Digital Machine and Process Monitoring of Industrial Vibrating Screens

This thesis contributes to developing an online monitoring system for industrial vibrating screens, known as the ¿I-STEP Workbench,¿ currently under development by the project's industrial partners. The research addresses two key questions: (A) How can vibration sensor data be used to provide novel insights into the screening process and increase its efficiency? (B) How can this data be utilized to detect screen failures in order to minimize unplanned downtimes? To answer these questions, the study integrates controlled laboratory experiments with two industrial case studies, where vibration sensors were installed on vibrating screens, data was collected over one year, and machine learning models were trained to predict process parameters. In Case Study I, a linear motion vibrating screen used for dewatering in a scrap processing plant, a machine learning model classified normal operating states with an accuracy of 88%, revealing that screen cleaning accounted for 10% of operational time, significantly affecting efficiency. Additionally, threshold-based methods effectively detected typical screen failures, including a loose screen deck, particle clogging, and bearing damage in the unbalanced drive. In Case Study II, a circular motion vibrating screen was used to classify angular gravel into aggregates of five product classes. A machine learning model predicted the volume feed flow rate with a precision of 90.6%. While no abnormal screen states were recorded, consistent particle size distribution and bulk density of the feed enabled reliable conversion of predicted volume flow rates into mass flow rates, with an average deviation of 0.15 t/h over eight consecutive operational days. Beyond model development, this study synthesizes insights from experimental and industrial data. The findings demonstrate that online vibration data from industrial vibrating screens can provide critical machine condition insights, detect failures, and enable the development of machine learning models for condition classification and feed flow rate prediction. This approach enhances the efficiency and reliability of vibrating screens, contributing to reduced downtime and optimized operation.