Power Consumption Prediction in Smart Manufacturing Using Machine Learning

Abstract

In industrial environments, real-time fault detection is essential for minimizing equipment failures, safeguarding personnel, and optimizing operational efficiency. This thesis examines the application of machine learning (ML) regression models to estimate power consumption during the critical roll testing phase of paper machine operations. The estimated power trend is displayed alongside the actual consumption, enabling operators to make rapid decisions when significant deviations occur which can potentially prevent hazardous events such as fires and mitigating the risk of asset loss and financial damage.

The research was motivated by an increasing demand for intelligent monitoring systems capable of enhancing fault detection capabilities. Leveraging key operational parameters including speed, temperature, and pressure, the study investigates the effectiveness of ML regression models in estimating power consumption in real time and identifying anomalous patterns indicative of system faults.

This investigation is guided by three primary research questions: the accuracy of machine learning-based power estimation, the comparative performance of different regression algorithms based on quantitative, qualitative, and business defined measures, and their computational suitability for real-time deployment. By integrating predictive analytics into the roll-testing workflow, this thesis demonstrates how intelligent systems can complement traditional monitoring approaches, reduce reliance on manual oversight, and enhance both safety and operational resilience. The findings contribute to the broader initiative of digital transformation in industrial environments, offering actionable insights into data-driven fault detection methodologies.