Many mechatronic parts are produced using the worse output behavior, a longer melt residence time in the injection molding technology. The product design of today, screw and to melt inhomogeneities. All these influences are not only in the automotive sector, requires highly integrated reflected in quality characteristics of the product. and high performance plastic parts with at the same time low production costs. To achieve this aim modern technologies III ACQUISITION OF PROCESS DATA have to be introduced to track the running production. The benefit of the online quality prediction is mainly based on the A Modelling in the running process reduction of the scrap rate. A deviation in parts dimensions for example is already noticeable while producing the lot and i md corrections can be instantly done to the machine parameter.process an mtelligent strategy for data acquisition and for a subsequent adaptation of the heuristic model must be implemented. These models describe the relationships I INTRODUCTION between process variables and quality features on base of Quality monitoring and qualitycontrolthatardata records obtained within proceeding production.Quality monitoring and quality control that are based on Statistical methods, such as multiple regression, and, more process models are not new in injection molding. Scientific ' ' s institutes all over the world have been doingresearch recently, artificial neural networks have proved to be useful institutes all over the world have been doing research on for the injection molding process [7]. this interesting field in al least the past 15 years [1, 2, 3, 4]. The research in the field of quality management in plastics The plastics injection molding process is exposed to processing is very application-oriented, because of the fact permanent scattering and disruptive disturbances that result that the development of new approaches or methods and the from different sources. To include these real production test in the shop floor are closely connected. The assurance stages the process data must be determined entirely from of the product quality which was not very long time ago, a the "natural" scattering behavior of the process during selling proposition is in our days nothing more than a normal production (including the corrections to necessity. The latest trend show that due to saturation of the manipulated variables that may become necessary during markets and the international competition the prices have to the observation period). The necessary parameterization of be lowered with at the same time demanded highest quality the prediction model can be carried out automatically in the standards.background by evaluating the scattering of the running This is especially critical when producing mechatronic process. The advantages of this method are the non These are in many cases highly integrated parts with disturbing implementation while production is running, and parts. gesetresmat often have te moledaroun furthermore the assessment of the actual series production complex...

All plastics processing companies have to fulfill the objectives of time, cost and quality. Against this background, those producing in high wage countries are especially challenged, because superior part quality is often the only possibility to prevail in competition. Since this leads to high expenses on quality assurance, for some time already efforts have been made to predict the quality of injection molded parts from process data using machine-learning algorithms. However, these did not yet prevail in industry, mainly for two reasons: First, because of the inevitable learning effort that is required to set up a quality prediction model and second, because of the complexity in the application. Current research in the field of transfer learning aiming to shorten learning phases addresses the first challenge. In this paper, we present a holistic approach for the data analysis steps that are necessary once process and quality data have been generated, aiming to minimize the application effort for the operator. This includes the development and application of suitable algorithms for automatic selection of data, process features as well as machine learning algorithms including hyper-parameter optimization and model adaption. Combining the two approaches could bring quality prediction one significant step forward to successful industry application. Beyond this, the presented approach is universally applicable and can therefore be used for other plastics processing methods as well.

No abstract

No abstract

No abstract