DEVELOPMENT OF FUZZY-NEURO APPROACH BASED INTELLIGENT SYSTEMS FOR PARAMETER RESETTING OF INJECTION MOULDING

Abstract

Plastic injection moulding is a commonly used manufacturing process due to its mass production capacity and relatively low production cost per part. However, process parameter setting for moulding defect correction is a complex and difficult task since a large number of factors ranging from part design to moulding process control are involved. The current practice in moulding industry is to adjust the parameters manually based on the products' defects in test-runs through trial and error. This process, however, requires enormous experience and is often time consuming. In this thesis, the development of a fuzzy-neuro approach for injection moulding process parameter resetting is presented. The proposed approach employs a hybrid paradigm of fuzzy sets and neural networks which is able to achieve the complex mapping between the moulding defects and the process parameter re-setting, i.e. { defects }?{parameter adjusters}. Based on this approach, two intelligent systems for parameter resetting are developed. One is called rule-based system as it is developed using the rules collected from expert knowledge and experience. In this system, seven commonly encountered moulding defects and two moulded part geometry variables are used as system input; nine process parameter adjusters are the system output. Both the input and output are described using fuzzy terms. The other system developed is called observation data-based system as it is developed using the rules from observed data collected from real moulding test-runs. It is an attempt to achieve an acceptable setting in one resetting cycle. In this system, five defects and three part geometry variables are used as system input described using fuzzy terms, and 20 process parameter adjusters on an injection machine are used as output in crisp values. Both the systems are designed to be used in the test-run of injection moulding and are able to predict the exact amount to be adjusted for each parameter automatically towards reducing or eliminating the observed defects. Testing in real cases showed that the two systems produced satisfying results and have their own particular characteristics and advantages.