system based on a fuzzy knowledge base has been tested. It is shown that automated patching under aesthetic aspects can be achieved by modelling the human wood worker’s implicit and explicit knowledge. Support Vector Machines (SVMs) are able to deal with the high dimensional registered image data and the associated non-linear classification problem that addresses the local aesthetics without the need for feature engineering. An expert system generates rectification instructions for the detected defects with respect to the final panel appearance and acts as a user interface to adjust the process. Satisfactory results in terms of aesthetically acceptable panels of Nordic Spruce patched with different types of solid and liquid fillers have been achieved.

The feasibility of machines being able to assess, preserve, modify or create aesthetics is demonstrated for the first time on wooden panels. The application in a productive, industrial environment has successfully been shown, therefore filling a gap in the automation of wooden panel production.

Date of Award	2016
Original language	English
Awarding Institution	Coventry University
Sponsors	Baumer Gmbh
Supervisor	Raymond Jones (Supervisor), John Owen (Supervisor) & James Shuttleworth (Supervisor)