Management of Early Failure Detection of Production Process: The Case of the Clutch Shaft Alignment using LSTM Deep Learning Algorithm
Artykuł w czasopiśmie
MNiSW
100
Lista 2021
| Status: | |
| Autorzy: | Przysucha Bartosz, Rymarczyk Tomasz, Wójcik Dariusz, Kowalski Marcin, Białek Ryszard |
| Dyscypliny: | |
| Aby zobaczyć szczegóły należy się zalogować. | |
| Rok wydania: | 2021 |
| Wersja dokumentu: | Elektroniczna |
| Język: | angielski |
| Numer czasopisma: | Special Issue 2 |
| Wolumen/Tom: | 24 |
| Strony: | 189 - 197 |
| Efekt badań statutowych | NIE |
| Materiał konferencyjny: | NIE |
| Publikacja OA: | TAK |
| Licencja: | |
| Sposób udostępnienia: | Otwarte czasopismo |
| Wersja tekstu: | Ostateczna wersja opublikowana |
| Czas opublikowania: | W momencie opublikowania |
| Data opublikowania w OA: | 12 czerwca 2021 |
| Abstrakty: | angielski |
| Purpose: In this paper the neural networks model based on long short-term memory (LSTM) for early failure detection of the clutch shaft alignment system is developed. This issue is of particular importance when assessing the condition of the tool and predicting its durability, which are keys to the reliability and quality of the production process. Design/Methodology/Approach: Based on real fault data of the measuring system, 500 clutch fault runs were simulated. Then, the time of failure was modelled with two neural networks, the conventional neural network of the ANN and the LSTM deep learning network. The study examined and compared the effectiveness and quality of both networks in the context of fault prediction. Practical Implications: In vibroacoustic diagnostics, we often deal with machines operating in various conditions, which makes it difficult to diagnose them using standard methods. In such cases, spectral methods require analysis of frequency bands, which may contain other components in addition to information about the diagnosed parameter. The algorithm for predicting impending failure gives the possibility to monitor the current degradation status of the device. This makes it possible to streamline planning processes in the areas of inspection, preventive replacement of parts, warranty, service, or storage of spare parts. Findings: The objective of the paper is to introduce an improved computational method for failure detection based on a deep learning algorithm. It was proven that LSTM networks are suitable for successfully solving this scope of tasks. Originality/Value: The research showed that the proposed LSTM algorithm is more effective and accurate than conventional artificial neural networks (ANN) based on the multilayer perceptron model. |
