This repository contains the code developed for the scientific paper Poka Yoke Meets Deep Learning: A Proof of Concept for an Assembly Line Application, published in Applied Sciences (MDPI), Vol. 12, Issue 21, Art. 11071.
The project proposes a deep learning-based Poka-Yoke solution to support operators in the assembly of speed reducers/multipliers for agricultural applications, with a particular focus on correct oil seal installation. The objective is to automatically detect three conditions:
- correctly installed oil seal,
- incorrectly installed oil seal, and
- missing oil seal.
The core of the solution is a Convolutional Neural Network (CNN) trained on a custom dataset of images depicting crankcases assembled with the components involved in the classification task, instantiated in the variants of interest.
The solution is part of a broader assembly line reengineering process that combines Lean Manufacturing tools (e.g., Muda/Mura/Muri analysis, layout analysis, Yamazumi assembly tasks balancing) and computer vision techniques to improve First Time Quality (FTQ).
Interestingly, the proposed CNN easily reaches a high accuracy, which can be interpreted as the new FTQ index of the assembly node. It also shows a high tolerance to false positive errors related to oil seal positioning, that is, cases where the oil seal is wrongly positioned but classified as correctly placed. This latter scenario is the most critical for the assembly line, as it allows the subsequent assembly steps to proceed on a defective component. In contrast, the opposite case (i.e., oil seal correctly positioned but classified as wrongly placed) does not degrade the FTQ performance of the line, since it simply recalls the operator’s attention, triggering an oil seal inspection and a new classification.
For more details about the assembly line improvement project and the Intelligent Poka-Yoke solution, please refer to the article Poka Yoke Meets Deep Learning: A Proof of Concept for an Assembly Line Application.