Intelligent Manufacturing of Polyurethane foam accomplishing surface inspection by Image Processing and Artificial Neural Networks techniques

摘要

Machine vision is becoming one of the more efficient emerging technologies for the fast and reliable control of different types of products. This technology allows obtaining a big amount of information, superficial and dimensional, of the pieces at high speed. The objective of the present article is to show a successful machine vision application in an industrial sector where this technology is hardly introduced. Quality assurance is a necessity in the industrial processes. Some of these processes are automated; otherwise, there are still many tasks that are accomplished manually and subjectively by an expert, such as these quality control tasks. The use of this manual control limits a lot the reliability and repetitivity of the results, since the human performance is not constant along the whole working day, and not only the localization of the defect but also its magnitude are relative depending on the person in charge of the identification. The automation of the processes solves these limitations. This happens in the manufacturing process of polyurethane foam. This is a long process throughout two days from the initial moment of pouring the component chemical substances in the mould designed for that, till obtaining the foam block in its final shape. There are some quantifiable parameters to guarantee the quality of the product, such as the hardness, permeability or number of pores per centimeter of the foam, but these are not obtained till the end of the process, and an expert in the foaming cycle only knows them. Inadequate proportions of the chemical constituents in the origin are detected in the end, when there is no possible solution, with the consequent economic loss for the company. Therefore, it is important to be capable of predicting which will be the final result from the earlier stage of the process. So, knowing the evolution of the process, the production variables can be immediately changed if needed. The selected parameter to focus the aim of the project presented in this article, is the number of pores per centimeter calculated over a small sample of foam. As result from the project, a machine vision system has been accomplished and placed in the first stage of the foaming machine. This system includes the expert knowledge, but in a more reliable and repetitive way. It carries out three main block tasks: 1. Acquisition of an image of a foam sample;2. Processing of this image by classical algorithm methods (morphologic, filters) to obtain output values (histogram, initial number of pores); 3.Introduction of the results of the previous phase in an artificial intelligence system based on neural networks. The output of this last will provide with the estimated number of pores that the foam will present in the end. This value will indicate the person in charge of supervision to know what to do next. The results achieved in this first approach are really promising and augur the chance of automating the system in the industrial plant.