Tesi etd-11172021-023540 |
Link copiato negli appunti
Tipo di tesi
Tesi di laurea magistrale
Autore
LEONE, GABRIELE
URN
etd-11172021-023540
Titolo
Multivariate Sensor Data Analysis for a Nonwoven Industry: An Unsupervised Approach
Dipartimento
INFORMATICA
Corso di studi
DATA SCIENCE AND BUSINESS INFORMATICS
Relatori
relatore Guidotti, Riccardo
Parole chiave
- clustering
- data mining
- discords
- machine learning
- motifs
- multivariate time series
- outlier detection
- unsupervised learning
Data inizio appello
03/12/2021
Consultabilità
Non consultabile
Data di rilascio
03/12/2091
Riassunto
Il lavoro di tesi si fonda sull'Unsupervised Learning. I dati su cui si sofferma l'analisi sono stati raccolti tramite più di 100 sensori installati sui macchinari di due siti produttivi che rilasciano il materiale grezzo usato per ottenere prodotti finali di nonwoven. Nel corso degli esperimenti sono state applicate diverse tecniche di clustering con focus su Toeplitz Inverse Covariance-Based Clustering e Bisecting K-Means. Tra gli aspetti sperimentati figurano l'Outlier Detection, il Motifs and Discords discovery e un approccio supervisionato basato sui cluster. Tramite queste analisi è stato possibile individuare diversi stati produttivi, riconoscere le anomalie, dedurre che i due siti produttivi analizzati si occupano di produzioni diverse.
This thesis is based on the Unsupervised Learning. Data is collected through more than 100 sensors installed on two machines belonging to two different sites, which relase the raw material used to obtain nonwoven final products. During the experiments different clustering techniques have been applied, with a focus on Toeplitz Inverse Covariance-Based Clustering and Bisecting K-Means. Other experimented areas have been Outlier Detection, Motifs an Discords Discovery plus a Supervised approach based on cluster labels. The analyses have made it possible to understand different production states, to find anomalies and to understand that the two sites produce different final products.
This thesis is based on the Unsupervised Learning. Data is collected through more than 100 sensors installed on two machines belonging to two different sites, which relase the raw material used to obtain nonwoven final products. During the experiments different clustering techniques have been applied, with a focus on Toeplitz Inverse Covariance-Based Clustering and Bisecting K-Means. Other experimented areas have been Outlier Detection, Motifs an Discords Discovery plus a Supervised approach based on cluster labels. The analyses have made it possible to understand different production states, to find anomalies and to understand that the two sites produce different final products.
File
Nome file | Dimensione |
---|---|
Tesi non consultabile. |