University of Twente Student Theses

Login

Automated Electronic Component Selection : a Machine Learning approach

Rizvi, Mohammad Abbas (2022) Automated Electronic Component Selection : a Machine Learning approach.

[img] PDF
1MB
Abstract:The research work in the study is conducted at Signify N.V., a multinational Lighting corporation providing lighting products and solutions to the consumer all over the globe. Signify is seeking ways to improve electronic component recommendation process, with the aim to reduce process complexity and product delivery time. The aim of this research is to automate the component selection process by developing a model to predict the best electronic component to be used in the design by designer/engineer based on their technical parameter requirement. The model should recommend the optimal component considering technical as well as commercial aspect of the component. A Hybrid supervised-unsupervised model approach is investigated for predicting the best electronic component. The Cross Industry Standard Process for Data Mining (CRISP-DM) is used across the study. Clustering techniques such as Hierarchical agglomerative, BIRCH, DBSCAN, OPTICS and Mean Shift are used to group similar components. After clustering, supervised learning algorithms such as Support vector machine, K-nearest neighbor, Random Forest, XGBoost and Naïve bayes are applied to predict the optimal component.Hierarchical agglomerative clustering and K-Nearest neighbor had the best result compared to the rest methods and hence were selected for model development. A hybrid model using combination of agglomerative clustering and KNN approach was developed. The predicted component results were evaluated and compared with the existing method. The model predicted 84% accurately for capacitor dataset and 81% accurately for resistor dataset. The model also predict substitute for obsolete component which would prevent the inclusion of obsolete component in the design. The model was also deployed on Heroku platform to complete the CRISP-DM methodology cycle. The proposed model would help component engineers and designer by saving a lot of their time which is needed for approving the design using traditional manual method. In this way, the overall product development or delivery time can also be reduced.
Item Type:Essay (Master)
Clients:
Signify N.V.
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Business Information Technology MSc (60025)
Link to this item:https://purl.utwente.nl/essays/90637
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page