Browsing by Subject "Predictive maintenance"

Now showing 1 - 4 of 4

Open Access
A comparison of state-of-the-art machine learning algorithms on fault indication and remaining useful life determination by telemetry data
(IEEE, 2021-11-15) Ünal, Aras Fırat; Kaleli, Ali Yücel; Ummak, Emre; Albayrak, Özlem; Younas, M.; Awan, I.; Unal, P.
Contemporary trends in the diffusion of artificial intelligence technologies has increased the number of studies on predictive maintenance, a recent focus of interest in many industrial domains. Despite the increased interest in the use of machine learning for predictive maintenance, few studies involve thorough comparisons of machine learning algorithms' performance on predictive maintenance applications. This work aims to predict the remaining useful life and machine failures and compares five different algorithms: Random Forest, Gradient Boosted Tree, K-Nearest Neighbors, Multilayer Perceptron and LightGBM. Our results suggest better performances for binary classification using Random Forest, and for regression using LightGRM comnared to other selected algorithms.
Open Access
Early diagnosis of breakdown through transfer learning
(2019-05) Özbek, Seren
Breakdown prediction of equipment is an essential task considering the management of resources and maintenance operations. Early diagnosis systems allow creating alerts on time for taking precautions on production. A significant challenge for diagnosis is to have an insufficient size of data, yet, transfer learning approaches can alleviate such an issue when there is a constrained supply of training data. We intend to improve the reliability of breakdown prediction when there is a limited quantity of training data. We recommend similarity correlation on Remaining Useful Life of these equipment. To do this, we offer learning a common feature space between the target and the source equipment, where we acquire prior knowledge from the source that has different measurements than the target. Within the learned joint feature matrices, we train our model on the vast amount of data of different equipment and finetune it using the data of our target equipment. In this way, we aim to obtain an accurate and reliable model for early breakdown prediction.
Open Access
Simultaneous prediction of remaining-useful-life and failure-likelihood with GRU-based deep networks for predictive maintenance analysis
(IEEE, 2021-08-30) Kaleli, Ali Yücel; Ünal, Aras Fırat; Özer, Sedat
Predictive maintenance (PdM) has been an integral part of large industrial sites collecting data from multiple sensors to reduce the maintenance power and costs with the advent of Industry 4.0. Two of the major problems in PdM used at large industrial sites are: (i) the prediction of remaining useful life (RUL); (ii) the prediction of the likelihood of failing within a predefined time period. While data oriented maintenance predictions were heavily focused on using classical techniques for such problems, recent interest shifted towards utilizing AI based solutions due to the better generalization capabilities of deep solutions. Among the time-sequence based deep networks, RNN, GRU and LSTM based networks are the most frequently used solutions. GRUs have demonstrated their faster learning capabilities with near or better prediction performance on certain tasks already. However, predicting multiple PdM tasks including both RUL and failure detection, simultaneously within the same network in an end to end manner with GRUs has not been much studied in the literature before. In this paper, we introduce a solution to predict those two tasks simultaneously within the same network based on GRUs. In our experiments we compare the performance of GRU layers to LSTM and RNN layers and report their performance on NASA dataset.
Open Access
A workflow for synthetic data generation and predictive maintenance for vibration data
(Molecular Diversity Preservation International (MDPI), 2021-09-22) Selçuk, Şahan Yoruç; Ünal, Perin; Albayrak, Özlem; Jomâa, Moez
Digital twins, virtual representations of real-life physical objects or processes, are becoming widely used in many different industrial sectors. One of the main uses of digital twins is predictive maintenance, and these technologies are being adapted to various new applications and datatypes in many industrial processes. The aim of this study was to propose a methodology to generate synthetic vibration data using a digital twin model and a predictive maintenance workflow, consisting of preprocessing, feature engineering, and classification model training, to classify faulty and healthy vibration data for state estimation. To assess the success of the proposed workflow, the mentioned steps were applied to a publicly available vibration dataset and the synthetic data from the digital twin, using five different state-of-the-art classification algorithms. For several of the classification algorithms, the accuracy result for the classification of healthy and faulty data achieved on the public dataset reached approximately 86%, and on the synthetic data, approximately 98%. These results showed the great potential for the proposed methodology, and future work in the area.