Rubber injection molding: Applying multivariate statistics to identify quality issues solely from process signals

Injection‐molded rubber parts are widely used in automotive, aeronautical, and industrial engineering applications Therefore, such rubber parts are often critical to the safe operation of the entire system, and part failure can result in significant human or environmental damage. To avoid shipping any parts of subpar quality, manufacturers need to continuously monitor the quality of their product. xIn this work, we apply a principal component analysis (PCA) based process monitoring method. This method is able to detect process fluctuations (faults) in real‐time solely from sensor data features, only requiring pretraining on data from about 10 in‐control cycles. Specific faults were set to critically affect the dynamic performance of the manufactured NBR rubber parts. Fisher discriminant analysis (FDA) was employed to automatically cluster individual molding cycles into those of being in control, those of defectives caused by unfavorable raw material storage and those of out‐of‐tolerance induced by an overheated mold, again solely from sensor data. Both PCA fault detection and FDA fault identification decisions were validated by oscillatory rheology and dynamic compression testing of the manufactured parts. This combined method approach is scalable, transferable, and can be implemented on standard industrial injection molding equipment.