Data-Driven Predictive Maintenance Approach for Spinning Cyber-Physical Production System

Abstract

The fundamental process of predictive maintenance is prognostics and health management, and it is the tool resulting in the development of many algorithms to predict the remaining useful life of industrial equipment. A new data-driven predictive maintenance and an architectural impulse, based on a regularized deep neural network using predictive analytics, are proposed successfully for ring spinning technology. The paradigm shift in computational infrastructures enormously puts pressure on large-scale linear and non-linear automated assembly systems to eliminate and cut down unscheduled downtime and unexpected stoppages. The sensor network designed for the scheduling process comprises different critical components of the same spinning machine frames containing more than thousands of spindles attached to them. We established a genetic algorithm based on multi-sensor performance assessment and prediction procedure for the spinning system. Results show that it operates with a relatively less amount of training data sets but takes advantage of larger volumes of data. This integrated system aims to prognosticate abnormalities, disturbances, and failures by providing condition-based monitoring for each component, which makes it more accurate to locate the defined component failures in the current spinning spindles by using smart agents during the operations through the neural sensing network. A case study has provided to demonstrate the feasibility of the proposed predictive model for highly dynamic, high-speed textile spinning system through real-time data sensing and signal processing via the industrial Internet of Things.