{"id":2758,"date":"2024-12-11T14:39:05","date_gmt":"2024-12-11T14:39:05","guid":{"rendered":"https:\/\/ptgeof.pl\/?page_id=2758"},"modified":"2024-12-11T14:56:41","modified_gmt":"2024-12-11T14:56:41","slug":"detecting-anomalies-in-numerical-stick-slip-cycles-with-the-unsupervised-algorithm-isolation-forest","status":"publish","type":"page","link":"https:\/\/ptgeof.pl\/?page_id=2758","title":{"rendered":"DETECTING ANOMALIES IN NUMERICAL STICK-SLIP CYCLES WITH THE UNSUPERVISED ALGORITHM ISOLATION FOREST"},"content":{"rendered":"

<\/a><\/a>DETECTING ANOMALIES IN NUMERICAL STICK-SLIP CYCLES WITH THE UNSUPERVISED ALGORITHM ISOLATION FOREST<\/span>
\nWykrywanie anomalii w numerycznych cyklach stick-slip przy pomocy algorytmu nienadzorowanego uczenia maszynowego Isolation Forest
\n<\/span><\/span><\/p>\n

Piotr Klejment<\/span><\/span><\/span>
\n<\/em>Przegl\u0105d Geofizyczny (2024) vol. 69, iss. 3-4, pp. 115-133<\/em><\/span>
\nhttps:\/\/doi.org\/10.32045\/PG-2024-050<\/span><\/p>\n

Streszczenie
\n<\/strong>Nienadzorowane uczenie maszynowe kryje w sobie du\u017cy, ale wci\u0105\u017c nie do ko\u0144ca wykorzystany potencja\u0142. W niniejszej pracy zastosowano algorytm Isolation Forest pochodz\u0105cy z rodziny algorytm\u00f3w nienadzorowanego uczenia maszynowego do wykrywania anomalii w numerycznym modelu cykli stick-slip. Do modelowania numerycznego zastosowano Metod\u0119 Element\u00f3w Dyskretnych, a w modelu uwzgl\u0119dniono obecno\u015b\u0107 warstwy granularnej. W symulacji zastosowano losowy mechanizm doboru czasu mi\u0119dzy kolejnymi wydarzenia typu slip, jak i gromadz\u0105cego si\u0119 stress, w efekcie osi\u0105gaj\u0105c nieregularny, bardziej realistyczny schemat cykli stick-slip. W ka\u017cdym z wydarze\u0144 typu slip dokonano rejestracji parametr\u00f3w opisuj\u0105cych stan uk\u0142adu w danym momencie. Pokazano, \u017ce w spos\u00f3b automatyczny Isolation Forest by\u0142 w stanie wyizolowa\u0107 anomalie dla ka\u017cdego ze slip. Miejscem gromadzenia si\u0119 anomalii by\u0142y warstwy skrajne i warstwa granularna. Potwierdza to istotn\u0105 rol\u0119 warstwy granularnej w charakterystyce cykli stick-slip. Do oceny wp\u0142ywu poszczeg\u00f3lnych parametr\u00f3w na dzia\u0142anie algorytmu zastosowano SHapley Additive exPlanations (SHAP). <\/span><\/p>\n

S\u0142owa kluczowe:<\/strong> cykl stick-slip, nienadzorowane uczenie maszynowe, Metoda Element\u00f3w Dyskretnych<\/span><\/p>\n

Abstract
\n<\/strong>Unsupervised machine learning has a large but still not fully exploited potential. In this work, the Isolation Forest algorithm, which comes from the family of unsupervised machine learning algorithms, was used to detect anomalies in the numerical model of stick-slip cycles. The Discrete Element Method was used for numerical modeling. The model took into account the presence of a granular layer, which is characteristic of mature faults. A random mechanism for selecting the time between subsequent slip events and the critical strength was used in the simulation, resulting in an irregular, more realistic pattern of stick-slip cycles. In each of the slip events, parameters describing the state of the system at a given moment were recorded. It was shown that Isolation Forest was able to automatically isolate anomalies for each slip. The places where the anomalies accumulated were the outer layers and the granular layer. This confirms the important role of the granular layer in the characteristics of stick-slip cycles. SHapley Additive exPlanations (SHAP) was used to assess the impact of individual parameters on the performance of the algorithm.<\/span><\/p>\n

Keywords<\/strong>: stick-slip cycle, unsupervised machine learning, Discrete Element Method<\/span><\/p>\n

<\/p>\n

Article is licensed under <\/span>CC BY 4.0<\/span><\/strong><\/a><\/p>\n

$\"Licencja$ <\/a><\/p>\n","protected":false},"excerpt":{"rendered":"