Smart manufacturing with transfer learning under limited data: Towards Data-Driven Intelligences

Abstract

Advanced cyber-physical production methods, often known as smart manufacturing or Industry 4.0, keep benefiting from the combination of cutting-edge artificial intelligence (AI) techniques with sophisticated manufacturing techniques. Yet, the degree to which the deployed predictive models can manage uncertainties and production data changes in the factory over time strongly influences such systems' “smartness” level. Traditional machine learning (ML) approaches often tend to perform badly in situations of configuration modification in a production process without enough fresh data. Conventional machine learning-based predictive modeling methods require large volumes of training data; however, accumulating large amounts of training data in real-world applications such as manufacturing is labor-intensive and costly, especially for small and medium enterprises. Consequently, limited data is a prevalent obstacle for ML techniques. The present study employs a transfer learning (TL) framework to address this issue. Due to its numerous benefits, electrical Discharge Machining (EDM) has been selected as an illustrative case. This study throws light on developing a new prediction model using TL to predict the material removal rate (MRR) of the electrical discharge machining (EDM) process. The input parameters are pulse on time, current, and voltage. The developed model is based on Deep Neural Network (DNN). The proposed method overcomes the limitations of traditional machine learning (ML) models, which require many experimental datasets for accurately predicting the responses. The results show that TL can be used to overcome the issue of limited data in the manufacturing process.