At SMART2023 on 4th July 2023, Dr. George Margetis, Senior RTD Engineer at Foundation for Research and Technology – Hellas (FORTH), Institute of Computer Science (ICS) presented the paper An elevator calibration recommender system for effective defect detection and prevention.
Authors: George Margetis, Nikolaos Dimitriou, Elpiniki Papageorgiou, Theodosios Theodosiou, Despoina Gavgiotaki, Konstantinos C. Apostolakis, Stavroula Ntoa, Dimitrios Tzovaras, Constantine Stephanidis
Brief description: Machine Learning and Recommendation Systems have had a significant impact on the manufacturing industry. A recommendation system is a class of machine learning that recommends items from a knowledge base utilising data filtering and analysis. It is designed to assist employees in making decisions, by improving their capacity to recognise the optimal option from a variety of alternatives. This work describes a Deep Neural Network model, trained using Kleemann’s dataset of the lift hydraulic press calibration unit test over a two-year period. The proposed model aims to provide the company’s operators with predictions about target parameters in terms of speed and pressure, during the hydraulic press calibration process, in order not to exceed a maximum level of noise.
This study was conducted in the context of OPTIMAI’s use case 2. The main objective of the implemented recommendation service is to help operators calibrate a lift to quickly obtain the optimal values in terms of pressure and speed that they regulate. The aim is to offer recommendations on whether to increase or decrease the pressure and/or speed to achieve the accurate noise level.
Findings: The results of the recommendation service demonstrate its effectiveness in predicting calibration parameters. “The results are very good. We have achieved strong linear correlations between predicted and actual values”, says Dr. Margetis.
Future work: Future work will focus on further improving pressure prediction using alternative models or data preprocessing methods. Future work will also consider additional factors, such as time of day or weather, that could enhance precision. A further consideration will be the integration of AI accountability measures, such as using blockchain.
Interested in finding out more? The paper will be published as an open access publication in the SMART 2023 conference proceedings.