Prognostics and Health Management (PHM) is getting more interest in the manufacturing field, since machine tools and equipment maintenance and failure costs represent a conspicuous fraction of the tool cost. Prognostics and Health Management, tries to reduce these costs as much as possible. The aim of this thesis work is to develop a hybrid prognostics solution for the adaptive prediction of tool wear in turning applications. The hybrid framework exploits the particle filter (statistical world) to update the structure of Multi-Layer Perceptrons (data-driven world). Such an approach, lowered the weaknesses and improved the robustness of the separated solutions. The proposed methodology allowed to adapt the Remaining Useful Life prediction of a turning insert to different operating regime conditions, thus different degradation rates of tools. The approach uses a really limited knowledge for training a single run to failure performed at 190 m/min (cutting speed) and 0,2 mm/rev (feed) was used as training datum. The solution was capable to adapt to unseen working conditions. The solution was tested on 9 run-to-failures with cutting speeds between 160 and 220 m/min and feed between 0,1 and 0,3 mm/rev. The solution was based on direct measurements of the tool flank wear. After applying a series of RUL predictions tests, absolute error percentages were observed. The maximum error, 57,1429% has been found in test number 8. Test 8 is the critical test with a combination of lower speed and lower feed rate.
Prognostics and Health Management (PHM) sta riscuotendo maggiore interesse nel campo della produzione, poiché i costi di manutenzione e guasto di macchine utensili e attrezzature rappresentano una frazione cospicuo del costo degli utensili. Prognostics and Health Management, cerca di ridurre il più possibile questi costi. Lo scopo di questo lavoro di tesi è sviluppare una soluzione prognostica ibrida per la previsione adattiva dell'usura degli utensili nelle applicazioni di tornitura. Il framework ibrido sfrutta il filtro antiparticolato (mondo statistico) per aggiornare la struttura dei Multi-Layer Perceptrons (mondo guidato dai dati). Tale approccio ha abbassato i punti deboli e migliorato la robustezza delle soluzioni separate. La metodologia proposta ha permesso di adattare il Remaining Previsione della vita utile di un inserto di tornitura a diverse condizioni di regime operativo, quindi diversi tassi di degradazione degli utensili. L'approccio utilizza una conoscenza davvero limitata per l'allenamento di una singola corsa al cedimento eseguito a 190 m/min (velocità di taglio) e 0,2 mm/giro (avanzamento) è stato utilizzato come dato di addestramento. La soluzione era in grado di adattarsi a condizioni di lavoro invisibili. La soluzione è stata testata su 9 run-to-failure con velocità di taglio comprese tra 160 e 220 m/min e avanzamento tra 0,1 e 0,3 mm/giro. La soluzione si basava su misurazioni dirette dell'usura sul fianco dell'utensile. Dopo aver applicato una serie di test di previsione RUL, sono state osservate percentuali di errore assoluto. L'errore massimo, 57,1429%, è stato riscontrato nel test numero 8. Il test 8 è il test critico con una combinazione di velocità e velocità di avanzamento inferiori.
Hybrid adaptive prognosis of tool wear in turning
Karakoc, Aykut
2020/2021
Abstract
Prognostics and Health Management (PHM) is getting more interest in the manufacturing field, since machine tools and equipment maintenance and failure costs represent a conspicuous fraction of the tool cost. Prognostics and Health Management, tries to reduce these costs as much as possible. The aim of this thesis work is to develop a hybrid prognostics solution for the adaptive prediction of tool wear in turning applications. The hybrid framework exploits the particle filter (statistical world) to update the structure of Multi-Layer Perceptrons (data-driven world). Such an approach, lowered the weaknesses and improved the robustness of the separated solutions. The proposed methodology allowed to adapt the Remaining Useful Life prediction of a turning insert to different operating regime conditions, thus different degradation rates of tools. The approach uses a really limited knowledge for training a single run to failure performed at 190 m/min (cutting speed) and 0,2 mm/rev (feed) was used as training datum. The solution was capable to adapt to unseen working conditions. The solution was tested on 9 run-to-failures with cutting speeds between 160 and 220 m/min and feed between 0,1 and 0,3 mm/rev. The solution was based on direct measurements of the tool flank wear. After applying a series of RUL predictions tests, absolute error percentages were observed. The maximum error, 57,1429% has been found in test number 8. Test 8 is the critical test with a combination of lower speed and lower feed rate.| File | Dimensione | Formato | |
|---|---|---|---|
|
Thesis_Aykut_Karakoc.pdf
non accessibile
Dimensione
6.01 MB
Formato
Adobe PDF
|
6.01 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/181894