Using Transfer Learning and Radial Basis Function Deep Neural Network Feature Extraction to Upgrade Existing Product Fault Detection Systems for Industry 4.0: A Case Study of a Spring Factory

Abstract

In the era of Industry 3.0, product fault detection systems became important auxiliary systems for factories. These systems efficiently monitor product quality, and as such, substantial amounts of capital were invested in their development. However, with the arrival of Industry 4.0, high-volume low-mix production modes are gradually being replaced by low-volume high-mix production modes, reducing the applicability of existing systems. The extent of investment has prompted factories to seek upgrades to tailor existing systems to suit new production modes. In this paper, we propose an approach to upgrading based on the concept of transfer learning. The key elements are (1) using a framework with a basic model and an add-on model rather than fine-tuning parameters and (2) designing a radial basis function deep neural network (RBF-DNN) to extract important features to construct the basic and add-on models. The effectiveness of the proposed approach is verified using real-world data from a spring factory.

Magnetic Characterization of MR Fluid by Means of Neural Networks

“…The coefficients c i , σ i , µ i can be calculated by using the pseudo-inverse operator, but this implies numerical instabilities. Thus, to reduce the computations required, the mathematical expression µ r = f (B) is calculated by means of a radial basis function neural network [47][48][49].…”