Comparison of Intelligent Approaches for Cycle Time Prediction in Injection Moulding of a Medical Device Product

TL;DR

This paper compares artificial neural networks and adaptive neuro-fuzzy systems for predicting cycle time in injection moulding of medical devices, aiming to improve process efficiency and quality.

Contribution

It evaluates the predictive performance of different ANN and ANFIS configurations on real industrial data for the first time in this context.

Findings

ANN performance varies with training methods and neuron count

ANFIS effectiveness depends on membership functions used

Both methods have strengths and limitations for online process optimization

Abstract

Injection moulding is an increasingly automated industrial process, particularly when used for the production of high-value precision components such as polymeric medical devices. In such applications, achieving stringent product quality demands whilst also ensuring a highly efficient process can be challenging. Cycle time is one of the most critical factors which directly affects the throughput rate of the process and hence is a key indicator of process efficiency. In this work, we examine a production data set from a real industrial injection moulding process for manufacture of a high precision medical device. The relationship between the process input variables and the resulting cycle time is mapped with an artificial neural network (ANN) and an adaptive neuro-fuzzy system (ANFIS). The predictive performance of different training methods and neuron numbers in ANN and the impact of model type and the numbers of membership functions in ANFIS has been investigated. The strengths and limitations of the approaches are presented and the further research and development needed to ensure practical on-line use of these methods for dynamic process optimisation in the industrial process are discussed.